COMBUSTION OF PULVERISED COAL IN SWIRL BURNERS

Transcription

1 COMBUSTION OF PULVERISED COAL IN SWIRL BURNERS Ilias Prassas DIPL.-ING. NTUA Thesis submitted for the degree of DOCTOR OF PHILOSOPHY in the University of London and for the DIPLOMA OF MEMBERSHIP OF IMPERIAL COLLEGE Imperial College Mechanical Engineering Department Thermofluids Section MARCH 1998

2 COPYRIGHT NOTICE The copyright of this thesis rests with the author and no quotation from it or information derived from it may be published without the prior written consent of the author. 2

3 To Demetra Prassas

4 Typeset in Adobe Garamond 12pt Formatted in Adobe PageMaker 6.5

5 Abstract In the context of combined cycles for power generation and, in particular, use of vitiated air to increase the cycle efficiency this thesis is concerned with the spatially-precise simultaneous measurement of velocity, size and temperature of individual coal particles in swirl stabilised gas piloted flames confined or unconfined, using atmospheric or high-temperature vitiated air as oxidant, the measurement of the volume flux of ensembles of particles and the measurement of the extinction limits of natural-gas swirl flames in vitiated air of high temperature. Novel optical instrumentation was developed for the present experiments and is based on a so-called shadow Doppler velocimeter (SDV), which is an imaging technique for sizing irregular particles, with dynamic range currently at 10:1, by measurement of the particle cross-sectional area with precision better than 10%, and on a two-colour pyrometer (TCP) for the measurement of coal particle char or soot mantle temperature. The SDV can measure an orthogonal particle velocity component with accuracy better than 15% using single-channel transmitting optics, does not suffer from the so-called trajectory ambiguity effect (which plagues phase Doppler anemometry) in the measurement of particle size and the TCP controlling software can discriminate between signals which originated from volatile flames or incandescent char by application of an amplitude-based criterion. Therefore, TCP can measure the temperature of single coal particles with accuracy and precision of +90K and ±6% respectively, with only few assumptions about their emission characteristics. The instrumentation was applied to gas piloted non-premixed nominally 10 kw flames using burners with D=16 and D=18 mm throat diameter and a quarl with length to exit diameter ratio of 2. Parameters included swirl numbers in the range of , bulk velocity at the exit of the burners in the range of m/s and oxidants containing % O 2, being at C temperature. No particle temperature or particle velocity correlations with particle size (from 10 to 100 µm) were found at z/d=2.7 just downstream of the quarl exit although the mean particle trajectory angle increased with particle size for S=0.41 and S=0.57, where the increasing deviation of particle away from the centreline of the burner with increasing size indicated that particles experienced mean centrifuging because of the swirling flow. Increase of the overall gas-flame equivalence ratio by 50% from 0.69 to stoichiometric and decrease of the primary to secondary air momentum ratio from 1/30 to 1/40 resulted in reduction of the number of burning char particles by about 40% and 80% respectively. Similar reductions, of the order of 70%, were also measured when vitiated air with 16.5% O 2 and of 400 C temperature was used instead of ambient atmospheric air. The measurements implied that the reductions are likely to be due to a combination of insufficient oxygen content or temperature of the oxidant and inadequate particle heat-up as a consequence of the reduced particle residence times inside the high-temperature recirculation zone. The sources of inadequate particle heat-up were also identified by a simple calculation of heat-up (from the differential equation which describes the heat transfer between a char particle and its surrounding fluid) and ignition of char particles suspended in a hot stream of gases with variable oxygen content and temperature. The measurements in purely gaseous flames using high-temperature ( C) vitiated air showed that smaller swirl numbers (around 0.55) improved the lean extinction limits by about 10% and that for a combination of oxygen content and temperature of the oxidant it is possible to improve the lean extinction limits relative to 20 C atmospheric air. For example, 500 C vitiated air with 16.5% O 2 had comparable extinction limits (in terms of air to fuel ratio) with 20 C atmospheric air but using 30% less fuel at extinction. The results are of sufficient detail for the evaluation of Computational Fluid Dynamics codes which can be useful tools for extrapolating the present results to industrial burner scales and for estimating the potential fuel economy by use of vitiated air and reduction of NO x emissions by optimization of the burner operational parameters. 5

6

7 Acknowledgements It is always difficult to estimate and acknowledge to the correct proportions the contribution of other individuals on one s work but I shall try to be as fair as possible. I am grateful to both my supervisors Dr. Alex Taylor and Prof. J. H. Whitelaw FRS for their scientific and financial support throughout this project. I am particularly indebted to Dr. Taylor whose constant advice and influence on PhD and personal matters has inevitably shaped my personality. My thanks are also due to Dr. Yannis Hardalupas for uncountable stimulating discussions in the lab and over lunch. Outside Imperial College, the support of Professors Masanobu Maeda and Koichi Hishida, particularly during my one-month visit in 1995 to their laboratory at Keio University, Japan, was important for this work. I benefited a lot from my interaction with their ex-student, now Dr. Hiroshi Morikita of the University of Tokyo, with whom I spent many hours discussing about optics, fluid mechanics and international affairs during his numerous visits at Imperial College. I also acknowledge the assistance at various stages of the experiments of Dr. Luis Moreira of Instituto Superior Technico, Portugal, Mr. Tobias Reichelt of University of Stuttgart, Germany, Mr. Suzuki of Keio University, and Mr. Nobuaki Abe of Shibaura Institute of Technology in Tokyo. Dr. Nikolaos Orfanoudakis introduced me to the secrets of optical sizing of irregular particles while he was striving to complete his PhD work whilst Mr. Fernando Israel briefly explained to me the principles of two-colour pyrometry. Technical support was provided by Messrs. Alan Finch, Paul Jobson, Gerry Rasmussen and Ian Wright. Of the other people in lab who helped a lot to create a friendly atmosphere special thanks to: Dr. Andy Moore, Dr. Katerina Sardi, Mr. Shane DeZilwa, Dr. Nicolas Carabateas, Mr. Pavlos Aleiferis, Dr. Rong-Feng Tsai, Dr. Yam-Wing Siu, Mr. Markus Keicher, Mr. Arndt Selbach. I am grateful to the Commission of the European Union for funding this research; to Honda R&D and particularly to Dr. Kiyoshi Ishii for additional financial support throughout the work and also during the writing up; lastly to British Steel for financial support at the early stages of writing up of this thesis. Finally I am indebted to my family and particularly to my mother, to whom this thesis is dedicated, for sympathising with my sorrows all these years. If I have not included somebody I should in this short list, it was due to plain forgetfulness from the stress of writing up. I have been many years in the lab and therefore interacted with many people; it is inevitable that not everyone has been mentioned in this note. 7

8

9 Table of Contents Abstract... 5 Acknowledgements... 7 List of Tables List of Figures List of Commonly Used Symbols CHAPTER 1 Introduction THE PROBLEM CONSIDERED NOX EMISSIONS Fundamentals NOx reduction Vitiated air (Supplementary) firing Reburning Air staging IN-SITU HOT GAS CLEANING IN DIRECT COAL FIRING PREVIOUS WORK ON OPTICAL PARTICLE SIZING THE PRESENT CONTRIBUTION THE STRUCTURE OF THIS THESIS CHAPTER 2 Measurement of Particle Velocity, Size and Flux by Shadow Doppler Velocimetry BACKGROUND OPTICAL ARRANGEMENT APPEARANCE OF PARTICLE IMAGES ON DETECTOR PLANE EFFECTS OF PARTICLE TRAJECTORY ANGLE DATA ACQUISITION AND PROCESSING Data acquisition Signal processor Simultaneous size and velocity measurement Validation sequence of SDV raw data Validation of Velocity Signal Creation of an Image Buffer for Storing the Sampled Images Separation of Raw Images and Cross-Correlation Estimation of the Area of the Shadow Image ACCURACY OF PARTICLE FLUX AND VOLUME FRACTION MEASUREMENT Introduction Method of flux and volume fraction measurement Size of the sampling-space of SDV Experimental results Accuracy of Estimation of the size of the Sampling-space

10 Flux and Volume Fraction Measurement Extension to residence-time based volume fraction measurement TRAJECTORY BIAS EFFECTS SOURCES OF DATA REJECTION SUMMARY OF EXPERIMENTAL UNCERTAINTIES Size Uncertainties in Open Isothermal Flows Size Uncertainties in Reacting or Confined Flows Uncertainty of Estimation of Sampling-Space Area and of Flux Measurement SUMMARY OF CHAPTER CHAPTER 3 Non-Intrusive Simultaneous Measurement of Velocity, Size, Temperature and Flux of Burning Particles INTRODUCTION Background Present contribution FUNDAMENTALS OF TWO-COLOUR PYROMETRY COMBINED SDV/TWO-COLOUR PYROMETER INSTRUMENT Optical arrangement of the two-colour pyrometer Optical arrangement of the combined SDV/pyrometer Data acquisition system Amplitude validation of pyrometer signals Uncertainties Estimation of burning fraction and flux of particles INTERPRETATION OF MEASUREMENTS OBTAINED WITH TWO-COLOUR PYROMETRY Background A criterion for discrimination between char and volatile flame signals Discrimination of char from volatile flames Use of the criterion and determination of temperature SUMMARY OF CHAPTER CHAPTER 4 Pulverised Coal Combustion in Open and Confined Piloted Swirl Stabilised Flames INTRODUCTION Background The Present Contribution FLOW CONFIGURATIONS AND EXPERIMENTAL METHOD Burner Geometries Metal and Ceramic Furnaces Pulverised Coal Analysis Experimental Conditions and Bulk Flow Quantities Instrumentation and Uncertainties Instrumentation RESULTS The Near-Burner Region Particle Aerodynamics in Open Flames Particle Aerodynamics In Confined Flames Combustion Characteristics of Particles in Open Flames The Far-Burner Region Experimental Method Flow Details

11 Measurement Strategy Results DISCUSSION: PARTICLE CENTRIFUGING Near Burner Region Far Burner Region SUMMARY OF CHAPTER Particle Aerodynamic Characteristics in the Near-burner Region Particle Burning Characteristics in the Near-burner Region Particle Aerodynamic Characteristics in the Far-Burner Region of Confined Flow CHAPTER 5 Gaseous and Pulverised Coal Reacting Flows in Vitiated Air INTRODUCTION Background Gaseous flames Pulverised coal flames The Present Contribution STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR Flow Configuration and Experimental Methods Boundary Conditions Bulk Quantities to Characterise the Flow Temperature and Stable Chemical Species Lean Extinction Limits Lean Limit as a Function of Unvitiated Air Temperature Lean Limits as a Function of Vitiated Air Temperature Lean Limit as a Function of Oxygen Mole Fraction in Vitiated Air Effect of Bulk Straining on the Lean Limit Stability of Gaseous flames: Discussion COMBUSTION OF PULVERISED COAL IN VITIATED AIR Experimental Method Uncertainties Results Ignition of char particles: Discussion SUMMARY OF CHAPTER Gaseous Flames Pulverised Coal Flames CHAPTER 6 Closure CONCLUSIONS Instrumentation Pulverised Coal Flames Combustion in vitiated air Combustion in atmospheric air Natural gas flames RECOMMENDATIONS FOR FURTHER RESEARCH Improvement of the Instrumentation Coal Combustion

12 APPENDIX I Derivation of turbulent stress tensor for Monte-Carlo simulation APPENDIX II Calibration of the Two-colour Pyrometer APPENDIX III Calculation of the spectral emissivity of a coal particle APPENDIX IV Flow Boundary Conditions APPENDIX V Determination of swirl number Flow Symmetry Check Influence of Orientation of Fuel Injector on Flame Symmetry Check of Secondary Air Flow Symmetry Effect of Coal Gun Orientation Effect of Coal Gun Misalignment APPENDIX VI Colour Plates References

13 List of Tables Table 2.1 Specifications of the SDV signal processor 67 Table 2.2 Estimated systematic uncertainties of the mass flux measurement at a point 87 Table 2.3 Error codes presented in figure 2.23 and their description 98 Table 2.4 Summary of experimental uncertainties with SDV 101 Table 3.1 Principal characteristics of the two-colour pyrometer 114 Table 3.2 Principal characteristics of the transmitting optics of the combined SDV and two-colour pyrometer instrument 116 Table 3.3 The characteristic uncertainties of the temperature of a single particle measured by two-colour pyrometry 123 Table 4.1 Proximate and ultimate analyses of Bentinck bituminous coal (UK). The table also shows details on the particle size distribution of the coal batch used in the experiments (except that of 4.3.2), measured by a Malvern instrument (Abbas 1994) 149 Table 4.2 Parameters of the optical configuration 153 Table 4.3 Summary of conditions of parametric study 155 Table 5.1 Cases examined, corresponding flow rates and oxygen concentration of exhaust gases 207 Table 5.2 Bulk strain rate as a function of vitiated air temperature

14 Table 5.3 Thermal power due to vitiated air H & T and potential heat release from combustion in the swirl burner at lean limit of operation for the multihole radial injector, except where otherwise indicated. Calculations are for swirl number of 0.5, at which the lower lean limits of operation of this burner were achieved 226 Table 5.4 Flow conditions of the present investigations with pulverised coal 235 Table 5.5 Constants used in the model described in Table II.2 Detectability limits of the pyrometer for sizes between 10 and 100 µm 271 Table IV.1 Principal characteristics of optical configuration of the LDV system used in the measurement of the swirl boundary conditions

15 List of Figures Figure 2.1 Optical setup of the shadow Doppler velocimeter instrument Figure 2.2 Appearance of shadows of spherical particles, moving across the centre of the probe volume, on detector plane as a function of particle trajectory in the LDV probe volume and their respective output signals. The signals are proportional to the irradiance detected from an array diode along the dotted lines (aa ) and (bb ) respectively Figure 2.3 Illustration of image appearance in image buffer: (a) Schematic of spherical particle moving at angle φ relative to the normal to the detector and (b) image in image buffer after acquisitions of a spherical particle moving at angle φ as in (a); l def and l pos denote image separation across and displacement along detector respectively. The spherical particle is moving on a plane (i.e. the plane of the paper) normal to that which contains the beams Figure 2.4 Example of sampling and image reconstruction of the shadow image of a defocused irregular particles, showing how the sampled one-dimensional slices are juxtaposed to form the two-dimensional image Figure 2.5 Examples of the reconstructed recorded shadow image of a spherical particle in the image buffer showing the effects of over- and undersampling Figure 2.6 Block diagram of the SDV signal processor showing the components for processing the analogue output signal from one diode of the array. The electronic component defined as FPGA is the programmable integrated circuit which controls the hardware Figure 2.7 Block diagram of the electronics for simultaneous size and velocity measurement using SDV indicating the connections between the SDV signal processor and the zero-crossings counter to the host computer

16 Figure 2.8 Schematic of the algorithm for separation of the two overlapping images, originating from each incident laser beam, in the image buffer. The (a) original image in the buffer is (b) divided in two by cutting across the boundary image intersection points; the bright shadow from each beam is (c) distinguished from the dark after division and (d) each bright shadow is blended with a copy of the dark shadow to form (e) the actual shadow image of the particle originating from each laser beam Figure 2.9 Details of the SDV sampling-space through which the particle volume flux was measured indicating the coordinate system and the dimensions defined in the text. In addition, examples of shadow images on the detector plane are also shown as a function of particle position in the LDV probe volume Figure 2.10 Variation of the calculated width of the sampling space as a function of particle size Figure 2.11 SDV image in image buffer for different cases: spherical particle at (a) 90º and (b) other than 90º; ellipsoid in-focus (c) and (d) and defocused (e) and (f). The width of the frame surrounding each image represents the width of the active diodes of the array. The height represents samples of the array as a function of time Figure 2.12 (a) mean and (b) rms SDV-measured defocus distance of a pinhole shadow image for four trajectory angles relative to the axis of the array Figure 2.13 Schematic of the water-flow channel Figure 2.14 Location of measurement points in a cross section of the flow channel of figure 2.13 used for the integration of the local flux Figure 2.15 Variation of the particulate mass flow rate ( ) as a function of the defocus limitation z def,lim, in comparison with the true flow rate (solid line) Figure 2.16 Comparison between the measured ( ) against the actual (solid line) integrated particulate volume fraction. Open circles correspond to the fraction corrected by the signal error rate of the SDV given in figure Figure 2.17 Signal error rate as a function of volume fraction Figure 2.18 Schematic of residence time calculation as a function of threshold levels, for PDA and SDV respectively Figure 2.19 Velocity bias effects due to SDV; (a) shows a schematic of the image of the LDV probe volume on the diode array and a particle trajectory through the probe volume at angle φ relative to the normal to the array, (b) the range of experimentallydetermined measurable trajectory angles φ and (c) the range of non-measurable velocities normal to the array which are missing from the velocity pdf

17 Figure 2.20 Variation of the normalised calculated mean axial and mean radial velocity, the normalised rms radial and the normalised velocity cross-correlation as a function of the axial rms velocity for a range of mean axial velocities, using Monte-Carlo simulation. Symbols as in legend Figure 2.21 Calculated axial velocity pdf as compared to the sampled Gaussian, for conditions: mean axial velocity U=1 m/s, mean radial V=1 m/s, rms axial velocity u=8 m/s and rms radial v=2 m/s Figure 2.22 Comparison between axial velocity pdfs measured in the burner of Chapter 4, for swirl number S=0.41 at z/d=2.67 and r/d=0, processed to include data which were velocity- but not size-validated (top) and size- and velocity-validated data (bottom) Figure 2.23 Sources of rejection of measurements by SDV along a radial profile, determined from measurement in the burner of Chapter 4 for S=0.41, at z/d=2.67 (top) and z/d=4 (bottom). The values of the ordinate are explained in Table Figure 3.1 Isometric schematic view of the transmission and receiving optics of the two-colour pyrometer Figure 3.2 Isometric schematic view of the transmission and receiving optics of the combined SDV and two-colour pyrometer instrument Figure 3.3 Arrangement of the electronics for data acquisition using the combined SDV and two-colour pyrometer instrument Figure 3.4 Typical signal records as digitised by the transient recorder corresponding to the two operating wavelengths of the pyrometer for two cases explained in the main text: (a) signals exhibiting a single peak and (b) signals exhibiting multiple peaks. The figure also defines characteristic quantities used in the text for the description of the signal validation criteria Figure 3.5 Typical plot of signal amplitude in channel 2 versus measured particle temperature calculated on the assumption of grey emission. Curves A and B define, for a given temperature, the ceiling and floor values of amplitude of signals that originated from soot cloud and char particles respectively Figure 3.6 Flow chart of the discrimination criterion Figure 3.7 Replot of curves A and B of figure 3.5 as a function of the voltage output of pyrometer channel Figure 4.1 Drawing of the 10 kw swirl burner I indicating the method of swirl generation. The throat diameter of this burner is 16 mm Figure 4.2 Detail of the tip of the gas and pulverised coal injector. Gas was transported through the annulus and was injected via 6 holes of 1 mm radially placed whilst pulverised coal was fed through the central 3 mm tube

18 Figure 4.3 Drawing of Burner II a modified version of the burner depicted in figure 4.1 along with detail of the tip of the fuel injector used with this burner Figure 4.4 Drawing of the measured metal laboratory furnace in elevation and in plan view showing the cartesian velocity components u,v. The windows which provided optical access for the measurements are also indicated in the figure Figure 4.5 Elevation of the ceramic furnace designed to operate in pressures up to 5 bar. The figure shows the various components of the furnace, i.e. the base and end wall plates, the two pieces of the metal casing surrounding the ceramic the access ports for optical instrumentation and the exit ports which were symmetrically placed. Only one exit port was used at a time Figure 4.6 Block diagram of the flow configuration and the SDV Figure 4.7 Radial profiles of net and positive particle volume flux corrected with the measurement validation rate, for (a) S=0.57 (flow 1 in Table 4.3), (b) S=0.41 (flow 2) and (c) S=0.57 and 400 lt/min secondary air (flow 3), for three particle size classes indicated in the legend Figure 4.8 Radial profiles of the (a) mean and the (b) rms axial velocity, as well as the (c) mean and the (d) rms radial particle velocity as compared to those of the gaseous phase, for S=0.57 (flow 1 in Table 4.3) at z/d=2.67, for three size classes. Symbols: (ç) µm, (á) µm and (ó) µm and (ò) gas phase Figure 4.9 Radial profiles of the (a) mean and the (b) rms axial velocity, as well as the (c) mean and the (d) rms radial particle velocity, for S=0.41 (flow 2 in Table 4.3) at z/ D=2.67, for three size classes. Symbols: (ç) µm, (á) µm and (ó) µm and (ò) gas phase Figure 4.10 Radial profiles of the (a) mean axial velocity, and the (b) mean radial particle velocity, for S=0.57 and 400 lt/min at z/d=2.67, for three size classes. Symbols: (ç) µm, (á) µm and (ó) µm Figure 4.11 Radial profiles of the mean particle trajectory angle relative to the centreline of the burner, multiplied with the sign of the particle axial velocity for (a) S=0.41 and (b) S=0.57 and 400 lt/min, for three size classes. Positive abscissa denotes trajectories away from the centreline of the burner. Also, size-trajectory angle correlation for (c) S=0.41 at r/d=0.9 and (d) S=0.57 and 400 lt/min at r/d=1.2. Vertical bars indicate the rms values Figure 4.12 Radial profiles of particle (a) mean and (b) rms axial velocity, (c) mean and (d) rms radial velocity and (e) rms value of the trajectory angle for S=0.57 at z/d=4. Symbols: (ç) µm, (á) µm and (ó) µm Figure 4.13 Radial profiles of the mean and rms axial and tangential velocity components as well as the AMD and SMD diameters inside the ceramic furnace fired at atmospheric pressure for S=0.57, at z/d=3.22. Symbols for velocity profiles: (ç) µm, (á) µm and (ó) µm. Symbols for diameters: (ç) AMD, (á) SMD

19 Figure 4.14 Radial profiles of the mean gas temperature, measured in the absence of particles at z/d=2.67, for (í) S=0.41 and (ó) S= Figure 4.15 Radial profiles of the fraction of burning particles, measured at z/d=2.67, for (a) S=0.41 and (b) S= Figure 4.16 Comparison between velocity pfs for burning particles and the ensemble, measured at z/d=2.67 for (a, b) S=0.57 and (c-e) S=0.41 at the following locations: (c) r/d=0, (d)=0.22, (a, e) r/d=0.44 and (b) r/d= Figure 4.17 Radial profiles of the measured size distribution of burning particles and the ensemble, measured for S=0.41 at z/d= Figure 4.18 Radial profiles of the measured temperature distribution of burning char particles, measured for S=0.41 at (a) z/d=2.67 and (b) z/d= Figure 4.19 Radial profiles of the mean temperature of the gaseous flame in the absence of particles as a function of the overall gas equivalence ratio φ. Symbols: (í) φ=0.69 and (ó) φ= Figure 4.20 Radial profiles of the fraction of burning particles, measured at z/d=2.67 and 4, for gas equivalence ratios φ=0.69 (left) and 1.0 (right) and momentum ratio MR=1/ Figure 4.21 Radial profiles of the fraction of burning particles, measured at z/d=2.67, for gas equivalence ratio φ=0.69 and momentum ratios MR=1/30 (top) and 1/40 (bottom) Figure 4.22 Radial profiles of the particle volume flux, measured at z/d=2.67 and 4, for gaseous equivalence ratios 0.69 (left) and unity (right). Positive flux takes particles with positive velocities into account only. The ensemble consisted of all particles, whether pyrometer-detectable burning or not. Symbols: (à) positive flux of the ensemble, (ò) net, (á) positive flux of burning particles only, (ó) net Figure 4.23 Radial profiles of the mean and the rms temperature of incandescent char particles, measured at z/d=2.67 and 3.05 for gas equivalence ratio φ=0.69, and at z/d=2.67 and 4 for φ=1. The temperatures are plotted for three particle size classes, namely 12-24, and mm. Symbols: (ç) µm, (á) µm and (ó) µm Figure 4.24 Scatter plots of the size of the incandescent char particle as a function of its instantaneous temperature, measured across a radial profile at z/d=2.67, for gas equivalence ratios of 0.69 (left) and 1.0 (right) Figure 4.25 (a) Alignment of the SDV photodiode array in the two adjacent sub-regions. Note that the arrows indicate alignment of the normal to the axis of the photodiode array; (b) Cartesian components u, v of measured particle velocity when the normal to the photodiode array is aligned at ψ and the velocity vector is at angle φ to the normal

20 Figure 4.26 (a) Scatter plots of Vm vs. diameter, diameter vs. φ, Vm vs. φ and u vs. v at z/ro=-1 and 34.5 mm upstream of the exit, and (b), as in (a), but at z/ro= Figure 4.27 (a) Mean, and (b) rms axial velocity component, u, of particle as compared with that of air for three different particle diameter classes 12-24, and µm at 34.5 mm upstream of the exit; (c) and (d) are mean and rms, respectively, for the radial velocity component, v, at the same location and for the same particle diameter classes. Symbols: (ç) µm, (æ) µm and (õ) µm and (ì) gas phase Figure 4.28 Particle size distribution at three vertical locations at 34.5 mm upstream of the exit. Symbols: (ç) z/r 0 =1, (æ) z/r 0 =0 and (õ) z/r 0 = Figure 4.29 Vertical profile of (a) air and (b) particle ensemble velocity vectors at 34.5 mm upstream of the exit as measured by LDV and SDV, respectively Figure 5.1 Detail of the tip of the axial (a) and the multihole radial (b) injector, showing the dimensions of the holes through which natural gas was injected Figure 5.2 Block diagram of the exhaust gas supply system, showing the tube connections between its various parts and the air and gas supplies Figure 5.3 Radial profiles of mean (a) and rms (b) temperatures measured for ambient air at 300 lt/min, for a swirl number of 0.68 and equivalence ratio 0.7 and 0.8. The profiles were measured at z/d= Figure 5.4 Radial profiles of mean (a) and rms (b) temperatures measured for vitiated air of 400 C and 16.5% oxygen mole fraction (case C) at 300 Normal lt/min, for a swirl number of 0.68 and equivalence ratios 0.73 and The profiles were measured at z/d= Figure 5.5 Comparison between radial profiles of mean temperatures corresponding to a swirl number of 0.68 and heat releases of 8.5 and 8.2 kw. The profiles were measured at z/d= Figure 5.6 Radial profiles of mean concentrations of CO2, O2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured at z/d=2.2, for the case of vitiated air of 400 C and 16.5% oxygen mole fraction (case C), at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used Figure 5.7 Radial profiles of mean concentrations of CO2, O2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured at z/d=2.2, for the case of vitiated air of 400 C and 16.5% oxygen mole fraction (case C), at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used

21 Figure 5.8 Axial profiles of mean concentrations of CO2, O2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured along the centreline starting from the quarl exit, for the case of vitiated air of 400 C and 16.5% oxygen mole fraction (case C), at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used Figure 5.9 Axial profiles of mean concentrations of CO2, O2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured along the centreline starting from the quarl exit, for the case of vitiated air of 400 C and 16.5% oxygen mole fraction (case C), at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used Figure 5.10 Radial profiles of mean concentrations of CO2, O2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured at z/d=2.2, for the case of ambient air at 300 lt/min, at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used Figure 5.11 Axial profiles of mean concentrations of CO2, O2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured along the centreline starting from the quarl exit, for the case of ambient air, at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used Figure 5.12 Extinction limits for the ambient air case and 300 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as an independent variable, for the multihole radial injector. Shaded areas correspond to stable operation of the burner at secondary air temperatures indicated on the graph. Lean limits of operation correspond to the right-hand side in (a) and the left-hand side in (b). Symbols indicate the measured conditions and arrows the extension of the stability limits. Dotted lines indicate the boundary of the measured shaded area where it is not obvious in the figure Figure 5.13 Extinction limits for the ambient air case and 250 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as an independent variable, for the multihole radial injector. See also figure Figure 5.14 Extinction limits for the ambient air case and 300 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as an independent variable, for the axial injector. See also figure Figure 5.15 Extinction limits for case C, with a 16.5% oxygen mole fraction in vitiated air and 300 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as an independent variable, for the multihole radial injector. See also figure

22 Figure 5.16 Extinction limits for case D, with a 15.5% oxygen mole fraction in vitiated air and 250 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as a parameter, for the multihole radial injector. See also figure Figure 5.17 Extinction limits for case C, with a 16.5% oxygen mole fraction in vitiated air and 300 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as a parameter, for the axial injector. See also figure Figure 5.18 Comparison between the axial and the multihole radial injectors, for 16.5% oxygen mole fraction (case C) and 300 lt/min and 500 C. See also figure Figure 5.19 Comparison between extinction limits for 21% and 15.5% (case D) oxygen mole fraction, at 250 lt/min and 500 C temperature for the multihole radial injector. See also figure Figure 5.20 Comparison between extinction limits for bulk strain rates corresponding to 250 (case F) and 300 lt/min (case C), at 400 C and 16.5% oxygen mole fractions for the multihole radial injector. See also figure Figure 5.21 Maximum calculated flame temperature of a laminar diffusion flamelet plotted as a function of the logarithm of the inverse of the scalar dissipation at the stoichiometric mixture fraction for three different oxidant conditions: Blocked squares: air at 20 C, open circles: vitiated air with oxygen mole fraction and 400 C and open squares: vitiated air with oxygen mole fraction and 500 C. Vectors indicate the point of extinction Figure 5.22 Radial profiles of the mean axial velocities for flow conditions A-C, measured at z/d=2.6 for three particle size classes Figure 5.23 (a) Radial profiles of the mean axial velocity of burning particles and the ensemble in flow A; Comparison between axial velocity probability density functions of burning particles and the ensemble in flow A and at (b) r/d=0.0 and (c) r/d=0.75; Comparison between particle size probability density functions of burning particles and the ensemble in flow A and at (d) r/d=0.0 and (e) r/d= Figure 5.24 Radial profiles of the burning fraction for flows A-C Figure 5.25 Radial profiles of the positive and net volume flux of burning particles and the ensemble (a)-(c) for flow conditions A-C respectively Figure 5.26 Time until ignition or until a char particle reaches 1700 K as a function of particle size on the assumption that the char particle initially entered the recirculation zone and escaped it after time t RES for two different conditions simulating (a) flow conditions A and (b) flow conditions C. Horizontal lines indicate the transit time of a particle from point of injection to the measurement location assuming rectilinear motion with the bulk injection velocity

23 Figure II.1 Optical arrangement for calibration of the two-colour pyrometer Figure II.2 Pyrometer calibration curves as a function of the voltage applied to the photomultipliers. The curves correspond to the response of the pyrometer to radiation emitted by a grey body Figure III.1 Geometry of a single char particle surrounded by a soot cloud. The receiving optics are located in the bottom part of the figure Figure IV-1 Radial profiles of the mean and the rms axial and tangential velocity components of the isothermal flow, measured at 2 mm (z/d=0.02) downstream of the exit of the burner in the absence of the quarl, for a range of derived swirl numbers indicated on the graph. Symbols: (à) Mean axial velocity, (ò) Mean tangential velocity, (á) rms axial velocity and (ó) rms tangential velocity Figure IV-2 (a) Cumulative swirl number along a radial profile at z/d=0.02 in the absence of a quarl for a range of swirl numbers between 0.16 and 0.57 as indicated on the graph. (b) Swirl number as a function of the fraction of the total mass flow rate which is supplied through the tangential inlets in the burner. Results correspond to measurements of figure IV Figure IV-3 Radial profiles of the mean axial velocity component measured at the exit of the burner (z/d=0.02) in the absence of the quarl for isothermal conditions which corresponded to S=0.57. Symbols: (á) mean axial velocity measured along a profile denoted X and (à) mean axial velocity measured at right angles to X profiles Figure IV-4 Radial profiles of (a) mean (á, à) and rms (ç, æ) axial velocity component for swirl numbers 0.57 and 0.41 respectively and φ=0.69, (b) mean (á, à) and rms (ç, æ) axial velocity component for overall gas equivalence ratios of φ=1.0 and 0.69 respectively, and (c) mean (á, à) and rms (ç, æ) tangential velocity component for swirl numbers 0.57 and 0.41 respectively and φ=0.69, all measured at z/d=3.05 in the absence of primary air Figure IV-5 Radial profiles of particle Sauter mean diameter (ç) and volume flux G (ì), and radial profiles of the mean axial velocity for particles of the µm (á) and the µm (ó) size class. All quantities were obtained from a single measurement profile for S=0.41 and overall gas equivalence ratio φ=0.69 at z/d= Figure V.1 Radial profiles of the mean temperature of vitiated air preheated at 600 ºC in the absence of combustion, measured at z/d=2.7 along traverses at right angles explained in the adjacent drawing, and were (à) along S-N and (á) along E-W. 289 Figure V.2 Radial profiles of the mean temperature of a natural gas swirl flame with swirl number S=0.41 and equivalence ratio φ=0.69, measured at z/d=2.7, along traverses at right angles, as explained in the figure, and corresponded to (à) S-N and (á) E- W directions; in (b) the measurements were made after the coal gun was rotated through 120º in reference to results of (a)

24 Figure V.3 Re-plot of data presented in figure V.2 as a function of coal gun orientation where the coal gun was aligned to the (à) north direction and (á) south direction; (a) presents measurements along a S-N traverse, and (b) along a E-W traverse Figure V.4 Radial profiles of the mean temperature of a natural gas swirl flame with swirl number S=0.41 and equivalence ratio φ=0.69, measured at z/d=2.7, along traverses at right angles, as explained in the figure, and corresponded to (à) S-N and (á) E- W directions. The coal gun was intentionally misaligned by 0.3 mm along S-N direction Figure V.5 Radial profiles of the mean temperature of a natural gas swirl flame with swirl number S=0.41 and equivalence ratio φ=0.69, measured at z/d=2.7, along a (a) S- N and (b) a E-W traverse, for three thermocouple orientations PLATE 1 PHOTOGRAPH OF THE TRANSMITTING OPTICS OF THE COMBINED SDV AND TWO-COLOUR PYROMETER INSTRUMENT. THE AR+LASER AND THE DISA SINGLE BRAGG-CELL UNIT ARE SHOWN ON THE PHOTOGRAPH PLATE 2 PLAN VIEW OF THE RECEIVING OPTICS OF THE COMBINED SDV AND TWO-COLOUR PYROMETER INSTRUMENT PLATE 3 DETAIL OF THE RECEIVING OPTICS OF THE COMBINED SDV AND TWO-COLOUR PYROMETER INSTRUMENT. THE PHOTOGRAPH SHOWS (FROM RIGHT TO LEFT) A F/500 ACHROMAT, THE RECTANGULAR ALUMINIUM-COATED MIRROR, THE F/300 ACHROMAT AND THE 40X MICROSCOPE OBJECTIVE LENS. AT THE BOTTOM OF THE PHOTOGRAPH THE SECOND COATED MIRROR IS ALSO SHOWN PLATE 4 FRONT VIEW OF THE MODULE CONTAINING THE PYROMETER OPTICS, AS WELL AS OF THE MODULE, THE LINEAR PHOTODIODE ARRAY (ON THE RIGHT OF THE PHOTOGRAPH). THE F/300 ACHROMAT, LOCATED IN FRONT OF THE 150 µm PINHOLE IS SHOWN. OBSERVATION ALONG THE AXIS OF THE OPTICS, FROM A LOCATION DOWNBEAM OF THE MICROSCOPE OBJECTIVE LENS PLATE 5 PHOTOGRAPH OF THE METAL FURNACE SHOWING MAINTENANCE ACCESS PORTS PLATE 6 PHOTOGRAPH OF THE CERAMIC FURNACE, SHOWING METAL CASING AND MAINTENANCE ACCESS PORTS PLATE 7 PHOTOGRAPH OF GAS-SUPPORTED PULVERISED COAL FLAME FOR SWIRL NUMBER 0.57 AND GAS EQUIVALENCE RATIO OF PRIMARY AIR: 10 LT/MIN AND SECONDARY AIR: 300 LT/MIN. CAMERA SETTINGS: F/5.6 AND 1/ PLATE 8 PHOTOGRAPH OF GAS-SUPPORTED PULVERISED COAL FLAME FOR SWIRL NUMBER 0.41 AND GAS EQUIVALENCE RATIO OF PRIMARY AIR: 10 LT/MIN, SECONDARY AIR: 300 LT/MIN. CAMERA SETTINGS: F/5.6 AND 1/ PLATE 9 PHOTOGRAPH OF GAS-SUPPORTED PULVERISED COAL FLAME FOR SWIRL NUMBER 0.57 AND GAS EQUIVALENCE RATIO OF PRIMARY AIR 10 LT/MIN AND SECONDARY AIR: 400 LT/MIN. CAMERA SETTINGS: F/5.6 AND 1/

25 PLATE 10 PHOTOGRAPH OF GAS-SUPPORTED PULVERISED COAL FLAME FOR SWIRL NUMBER 0.41 AND GAS EQUIVALENCE RATIO OF PRIMARY AIR: 10 LT/MIN, SECONDARY AIR: 300 LT/MIN. CAMERA SETTINGS: F/5.6 AND 1/ PLATE 11 PHOTOGRAPH OF GAS-SUPPORTED PULVERISED COAL FLAME FOR SWIRL NUMBER 0.57 AND GAS EQUIVALENCE RATIO OF PRIMARY AIR: 7.5 LT/MIN, SECONDARY AIR: 300 LT/MIN. CAMERA SETTINGS: F/5.6 AND 1/ PLATE 12 PHOTOGRAPH OF THE VITIATED AIR FACILITY, SHOWING THE SYSTEM FOR GENERATION OF EXHAUST GAS, THE BURNER WRAPPED IN FIBRE INSULATION, AND A CYLIDRICAL FURNACE WITH ACCESS PORTS MOUNTED ON THE BURNER, USED FOR TEST EXPERIMENTS PLATE 13 DETAIL OF COLOUR PLATE 12 SHOWING THE TWO FURNACES WHICH PRODUCE THE VITIATED AIR SUPPLY TO THE SWIRL BURNER PLATE 14 PHOTOGRAPH OF THE UNITS FOR METERING OF AIR AND GAS SUPPLIES TO THE SYSTEM FOR GENERATION OF EXHAUST GASES, AND THE ELECTRONIC UNITS FOR THE CONTROL OF THE FLAMES IN THE PREMIX BURNERS PLATE 16 PHOTOGRAPH OF GASEOUS FLAME OF 0.80 EQUIVALENCE RATIO, STABILISED USING VITIATED AIR CONTAINING 17.6% OXYGEN AS OXIDANT PLATE 15 PHOTOGRAPH OF GASEOUS FLAME OF 0.65 EQUIVALENCE RATIO, STABILISED USING AIR AS OXIDANT

26

27 List of Commonly Used Symbols A size of SDV sampling space (m 2 ) A 0 pre-exponential factor (kg m -2 s -1 [atmo 2 ] -1 ) C 1, C 2 Planck s first and second constant C p heat capacity of particle (kj Kg -1 K -1 ) C v (i) particle volume fraction of size class i ( ) d particle size (m) D burner diameter (m) d w width of sampling-space (m) e l (spectral) emissive power of radiation (J s -1 m -2 µm -1 ) E 0 activation energy (kj kmol -1 ) f mixture fraction ( ) f b fraction of burning particles ( ) F Sampling rate (Hz) g acceleration of gravity (m s -2 ) G particle volume/mass flux (m 3 m -2 s -1 /kg m -2 s -1 ); optics magnification ( ) i intensity of radiation k g gas thermal conductivity (Wm -1 K -1 ) K s, K d surface and diffusion reaction coefficient (kg m -2 s -1 [atmo 2 ] -1 ) l pos, l def displacement of image in image buffer along and across detector (m) m particle mass (kg) N number of particles Nu Nusselt number, (=2) q char combustion rate (kg m -2 s -1 ) Q heat flux (W m -2 ) r distance travelled by shadow in units of diode pitch ( ); radial distance (m) R universal gas constant R s radius of soot cloud s strain rate (s -1 ) S p area of particle projected image (m 2 ) 27

28 S swirl number t time (s) T temperature (K) T a T g T s temperature of surroundings (K) gas temperature (K) sampling time (s) u particle velocity; flow rms velocity (m s -1 ) U flow instantaneous axial velocity; particle axial velocity (m s -1 ) U 0 secondary flow bulk velocity (m s -1 ) U T terminal velocity (m s -1 ) v flow rms velocity (m s -1 ) V instantaneous transverse velocity; particle transverse velocity (m s -1 ) f stoich stoichiometric fuel mixture fraction ( ) Y OX x oxygen mass fraction oxygen mole fraction s strain rate (s -1 ) St Stokes number V particle volume (m 3 ); Photomultiplier voltage (V) X pyrometer gain (V m -2 ) z axial distance (m) z def defocus distance (m) Greek α absorptivity ( ) γ drift parameter ( ) H heating value of coal (kj kg -1 ) ε emissivity ( ) J laser beam half-angle ( ) l wavelength (m) µ viscosity (kg m -1 s -1 ) ρ density (kg m -3 ) t particle residence time (s) φ trajectory angle ( ); overall equivalence ratio ( ) χ scalar dissipation (s -1 ) ω angular velocity; angle ( ) 28

29 Subscripts normal component b black body; burning g gas grey grey emitter p particle stoich stoichiometric t total l spectral Superscripts rms velocity (m s -1 ); diacritical( ) ² diacritical ( ) p particle s soot Abbreviations AASB ASPB ASTM CFD EFCC IFRF IFSB LDV MACT PDA pf PFBC PPCC RUN-1DL SCR SDV SNCR Aerodynamically Air Staged Burner Air-Staged Precombustor Burner American Society for Testing and Materials Computational Fluid Dynamics Externally Fired Combustion Cycle International Flame Research Foundation Internally Fuel Staged Burner Laser Doppler Velocimetry Mitsubishi Advanced Combustion Technology Phase Doppler Anemometry Probability Function Pressurised Fluidised bed Combustion Pressurised Pulverised Coal Combustion The Universal Laminar Flame and Flamelet Code Selective Catalytic Reduction Shadow Doppler Velocimetry Selective Non-Catalytic Reduction 29

30

31 CHAPTER 1 Introduction

32 1.1 THE PROBLEM CONSIDERED 1.1 The Problem Considered According to a directive on limitation of emissions of pollutants into the air from large combustion plants (directive 88/609), commissioned by the European Union (EU) in 1988 then European Economic Community (EEC), most members of the Union have to reduce NO x emissions from plants with output greater than 50 MWt, so that by 1998 the overall EU reduction is 30% relative to 1988 values (Hjalmarsson 1990). Comparable reductions are also required for the other major pollutants, namely sulphur dioxide (SO 2 ), carbon dioxide (CO 2 ) and particulates. Reduction of emissions can be achieved either by (i) removing the tail-pipe pollutant from the fuel or the flue gas, or (ii) avoiding pollutant generation by use of better combustion technologies. There is currently preference towards combustion of natural gas, not only because of lower tail-pipe pollutants (here CO 2 ) but also because of low construction and maintenance costs of natural gas-fired power plants. However it is known that economically exploitable coal reserves will last for about 200 years, as opposed to the estimated 40 for oil and 65 for gas, at current rates of consumption (de Sampaio Nunes 1997) and so, in the medium to long term, effective measures for clean coal combustion must be developed. In any case, coal reserves are more uniformly distributed around the globe, as compared with the geographically concentrated gas and oil reserves and market mechanisms will both ensure that use of coal will persist in the future, particularly in developing countries, and so will the requirement (through mechanisms such as the Rio Treaty) for lower emissions through better combustion technologies. Of all of the major pollutants, sulphur dioxide can be removed only by use of chemicals, either directly incorporated into the fuel stream (if the fuel is gas or light oil), or the flue gas (for pulverised coal) by injection of limestone, because the relatively fast reaction of sulphur with oxygen, compared with that between carbon and oxygen, does not allow for reduction through practically applicable combustion modifications. Use of limestone scrubbers constitutes the most cost-effective technology for the reduction of SO 2 emissions, used in 90% of all flue gas cleaning systems (Michelfelder 1997) and, hence, reduction of SO 2 emissions is beyond the scope of this thesis. Carbon dioxide emissions can be reduced either by introduction of low-carbon fuels, which is currently infeasible, or increase of the overall efficiency of the plants so that smaller amount of carbon is consumed per MW of electric power produced (Michelfelder 1997). The latter is achieved by use of advanced combustion cycles for power generation that promise higher efficiencies than conventional cycles (Reichert et al. 1990). Depending on coal type (brown coal, anthracite, etc.) conventional supercritical cycles achieve efficiencies of up to 47% (Wang and Leithner 1995). On the other hand, combined cycles that include a topping gas turbine cycle, such as certain fluidised bed combustion and gasification cycles, supplementary firing 1 32

33 CHAPTER 1 INTRODUCTION (see, for example, Haywood 1991), or pressurised pulverised coal combustion (PPCC) achieve higher efficiencies. For example, PPCC can currently go up to above 50% (Wang and Leithner 1995) assuming a gas turbine inlet temperature of 1300 C. This temperature limit is imposed by the resistance to heat of current materials of turbine blades and cycle efficiency can increase with developments in materials of turbine blades permit higher inlet temperatures. The major drawback of PPCC is that the exhaust gas contains particulates in the form of fly ash and also alkali contaminants (contained in coal) which are highly erosive and corrosive for the turbine blades and effective methods for cleaning the exhaust gas must be used before the exhaust gas can be fed into the turbine. This problem can be alleviated by use of ceramic filters (but their use is currently limited to temperatures of up to 1000 C, Weber ), pebble beds (Benesch et al. 1995), or aerodynamic separation (Bannister and Newby 1996; Mattson and Stankevics 1985). A variant of PPCC which achieves comparable cycle efficiency and avoids the problem of erosion of the turbine blades is the externally fired combustion cycle (EFCC, LaHaye and Bary 1994), where the working medium in the turbine is air, heated in a heat exchanger by the exhaust gases from prior pulverised coal combustion. Although in such a cycle the turbine does not suffer from erosion, the problem of fouling of the heat exchanger as a result of the aerodynamic motion of fly ash particles contained in the exhaust gases, inside the exchanger, their impingement on the tubing and their subsequent deposition on the materials must be addressed. Most efforts for the reduction of NO x emissions are concentrated on use of combustion technologies which control the motion and, hence, burning of the pulverised fuel in the nearburner region (reviewed, for example, by Hesselmann and Irons 1992; Hjalmarsson 1990; Soud and Fukasawa See also 1.2.2). Such technologies are preferred to chemicallybased reduction methods (such as selective catalytic, SCR, or non-catalytic reduction, SNCR) because of the lower cost of replacing burners and retrofitting existing plants as compared with building new plants needed for the implementation of SCR and SNCR. Combustion technologies focus on the near-burner region, because it is there that pulverised coal releases the bulk of the NO x -precursors (see below) which are subsequently converted into NO x. In section 1.2 the relevance between particle aerodynamics and combustion in the nearburner region and NO x emissions is established by explanation of the fundamentals of NO x generation and reduction and review of past work and current practice, where application of combustion technologies, which consist of certain modifications of particle aerodynamics and, hence, burning, are correlated with low NO x emissions. Section 1.3 focuses on the subject of 1 The term is common among power plant technologists and those involved in research in thermodynamics and refers to the process described in this thesis as combustion in vitiated air, or flue gas recirculation. 2 It is likely that to date there are ceramic filters which can be used at higher temperatures in industrial-scale installations. 33

34 1.2 NOX EMISSIONS clean-up of the high-temperature exhaust gases before, for example, these enter and foul the blades of a gas turbine in direct-fired combined cycles (e.g. PPCC), or the banks of the heatexchanger in indirect firing (e.g. EFCC). Both problems addressed in 1.2 and 1.3 are related with the interaction between pulverised coal particles and turbulent continuous flow, therefore, development of instrumentation for non-intrusive in-situ measurements of particle aerodynamics (velocity, size) is important. For this reason, the subject of optical particle sizing of irregular particle, such as pulverised coal, is briefly reviewed in 1.4. This chapter concludes with a detailed presentation of the contribution of this work ( 1.5) and an overview of the material presented ( 1.6). 1.2 NO x Emissions Fundamentals There is agreement among researchers that the three main mechanisms for NO x formation in burners 3, particularly those fired on fuel containing nitrogenous species, are due to: (i) thermal fixation of atmospheric nitrogen of the oxidant according to the Zeldovich mechanism, also termed thermal NO x, (ii) the release and subsequent oxidation of fuel-bound nitrogen in coal, known as fuel NO x and (iii) hydrocarbon radical concentrations, resulting in the so-called prompt NO x. The contribution of fuel-no x to total NO x emissions from coal-fired burners is over 75% (Pershing and Wendt 1977), whilst thermal NO x becomes important at temperatures above 1800K and prompt NO x is the main contributor (over 50% of total NO x ) in fuel-rich premixed- and in diffusion-hydrocarbon flames (Bowman 1992). For this reason analysis will follow only for fuel NO x mechanisms, which are dominant in coal combustion. When coal particles are heated up, their macromolecular structure disintegrates and they release a mixture mainly of CO, CO 2, CH 4, sulphur and nitrogenous compounds, as well as H 2 and tar (Saxena 1990). The process is called primary devolatilisation, as opposed to secondary, during which the tar, which is a mixture of heavy aromatic hydrocarbons 4, is converted into soot, acetylene and additional CO, H 2 (Marlow et al. 1992). Volatile matter from bituminous coal contains, on average, about 1.5% nitrogen and the amount of nitrogenous compounds 3 There are recent publications on a new mechanism for NO production in flames burning in air, termed the NNH mechanism (Bozzelli and Dean 1995 quoted by Harrington et al. 1996), but it is unlikely that it contributes to NOx production in coal flames. Besides, this mechanism is a new suggestion and its manifestation has not been rigorously tested in turbulent flames. 4 but also contains heterocyclic nitrogen-containing structures, such as pyridine, pyrrole and amines (Smart and Weber 1989). 5 Another important pollutant related to nitrogen compounds is nitrous oxide (N 2 O), of which only a small amount originates from combustion sources and particularly from fluidised bed combustion (Bowman 1992). Because fluidised bed combustion is beyond the scope of this thesis, the chemistry and review of previous work on N 2 O emissions have not been included here. 34

35 CHAPTER 1 INTRODUCTION that could be converted to NO 5 is therefore dependent on the amounts of volatiles released during primary devolatilisation, tar conversion during secondary pyrolysis and that which remains in the char. The weight loss of coal particles during heat-up and, thus, the total volatile yield depends primarily on the peak temperature of heat-up (Anthony et al. 1976; Blair et al. 1976; Pohl and Sarofim 1976) rather than the heat-up rate, as claimed by other researchers (e.g. Smart and Weber 1989), although it is usually difficult to separate the influence of the two parameters (Saxena 1990). It has been experimentally found by Anthony et al. (1976) that the heating rate does have a small effect on the total volatile yield, and the authors attributed strong correlations between the volatile yield and high heating rates reported by others to the use of smaller particles and particle dispersion. Of the nitrogenous species bound in coal, hydrogen cyanide (HCN) is the principal product of devolatilisation when nitrogen is bound in aromatic rings and ammonia (NH 3 ) when it is in the form of amines 6 (Miller and Bowman 1989). For temperatures higher than 800 C, and especially above 1200 C, which are representative of temperatures present in swirl burners (e.g. Milosavljevic 1993), HCN yield dominates over NH 3 yield, regardless of the type of coal used (Nelson et al. 1992). For this reason, attention in the analysis below shall be focused on HCN chemistry, without harming the generality of the conclusions. The main paths of oxidation of HCN which lead to NO formation are outlined below (following Miller and Bowman 1989). First, HCN reacts with O atoms to produce NH or NCO according to: +& & & &2 (1.1) NH and NCO subsequently react with H atoms and produce N atoms. Subsequently, there are two potential paths for N atoms as follows: (1.2) There is also the possibility that HCN is removed by OH radicals as, for example: &1 & & &2 + + (1.3) where NCO will subsequently react as described earlier. Although the above scheme does not capture the complexity of nitrogen chemistry, equation (1.2) clearly shows that N atoms can be reduced to molecular nitrogen and, thus, reduction of NO emissions can be achieved by 6 Ammonia and hydrogen cyanide are generally termed, in NO x chemistry, NO x precursors. 35

36 1.2 NOX EMISSIONS providing suitable conditions so that the reduction chemical path in (1.2b) is preferred over oxidation path of (1.2a). Alternatively, as demonstrated by Myerson and co-workers first in 1957, it is possible to reduce NO by reaction with CH radicals. Reaction of hydrocarbon radicals CH with NO yields (Myerson 1975): & &2 + 1 & &1 + 2 (A) (B) (1.4) where HCO formed in (A) will subsequently break down to H, CO and atomic N due to the exothermicity of equation (1.4), producing more atomic N which can reduce the existing NO according to path (1.2b). A small quantity of HCN produced in (1.4) will be converted to NO and will be subsequently subjected to the NO reduction path of (1.2), provided that suitable gaseous fuel stoichiometry and temperature prevail locally. Findings from previous experimental work will be presented in describing empirically-determined suitable local conditions for NO reduction. The simplified NO chemistry model described above holds for the case of fully-mixed gases and not for diffusion-limited processes such as, for example, surface reaction on char particles 7. That is because this gaseous phase chemistry assumes that a mixture of gases reacts in a process that is controlled by the time scale of the chemical reactions, as opposed to the heterogeneous chemistry controlled by the mechanical time scale of diffusion of gases into the porous surface of the char. This fact is of great importance in swirl burners where NO x reductions can be achieved by modifications on the flow field and, hence, the combustion characteristics of the fuel. Generally this means that the amount of NO x produced depends on (i) the volatile yield 8, (ii) the proportion of nitrogen-bound species in the yield that is converted to NO x and (iii) the proportion of char-n that is converted to NO x, the latter being the major contributor under fuel-rich conditions (Phong-Anant et al. 1985) that prevail in the early phases in staged 9 combustion. Of the three variables, volatile yield depends on the final temperature and the holding time at this temperature. Yield in the literature is reported as percentage of the proximate value, defined as the yield under slow controlled heating documented in ASTM 10 standards. Typical proximate yields are about 40% (the amount of weight loss of a sample mass) for bituminous coals, but can be as high as 80% depending on the type of coal and the final temperature for devolatilisation under more intense heat up conditions (Annamalai and Ryan 1993). For example, it was found that increasing the heat-up temperature from 1250 K to 2100 K the volatile yield increased from 30% to 63% for a given coal (Kobayashi et al. 1977), meaning that the actual yield increased by a factor of 2. Such experimental findings have been confirmed 36

37 CHAPTER 1 INTRODUCTION by other researchers. The retention of nitrogen in char, at least for 40 µm bituminous coal particles, submerged in the helium atmosphere of an electrically-heated furnace, decreased with increasing furnace temperature and increasing residence time in the furnace (Pohl and Sarofim 1977). For example, about 80% of the nitrogen remained in the char at a furnace temperature of 1740 K and residence times in the range ms, but this value dropped to 30% for a furnace temperature of 2100 K. Generally, devolatilisation at temperatures of the order of 2000 K yield volatile matter larger than the ASTM-determined proximate value by a factor of for high bituminous coals, and more than 2 for lignite (Wall 1987). It is the remaining two variables the fate of nitrogen released and of that remaining in the char that are the most interesting aspect of the process. The nitrogen species released with the volatiles can be reduced to molecular nitrogen (by combustion technologies such as those described in 1.2.2), whilst those remaining in char are responsible for the NO x emissions from low-no x burners and fluidised beds. The slow oxidation of char (hence oxidation of nitrogen species in char) does not allow for control of the reaction paths outlined earlier in this section towards N 2, rather than NO, formation. Consequently, in combustion processes where the volatile yield is not high and a large proportion of nitrogen remains in the char, high NO x emissions are produced. This explains why in a fluidised bed reactor, operating in the temperature range K, most of the NO produced was due to char-n conversion (Tullin et al. 1993). By defining that char combustion commences at the point when 20% of char carbon has been oxidised, they found that for particles, 1-2 mm in diameter burning in 4% oxygen, 25% of NO was the product of volatile-n at 1073 K which dropped to 10% when the temperature was increased to 1123K. Another parameter which comes into question is the coal type/rank. Fuel NO x, assuming no combustion modifications to minimise NO x emissions, is not only a function of the total nitrogen content in the coal, but is also directly dependent on the amount of nitrogen released with volatiles, and is highest for those coals which released most nitrogen with volatiles and lowest for those which retained nitrogen in char (Chen et al. 1982). Chen and co-workers (1982) tested a number of about 50 coals from different places in the world in a vertical, refractory-lined furnace and they found that two different coals containing both 1.05% nitrogen produced NO emissions which varied by 60% under excess air combustion (condition which did not promote the NO reduction mechanisms mentioned above), presumably due to the different amount of N released in volatiles. This result implies that the effectiveness of combustion 7 Depending on available oxygen, the temperature and the size of a coal particle, reaction can be chemically controlled or diffusion limited. For example, at high temperature and for adequate oxygen mole fraction, the reaction is diffusion limited, since chemical reactions are much faster. 8 The reader should bear in mind that in this thesis we are interested in swirl burners where residence times in the flame are small compared with char oxidation. In the case of fluidised bed combustion, the time scale of char oxidation is comparable to the residence time in the bed and, thus, the process of NOx production (and reduction) is different as will be briefly described below. 9 The term is explained below. 10 American Society for Testing and Materials. 37

38 1.2 NOX EMISSIONS 38 technologies for low NO x emissions depends on the partitioning of the total nitrogen in coal in volatiles and char. These authors also found that bituminous coals released most nitrogen during pyrolysis in the form of HCN, whilst low-rank coals (e.g. lignite) released mostly NH 3. However, in experiments conducted by Nelson et al. (1992) with a range of Australian coals pyrolised in a fluidised bed combustor at temperatures C, the proportion of nitrogen in coal released in the form of NO x precursors (e.g. HCN), was independent of coal type, or amount of nitrogen in volatiles, contrary to the results of Chen et al. (1982). Other researchers could not correlate the amount of NO produced with the amount of N contained in coal (Pershing and Wendt 1977) or found coal-n conversion to NO varied between 45 and 100% depending on coal type, when particles were fired in an arc-jet entrained flow reactor (Haussmann and Kruger 1990). The experimental findings reviewed above indicate that although the amount of nitrogen, which will be released from different coal types under the same experimental conditions, depends on coal type, the amount of NO which may be produced from nitrogen evolved in volatiles depends on the firing conditions rather than the fuel. In addition, provided that one can limit the extent of conversion of volatile nitrogen to NO (presumably providing the conditions so that the chemical paths towards formation of N 2 and destruction of NO, presented in equations , prevail), then char nitrogen is the greatest NO x contributor, because of its slow reaction with oxygen relative to its residence time in the flame, which renders char nitrogen conversion insensitive to combustion modifications in the near-burner region. Char burnout time in a furnace, i.e. the time taken until the carbon of an ignited char particle is consumed and the particle is reduced to fly ash, is of the order of seconds (e.g. Wall 1987; Field et al. 1967). Conversion of char nitrogen is an increasing function of carbon oxidation, therefore the time scale of char nitrogen conversion to NO is comparable to the time scale of char oxidation (Tullin et al. 1993). Particles injected at 10 m/s in a 1 m diameter swirl burner are bound to remain in the flame region for about 100 ms and, thus, it is unlikely that within their residence time in the flame, which can be controlled by modification of the near-burner aerodynamics, char-bound nitrogen conversion can be controlled. As explained in the next section, it is the conversion of nitrogen in volatiles to NO x which can be modified. The literature suggests that effective NO x reduction can be achieved by combustion technologies which maximise volatile yield (and, hence, minimise nitrogen retention in char) and ensure that the evolved nitrogen reduces to N 2. Such technologies shall be reviewed in the following section. 11 It is useful, at this point, to explain why NO reduction occurs in rich combustion systems in the presence of hydrocarbon radicals. The bulk of heat during combustion is produced during oxidation of carbon monoxide (CO) to carbon dioxide, via OH radicals (see Griffiths and Barnard 1995). A series of reactions occur until CO oxidation, that result in breaking of the carbon backbone of heavier hydrocarbons down to methane (Skevis 1996) provided, of course, that the fuel is heavier than methane. Other reactions also result in generation of hydrocarbon, as well as H, O and OH radicals. Radicals initiate and maintain the reaction (at least in the high-temperature chemistry of combustion phenomena). CO oxidation by OH has a competing reaction, in the form of hydrogen abstraction (removal of a hydrogen atom) from the hydrocarbon fuel by OH, which is of the order of 10 times faster than CO oxidation. For example, the hydrogen abstraction reaction for methane is: & & , and results in production of CH 3 radicals (which will further react and also produce smaller radicals down to CH, which is an effective NO reduction agent). This last reaction will occur provided that, for a uniform mixture, local conditions are fuel-rich so that some fuel remains unreacted after ignition. Of course, there is a limit to the excess of fuel in the mixture, because excessively fuel-rich conditions result in unacceptably high CO emissions, and extinction problems, due to temperature drop (owing to smaller heat release from CO oxidation).

39 CHAPTER 1 INTRODUCTION NO x reduction It is already outlined in that NO can be reduced to N 2 either by reaction with N (equation 1.2) or, even more effectively, by reaction with hydrocarbon radicals (equation 1.4). The reduction of NO to N 2 by hydrocarbon radicals that occurs in rich combustion systems 11 (Miller and Bowman 1989), has been suggested by many researchers as method for reduction of NO x emissions (e.g. Reed et al. 1969; Wendt et al. 1973; Myerson 1975; Takahashi et al. 1981). Of the major techniques to reduce NO x emissions in burners utilising pulverised fuel, the most common are air staging and fuel staging, or reburning as the latter is better but perhaps misleadingly 12 known. Vitiated air firing has also been found to result in reduction of NO x emissions with the additional merit of increased cycle efficiency when a conventional pulverised coal cycle is combined with a gas turbine topping cycle. In air staging, only part of the oxidising air is supplied early with the fuel, resulting in a substoichiometric 13 region, whilst the rest of the oxidiser is provided downstream in order to complete combustion. Reburning is a similar process whereby fuel is provided in two stages. Although in practice reburning methods apply to boilers, rather than to individual burners 14, with the additional fuel added downstream of a matrix of burners, the process of NO x reduction by air and fuel staging corresponds to two different implementations of combustion modifications, based on the same NO-destruction principle. Vitiated air (Supplementary) firing A notable example of modification of the combustion process (and the thermodynamic cycle in a power plant) which can also reduce NO x emissions is use of the exhaust gas of a combustion process mixed with atmospheric air as oxidant. The purpose, of course, of utilising turbine exhaust gas (vitiated air) 15 is to increase the overall efficiency of the cycle (a subject which is treated in the next section, with emphasis on the hot gas clean-up of the flow from particulate matter), rather than primarily reduce NO x emissions. This situation is thus to be contrasted with automotive practice. 12 Misleadingly because the fundamental reaction, equation (1.4), is NOT a combustion reaction. It is a reduction mechanism. 13 So far the term fuel-rich was used, but in reference to a homogeneous mixture of gases. Here, we now consider an inhomogeneous mixture of solid particles and air and, in order to avoid confusion, the term substoichiometric (air) was preferred. This describes a combustion zone (or an inhomogeneous mixture in general), where the amount of air is less than the overall stoichiometric requirement for complete oxidation of the fuel. 14 in which case the more proper term fuel staging should be used, although in their original publication in 1973 Wendt and his co-workers proposed the term reburning for this two-stage fuel injection process. 15 Combustion using turbine exhaust gas (TEG) term commonly used by the IFRF is also known as supplementary firing (Haywood 1991) or flue gas recirculation (Soud and Fukasawa 1996). The term exhaust gas recirculation (EGR) is commonly used to describe this process in an internal combustion engine. 39

40 1.2 NOX EMISSIONS In applications other than power generation, vitiated air 16 has been used in a cement kiln to reduce NO emissions. Hunter and Benson (1985) reported a 20-30% reduction in NO when the oxygen content in primary air decreased from 21% to 13%, without deterioration of the process in the kiln. However, what is of more interest to the work of this thesis is that, in a furnace using a 500 kw swirl burner, reductions between 30% and 70% were measured, depending on the type of the injector and oxygen concentrations (Abbas et al. 1997). Unfortunately but predictably, the stability limits 17 became increasingly narrower with decreasing oxygen concentration, which limits the potential of application of the process. Less encouraging evidence for the universal efficacy of vitiated air to burn pulverised fuel were the studies by Smart and van de Kamp (1994) and van de Kamp and Smart (1992) who found that its use did not necessarily result in NO x reductions. When they took into account the increased volumetric flow rate through the furnace due to increased temperature, they discovered that NO x output was a non-linear function of the oxygen concentration in the vitiated air. Especially in the case of a low- NO x flame, utilisation of vitiated air of 15% oxygen mole fraction increased the NO x output (actual values depended on coal type but the average increase was about 20%) relative to the case of atmospheric air. Another side effect of vitiated air was the fivefold increase of the unburned fraction of carbon for vitiated air with oxygen mole fraction of about 11-13%. It can be speculated here that one of the reasons for the decrease of the carbon unburned fraction is the combination of decreased oxygen concentration (which results in smaller carbon reaction rates) and decreased particle residence time in the hightemperature regions (owing to the higher particle velocities due to the increased volumetric flow rates through the burner). One can argue, of course, that the increased temperature of the oxidant due to the enthalpy of the vitiated air should compensate for the decreased residence times. Accordingly, detailed measurements in vitiated air pulverised coal flames using a pyrometer/velocimeter can reveal the balance between the two variables. In spite of the possibility of having undesirable side effects on the combustion process due to the vitiated air, it is possible to incorporate turbine exhaust gas in a burner which burns synthetic gas produced from gasification of coal, such as that reported by Westinghouse Corporation (Bachovin and Domeracki 1996), in conjunction with pulverised fluidised bed, in a combined combustion cycle developed in collaboration with Foster Wheeler Development Corporation (e.g. Robertson et al. 1993). Bachovin and Domeracki used a low-no x burner to burn the synthetic gas, whilst the pressurised fluidised bed consumed the char remaining from gasification and, by optimising the stoichiometry and temperature of the burner primary combustion zone, they achieved NO x reductions of the order of 50% compared with non-optimised conditions. The advantages of their method is that gas combustion is a process more easily controlled than combustion of pulverised fuel and no flame stability problems are experienced. One disadvantage of the proposed cycle is that gasification is generally expensive and unreliable, 16 see Chapter range of conditions for which stable flames existed. 40

41 CHAPTER 1 INTRODUCTION and there is less experience compared with direct coal firing in vitiated air. If the limitations of gas combustion in vitiated air regarding flame stability at partial load for different operational parameters of the burner (and fuels) are established, this method can be an alternative to conventional supplementary firing. Reburning Early experiments by Wendt et al. (1973) showed that there is a potential for reduction of NO x by addition of gaseous fuel as a reducing agent downstream of the primary combustion zone. Measurements in a flat premixed flame with 10% excess air showed that, addition of about 100% CH 4 relative to the primary fuel reduced NO x emissions by 50% compared with the non-reburning case. The authors concluded that NO x was formed in the primary zone and destroyed through reburning. They suggested that this method was more suitable than airstaging, particularly in combustion of dirty fuels, because soot generation was suppressed in the primary zone, owing to combustion in excess air. A similar method for NO x destruction, heuristically explained, was suggested by Reed et al. (1969). Contrary to the claims of Wendt et al. (1973), they suggested that the primary combustion zone contained substoichiometric air, and this condition did not result in smoke emissions. Although reburning is thought to reduce NO x emissions through the reaction of hydrocarbon radicals (equation 1.4), Spliethoff et al. (1996) claim that the reduction is achieved through N atoms. They burned coal in an electrically-heated entrained-flow reactor and used a range of gaseous reburn fuels with variable hydrocarbon content. Of all reburn fuels the highest reduction efficiency was achieved by pyrolysis-gas, which mostly consisted of H 2, some CO and hydrocarbons and N-containing species. Unfortunately, reburning is by itself complicated because the amount of available oxygen in the reburning zone is crucial for the effectiveness of the reduction, as is the temperature of gases. A well-balanced air-to-fuel ratio in the reburning region is required, in order to produce enough hydrocarbon radicals for the reduction (Myerson 1975). In a series of experiments in an electrically-heated reactor kept at temperatures in the range K, Myerson (1975) supplied simulated exhaust gas and created a reburning zone using mixtures of oxygen and butane (i-c 4 H 10 ). Oxygen at high concentrations completely oxidised the fuel and no reduction was observed, whilst the reactions were too slow in the case of insufficient oxygen concentration and produced inadequate quantities of hydrocarbon radicals to have any destructive effects on the NO of the initial mixture. One of the earliest practical applications, where full-scale burners and pilot-scale boilers were utilised to produce a reburning facility, was the MACT (Mitsubishi Advanced Combustion Technology) system designed by Mitsubishi Heavy Industries (Takahashi et al. 1981). Reductions of the order of 50% were achieved both in a 0.8 m diameter furnace and a 20 MW 41

42 1.2 NOX EMISSIONS corner-fired pilot furnace using in the reburn zone 10% fuel of that of the primary zone. In both cases, a low-no x burner was used that utilised offset combustion by co-injecting alternating streams of fuel and air at small angles relative to each other, thus achieving premixing of fuel and air (Takahashi et al. 1979; Soud and Fukasawa 1996). A temperature of the exhaust gases for reburning of at least 900 C (so that hydrocarbon breakdown is initiated) and typically 1200 C was necessary, in agreement with the experiments of Myerson (1975). From the temperatures required to ensure that the method is effective it can be argued that the reburning process as described, for example, by Takahashi et al. (1981) is not self-sustained combustion, rather a high temperature process. One of the disadvantages of the method is that, except from the temperature, the stoichiometry of the primary and reburning zones as well as the residence time of the fuel in the reburn zone must be carefully controlled so that the effective NO x reductions can be achieved. For example, in the experiments by Takahashi et al. (1981), propane was burned in a 0.8 m furnace with 1% excess O 2 whilst 2% O 2 existed in the reburn zone rendering it a fuel-rich zone. More air, of course, had to be added downstream of the reburn zone in order to complete the oxidation of the hydrocarbons (and of the CO) which resulted in production of additional NO. Tail-pipe NO x reductions of up to 80% were achieved by this method and they were independent of the fuel used. The only difference with coal firing is that for the reburn zone coal was pneumatically transported by N 2, presumably to avoid reactions during transport. It can be also argued (although Takahashi et al did not report any data) that if coal is used as reburn fuel, particle dispersion, heat-up and pyrolysis are important for generating the local conditions suitable for NO reduction. Other than the trials of Takahashi et al. (1981) reburn systems using gaseous fuels have been applied in full-scale boilers, such as, for example, that reported by Spliethoff (1991), in which potential NO x reductions of the order of 65% were reported by use of coal gas containing 50% H 2 and 25% CH 4. Unlike gaseous reburn fuels, use of pulverised coal is not as extensive and tests are required to optimised the boiler operational parameters, such as the stoichiometry of the reburn zone, the particle size distribution and the fuel injection system. Combination of test-scale data and calculations of the flow inside a 320 MWe boiler indicated that 20% of the fuel should be used in the reburn zone and that there was no necessity for micronised fuel for expected reduction of NO x emissions of up to 60%, whilst achieving high char burnout (Antifora et al. 1997). In a study by Hesselmann et al in a 160 kw coal-fired with coal as reburn fuel, maximum reduction was achieved with reburn air ratio of 0.9. Independent of the stoichiometry, a particle residence time in the reburn zone of 0.8 seconds was required for maximum NO x reduction. Hesselmann et al. (1997) claim that the reburn fuel injection velocity and momentum and the number of injection nozzles are the key parameters influencing the mixing between the reburn fuel and the exhaust gas from the primary zone. This suggests that although for different reasons compared with the case of NO x formation in the nearburner region knowledge of the interaction between particles and the turbulent flow allows 42

43 CHAPTER 1 INTRODUCTION control of particle motion in the reburn zone, which is important for achieving maximum NO x reduction. Reburning has been applied in a 2.5 MW swirl burner at the International Flame Research Foundation (IFRF), in the form of an internally fuel-staged burner (IFSB), where nozzles protruding from the quarl exit injected secondary (reburn) fuel downstream of the coal flame, thus creating a reburning zone (Knill et al. 1989). They found that NO x emissions decreased from 1000 to 250 ppm by reducing the reburn zone stoichiometry 18 from 1.1 to This agrees with expectations that excess quantities of fuel are required to produce the NO-reducing hydrocarbon radicals. The idea behind the IFSB resembles that of the 2.5 MW aerodynamically air-staged burner (AASB, Smart and Weber 1987) where the air, rather than the fuel, is staged. Detailed NO x emissions profiles on the centreline of the burner showed that, although NO x emissions in the case of the unstaged burner increased for distances in the furnace longer than 0.3 m (diameter of quarl exit is 0.85 m), in the AASB not only was the overall reduction in emissions more than 60% 19, but also the maximum NO x concentration occurred at 0.3 m and decreased farther downstream. In view of the reductions achieved by the IFSB, it is likely that the observed reduction with the AASB downstream of the 0.3 m axial station from the burner was due to the slow reburning effect of the char. The latter has been observed in fluidised beds, where 25% NO reduction was measured for a stoichiometric air coal mixture with 50% of devolatilisation already completed (Song et al. 1981). Air staging The purpose of air staging is to suppress NO formation by reduction of nitrogenous species in the flame to N 2, as explained in 1.2.1, and relies on the generation of a rich combustion zone (by feeding only part of the air in the early phase of combustion, thus air staging ) where reduction of NO to N 2 is preferred over oxidation of N to NO. Well-documented designs of air-staged burners are those developed at IFRF (e.g. Smart and Weber 1987 and 1989). The two main designs are the air-staged precombustor burner (ASPB) and the aforementioned aerodynamically air-staged burner. The former relies on the physical separation between the secondary and tertiary air streams achieved by use of a precombustor, i.e. a confined region within the furnace near the burner exit that is narrower than the furnace itself. The purpose of the secondary air stream is to supply the bulk of air required for combustion 18 In accordance to footnote 13, stoichiometry is defined as the ratio of the actual volumetric supply of air and the stoichiometric requirement, and takes, therefore, values smaller than unity in the case of air supply smaller than the stoichiometric requirement. This term should not be confused with the fuel equivalence ratio, defined in Chapter 4, which takes values larger than unity in the case of a substoichiometric mixture. 19 Explained in the next section. 43

44 1.2 NOX EMISSIONS but to maintain a substoichiometric primary combustion zone, whilst the tertiary stream is supplied downstream of the secondary to help to complete combustion. The most important parameter that affected emissions was found to be the particle residence time in the precombustor. Although smaller NO x emissions were achieved for lower stoichiometry of the primary zone, without any significant effect on the carbon burnout from the furnace, reductions were stronger with increasing residence times in the precombustor. For example, strong reductions of the order of 40% were achieved for an estimated residence time of 100 ms when stoichiometry was reduced from 0.8 to 0.6, but became less pronounced for a further lowering to 0.4. The reductions became more significant with increasing residence times, although the rate of decrease depended on the stoichiometry: emissions from flames of stoichiometry closer to unity were more sensitive to variation of residence time than substoichiometric ones. The AASB was an alternative design whereby the fuel injector was inserted in the flow and the fuel was thus introduced inside the recirculation zone, rather than upstream. The minimal particle dispersion towards the oxygen-rich shear layer of the recirculation zone due to the initial momentum of the coal jet, and the late combustion of particles inside the oxygendeficient recirculation zone, resulted in reductions of the order of 50% when the injector was inserted by about mm (diameter of exit of quarl: 850 mm) compared with the case of fuel injector at its normal position (Smart and Weber 1987 and 1989). The influence of insertion of the injector in the flame on particle dispersion was also confirmed from detailed simultaneous measurements of particle velocity and size in a laboratory-scale, 10 kw swirl burner (Orfanoudakis 1994). The reduction in NO x emissions observed by Smart and Weber (1987 and 1989) is correlated with late heat-up of the gases in the penetrated flame, as suggested by spatially-averaged temperature measurements (Woycenko and Smart 1993). Combustion occurred at about 1.7 diameters downstream of the burner throat as opposed to the 1400 K temperatures measured at axial distances smaller than 0.5 m in the non-penetrated flame. It would be useful therefore to measure spatially-precise particle temperature simultaneously with particle velocity and size in order to identify if any the mechanism which would lead to a particle-flow pattern similar to that manifested in the flows of Smart and Weber (and which was created by the mechanical insertion of the coal injector into the flow) and which was responsible for the low-no x emissions. Comparable reductions in NO x were also accomplished by using smaller swirl numbers in a 500 kw furnace, fired on a burner with annular coal injection and swirling primary and secondary air streams (Schnell et al. 1993). When the swirl number decreased from about 1.5 to about 0.5, reductions ranging from 30% to 40% were achieved for coarse and fine grindings respectively, for a penetrated flame (see Smart and Weber 1987 for definitions on the type of flame). Calculations accompanying the measurements confirmed what is intuitively expected, namely that in the case of lower swirl numbers which produced lower emissions (compared with higher swirl numbers), particles remained inside the recirculation zone. On the contrary, 44

45 CHAPTER 1 INTRODUCTION the stronger the swirl, the higher the proportion of particles which escaped towards the shear layer of the recirculation zone where more oxygen was available (as, for example, indicated by the measurements of Milosavljevic 1993). Although no experimental confirmation of the calculation was provided, Orfanoudakis (1994) has measured similar trends on particle dispersion in his 10 kw swirl flames. In their 500 kw furnace, Abbas et al. (1993) found that NO x emissions increased with increasing swirl only for swirl numbers larger than unity, whilst for smaller than unity the trend was reversed. It is likely that the latter observation was due to aerodynamic interaction between particles and the weak flow for swirl smaller than unity, showing the complexity of particle motion in a swirling flow. The effect of high swirl number on NO x emissions agrees with current designs of low-no x swirl burners (e.g. Beér 1995; Soud and Fukasawa 1996) in which the secondary air stream is divided into multiple streams supplied in turn to the main flow. Apart from control of the local stoichiometry, such burners are most likely to modify the aerodynamic response of particles to the flow in a way which promotes particle motion into the recirculation zone. The latter was the reason for the 30% reductions measured in a 500 kw furnace, when a central orifice injector was utilised in place of annular injection 20 (Abbas et al. 1993). The calculated mean trajectories in the case of central orifice injection indicated that the particle residence time in the recirculation zone, which provides conditions suitable for NO x reduction, is at least twice as long as the particle residence in annular injection (Abbas et al. 1994). Comparison between the quantitative experimental results and qualitative findings from numerical modelling of the flow suggest that the type of particle trajectories, which result in a range of residence times in the primary combustion zone, is strongly correlated with NO x emissions. Published experimental investigations in swirl burners (Orfanoudakis and Taylor 1995; Orfanoudakis 1994) reported correlations between size and axial particle velocity for a range of swirl numbers and overall burner stoichiometry and identified that small particles dispersed inside the recirculation zone, whilst big ones went through it, because of their larger momentum. Their extensive studies were limited by the fact that no particle temperature was measured. However, the previous work reviewed in this section implies that low-nox emissions are correlated with particle temperature/ ignition probability, which is a consequence of particle dispersion in the high-temperature region of the recirculation zone and this assumption must be experimentally confirmed for local flow conditions simulating an air-staged flow. 1.3 In-situ Hot Gas Cleaning in Direct Coal Firing As mentioned in 1.1, it is possible to reduce NO x emissions by modification of the combustion process. Experimental, as well as computational, work reviewed in showed that particle motion is influenced by modification of design and operational parameters, such as the swirl 20 which is common in full-scale burners, Wall

46 1.3 IN-SITU HOT GAS CLEANING IN DIRECT COAL FIRING number, with consequences on the generated NO x emissions. As explained in 1.1 regarding CO 2 emissions, increase of the efficiency of cycles (say by a topping cycle) will result in less specific production of CO 2 (kg/mwh of electric power). The incorporation of a gas turbine topping cycle, though, imposes an additional problem to the design of certain cycles (see below) in the form of the coal particles in the hot gas fed into the turbine which will, at some point, destroy the turbine blades. Although the purpose of this thesis is not to investigate the thermodynamics of such cycles, it is important to ensure that the gas is somehow cleaned insitu (i.e. at temperatures of about 1300 C) from the particles before entering the turbine and that involves designing the aerodynamic interaction between the continuous and the dispersed phases. Generation of electricity from burning solid fuels in high-pressure combined cycles, which include Rankine- and a topping gas turbine (Brayton)- cycles, offers thermal efficiencies higher that those obtained from conventional steam cycles (Reichert et al. 1990). From the most important combined-cycle technologies, namely pressurised fluidised bed combustion (PFBC), pressurised gasification and pressurised pulverised coal combustion (PPCC), the latter offers thermal efficiencies comparable to those currently delivered by natural gas powered electricity generation (Reichert et al. 1990; Benesch et al. 1995). In addition, PPCC -unlike PFBCmakes it possible to take advantage of the gas turbine developments regarding the increasing maximum turbine inlet temperature (Preusser and Spindler 1988) and therefore constitutes a promising technology for future developments. The requirement for particulate-free exhaust gas from the pressurised combustion chamber on entry to the gas turbine, necessitates inline hot gas clean-up of the ash from the gases to a very stringent level, which is particularly difficult at the exit temperatures of a PPCC system (Weber 1988; Reichert et al. 1990). One potential solution is to use the externally fired combustion cycle (EFCC, LaHaye and Bary 1994), where a heat exchanger is used to heat up air which will subsequently enter the gas turbine. 21 This, of course, reduces the thermodynamic efficiency. 22 Known particle retention technologies, like electrostatic precipitation and ceramic bag filters, cannot be used at the high temperature (around 1300 C) of the exhaust gases, therefore there is a need for efficient hot gas clean-up in the combustor. Because of the high temperature and pressure of the exhaust gases, efficient particle retention is difficult. Therefore, only a few pilot PPCC systems have been built to date, most notably the 3.5 MW 6 bar slagging combustor developed by Westinghouse (Mattson and Stankevics 1985; Bannister and Newby 1996), 21 Although hot gas clean-up is not required for the turbine, minimisation of heat exchanger fouling is required and thus clean-up of some kind remains important. 22 Another difficulty associated with direct firing (including PFBC) is corrosion due to alkali. EFCC also avoids this difficulty. 46

47 CHAPTER 1 INTRODUCTION models of which were investigated by Moore (1998), and the 1 MW slagging combustor operated at pressures up to 20 bar at the IUVT institute in the University of Essen, Germany (Reichert et al. 1990). In full-scale PPCC furnaces, one relies on the formation of a sticky slag film on the furnace walls, owing to high temperatures (~1600 C), which retains particles either by adventitious collision due to turbulence or by aerodynamic (inertial) centrifuging, owing to mean streamline curvature, in order to clean-up the exhaust gas by removing the particulate phase. The particular example of aerodynamic clean-up considered here arises owing to the abrupt streamline convergence at the furnace exit. Aerodynamic centrifuging of particles is a common method for efficient particle-laden flow cleaning-up and it is widely used in cyclone separators. Centrifuging can be used in coal combustors, which must be designed to obtain high ash retention (Reichert et al. 1990). In principle, this means that a suitably defined centrifuge Stokes number, which is a mean flow time scale divided by the particle relaxation time scale (e.g. Hardalupas et al. 1992), must be smaller than unity, so that the particles are centrifuged owing to the mean streamline curvature. It is likely that centrifuging owing to turbulence is also useful, particularly near mean recirculation zones (Siu 1996). Particle relaxation time depends on its inertia (Hardalupas et al. 1992) which is a function of particle size; hence development of basic knowledge and of particular designs would benefit from simultaneous measurement of velocity and size of particles and of the continuous flow. Using such measurements one can identify particle leakage from a desired trajectory, particularly due to turbulence. This means in turn that optical instrumentation for spatially-precise in-situ measurements of the velocity and the size of a single particle of arbitrary shape (such as coal) must be developed. 1.4 Previous Work on Optical Particle Sizing In contrast to the extended development of the phase Doppler anemometry (PDA) method for sizing spherical particles and droplets, techniques for inflow sizing an irregular dispersed phase are still at an early stage. A number of methods have been developed based on principles such as, for example, holography (e.g. Trolinger and Heap 1978), the amplitude of diffractivelyscattered light (e.g. Allano et al. 1984; Fristrom et al. 1973; Hishida et al. 1982; Holve 1983; Maeda et al. 1986; Morikita et al. 1994; Orfanoudakis 1992; Orfanoudakis and Taylor 1995), the visibility of a Doppler signal (e.g. Farmer, 1972; Hong and Jones 1976) or imaging (e.g. Hardalupas et al. 1994; Ren et al. 1995) but the number of optically-based instruments for simultaneous measurement of size and velocity is small and mostly based on the amplitude of diffractively-scattered light. Such an instrument has been used for an extensive study of unconfined and confined, swirling, small-scale (10kW), gas-piloted pulverised-coal flames 47

48 1.4 PREVIOUS WORK ON OPTICAL PARTICLE SIZING (Orfanoudakis 1994). In most practical situations, and frequently even in laboratory scale models, pulverised-coal flames are confined and the sizing beam has to pass through glass windows subject to uncontrolled movement due to thermal expansion, making the maintenance of optical alignment itself, unfortunately, a difficult task (Holve 1983). In addition, window fouling due to particles sticking on it during measurement deteriorates substantially the accuracy of sizing (Orfanoudakis 1994). Ultimately in his measurements of confined flames, Orfanoudakis (1994) had to make use of small window-less slots cut into the confining furnace wall to permit unfouled access to the flame by the laser beams. Because of the complexity of developing and applying optical instrumentation for simultaneous measurement of particle velocity, size and even temperature, only few studies exist of particle behaviour in non-unidirectional flows, other than those of Orfanoudakis (1994) and Orfanoudakis and Taylor (1995), albeit either for a particle ensemble instead of a isolated particle or for measurements of only particle velocity. Measurements of size and velocity of coal particles have been reported in an 85 MW tangentially-fired furnace (Bonin and Queiroz 1991) at locations far downstream of the burner region but because they were ensemble, rather than single, particle measurements they do not provide information on the aerodynamic behaviour of particles with size. In full-scale furnaces, many optical- and physical access limitations severely restrict the measurements and, for example, in a 37 MW burner of a 2000 MW power station, limited velocity measurements were obtained by using laser Doppler velocimetry (Ereaut and Gover 1991), particle size was not resolved and the flow parameters could not be varied for a detailed investigation. Of significantly more complexity is, naturally, the simultaneous measurement of velocity, size and temperature 23 of isolated coal particles and to the author s knowledge, there is no such study to date for the case of recirculating flows, such as swirl flames. One of the few studies in a laboratory swirl burner concerns the measurement of size, velocity and temperature of kerosene droplets (Israel et al. 1995), a much easier task because droplet sizing is simpler. In the case of coal particles there is published data regarding velocity-size and temperature correlations in drop tube furnaces, because the particle sizing techniques required in unidirectional laminar flows are simpler. More details on this will be presented in Chapter 3. Numerous researchers have measured single particles undergoing combustion in drop tube furnaces, such as, for example, particle temperature using two-colour pyrometry (e.g. Panagiotou and Levendis 1994) or simultaneous particle temperature and morphology (shape) measurements (Schroeder et al. 1992). In drop tube furnaces the flow is laminar, unidirectional and the flow conditions easily controlled, unlike turbulent flows such as investigated in this thesis. Review of studies in drop tube furnaces is beyond the scope of this thesis. 23 A brief review of particle temperature measurements is presented in Chapter 3. 48

49 CHAPTER 1 INTRODUCTION 1.5 The Present Contribution The existing knowledge of chemical mechanisms for NO x generation and destruction during coal combustion, and the experimental evidence on the potential for reduction of NO x emissions suggests that the aerodynamic interaction between pulverised fuel particles and the turbulent flow in swirl stabilised burners/boilers and presumably other designs (such as tangentiallyfired burners) and, consequently, the particle temperature history are strongly correlated with emissions. It is well documented that, for example, reducing the swirl number yields smaller NO x emissions because particles disperse increasingly into the recirculation zone, the oxygen-deficient environment of which promotes smaller NO x generation, as well as NO x destruction. The velocity and size of particles have been found to be directly or indirectly correlated with NO x emissions, presumably because particles follow trajectories which depend on their instantaneous interaction with the turbulent flow. This process is a function of size and velocity of an individual particle, as well as the instantaneous velocity of the continuum. Emissions depend on particle dispersion and heat-up in the flow. The heat-up of the particle (i.e. the particle temperature history) depends on the local conditions that a particle encounters along its trajectory and, accordingly, the probability of particle ignition and subsequent burning, as well as the location of ignition depend on the temperature history of a particle. Particle dispersion (which depends on particle velocity and size, as well as flow velocity) cannot be separated from particle heat-up (and combustion the temperature history which, as explained, also depends on particle size and velocity) and, therefore, knowledge of particle velocity, size and temperature simultaneously is important. Most previous studies have been conducted in furnaces of the scale of 500 kw up to 2.5 MW which, although provide scales closer to those found in full-scale plants, impose limitations on the measurements that can be taken. Development of optical instrumentation for non-intrusive measurements which can provide simultaneous data of the velocity, size and temperature of individual particles in a turbulent flow like those already mentioned is by itself a formidable task. In addition, optical instrumentation requires uninterrupted optical access to the flow and this is not always feasible in industrial scale. It is inevitable, particularly after review of the limited information that has been obtained in the past in semi-industrial and industrial scales, that understanding of the aerodynamic interaction between individual particles and the flow, as well as understanding of particle heat-up can only be obtained in laboratory scales under controlled flow conditions, by measurements of high spatial resolution and accuracy. Extrapolation of the results in lab-scales to industrial scales rests with CFD tools (usually validated on detailed lab-scale measurements). Because of these reasons, there is a gap between less-detailed experimental data obtained in semi-industrial scale swirl flows and those of high accuracy and resolution obtained in controlled laminar, unidirectional flows, as those in droptube furnaces. This gap motivated the measurements in laboratory-scale swirl burners, presented in this thesis. 49

50 1.6 THE STRUCTURE OF THIS THESIS The purpose of the present contribution is to develop novel instrumentation for the simultaneous localised measurement of velocity, size and temperature of coal particles, as well as the volume flux of a particle ensemble, and use it to study the aerodynamic and combustion characteristics of coal particles in swirling reacting flows. The experimental data not only extend our understanding of the way particles disperse and ignite in swirl flames, but also provide a basis for evaluation of computer codes. 1.6 The Structure of this Thesis CHAPTER 2 describes in detail part of the instrument concerned with size and velocity measurement, namely the shadow Doppler velocimeter (SDV). It also evaluates its accuracy and precision of measurement of the volume flux of an ensemble of particles. The SDV is a novel imaging technique for sizing irregular particles and can also simultaneously measure two particle velocity components. This chapter describes the optical and electronic configuration, and explains the processing of the signals and the potential sources of data rejection. Its accuracy of measuring particle volume flux was assessed in a vertical turbulent water channel flow. CHAPTER 3 describes the part of the instrument concerned with temperature measurement of a radiant body, based on the principle of two-colour pyrometry, as well as the optics and electronics of the current two-colour pyrometer. In addition, the SDV was combined with the two-colour pyrometer in terms of optics, electronics and control software to produce a novel instrument which can simultaneously measure the velocity, size and temperature of single particles of arbitrary shape, and the flux of an ensemble. A theoretical analysis is also presented which lead to formulation of a criterion for the interpretation of the pyrometer signals and thus increased the reliability and accuracy of the pyrometer measurements. CHAPTER 4 presents results obtained in the near-burner region of a 10 kw natural-gas supported swirl burner, confined or unconfined, and in the far-burner region of a similar confined burner. Results are reported on the velocity, size and temperature of coal particles in the nearburner region, and size and velocity in the far region, from which, centrifuging effects on particles were assessed. Also, the influence of burner design and operational parameters, such as swirl number, primary to secondary air momentum ratio and the overall stoichiometry of the support flame on particle velocity and temperature were measured, and their potential impact on NO x emissions was discussed. CHAPTER 5 presents results from studies on gaseous and pulverised-coal swirl-stabilised gassupported flames where vitiated air was used as oxidant. The first part presents a detailed study of the effect of oxygen concentration, temperature and strain rate on the stability limits of the 50

51 CHAPTER 1 INTRODUCTION gaseous flames, whilst the second investigates the effect of oxygen concentration in the oxidant on particle combustion characteristics. In addition, a simple theoretical model on particle heat-up which implemented an ignition criterion based on Semenov s thermal explosion theory was utilised. The analysis simulated the heat-up and combustion processes in the present flow and was used to estimate the time scales of combustion, in order to explain the measured trends. CHAPTER 6 is a summary of the most important conclusions from this work and suggestions for further research. 51

52

53 CHAPTER 2 Measurement of Particle Velocity, Size and Flux by Shadow Doppler Velocimetry

54 2.1 BACKGROUND 2.1 Background Shadow Doppler velocimetry (SDV) is a novel imaging technique for sizing particles of arbitrary shape and arbitrary refractive index and was first proposed by Hardalupas et al. (1994). The method is based on the collection of the light diffracted by a particle illuminated by two laser beams 1, thus the term shadow 2, from which the cross-sectional area of the particle can be directly measured. In contrast to existing sizing methods based on the collection of diffracted laser light scattered by a particle which were reviewed in Chapter 1, such as that of Orfanoudakis (1994), SDV is not amplitude-based as explained below. The independence of the method from the amplitude of diffractively-scattered light means that no elaborate calibration of an instrument based on the technique is required before measurements, and reliable results can be obtained in confined two-phase reacting flows, where it is necessary to access the flow through optical windows which are susceptible to particle accretion and random movements due to thermal stresses developed on the surrounding metal parts confining the flow. In addition, particle mass flux measurements of accuracy higher than those obtained, for example, with amplitude-based sizing techniques or phase Doppler anemometry can be made due to the strictly geometrical definition of the area of the sampling space, through which flux is measured, in contrast to the irradiance-dependent estimation of the effective probe volume in phase Doppler systems. More details on the method and the accuracy of the measurement of particle mass flux are presented in Optical Arrangement The generic optical setup of the SDV instrument used in the present work is presented in figure 2.1. The transmitting optics were those of a conventional laser Doppler velocimeter and consisted of an Ar+ laser operating at 488 nm and power output of mw and a beamsplitter and optical shifter unit which in the present work was a commercial system (Model DISA 55X, DANTEC A/S) with optical frequency shifting provided by a single Bragg cell. 3 The beams were focused by a f/600 singlet and formed the LDV probe volume. The collection 55 1 The method is independent of the number of laser beams by which a particle is illuminated and in its simplest form only one beam can be used (see, for example, figure 10.1 of Hecht 1987). An arrangement where the two beams of a laser Doppler velocimeter are used offers the advantage that two components of particle velocity and the particle mass flux can be simultaneously measured. In addition, particle location in the probe volume relative to the focus of the receiving optics can be measured. 2 The same way the shadow behind objects is formed when they are illuminated by non-coherent polychromatic light, such as sunlight. In such a case, the total dark shadow behind an object is called umbra, whilst the lighter shadow surrounding the umbra is called penumbra. 3 Initially a diffraction-grating-based unit was used but the small size of the resulting LDV probe volume and non-uniform irradiance distribution in the probe volume made the measurements difficult and that unit was abandoned.

55 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY $OXPLQLXPFRDWHG PLUURU I ;RU; REMHFWLYH 3DUWLFOHVKDGRZ LPDJH /'9SUREH YROXPH I I I 3KRWRPXOWLSOLHU 6LQJOH%UDJJ&HOO /'9WUDQVPLWWLQJ2SWLFV 'LUHFWLRQRI SDUWLFOHPRWLRQ 'LRGHDUUD\ 'LUHFWLRQRI SDUWLFOHPRWLRQ Figure 2.1 Optical setup of the shadow Doppler velocimeter instrument. optics 4 consisted of an 80 mm-diameter f/500 (Melles Griot Inc., USA) and an 82 mm f/300 (Spindler & Hoyer GmbH & Co., Germany) achromats which collected and focused the laser beams respectively on an 40X microscope objective lens (Edmund Scientific Co., USA), interchangeable with other lenses of smaller magnification ratios. The focal plane of the f/500 lens was on the intersection region of the two laser beams. A 25x35 mm aluminium-coated elliptical mirror was placed behind the f/500 lens and reflected part of the collected light which was subsequently focused on a photomultiplier for the measurement of particle velocity 5. The purpose of the microscope objective lens was to magnify the image of the LDV probe volume and to project it onto a linear photodiode array (Model S , Hamamatsu Corp., Japan) 6. A linear photodiode array was selected as detector in the present implementation of the instrument because it offered the advantage over CCD cameras of high sampling rates, and smaller amount of information for processing, which were required for measurements in the velocity ranges encountered in the flow configurations of this work. The total optical magnification of the collection optics, which was experimentally determined from measurement of the size of the image of a 100 µm pinhole placed at the centre of the LDV probe volume on the detector plane, was not only a function of the focal lengths of the lenses and the magnification factor of the microscope objective lens, but also an indefinitely 4 The collection optics for the combined SDV/two-colour pyrometer instrument were a modified version of those presented in this section and will be described later. 5 In this thesis modifications have been applied to the optical setup for the purpose of simultaneous measurement of particle velocity, size, flux and temperature and these are presented in Chapter 3. 6 As manufactured, the linear array has 35 diode segments. The current SDV instrument only uses 32 segments and, hence, the array is referred to as a 32-diode array. 55

56 2.3 APPEARANCE OF PARTICLE IMAGES ON DETECTOR PLANE increasing function of the distance between the objective lens and the image plane, where the linear photodiode array was placed. The magnification ratio prescribed the dynamic sizing range of the instrument as shown in following section, and upon selection of the collection optics from readily available lenses the magnification ratio could be increased by moving the photodiode array along the axis of the optics away form the objective lens and vice versa. 2.3 Appearance of Particle Images on Detector Plane Before describing the details of the acquisition and processing procedures it is necessary to explain what could be observed by naked eye on the detector plane when a stationary particle was placed in the probe volume. The way shadow images appear is important because it prescribed the design of the data processing hardware and the software validation schemes. For the purposes of explanation, assume that a spherical particle was mounted on a rotatable optical flat with axis of rotation parallel to the beam bisector, which could be traversed along the axis of the probe volume 7, i.e. parallel to the bisector and was located on the volume s centre and the particle was initially kept stationary in the probe volume. The illuminated particle casts a shadow on the detector plane which is equal to the particle cross-sectional area multiplied by the magnification ratio of the receiving optics (see 2.5.1). Figure 2.2 shows the possible images one could observe on the detector plane as a function of axial position along with the temporal signals if particles were moving due to rotation of the flat. The analysis will be initially confined to stationary particles and thus the temporal signals should be ignored. Provided that the particle was located in the centre of the probe volume, its shadow would appear on the detector plane as a dark region, figure 2.2(a), surrounded by a bright area and the latter corresponds to the laser light which was not blocked by the particle. Assume now that the particle is displaced off-centre along the axis of the optics. This displacement distance away from the centre of the probe volume along the optical axis, is called defocus throughout this work. Defocusing results in the appearance of two particle shadow images on the image plane, as shown in figure 2.2(b), result of illumination of the particle by two laser beams. The two shadow images partly overlap and separate gradually with increasing defocus distance until they disappear when the particle is completely outside the probe volume. On the detector plane, figure 2.2(b), the overlapping region appears darker than the remainder and Hardalupas et al. (1994) found that the irradiance level on the detector plane measured over the nonoverlapping region is half that over the bright region, i.e. the region where no shadow was present. 57

57 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY 3DUWLFOH7UDMHFWRU\ 7UDMHFWRU\$ PHDVXULQJ YROXPH 'HIRFXV 7UDMHFWRU\% 6KDGRZDWWKHGHWHFWRUSODQH D GLVWDQFH] E 2XWSXW6LJQDO 6LJQDO/HYHO 7UDMHFWRU\$ D D 9 E 7UDMHFWRU\% E 9 9 'LVWDQFHDORQJGRWWHGOLQH D 7UDMHFWRU\$ E 7UDMHFWRU\% Figure 2.2 Appearance of shadows of spherical particles, moving across the centre of the probe volume, on detector plane as a function of particle trajectory in the LDV probe volume and their respective output signals. The signals are proportional to the irradiance detected from an array diode along the dotted lines (aa ) and (bb ) respectively. If we consider that each beam has a value of maximum irradiance I b, then in absence of particles in the probe volume, the maximum irradiance measured at the detector plane would be 2I b, and 0 in the region of the shadow image of a particle placed on the centre of the probe volume. For defocused particles, the overlapping region on the detector plane is the image of the region of the probe volume which is blocked by both beams whereas the non-overlapping shadow region is the image of the area of the probe volume in which a particle blocks part of one beam but allows the other beam to freely propagate. In such a case therefore, the measured irradiance on the detector plane would be 0 (blocked beam) + I b (unblocked beam), i.e. half the maximum value of 2I b. If we normalise those values with the maximum 2I b, which was the sum of the maximum irradiance of each of the two beams, the normalised irradiances (in the general case where a particle is defocused) correspond to the three regions of dark overlapping ( dark shadow ) 8, non-overlapping shadow ( bright shadow ) and bright regions would assume values of 0, 0.5 and 1 respectively. This theoretical result has been experimentally confirmed by Hardalupas et al. (1994) and has the consequence that since the presence of a shadow is a binary phenomenon 9 the technique can be rendered amplitude-independent, provided that the value of 2I b is experimentally determined as described below. 7 which coincides with the axis of optics. 8 See also footnote 2. The dark shadow corresponds to umbra and the bright to penumbra. The English, descriptive terms have been preserved in this thesis in place of the Latin terms. 9 The word binary refers to the fact that it is generally true that a shadow is either present or absent on the detector plane and, despite the distinction of shadows into bright and dark in the case of defocused particles, it is only necessary, during data processing, to distinguish between the binary possibility of the presence or absence of a shadow. This fact has been exploited in the design of the signal processor used in the open flame and nearburner confined flow measurements of Chapter 4 and the two-phase flow measurements of Chapter 5. 57

58 2.4 EFFECTS OF PARTICLE TRAJECTORY ANGLE It must be mentioned here that at sufficiently large defocus distance, the two shadow images will cease to overlap and therefore no dark shadow will be observed on the detector plane. This case was not considered in the analysis of the previous paragraph because, as shown in 2.5.2, the signal processor used in most experiments was designed to ignore particle trajectories for which no dark shadow occurred. The original assumption in the preceding description that the particle was stationary in the LDV probe volume does not affect the generality of the observations. As particles travel through the probe volume, a snapshot of the shadow image on the detector plane during particle transit would appear as described in the previous paragraph for stationary particles. Figures 2.2(a) and (b) also show idealised representations of the output signal from a photodiode that recorded the transit of the particle through the probe volume for the two defocus cases presented earlier. The idealised voltage output was not a function of particle diameter. For an in-focus particle, 10 the signal had the form of a trapezoidal wave, where the minimum corresponded to passage of the shadow over the detector and the maximum to absence of shadow. When the particle was defocused the signal also assumed an intermediate voltage level which, as explained earlier, was half the maximum and corresponded to the bright shadow of the particle, whilst the minimum corresponded to the dark shadow. The shapes of the signals presented in figure 2.2 are idealised and shallower voltage level changes were recorded by Hardalupas et al. (1994) in their experiments during passage of particle image over the detector, which necessitated use of experimentally-determined voltage threshold levels to distinguish between bright and dark shadows. Hardalupas et al. (1994) selected voltage levels of 40% and 75% of maximum, which means that any signal between 40 and 75% of maximum (i.e. maximum laser irradiance on the detector) corresponded to a bright shadow whilst smaller and larger outputs to dark shadow and no shadow respectively. 2.4 Effects of Particle Trajectory Angle It has been assumed for simplicity of explanation that a particle and, thus, its shadow image travels normal to the array. Generally, almost no region of particle unidirectional motion exists in flows of practical interest, like swirling reacting flows and it is most likely that the condition of normality is rarely satisfied. Figure 2.3(b) illustrates how images appear in the image buffer 11, when particles travel at angle φ relative to the normal to the axis of the array, figure 2.3(a). For convenience only, it is assumed that particles were spherical. The shapes of the circular shadows have been intentionally distorted to demonstrate the effect of trajectory angle, defined on a plane normal to the plane of the beams. Provided that particles are sufficiently (see below) 10 Particle moving across the centre of the probe volume 11 This term refers to the host computer memory where the image of a particle is stored after sampling. 59

59 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY φ O SRV /'93UREHYROXPH O GHI Figure 2.3 Illustration of image appearance in image buffer: (a) Schematic of spherical particle moving at angle f relative to the normal to the detector and (b) image in image buffer after acquisitions of a spherical particle moving at angle f as in (a); l def and l pos denote image separation across and displacement along detector respectively. The spherical particle is moving on a plane (i.e. the plane of the paper) normal to that which contains the beams. defocused, the two images are displaced along and across the detector by l pos and l def respectively. The former is a result of the trajectory angle, whilst the latter is due to the defocus distance z def. Using simple geometry, the defocus distance and the trajectory angle are deduced from the following relations: z def ϕ = tan ldef = 2tan( ϑ) 1 l l pos def (2.1) where J is the laser beam half-angle of intersection and l pos and l def are measured quantities. 2.5 Data Acquisition and Processing This section is divided into two subsections. The first deals with the recording of the particle shadows using the linear photodiode array and hence the custom-built processor is described and the second explains the software validation sequences applied to the raw data. 59

60 2.5 DATA ACQUISITION AND PROCESSING Data acquisition According to 2.3, a particle illuminated by a laser beam casts a shadow on the detector plane which equals the size of the cross-sectional area of the particle multiplied by the magnification factor of the collection optics. Particle size can be therefore deduced from measurement of the area of the two-dimensional projected shadow image and the detector used for this purpose is described below. The photodetector used in this work was a linear photodiode array comprising 32 elements (refer to footnote 6) 1 mm wide and 4 mm high, capable of continuously recording data at sampling rates of the order of 10 MHz. Although the shadow is a two-dimensional image, and a CCD camera seems a natural choice as detector, particular as to date CCD cameras can be gated down to a few milliseconds, which is adequate to freeze the moving image of the particle. However, the repeated sampling rate of common CCD cameras is limited to television frame rates of 25 Hz (and up to 50 or 60 Hz), much lower than the Kilohertz-rates required in the flows investigated in this work. In addition, the amount of data transferred to the host computer per sampling cycle in the case of the linear array is drastically smaller than the case of a CCD camera due to the larger number of elements on the latter, typically 512x512. On the other hand, use of a linear photodiode array implies that the moving shadow image must be sliced-sampled continuously and one-dimensional records must be used to reconstruct the two-dimensional shadow image. The way in which one-dimensional samples were taken and their relationship to the actual two dimensional shadow image is explained in figure 2.4, which shows the passage of the shadow image of a defocused particle over a photodiode array with axis normal to the plane containing the laser beams, as in figure 2.1, and sampled continuously at fixed sampling rate. The projected Ferret diameter 12 of the particle under consideration was smaller than the width of the detector, because if its Ferret diameter were larger then part of the image would have been clipped (vignetted) and the particle size would have been underestimated by an indeterminable factor. According to 2.3, a defocused particle crossing the LDV probe volume cast an image on the detector which consisted of a dark and a bright shadow, appearing as shapes of different grey shades in figure 2.4. During particle passage, the array sampled only a one-dimensional slice of the moving image at each instant, and the proportion of that slice occupied by the shadow varied between samples, because the particle location had changed as a result of its velocity. 13 In figure 2.4 the location of an image relative to the array at three instants t 1, t 2 and t 3 is presented, and it has been assumed that the particle moved with an arbitrary velocity and direction relative to the axis of the array. Provided that sampling rate was The Ferret diameter is defined as the maximum dimension of the particle cross-section normal to the direction of motion. The Projected Ferret diameter is the one measured on the detector plane, i.e. after optical magnification has been taken into account. 13 It mentioned in 2.7 it was necessary that the velocity of a particle be larger than 0.5 m/s, so that its shadow image did not appear stationary on detector plane, as this measurement would have been rejected.

61 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY 2ULJLQDOWZRGLPHQVLRQDO SURMHFWHGLPDJHRIWKHSDUWLFOH 'HIRFXVHGLPDJH W W W W 'LRGHDUUD\ W W W W W W &KDQQHO W W 6DPSOLQJ &KDQQHO &KDQQHO Figure 2.4 Example of sampling and image reconstruction of the shadow image of a defocused irregular particles, showing how the sampled one-dimensional slices are juxtaposed to form the two-dimensional image. sufficiently high -sufficiency will be defined in following text- a time-series of one-dimensional slices was generated. An example of juxtaposed time series is presented in the cartesian grid of figure 2.4, where the horizontal grid corresponds to the length of the detector and each vertical grid line to the time series of the sampled output of each diode segment. Although figure 2.4 describes the principle behind one-dimensional sampling, it does not take into account the ratio of the (finite) width of the detector and the distance normal to the array travelled by the shadow image between consecutive samples. The effect of the latter can be better illustrated in mathematical form as follows. Assume, without loss of generality, a particle travels across the SDV probe volume on the plane containing the beams and that the detector is sampled at a rate F [MHz] and a time t has lapsed between consecutive samples: t = 1 10 F 6 [s] (2.2) If the velocity component of the particle normal to the fringes were U, then the distance travelled by the particle in the SDV probe volume in that direction during time t would be: l = U t [m] (2.3) 61

62 2.5 DATA ACQUISITION AND PROCESSING and therefore, the shadow image of the particle would have travelled a distance L in the image (i.e. detector) plane during time t over the detector distance: L = G l [m] (2.4) where G was the magnification factor of the collection optics and all quoted distances are in meters. Combining equations (2.2)-(2.4) it is concluded that between successive onedimensional slices a particle shadow image has moved over the detector by: 1 L = G U [ µ m] F (2.5) A consequence of equation (2.5) is that the distance travelled by the shadow image between samples was inversely proportional to the sampling rate and, for example, doubling of the sampling rate resulted in halving of the distance travelled by the shadow for given particle velocity. Assume for simplicity that the particles are spherical of diameter d, and that their velocity vector is normal to the axis of the linear array. The size of the shadow image of such particle in the detector plane is: L = G d (2.6) Division of equation (2.6) by (2.5) results in the following formula which gives the number of samples necessary to sample the whole shadow of a moving spherical particle of diameter d travelling with velocity U, when the sampling rate is F: N = L L F d = U (2.7) In figure 2.4, N is the number of horizontal grid lines which contain part of the shadow of the particle. In the case of irregular particles the same principle applies, but the distance travelled by the shadow image across the detector depends on the orientation of the particle, since the particle aspect ratio is not necessarily one as, for example, is the case for the particle of figure

63 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY Equation (2.7) shows that the number of samples taken during particle passage is proportional to the sampling rate. If the sampling rate is small and the numerator of equation (2.7) equals the denominator, then only one sample is taken during particle passage. This way, particle shape in the direction normal to the axis of the array (and, hence, size) would not be resolved from just one one-dimensional slice. On the other hand, if sampling rate is sufficiently large the number of samples will be adequate to reconstruct the shape of a particle in the direction normal to the axis of the array. Between these extremes an optimum exists (see below) and is defined as the number of samples obtained at a sampling rate such that the distance travelled by the shadow image over the linear array (i.e. in the detector plane) between successive samples is equal to the diode pitch, 1 mm for the photodiode array used in this work. The optimum is thus defined as the sampling rate at which the resolution of the sampled shadow imaged in the direction normal to the array is equal to that along the axis of the array. 14 On the assumption that a spherical particle of diameter d and velocity U is sampled, the number of samples taken under the optimum condition is: 3 10 d N = U (2.8) where d is in [µm] and U in [m/s]. Substitution of the value of N from equation (2.8) into equation (2.7) yields the appropriate sampling rate. Use of the sampling rate estimated according to this condition implies that enough samples were taken to ensure that the particle dimension which corresponded to the direction normal to the array (along a vertical grid line in figure 2.4) was resolved with precision no worse to that of the dimension parallel to the array axis. Accordingly the cells in figure 2.4 have been depicted as squares and if all one-dimensional slices were placed next to each other in the manner of figure 2.4, the sampled circular shadow image would appear circular overall (with castellated perimeter). It must be noted at this point that the assumption of spherical particles does not impose restrictions on the analysis and was made only for simplicity. In general, particles travel with a distribution of velocities whilst the sampling rate of the electronics is fixed 15, and the number of samples taken is different from the calculated optimum. Figure 2.5 presents examples of the reconstructed image of a spherical particle from samples made at rates higher and lower than optimum, termed oversampled and undersampled respectively. In drawing the reconstructed images it has been assumed that the particle velocity was such that the raw shadow image moved across the detector and normal to its axis by 1 mm (the diode pitch) between successive samples. The reconstructed shadow image of an oversampled spherical particle appears longer than the raw whilst an undersampled image appears shorter, 14 According to this statement, if the sampling rate is higher than the optimum the image resolution in the direction normal to the axis of the array will be higher than that along the axis of the array; this, though, would not result in higher precision of sampling of the entire image (which is limited by the worst resolution, i.e. the resolution along the axis of the detector) and, thus, would not be an optimum sampling rate. 15 Which is indeed the case with the signal processor. 63

64 2.5 DATA ACQUISITION AND PROCESSING RQHÃVDPSOH VOLFH H G R L Ã G H Q R 25,*,1$/63+(5,&$/ 3$57,&/( 29(56$03/(' 81'(56$03/(' Figure 2.5 Examples of the reconstructed recorded shadow image of a spherical particle in the image buffer showing the effects of over- and undersampling. because the raw image translation over the array was smaller and larger, respectively, than assumed during reconstruction of these shadows. The calculated aspect ratio of the reconstructed image would generally be different from that of the raw shadow image on the detector plane (before corrections in software are applied, see 2.5.4) as a consequence of a combination of particle velocity and sampling rate different to the corresponding optimum from equation (2.8) assumed in drawing the images. The reader should refer to below, where equation (2.9) presents mathematically how oversampling and undersampling occur (and how they are corrected in software). Hence, the question arises is as to the correspondence between the reconstructed image (formed if all of one-dimensional slices were placed next to each other in the manner of figures 2.4 and 2.5) and a snapshot of the raw moving shadow image (e.g. photographed by a camera on the detector plane). Image distortion due to under- or oversampling is alleviated by simultaneous measurement of the velocity of a particle which can be introduced in the system of equations (2.2)-(2.7) and be applied to correct the aforementioned under- or oversampled image. The details of the correction sequence applied during reconstruction of the actual images from the sampled data are described in

65 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY $QDORJXHLQSXW IURPDGLRGH VHJPHQW Window comparator and transient recorder (for one channel) %ULJKW $PSOLILHU &RPSDUDWRU Dual - '$&RQYHUWHU 'DUN Port &RPSDUDWRU Static RAM Digital signal D/A level configuration +RVW &RPSXWHU ISA-Bus Buffer * Controller (FPGA) 32 [ch] Master Controller * no relation with the so-called "Image Buffer" Figure 2.6 Block diagram of the SDV signal processor showing the components for processing the analogue output signal from one diode of the array. The electronic component defined as FPGA is the programmable integrated circuit which controls the hardware Signal processor The requirement from the signal processor of the SDV signals was to sample the output signal of each diode segment and transfer the data to the host computer for processing which would include recognition of areas corresponding to dark and bright shadows, and to no shadows at all. The processor used by Hardalupas et al. (1994) and Morikita et al. (1995) was based on A/D conversion of the output of each diode and software processing of the converted signals. This method permitted simple electronic circuitry based on 8 bit A/D converters but resulted in large amounts of data being transferred to the computer with consequent low particle image acquisition rates which were inadequate for particle flux and volume fraction measurements. The design was later improved and all measurements in this thesis, except those at the downstream region of the confined flow in Chapter 4, were measured with the signal processor described below. In order to make flux and volume fraction measurements feasible, the speed of the processor was increased from the configuration used by Morikita et al. (1995) and the amount of data transferred to the host computer was reduced by adopting a different method of processing. A new data recorder was developed, consisting of 2-bit analogue window comparators and digital 65

66 2.5 DATA ACQUISITION AND PROCESSING transient recorder, shown in figure 2.6. The window comparators were aligned to distinguish three different irradiance levels, two of which corresponded to the dark and bright shadows, and the third to the absence of shadow on the detector. The threshold levels of the comparators were set empirically at 75 and 40% of the maximum amplitude (Hardalupas et al. 1994): the maximum amplitude corresponded to irradiance levels on the detector in the absence of particles in the probe volume, and the two threshold levels were used to discriminate between dark shadows (irradiance levels smaller than 40% of maximum 16 ) and bright shadows (between 40 and 75%). 17 The output signal of each comparator (i.e. the comparators for the bright and the dark shadows) were passed through a latch, the purpose of which was to combine together the digital outputs of the comparators and synchronise them with the clock of the SDV processor. 18 This arrangement allowed faster, hardware processing of the images, performed by the comparators rather than software processing of the A/D converted signals, and also the amount of the digital data transferred to the host computer was reduced by 75% compared to the 8 bit A/D converted signals. 19 In addition, the data transfer rate from the processor to the host computer was increased by a factor of 10 compared to all previous work 20 and, hence, the maximum measurable particle number density was further increased. The processor was designed to be triggered only by dark shadows -thus only by particles at small defocus distances- as this kept the sizing uncertainty to less than 10% (Hardalupas et al. 1994). The processor had two independent sets of memory (dual-port static Random Access Memory), with 4096 samples/channel for 32 channels for storage of the signals from the comparators. The reason for two sets of memory is that when the SDV stores the processed The maximum was measured in the absence of particles and this referencing procedure allowed estimation of the signal output corresponding to irradiance 2I b mentioned in 2.3. The reference voltage served to normalise signals and thus, allowed distinction between various types of shadows as explained. The referencing procedure was performed at the beginning of each experiment or whenever irradiance on the photodetector decreased as, for example, the case of beam extinction through soiled windows in confined measurements. 17 The digital threshold levels of the comparators were set by the D/A converters. The full analogue scale (output of each segment of the linear array) corresponding to voltage scale between dark shadow and absence of shadow was divided by 256 (8-bit D/A, 2 8 =256) and the digital levels corresponding to 40% and 75% threshold levels were set by the D/A converters for subsequent use by the comparators. 18 The 2-bit digital output of the latch was either 00 (irradiance levels below 40%, i.e. dark shadow), 01 (irradiance levels between 40% and 75%, i.e. bright shadow) and 11 (irradiance levels above 75%, i.e. no shadow). Output 10, which would have meant that irradiance levels were below 40% and above 75% is, of course, not possible. 19 Since images were digitally stored, a record from each diode at any instant that could be either covered by a shadow or not, could be represented by a binary digit, assuming values of 0 (no shadow) and 1 (shadow). The advantage was smaller storage space and higher processing speed. 20 Instead of using an interface card which performs DMA transfers from peripheral devices attached to the computer to the main computer memory, the current SDV processor was connected to the computer directly on the computer ISA bus. This way, the memory in the SDV processor was attached to the main computer memory and was seen by the computer as extended memory. The computer was then able to perform transfers between the peripheral (SDV) memory and its main memory at rates about 10 times faster than through DMA. This design allows increasingly faster data transfers between peripherals (such as the SDV processor) and computers as the computer system buses become faster. One, for example, can design an interface card which will utilise the PCI bus of modern personal computers and achieve data transfers more than an order of magnitude faster than those quoted in Table 2.1.

67 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY signal from the linear array in one set of memory, the contents of the other memory set can be transferred to the main computer memory. Upon completion of this storing/transfer procedure of the SDV processor, the two memory sets can be swapped and this procedure can be reinitiated. This design minimises the idle time of the processor and maximises the acquisition rate of the instrument. The operation of the SDV processor was based on pretrigger recording, to allow capturing of the part of the image which preceded the trigger event. The processor sampled continuously (and stored the processed signal in the one set of memory) and if a trigger event occurred, the storing/transfer procedure mentioned above was initiated, upon completion of sampling of a predefined number of samples. Table 2.1: Specifications of the SDV signal processor Input Channels 32 (can be extended up to 128) Sampling Frequency for Shadow [MHz] 10 Sampling Frequency for LDV [MHz] 1 10 Data transfer rate [Mbytes/s] 2.0 Interface IBM-PC/AT (ISA) The detailed specifications of the processor are listed in Table 2.1. The maximum data rate of the present hardware was about 300 particles per second and was limited by the data transfer (the capability of the ISA bus) to the host computer rather than by the sampling rate. The sampling frequencies for the LDV processor and for the shadow are limited by the design of the current circuitry Simultaneous size and velocity measurement The requirement to measure particle velocity simultaneously with size necessitated synchronisation of the velocity and size signals and processing of the Doppler signal by a zerocrossings counter 22. The synchronisation was achieved by monitoring the output of the central diode of the array and the Doppler signal and adjusting the alignment of the photomultiplier 21 Although the processor has built-in A/D converters for sampling Doppler signals, the maximum sampling frequency of 10 MHz of the present electronics was considered inadequate to resolve the Doppler frequencies of velocities encountered in the near-burner region of the swirl burners examined in this thesis,. An arrangement with a zero-crossings counter, described in the following section, was preferred. Nevertheless, in the water channel flow used for the estimation of accuracy of flux measurement of 2.6.4, the flow velocity was about 4 m/s which, for the optical setup of that experiment, could be resolved by FFT of the signal sampled at 10 MHz. 22 In the measurements in the far-burner confined flow of Chapter 4 a FFT processor was utilised instead of a zero-crossings counter, because of unavailability of the counter at that time. Synchronisation was achieved by monitoring the trigger pulse which was the same for both (SDV and Doppler) signals. 23 The number and location of trigger diodes is set by the operator of the instrument. 67

68 . 2.5 DATA ACQUISITION AND PROCESSING Synchronisation 6'96,*1$/352&(6625 *DWHÃSXOVH *$7(2XW 6KDGRZ,1 no shadow "bright" shadow 6KDGRZ "dark" shadow 'RSSOHU =(52&5266,1*6&2817(5,1 ([WHUQDO5HTXHVW ISA BUS Analog signals '233/(5 ) /7,3/,(5 6,*1$/ )520 $55$< +267&20387(5 '$7$$&48,6,7,2162)7:$5( Scattered Light Shadow Light Figure 2.7 Block diagram of the electronics for simultaneous size and velocity measurement using SDV indicating the connections between the SDV signal processor and the zero-crossings counter to the host computer. until the maximum of the Doppler- coincided with the minimum of the shadow signal. Figure 2.7 presents block diagram of the electronic arrangement which ensured simultaneity between size- and velocity data acquisition. The trigger event was provided by the SDV processor which issued a TTL pulse upon detection of a dark shadow by a number of trigger diodes (arbitrarily set to eight or ten in this work) symmetrically located in the centre of the array 23. The TTL pulse was fed into a 20 MHz Wavetek function generator (Model 145, Wavetek Corp.) 24 which produced a variable-width TTL gate pulse that was connected to the external request input of the custom-built phase Doppler zero-crossings counter (so-called Model 1), operating in single-channel mode. Hence, the counter was only processing input Doppler 24 The function generator was later replaced by custom-built electronic circuitry placed inside the SDV processor to reduce the complexity of connections. 69

69 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY signals, high-passed at 1.5 MHz, while the external request input was fed with the gate pulse. The width of the gate pulse was previously adjusted by trial-and-error procedure that maximised the data rate during measurements in the regions of reverse flow velocity of the burner of Chapter 4, so that no velocity measurements were rejected because of an inadequate number of detected zero-crossings during gating Validation sequence of SDV raw data Once data were transferred from the SDV processor to the computer memory, a sequence of validation criteria were applied to the raw signals with the purpose of extracting validated shadow images and calculating their area, from which a characteristic particle diameter was defined, as described later in this section. The steps followed until a valid shadow image was extracted from the one-dimensional sampled slices were the following: Validation of Velocity Signal The validation procedure continued provided that particle velocity was valid (see 2.5.1). On invalid velocity, the current shadow measurement was abandoned and sampling was re-initialised with the next particle. Creation of an Image Buffer for Storing the Sampled Images A number of 1024 photodiode voltage samples were taken, on average, per particle measurement, at sampling rates F [khz] varied between 1250 and 5000 khz in the present experiments. The sampling rates were empirically selected to satisfy the optimum criterion explained in (and mathematically described by equation 2.9). As shown earlier in equation (2.5), the distance travelled by the shadow image across the detector between successive samples depended on particle velocity and was thus different from the diode pitch. This distance was: 1[mm] F[kHz] r = U [m / s] [no units] (2.9) In this equation 1 mm is the diode pitch and U the particle velocity normal to the axis of the photodiode array, measured simultaneously with the sampling of the shadow image. Variable r gives the distance travelled by the shadow measured in units diode pitches and, for example, 25 Phase Doppler counter model 1 of the Thermofluids Section at Imperial College was design to accept only those signals for which a minimum of 9 zero-crossings were detected. Detailed description of the hardwareand software validation procedures is beyond the scope of this thesis and the reader should refer to Hardalupas and Laker (1993). 69

70 2.5 DATA ACQUISITION AND PROCESSING r=0.5 means that between samples the shadow image moved by 1mm / r, i.e. 2 mm across the detector and, therefore, the image was undersampled. In case of r=1 then the sampling rate was optimum according to definitions of The calculated value for r was used to correct the image for under- or oversampling and restore the correct aspect ratio which was distorted due to sampling rate unequal to the optimum. The correction was performed by scaling the distance between two horizontal gridlines of figure 2.4, using the value of 1/r, to be equal to 1 mm (the diode pitch), so that the cells of figure 2.4 are square (which is not the case before correction, figure 2.5). The corrected image was stored in an image buffer for further processing. 26 Because the image has been corrected with parameter r, all lengths are henceforth quoted as multiples of diode pitch. The size of the image buffer was 32x96 27 (diode pitches 28 ), thus a maximum time record from sampling the shadows, which was equivalent to three-times the width of the array was allowed for storing the image. In other words, the longest particle dimension across the detector could not exceed 96 mm / (magnification ratio). The size of the image buffer was selected to maximise the data and processing rate and was sufficient for most particles encountered in practical reacting two-phase flows; only in the case of long fibres, for example, could it have proven inadequate. 29 Separation of Raw Images and Cross-Correlation Once the image buffer, which consisted of the sampled one-dimensional slices (the image buffer can be visualised as the grid of figure 2.4), was created multiple scatterer events were removed, by identifying a contiguous area of bright and dark shadows, surrounding the one-dimensional slice which corresponded to the trigger event. In the general case, single particles cast two bright shadow images on the detector which overlapped in part, forming a dark shadow. 30 The purpose of the separation was to distinguish the two shadows originating from each laser beam and be able subsequently to perform a cross-correlation between them. The procedure for division of the image and separation into Before correction an image in memory would occupy, for example, 32x1024, if 1024 were the number of samples per particle. Obviously a lot of memory is inevitably consumed when a particle is small and moves fast, but the number of samples is so selected as to enable complete sampling of particles of different diameters and velocity. 27 Fixed in software in the current SDV design and by no means a limiting feature of the instrument. This selection resulted from optimization between speed of processing and ability to sample the coal particles of the present work which are unlikely to have aspect ratios larger than 3 (96:32). 28 Literally bits, but each bit had an equivalent length (in world units) of 1 diode pitch (=1 mm). Binary arithmetic was used in the storing and processing of the shadow images that allowed faster data acquisition rates than if integer arithmetic were used. 29 As shown in 2.7, the same applied for the case of particles travelling with velocities smaller than about 0.5 m/s. 30 It is recalled that if defocus was large enough and no overlapping occurred the SDV signal processor would not be triggered. 31 Although it may not be clear to the reader, performing a cut was the computationally fastest method for separating the images.

71 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY &XW DRULJLQDOLPDJHLQ EXIIHU ELPDJHGLYLVLRQ FVHSDUDWLRQRI GDUNDQGEULJKW VKDGRZV + + GEOHQGLQJRIHDFKVHSDUDWHGEULJKW VKDGRZDQGWKHGDUNVKDGRZ HILQDOVKDGRZLPDJHV RULJLQDWLQJIURPHDFKODVHU EHDP Figure 2.8 Schematic of the algorithm for separation of the two overlapping images, originating from each incident laser beam, in the image buffer. The (a) original image in the buffer is (b) divided in two by cutting across the boundary image intersection points; the bright shadow from each beam is (c) distinguished from the dark after division and (d) each bright shadow is blended with a copy of the dark shadow to form (e) the actual shadow image of the particle originating from each laser beam. two shadows is illustrated in figure 2.8. The original image (a) containing dark and bright regions which were digitally stored and therefore their boundaries were sharp and directly distinguishable, was separated in (b) by a cut through the weighted centre of the dark shadow 31, and (c) the area that corresponded to dark shadow was removed, resulting in three parts 71

72 2.5 DATA ACQUISITION AND PROCESSING constituting the original image which were (d) recombined into pairs of bright shadow/ dark shadow images to result in (e) two distinct images originating from each laser beam. After division, the images were bitwise cross-correlated to determine whether they were actually shadows of the same particle illuminated by two laser beams, rather than the result of noise. Additional information obtained by cross-correlation were the displacements l pos and l def in figure 2.3, which were related to the particle trajectory angle and defocus distance, respectively, according to equation (2.1). The resolution of calculation of l pos was equal to the diode pitch of the image of the array in the sampling space (i.e. 1 [mm]/ magnification ratio), whereas for the l def it was z, according to: 1[mm]/ G z = 2tan( ϑ) (2.10) The value of the latter is order-of-magnitude higher than that of the former and would yield to unacceptable random uncertainty in calculation of the trajectory angle. In order to improve the accuracy of the calculation, an interpolating technique, based on the adjusted Gaussian fitting procedure (Kobashi et al. 1990) was used to provide better accuracy for the measurement of the defocus distance and, thus, the trajectory angle. Estimation of the trajectory angle required a minimum defocus equal to z (i.e. the minimum resolvable defocus distance), otherwise with zero defocus equation (2.1) became indefinite. The trajectory angle was estimated with higher accuracy as the defocus distance increased, due to the resolution defined by equation (2.10), but a ceiling was placed on the defocus distance equal to 500 µm, because according to Hardalupas et al. (1994) sizing errors higher than 10% in higher defocus distances were the result of deterioration of image quality, caused by the finite depth-of-focus of the receiving optics. This defocus limitation was applied by software in the post-processing of the measurements and was independent of data acquisition. Estimation of the Area of the Shadow Image Two independent calculations were performed. The first was based on calculation of the number of diodes in the image buffer covered by a shadow. If we denote by S U the area covered by a dark shadow and by S L that by a bright one, then the average area, S P, covered by the shadow from one beam (see figure 2.9, for example) is: S p = 2S L + S 2 U (2.11) 73

73 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY A common parameter defined to characterise the size of a non-spherical particle is the diameter of a spherical particle with the same cross-sectional area (e.g. Morikita et al. 1995; Orfanoudakis and Taylor 1995), particularly when measuring by (diffracted light) amplitude-based techniques (as in the work of Orfanoudakis 1994), which has been followed here. Hence, the diameter of spherical particle with the same cross-sectional area as the area of the measured shadow is d = 4 π p S p (2.12) The second method for the estimation of S p was based on cross-correlation and was used for cross-checking and identification of images from multiple-scattering. The area was determined by measuring the area common to both shadow images, once they were shifted until they overlapped in the image buffer; shifting was performed until the cross-correlation was maximised. Areas calculated by both methods were subsequently compared and the measurement was rejected if the area difference was larger than 50%, an empirically-determined limit (Morikita 1996). The latter criterion allowed images due to multiple scattering to be rejected. 2.6 Accuracy of Particle Flux and Volume Fraction Measurement Introduction The measurement of flux and volume fraction of a dispersed phase represents an outstanding difficulty for optical single particle counters. For both cases of spherical and irregular particles, which are typically measured by phase Doppler anemometry (PDA) and diffracted light amplitude (e.g. Hirleman et al. 1982; Morikita et al. 1994; Orfanoudakis and Taylor 1992 & 1995; Yeoman et al. 1982) respectively, the determination of the local particle flux or concentration is associated with substantially higher uncertainties than is the case for the size and the velocity. This is because flux and concentration are not directly measured, but calculated from the particle velocity and size and a corresponding cross-sectional area, or volume, of the optical probe volume. It is the uncertainty in the latter two quantities, and particularly their dependence on particle size due to the Gaussian irradiance of the incident laser beams, which are the major contributors to the uncertainties (see Taylor 1995, for example, for a review of the definition of flux and concentration of a dispersed phase measured by optical methods). 73

74 2.6 ACCURACY OF PARTICLE FLUX AND VOLUME FRACTION MEASUREMENT In the case of amplitude-based techniques which use diffracted light, the sizing error due to the shape of the particles is typically 20% and can be as high as 70% for 20 µm ellipsoids of aspect ratio of 2.0 (Hardalupas et al. 1995). There is little recent published work on flux measurements by such a technique with the exception of Orfanoudakis (1994), who used the work by Hardalupas and Taylor (1989), as developed for phase Doppler anemometry (PDA). The accuracy of flux measurement with methods as that of Orfanoudakis is expected to be similar to that for PDA, discussed below. In PDA, the size (i.e. area) of the optical probe volume is preferably determined in-situ from each measurement (e.g. Saffman 1987; Hardalupas and Taylor 1989; Schöne et al., 1994). The extensive experience in the use of PDA has proven that the uncertainties are usually of the order of several tens of percent (e.g. Hardalupas et al. 1994b; Maeda et al. 1996), although these may be as low as 5% under ideal conditions (Qiu & Sommerfeld 1992; Sommerfeld & Qiu 1995). Theoretical investigations show, in addition, that the so-called trajectory ambiguity effect (Gréhan et al. 1992), also due to the Gaussian irradiance, and the so-called slit effect (Xu and Tropea 1994) can result in an effective probe volume different from that estimated by the formulae of, for example, Saffman (1987), and this error source is a large potential contributor to the observed discrepancies. In addition, determination of the probe volume dimensions can be erroneous in the case of measurement of a two-dimensional flow using a single-channel PDA, because in the latter case the usual formula applied for the size of the probe volume assumes that the flow is always normal to the fringes (Saffman 1987). Due to the fact that diffracted amplitude-based size measurement of irregular particles is less straightforward and accurate than for spherical ones, the imaging technique developed and described in this thesis is valuable. The sizing accuracy of the shadow Doppler velocimeter (SDV) instrument has been evaluated by comparison with microscope measurements and it was found that the sizing accuracy was better than 10% with a further ±5% inaccuracy caused by so-called defocusing of the particle from the focal plane of the receiving optics by up to ±500 µm (Hardalupas et al. 1994). The instrument was later improved (Morikita 1995) by reducing the response time of the photodiode array and by increasing the number of active segments of the detector, thereby increasing the maximum measurable particle velocity and the particle size dynamic range, respectively. These features of the technique were exploited to measure the location of the particle during its trajectory through the probe volume and to measure the angles of particle trajectories relative to the axis of the photodiode array (Morikita et al. 1995). Additionally, the accuracy of size measurement made by this improved version was assessed in the case where the beams had to pass through optical windows and through the variable refractive index fields produced by flames, situations where amplitude-based instruments can yield unacceptably large sizing errors due to the uncontrollable thermal movement of the windows. Hishida et al. (1995) found that the sizing uncertainties do not exceed -12% in the former case and 15% in the latter. In this thesis (Chapter 5) it is also demonstrated that the instrument can measure the two-dimensional motion of burning coal 75

75 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY particles near the exit of an asymmetrical laboratory-scale confined coal combustor, although no particle flux or concentration is reported from this flow. For particle mass flux and concentration measurement using SDV, the size of the effective probe volume (hereafter termed as sampling-space to indicate that it is explicitly userdelimited, rather than set by a combination the Gaussian irradiance profile and electronic trigger levels, with the resulting sampling-space size dependence estimated from on-line measurement, as in the case of PDA) must be accurately defined. However, the procedure of determining the sampling-space suggested by Morikita et al. (1995), mathematically described below by equations (2.17)-(2.20), is complicated for non-spherical particles in omni-directional flows due to the dependence of the sampling-space size on the particle size, shape, orientation and trajectory, which could cause a systematic bias in the validation rate, especially for trajectories near the edge of the sampling-space. In addition, the design of the signal processor they used implies increasing probability for particles to trigger the instrument with increasing particle size because they used a single diode as a trigger-channel. Although correction for this bias was included in the post-processing of their data, it is not a practically convenient design for the measurement of polydisperse particles. Therefore simplifications in the procedure for the calculation of particle mass flux and concentration are required and should be tested experimentally in order to enable accurate flux and concentration measurements in complex flows. The purpose of 2.6 is to suggest improvements on the procedure for the calculation of the size of the sampling-space of the SDV and, using this procedure, to assess the accuracy of the flux and volume fraction measurement in a unidirectional turbulent channel water flow. Monodisperse spheres with density similar to that of water were used in the experiment, and the measurement accuracy was tested as a function of particle volume fraction. The procedure to determine the size of the sampling-space was simplified relative to Morikita et al. (1995) by providing a user-defined spatial limitation to eliminate the effect of the particle size, shape, orientation and particularly trajectory angle. As explained in 2.5.2, the signal processor was redesigned to achieve high data acquisition rates and, hence, allow measurement of particle flux and concentration in the flow Method of flux and volume fraction measurement The mean mass flux of particulate phase, G [kg m -2 s -1 ], a vector defined in the direction normal to the interference fringes in the LDV probe volume, is given by the amount of particle mass crossing unit area per unit time (Hardalupas et al. 1994b): 75

76 2.6 ACCURACY OF PARTICLE FLUX AND VOLUME FRACTION MEASUREMENT G ρ = T p s i V(i) A (i) n(i) u (i, j) u (i, j j= 1 (2.13) where T s is the total sampling time of the measurement, ρ p is particle density, V(i) is volume of the particle, u (i,j) is velocity of particle sample number j in size class i, n(i) is the total number of particles in size class i and A (i) is the area of sampling-space through which particles flow with unit normal perpendicular to the fringes. The cross-sectional area A (i) is a function of the particle size and the method for its calculation will be described in The particle volume V(i) in the case of irregular particles can be estimated only from the projected area equivalent diameter, d p ; however, spherical particles have been used in the present study to avoid this additional uncertainty in the calculation of particulate volume fraction. There are two different methods to determine the mean particle volume fraction with LDVbased techniques. The first, using the LDV measured particle velocity perpendicular to the plane of the fringes, U, assumes predominantly unidirectional flow normal to the plane of the fringes from which the particle volume fraction C v (i) is obtained as follows C (i) = v 1 T s i A V(i) (i) U (i) (2.14) where U (i) is the ensemble-averaged velocity for size class i, U (i) = n(i) j u n(i) (i, j) 1 (2.15) where u (i,j) is velocity of particle j in size class i and n(i) is the total number of particles in size class i, measured over T s. This is an approximation which can yield indefinitely large errors in recirculating flows near a stagnation point, because the denominator in equation (2.14) there approaches zero. Hardalupas and Taylor (1989), based on the suggestion of Capp (1983), proposed an alternative method to overcome this problem. This method uses the residence time of a particle in the samplingspace: 77

77 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY C (i) = v i n(i) j= T s τ(i, j) V s(i) 1 (2.16) where V s (i) is the volume of the sampling-space for particles in size class i and τ(i,j) is the particle residence time of the j th particle of this size class in the sampling space. The definition in equation (2.16) is, in principle, a better estimation of the volume fraction than equation (2.14), because it represents the duration of particle occupancy in the probe volume, which is equivalent to volume fraction independent of the velocity direction. However, if the flow is predominantly one-directional, the former equation is equivalent to the latter. Since the present measurements were made in a uni-directional flow, the former method was preferred. An extension of the residence time method suitable for SDV measurements is proposed in Size of the sampling-space of SDV To measure the correct size distribution, Morikita et al. (1995) proposed a procedure to correct the biasing caused by the dependence of the size of the sampling-space inside the probe volume on particle size and defocus. The sampling-space is defined as that region of space over which a particle generated two superimposed shadows, the overlapping region of which, over some part of its trajectory, passes over the trigger diodes and which is not vignetted by the finite width of the linear array. These statements are quantified below but Morikita et al. (1995) showed that the sampling-space has, in principle, elliptical shape in the y-z plane (figure 2.9). Their definition was, however, appropriate only for spherical particles travelling normally to the axis of the diode array when only one trigger channel of the diode array was used. The equations proposed by Morikita et al. (1995) for the dependence of the width of the elliptical sampling-space on particle size, assuming spherical particles, are here extended to include the effects of multiple (eight contiguous 32 ) trigger channels, of the width of the samplingspace as a function of the defocus distance z def, the particle diameter d p, the laser beam halfangle of intersection γ, and d p, the width of the image of the triggering area in the LDV probe volume (given by the width of the eight trigger elements of the diode array divided by magnification of the collection optics, d t =1[mm]/G 8= µm). Assume, for the 32 Eight channels were used in the measurements of this chapter. In Chapters 4 and 5, ten channels were empirically selected given the size range of the particles. 77

78 2.6 ACCURACY OF PARTICLE FLUX AND VOLUME FRACTION MEASUREMENT 6KDGRZLPDJH 'HWHFWRUSODQH 'LRGHDUUD\ &ROOHFWLQJ/HQV 3DUWLFOH /DVHU%HDPV [ SUREHYROXPH ] VDPSOLQJVSDFH ' \ ' \ ] GHIOLP ' ] ' ] \ OLP \ SDUWLFOHWUDMHFWRU\ ] F E H D G E ] GHI F \ 6KDGRZLPDJHVLQGHWHFWRUSODQH D E F 'LRGH$UUD\ G H EULJKWVKDGRZ6 + GDUNVKDGRZ6 / Figure 2.9 Details of the SDV sampling-space through which the particle volume flux was measured indicating the coordinate system and the dimensions defined in the text. In addition, examples of shadow images on the detector plane are also shown as a function of particle position in the LDV probe volume. 79

79 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY moment, that the linear array and that D y (figure 2.9) are indefinitely large. Then the width of the cross-sectional area of the sampling-space d w (which depends on particle size and the number of trigger diodes, rather than on the width of the laser beam interference region) is: w p def 2 p 2 def d (d, z ) = d 4z tan γ + d t (2.17) Figure 2.10 shows the increase of d w with particle size, calculated for defocus distance z def of 0 and 300 µm using equation (2.17) for the present optical parameters. Owing to the small beam intersection angle 2γ, d w is almost independent of particle size, for sizes over about 50 µm, and of the defocus distance, z def. We now take into account that the diode array has finite length and thus can image only a finite portion of the optical probe volume. Let W be the width of the image of the detector 33 in the sampling-space of the velocimeter and which depends on the magnification of the collection optics. Morikita et al. (1995) showed that the width of the cross-sectional area of the sampling-space, d w, decreases linearly with particle size to avoid vignetting: d w (d p ) = W d p (2.18) > P@ DQGG :LGWKRIVDPSOLQJVSDFHG Z Z 3DUWLFOHVL]HG S > P@ d w, equation (2.18) d w, equation (2.17), at z def =0 [µm] d w, equation (2.17), at z def =300 [µm] Figure 2.10 Variation of the calculated width of the sampling space as a function of particle size. 79

80 2.6 ACCURACY OF PARTICLE FLUX AND VOLUME FRACTION MEASUREMENT Note that d t does not appear in this equation. The calculated values for d w from equation (2.18) are also plotted, as open circles, in figure The net value of the width of the sampling-space, d w, for any particle size is given by w ( d,z ) = min( d ( d,z ),d ( d ) d p def w p def w p (2.19) An implication of equations (2.17) and (2.18) is that the area A (d p ) of the sampling-space for particle size d p with defocus limitation at z def,lim, A z def,lim (dp) = dw z def,lim ( d,z ) p def dz def (2.20) is a non-monotonic function of particle diameter. The variation of the area of the samplingspace was estimated to be about 20% between 80 and 120 µm particle sizes in the experiment described below, with a triggering width, d t, of 98 µm and a defocus limit, z def,lim, of 300 µm, for reasons given in In general, the validatable sampling-space will be smaller than that predicted by equation (2.20) if there are particles of random shapes and with random orientations (which was not the case in the experiment of 2.6.4) and particularly if the trajectories are randomly oriented relative to the axis of the photodiode array (which is true in any turbulent flow). The latter will be vignetted by the detector, and consequently rejected by software, which is not accounted for in equations (2.18) and (2.20) (Hardalupas et al. 1994). Figure 2.11 shows a pseudoimage, i.e. the output of the linear detector array as a function of time, which illustrates these cases: in figures 2.11 (a) and (b) a spherical particle passed through the sampling-space at trajectory angles relative to the axis of the array of 90, case (a) and other than 90, case (b). The trajectory in figure 2.11(a) will be accepted but that of figure 2.11 (b) will be rejected by software because during the course of trajectory, part of the shadow image was vignetted and not recorded, although the particle, at the time of triggering, lies within the area defined by equation (2.20). Figures 2.11 (c) and (d) illustrate the first of two possible effects of particle shape and orientation for the particular case of identical ellipsoids with an aspect ratio of about two and trajectory normal to the diode array axis, passing through the centre of the probe volume. The choice of ellipsoids is for the sake of example only and does not limit the generality of application of the conclusions drawn here. Inspection shows that the width of the validatable cross-sectional area of the sampling-space in figure 2.11 (c) is larger than in figure 2.11 (d) corresponding to the 30 segments of the diode array.

81 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY 7ULJJHUDUHD^ Figure 2.11 SDV image in image buffer for different cases: spherical particle at (a) 90º and (b) other than 90º; ellipsoid in-focus (c) and (d) and defocused (e) and (f). The width of the frame surrounding each image represents the width of the active diodes of the array. The height represents samples of the array as a function of time. Figure 2.11 (e) demonstrates the second effect, which is to limit the maximum defocus distance up to the point at which the two shadows just overlap. Particles with the orientation of figure 2.11(e) can have larger maximum defocus distances than with the orientation of figure 2.11(f) before they fail to trigger the instrument, and the length of the validatable area of the samplingspace will be therefore bigger. The magnitude of these two effects depend on the details of particle shape and orientation. In order to avoid biasing of the flux and volume fraction measurements due to the effects illustrated in the preceding paragraph, an additional spatial limitation was introduced in postprocessing software to restrict the sampling-space along the y axis, which also simplified the shape of the sampling-space to a rectangle (figure 2.9). Thus, only images, the centre of which lay within y<d w, were accepted in the post-processing software, where d w was calculated from: d w = 2y lim (2.21) and y lim is a limiting distance along the y axis, measured from the centre of the array. This dimension was set at 100 µm, empirically, by finding the area of the sampling-space over which the measured particle arrival rate (number of measurements made over a given time) was uniform. By introducing the spatial limit, y lim, the area of the sampling-space varied by only 15% over the measured size range in the experiment described below. 81

82 2.6 ACCURACY OF PARTICLE FLUX AND VOLUME FRACTION MEASUREMENT Experimental results Accuracy of Estimation of the size of the Sampling-space In order to apply the method described in for the calculation of the sampling-space, it is first necessary to establish the error of measurement of the location of the particle in the sampling-space. The particle position in the y direction of figure 2.9 can be measured from the position of the recorded two dimensional image, by calculating the position of the weightedcentre of the shadow image. The uncertainty was due to the discretisation error, corresponding to the half-pitch of the detector, in determining shadow displacements in the y direction: for the magnification used in this experiment, this corresponded to an uncertainty of ±6 µm in the present configuration, or ±3% of the width of sampling-space, from equation (2.21), used in the present experiments. The position in the z direction (defocus, z def ) was deduced from the measurement of the separation between the two shadow images described earlier. However, the uncertainty of the measurement of the defocus was expected to be worse than that in the y direction due to the resolution z, as a result of the small intersection angle. 34 In order to quantify the uncertainties, an experiment was carried out using a 100 µm pinhole by Morikita (1996) and his results are presented here. The size of the pinhole was arbitrarily chosen because the error on the defocus measurement is dependent only on the separation between the two shadow images. The pinhole image, which extended over approximately 8 segments of the diode array, was fixed on an optical flat on a rotating disk driven by a servomotor mounted on a three-dimensional traverse to provide a stable and reliable signal source. The pinhole was rotated with a velocity about 1.5 m/s and the rotational drifting was smaller than 2%. This technique was proposed by Hovenac and Hirleman (1991). Figures 2.12 a and b compare the mean and rms of the SDV-measured defocus with the true defocus for four trajectory angles, which were varied by traversing the rotating disk. The measured mean pinhole defocus distance, over 100 measurements, collapsed on a single curve for defocus distances between 100 and 500 µm, with systematic error smaller than 8% except in the region with defocus smaller than 100 µm. This systematic error was probably due to error in measurement of the intersection angle which, for practical reasons, had to be measured from triangulation limited to the transmitting optics side. The large departures from the actual defocus distance in the region of 0 to 100 µm were due to the response time of the detector array, which was not fast enough to detect the sharp signal changes from the shadow passage. The random error was always smaller than 50 µm and typically 20 µm, which corresponded to 7% if the sampling-space dimension in z direction was larger than 600 µm. In any case, a random error does not affect the measurement of G (less than 1% if the sample size was more than 100) and only the systematic error of 8% was considered in the further discussion. As a consequence, the systematic error in the determination of A was 8%. 83

83 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY D E 6'9PHDVXUHGGHIRFXVGLVWDQFH> P@ 6'9PHDVXUHGGHIRFXVUPV> P@ Trajectory 0 [ 0 Angle ] DUWLFOH]SRVLWLRQ> P@ Trajectory Angle 0 [ 0 ] DUWLFOH]SRVLWLRQ> P@ Figure 2.12 (a) mean and (b) rms SDV-measured defocus distance of a pinhole shadow image for four trajectory angles relative to the axis of the array. 34 Large intersection angles are generally prohibited by the diameter of the receiving lenses and the requirement to collect the laser beams which contain the shadow information, as described in the optical arrangement ( 2.2). 83

84 2.6 ACCURACY OF PARTICLE FLUX AND VOLUME FRACTION MEASUREMENT 6 & 3 33XPS &&RQWUROOHURISXPS 6:DWHUOHYHOVHQVRU Figure 2.13 Schematic of the water-flow channel. Flux and Volume Fraction Measurement The accuracy of flux measurement was estimated by comparison with measurements in a flow with known particulate mass flux and volume fraction. In the present study, a water channel flow laden with polyethylene spheres was used, as shown in figure A closed water circuit was arranged by using a vertical rectangular duct 550 mm long with mm 2 square crosssection connected to an electronically controlled pump. The bulk water velocity was 0.16 m/ s and the corresponding Reynolds number Pre-weighed batches of monodisperse polyethylene particles (SB-100S, Asahi Kasei, Japan) of 100 µm nominal, company-quoted diameter were mixed in a known water volume and thoroughly stirred to form a uniform mixture. The uncertainty of the precision balance was smaller than 0.1% of the total particulate mass suspended in the water. The drift velocity between the continuous and the dispersed phase was expected to be negligible, because the density of the particles was 1055 kg/m 3, very close to that of water and it is therefore a reasonable approximation to assume that the velocity of the continuous phase was that of the particles. Particle mass flow rate was evaluated by integration of the local flux at 20 points over the cross sectional area at 450 mm below the inlet of the duct, as illustrated in figure 2.14, for the case of bulk volume fraction, or loading, of 0.005%. The error due to numerical integration, which was estimated from comparison between best and worst results using numerical 85

85 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY 0HDVXUHPHQW3RLQW \ ] Figure 2.14 Location of measurement points in a cross section of the flow channel of figure 2.13 used for the integration of the local flux. integration schemes of first and second order of accuracy, was estimated to be smaller than 12% and was considered to be random. In addition, from the integration procedure it was estimated that the measured particle loading on the centreline was larger than the bulk value by a factor of 1.3 as might be expected from the ratio of centreline to bulk velocity in a turbulent pipe flow. The bulk volume fraction was then derived by dividing the mass flow rate by the bulk velocity, based on the bulk particle velocity established by integration and the assumption that the particle and continuous phase velocity were identical, and the known cross-sectional area of the duct. The bulk particle volume fraction was estimated from a single flux measurement in the centre of the duct at all other loadings, by correction with the empirical factor estimated for the case of 0.005% volume loading. This was a reasonable approximation, since the volume fraction profiles were expected to be similar for the range of volume loadings considered here because particle-particle interaction is negligible. For the determination of the local flux, equation (2.13) was applied. The size of the area of the sampling-space was calculated from equation (2.21) but, following the analysis of 2.6.3, y lim in equation (2.21) was chosen to be 100 µm and that resulted in a rectangular cross-sectional area for the samplingspace which satisfied the requirements for minimum width of the sampling-space from equations (2.17), (2.18) and (2.21). In principle, the measured mass flux should be independent of the size of the sampling-space, because the particle density is presumably uniform within the probe volume. However, because of the finite response time of the detector and the so-called cockscomb appearance of the 85

86 2.6 ACCURACY OF PARTICLE FLUX AND VOLUME FRACTION MEASUREMENT leakage error associated with the use of the adjusted Gaussian fitting functions for interpolation of the correlation-based measurement of defocus distance (Kobashi et al., 1990; Ibrahim et al., 1990), the measured flux varied with defocus limitation. Thus, for a uniform distribution of particles in the z direction, the probability density function of measured defocus distance would 7RWDO0DVV)ORZ5DWH>PJV@ WUXHIOX[ 'HIRFXVOLPLWDWLRQ] GHI > P@ Figure 2.15 Variation of the particulate mass flow rate ( ) as a function of the defocus limitation z def,lim, in comparison with the true flow rate (solid line). not be uniform but could have a quasi-sinusoidal cockscomb variation about the mean, with wavelength related to z. To minimise the error associated with this variation, it is necessary to make the value of z def,lim as large as possible. Figure 2.15 presents the measured particulate mass flow rate (circles), over 1000 measurements for each point, as a function of defocus limitation, compared to the mass flow rate (solid line, calculated from the particle mass and water volume) of 3.21 mg/s for the case of 0.005% volume fraction and mm -3 calculated number concentration. The raw data, stored in the computer, for this figure was the same for each point and the defocus limitation was applied in post-processing software: the calculated particulate flux was within -7% and +25% of the SDV-measured value and the variation is due to the cockscomb leakage effect. It is observed that the flow rate increases linearly with defocus for distances smaller than 200 µm, due to leakage effects and the use of z 100 µm, and then decreases towards the true value. For defocus distances larger than 500 µm, the sizing error becomes larger than the 10% random error in the estimation of the area of the sampling-space, which again results in high uncertainties in the calculation of the flux. As a rule of thumb, the minimum defocus distance, z def,lim, should not be smaller than that corresponding to separation between the two shadows equivalent to 2 diode segments (about 87

87 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY 250 µm in the present configuration) and not larger than 500 µm, because the sizing error becomes comparable to the error in calculating the sampling-space. A value of z def,lim of 300 µm was therefore adopted in this work. The uncertainties of the quantities related to the mass flux measurement are summarised in Table 2.2. The systematic uncertainty of the integrated particle volume was estimated from comparison of the SDV-measured mean diameter and the maker-specified diameter. The random uncertainties in the measurement of a particle diameter, the position along the detector and the defocus distance which were (0.5/8) or 6%, 3% and 7% respectively, do not contribute to the total uncertainty in the flux measurement, because volume flux was estimated from average quantities over a large sample and according to the central limit theorem, the variance of the mean (i.e. the standard error of the mean) decreases with increasing sample size. Table 2.2: Estimated systematic uncertainties of the mass flux measurement at a point. Particle volume V(i) 6% Area of sampling-space A (i) 8% Mass flux G (i) Particle volume fraction Cv 14% Figure 2.16 shows comparison between the SDV-measured (blocked circles) against the actual (solid line) particulate integrated volume fraction on log-log axes. The vertical uncertainty bands denote the expected systematic uncertainty according to Table 2.2. We chose to base the comparison on integrated particle volume fraction, rather than on mass flow rate, because the reference values were a known mass of particles, of known material density, and a known volume of water in the flow circuit, so that the bulk volume fraction is more accurately known than is the mass flow rate of particles in the duct. For dilute flows with volume fraction up to 0.005%, the discrepancy did not exceed 20% and the observed discrepancy in figure 2.16 is, taking into account the systematic errors listed in Table 2.2 together with 12% of the systematic error due to flux profile integration, no worse than is to be expected. However, the measured flux was consistently smaller than the actual value for volume fractions larger than 0.005%. This difference was caused by the turbidity of the flow (e.g. Kliafas et al. 1987) and most of the rejected signals were interpreted by the software as multiple occupancy of the sampling-space; it was even observable by eye that the incident laser beams were partly extinguished by their passage through the flow. The denseflow limitation of the instrument on the basis of the present measurements was estimated in the order of 0.005% in volume fraction: to provide a comparison, this would be equivalent to a mass loading of 5% for water droplets in an air flow in the duct. 87

88 2.6 ACCURACY OF PARTICLE FLUX AND VOLUME FRACTION MEASUREMENT 6'9PHDVXUHGLQWHJUDWHG YROXPHIUDFWLRQ&Y measured corrected by validation Figure 2.16 Comparison between the measured ( ) against the actual (solid line) integrated particulate volume fraction. Open circles correspond to the fraction corrected by the signal error rate of the SDV The larger discrepancy in figure 2.16 for integrated volume fraction higher than 0.005% is likely to be a result of beam extinction due to multiple particle occupancy along the beam path. As a consequence, the discrepancy should correlate with the number of rejected measurements. Indeed, figure 2.17 shows that the signal error rate (rejected signals / validated 6LJQDOHUURUUDWH>@,QWHJUDWHGYROXPHIUDFWLRQ>@ Figure 2.17 Signal error rate as a function of volume fraction. 89

89 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY signals, where signals were rejected either because the particle location was outside the user-set defocus limitation, z def,lim, or because the cross-correlation routine suggested multiple-occupancy of the probe volume). In figure 2.17, it should be noted that, for volume fractions below about 0.01%, the signal error rate of about 20-30% was overwhelmingly due to particles being beyond z def,lim and hence the signal error rate does not, in principle, contribute to an error in mass flux measurement. In contrast, the signal error rate increased rapidly with volume fraction above a volume fraction of 0.01% and this was overwhelmingly due to multiple occupancy. This result suggested that correction of the flux measurement by the signal error rate can provide a reasonable estimate of the flux particularly for fractions above 0.01% v/v; the open circles in figure 2.16 correspond to the corrected values of the measured fraction. As expected, better agreement for the dense limit of the present measurements was achieved from the correction procedure and the maximum discrepancy did not exceed 40% for the worst case, compared with the almost 100% for the uncorrected values Extension to residence-time based volume fraction measurement The features of the method described in earlier sections imply that it is possible to calculate residence time-based volume fraction using equation (2.16). In that case, the particle residence time τ(i,j) through the sampling-space must be determined from the analogue signal, i.e. the signal before passing through the window comparators, which in turn implies that this time will be a function of the threshold levels used for the discrimination of the shadows, figure The uncertainty in the determination of the residence time arises from the shape of the 3'$ $UELWUDU\VHWWLQJRI WKUHVKROGOHYHOV 8QFHUWDLQW\LQUHVLGHQFHWLPH 6'9 7KUHVKROG/HYHO 5HVLGHQFHWLPH Figure 2.18 Schematic of residence time calculation as a function of threshold levels, for PDA and SDV respectively. 89

90 2.7 TRAJECTORY BIAS EFFECTS analogue signal which rises and falls more sharply than the raw signals from LDV. Thus, according to the results of Hardalupas et al. (1994), the error in the estimation of a transit time will be exactly that involved in the determination of diameter and hence will not exceed 10% for irregular, powder-like particles. However, the method of must now be extended to calculate the volume V s (i) of the sampling-space as a function of the trajectory and the particle shape and size. The volume of the effective sampling-space of the SDV can be determined from the area of the sampling-space, following the analysis of 2.6.3, together with the relevant dimension in the x direction of figure 2.9. This dimension will be the size of a sampled particle, because the instrument is triggered as long as there is particle passing through the sampling-space. The volume thus is given by V (i) = D s y D d z p (i) (2.22) where D y =2y lim and D z =2 z def,lim are the sizes of the sampling-space in the y and z direction, respectively, as shown in figure 2.9 and d p (i) is the nominal diameter of particle size class i. As for single-component LDV and PDA systems, this analysis holds if the x velocity component is larger that the y and z components, or, in other words, if the flow is almost normal to the axis of the photodiode array. Again as in LDV and PDA systems, if the latter condition is not satisfied, large uncertainties in the measurement of particle volume fraction are likely to arise. When the particle is non-spherical, asphericity of the particle could cause measurement error on flux, similar to the effect illustrated in figures 2.11(c) and (d). To avoid this effect, use of the area-equivalent diameter, d p, in equation (2.12) will suffice, assuming that the particle passes the probe volume with random orientations. Although the volume fraction defined by equation (2.16) is equivalent to the result from equation (2.14) in the present condition, since the particle used in the experiment is spherical and the flow was uni-directional, the technique can be used to obtain residence-time based volume fraction measurement, as well as the velocitybased value. 2.7 Trajectory Bias Effects Since the sizing technique is based on a linear photodiode array, size-validated measurements can be obtained only when particle velocity exceeds about 0.5 m/s, a limit which has been experimentally determined (see figure 2.19). This limit is due to the fact that the shadow Which was aligned to coincide with the streamwise, or axial flow velocity except for the measurements in the downstream region of the confinement of Chapter 3, where the array was not aligned with any cartesian velocity components.

91 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY Trajectory angle, φ LDV probe volume particle velocity vector Axis of array Linear array Image of particle Direction of axial velocity, U (a) Not measurable &RQHÃRIÃDFFHSWDQFHÃIRUÃWKH WUDMHFWRU\ÃDQJOHÃφ 70 o 70 o pdf Not measurable Not measurable U (m/s) (b) (c) Figure 2.19 Velocity bias effects due to SDV; (a) shows a schematic of the image of the LDV probe volume on the diode array and a particle trajectory through the probe volume at angle f relative to the normal to the array, (b) the range of experimentally-determined measurable trajectory angles f and (c) the range of non-measurable velocities normal to the array which are missing from the velocity pdf. image of a particle must travel over the array, so that a finite number of one-dimensional slices is obtained before the particle projected image can be reconstructed. If the shadow image of a particle were travelling with velocity which was practically zero, as could have happened in the region of a recirculation zone, then the one-dimensional slices obtained during sampling would 91

92 2.7 TRAJECTORY BIAS EFFECTS correspond to the same position of the particle and no particle image could be reconstructed. In addition, particles which have mean velocity normal to the array 35 are likely to have finite mean velocity along the detector or follow the high fluctuations in the region of a stagnation point and, therefore, are likely to travel parallel to the detector end cover, at some point, the extreme diodes of the array. The latter is a cause for rejection as shown in 2.8. For a similar reason, a measurement will be validated, provided that the trajectory of a particle is not parallel, or near-parallel, to the axis of the photodiode array, which in practice means that only particles with trajectories at angles φ smaller than about 70 will be measured, figure 2.19b. A consequence of these two effects is that there is a combination of particle trajectories and velocities which will not contribute to the particle velocity statistics and it is therefore necessary to investigate theoretically the extent of velocity biasing caused by these two effects. A parametric investigation, based on a Monte Carlo simulation was selected for this analysis, because it allowed realistic representation of a turbulent fluid flow field and the angle- and velocity restrictions mentioned earlier. The underlining idea behind the simulation is the following: a flow field is assumed with mean axial and radial velocities U and V whilst 2 2 turbulent fluctuations in these directions are characterised by rms values u and v respectively, which are Gaussian-distributed around the mean velocities and also have a mean uv correlation. The simulation is performed by sampling randomly the Gaussian distributions for the axial and the radial velocity fluctuations and estimating the turbulent velocities from: u = (u 2 ) 1 2 uv v = 2 (u ) 1 2 P (u v uv ) P (u ) P 2 (2.23) the derivation of which is explained in Appendix I. In this equation, P 1 and P 2 are random variables, sampled from a Gaussian probability density function. Following sampling, the instantaneous velocities are: U = U + u V = V + v (2.24) and the constraints for the axial velocity and the trajectory angle mentioned above are imposed on U, V pairs for which both following conditions hold: U > 0.5 (2.25a) 93

93 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY V tan U < tan(70 o ) "cone of acceptance" (2.25b) were included in the calculation of the SDV-measurable velocity statistics. A total of samples were used which resulted in statistical uncertainty of less than 0.5% in the rms velocities. It should be noted here that for simulated conditions presented below, which correspond to those of the experiments, typically 10-20% of the samples were rejected because they did not satisfy the velocity and trajectory angle constraints, but the statistical uncertainty did not exceed 1% in most simulations. Figure 2.20 shows indicative results from the simulation to demonstrate the effect of increasing the true rms of the fluctuations of the axial velocity on the measured mean axial and radial velocity, the radial rms and the cross correlation, with the true mean axial velocity as parameter. The results presented in the figure are for mean radial velocity of 1 m/s, radial rms of 2 m/s and cross correlation coefficient of 0.3 which are typical values measured in the flows of Chapter 4. The calculated range of axial rms velocities was from 1 to 8 m/s, but values in the range of 2 to 5 m/s are likely to be measured. The calculation extended to values higher than those of practical interest in order to demonstrate the asymptotic behaviour of the quantities sampled by the Monte-Carlo method with increasing axial rms velocity. The selection of the axial mean velocities over which the calculation was performed was arbitrary, but nevertheless sufficient for the investigation, as it spans the range of mean axial velocities which were measured in the neighbourhood of the recirculation zone of Chapter 4, as well as axial velocities which are encountered away from it. Although results are presented for positive mean axial velocities, extension to negative ones is straightforward and results have shown that in those cases the deviations have the same sign as the ones presented, i.e. quantities which are overestimated when the mean axial velocity is positive are overestimated by the same amount when the mean axial velocity changes sign. The estimated (i.e. simulation of the measured) quantities are plotted in figure 2.20, normalised by the initial values of the Gaussian distributions. A feature of all plots in figure 2.20 is that only under conditions which correspond to a mean axial velocity of 10 m/s, are the computed and the actual values for the mean and the turbulent quantities within about 3% of each other, and deviations from the actual values increase with decreasing mean axial velocity and its rms. This is expected due to the two potential sources of bias in equation (2.25), and especially the effect of the cone of acceptance for the trajectory angles. In principle, this effect will be important when the instantaneous radial velocity is around 2.5 times larger than the instantaneous axial velocity. The probability of occurrence of such unfavourable instantaneous velocity ratios and thus, rejected instantaneous trajectory angles, decreases with increasing mean axial velocity and axial rms velocity. Moreover, the computed mean axial velocity as well as the cross-correlation coefficient are systematically 93

94 2.7 TRAJECTORY BIAS EFFECTS H X U W 9 O F D 9 F! H X Y W U! O F D Y F X! X!! H X Y W U X! O F D Y F X H X U W 8 O F D F 8 X! X! 8 ÃPV 8 ÃPV 8 ÃPV 8 ÃPV 8 ÃPV Figure 2.20 Variation of the normalised calculated mean axial and mean radial velocity, the normalised rms radial and the normalised velocity cross-correlation as a function of the axial rms velocity for a range of mean axial velocities, using Monte-Carlo simulation. Symbols as in legend. overestimated and the mean and rms radial velocities are systematically underestimated. Inspection of the curves for zero mean axial velocity suggests that the deviations from the actual values must be owing to the effect of the trajectory angle, equation (2.25b), since the gap in the velocity pdf due to unmeasurable small axial velocities due to equation (2.25a)--those 95

95 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY 5HVXOWDQWSGI 6DPSOHG*DXVVLDQSGI SGI $[LDO9HORFLW\8PV Figure 2.21 Calculated axial velocity pdf as compared to the sampled Gaussian, for conditions: mean axial velocity U=1 m/s, mean radial V=1 m/s, rms axial velocity u=8 m/s and rms radial v=2 m/s. between -0.5 and 0.5 m/s is symmetrically distributed in the pdf when the mean is zero (figure 2.19c). Thus, since the mean axial velocity is affected, as can be observed in figure 2.20 the effect must be due to equation (2.25b). Figure 2.21 presents example of sampled Gaussian and its resultant pdf, created after imposing the velocity and trajectory angle limitations from equation (2.25a), for a flow with 1 m/s mean axial and radial velocity and 8 m/s and 2 m/s for the respective rms, which are typical measured values in the region of the recirculation zone. The characteristic of the resultant pdf is that a part of the distribution is missing, due to the velocity and angle limitations, but it does not drop to zero, as one would expect for the velocity range from -0.5 to 0.5 m/s, due to the 2.5 m/s wide velocity bins used to construct the distribution. The maximum in the resultant pdf shifted from 1m/s towards larger velocities, because the limitation for the trajectory angle resulted in rejection of axial velocities smaller than 1 m/s, where the ratio V/U and, therefore, the angle became large. In fact, the measured axial velocity pdf in the region of the free stagnation point associated with the mean recirculation zone was never Gaussian in the experiments of Chapter 4 but bimodal. The effect of the aforementioned limitations on the measured axial velocity pdf is likely to be different to those presented in the previous paragraph, which were produced by simulating the velocity distribution as Gaussian. For this reason, statistical analysis of the measured raw data was carried out in order to demonstrate that the actual effect of the velocity 95

96 2.7 TRAJECTORY BIAS EFFECTS Figure 2.22 Comparison between axial velocity pdfs measured in the burner of Chapter 4, for swirl number S=0.41 at z/d=2.67 and r/d=0, processed to include data which were velocity- but not sizevalidated (top) and size- and velocity-validated data (bottom). and angle limitations on the measured particle velocity pdf. This, as shown in figure 2.22, was smaller than predicted by calculations, probably due to deviation of the measured particle axial velocity pdf from the assumed Gaussian in the simulation. Figure 2.22 presents comparison between (a) the axial velocity pdf constructed from all measurements, either size- and velocityor only velocity-validated and (b) the pdf constructed from only size- and velocity-correlated measurements, measured in the burner of Chapter 4, for swirl number S=0.41, at r/d=0 and z/d=0 (see Chapter 4 for definition of the coordinate system). This condition was arbitrarily selected from the experimental investigations of Chapter 4 and all size classes were included in 97

97 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY the processing. A gap in the region of zero instantaneous velocities is observed in (b), the result of the velocity-limitation, but the difference between mean axial velocities corresponding to each pdf is about of 5% providing confidence that the calculated values presented in figure 2.20 constitute worst-cases, and smaller effects of the aforementioned trajectory bias effects are expected in practice. 2.8 Sources of Data Rejection A statistical analysis of two arbitrarily-selected raw data sets was performed in order to estimate the contribution of the sources of rejection of measurements to the total rejection rate the reciprocal of the validation data rate identify mechanisms which contribute to a particular rejection source and suggest potential remedies to decrease rejection and thus increase data rates. The latter is of great importance to complex flows like those investigated here, because high rejection rates increase the duration of experiments and this is detrimental to the materials used in confined reacting flows. The analysis was performed by counting the rejected measurements in the raw data sets as a function of the error code produced by the software validation procedure described in The results from processing are presented in bar charts below, and Table 2.3 provides the nomenclature for the reported error codes. It is recalled that the error codes presented in Table 2.3 are due to application of software validation procedures and the reader should refer to for details on the validation criteria that the data failed to meet. Figure 2.23 presents statistical distribution of sources of data rejection from measurement of a complete radial profile measured in the swirl burner described in Chapter 4 for swirl number 0.41, at (a) z/d=2.67 and (b) z/d=4. It shows that the largest contribution in data rejection was from images which were clipped along the detector, and amounted to up to 20-30% of the total SDV-rejected data. The total number of rejected data decreased across the profile and had minima of 5% and 10% in the cases presented in figure 2.23(a) and (b) respectively. In view of the contribution of the almost constant contribution across a profile of about 5% due to multiple scattering, it is assumed that a large number of rejected measurements due to multiple scattering was actually rejected due to image clipping because of the validation sequence. From the other sources, rejections due to lack of peak in the cross-correlation contributed about 5%, probably a result of optical noise, rejections due to overflow of image buffer could contribute up to 10%, as in figure 2.23(b), whilst rejections due to image clipping along the detector, failure to separate the raw image and incorrect sampling rate had insignificant 97

98 2.8 SOURCES OF DATA REJECTION Table 2.3: Error codes presented in figure 2.23 and their description Error code name Error Description IB OVERFL Overflow of fixed-size Image Buffer where all one-dimensional slices are stored. Caused by a slow-moving large particle. IM CL TOP/BOTT IM CL Ch 1/32 IM DIV FAIL NO PEAK CC (1) Top/Bottom of Image was Clipped. In both cases this can be caused when the number of samples taken was to small and the image was not sampled properly, due to incorrect velocity measurement. Image was Clipped along detector near channel 1 or near channel 32 (corresponding to the extreme segments of the diode array). Both errors are caused when a large particle passes off-centre from the array, or beam wandering through flames causes irradiance on channel 1 or 32 to fall below reference levels, and this is misinterpreted as shadow. 36 It can also be caused if the particle velocity component along detector axis is large compared with the component normal to it. Failed to separate the defocused shadow image into two shadows because no "dark" shadow was captured near the trigger event, caused by either noise or presence of particles smaller than the equivalent of two diode pitches. No peak in the cross-correlation function was found, i.e. the two images after separation were unrelated. Could only be caused by noisy signals. NO PEAK CC (2) The cross-correlation coefficient was 0, i.e. after separation only one shadow was found; caused by noisy signals. WRONG SR MULTSCA PROBLEMATIC Not enough samples due to Wrong Sampling Rate. The source of this error is fundamentally different to "IM CL TOP" or "IM CL BOTT" errors and is caused when the normalised distance, r, travelled by a shadow between successive samples in the direction normal to the detector is less than an arbitrarily selected limit. In general, the sampling rate is selected so that r is around 1 or greater and, thus, occurrence of this error should be rare. Multiple scattering was found. 37 This error is caused when, applying correlation techniques to match the two images after separation, the non-overlapping area 50% is larger than that of the overlapping, implying presence of a parasite image. Problematic Measurement, rejected either because of noise, which resulted in a single diode appearing as covered by shadow, or because an area larger than the 1/3 of the Image Buffer was covered by shadow. The latter is usually caused in reacting flows by large beam wandering which can be misinterpreted as a large shadow by the signal processor. Can be also caused by measurement of a particle with shadow image larger than the width of the array, which was accidentally validated due to its alignment with the flow 99

99 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY (%) r/d= ,% Ã 2 9( 5 )/,0 Ã & /Ã ,0 Ã & /Ã%,0 Ã & /Ã & KÃ K Ã,0 Ã & /Ã&,0 Ã ', 9 Ã) $, /1 2Ã 3 ($. Ã & & 1 2Ã 3 ($. Ã & & &$ : 521*Ã 0 8/7Ã 3 52%/ ( 0$7,& (%) r/d= ,% Ã 2 9( 5 )/,0 Ã & /Ã ,0 Ã & /Ã%,0 Ã & /Ã & KÃ K Ã,0 Ã & /Ã&,0 Ã ', 9 Ã) $, /1 2Ã 3 ($. Ã & & 1 2Ã 3 ($. Ã & & &$ : 521*Ã 0 8/7Ã 3 52%/ ( 0$7,& Figure 2.23 Sources of rejection of measurements by SDV along a radial profile, determined from measurement in the burner of Chapter 4 for S=0.41, at z/d=2.67 (top) and z/d=4 (bottom). The values of the ordinate are explained in Table 2.3. contribution to the total rejection rate. It is evident that the most important source is rejection due to image clipping along the detector, and it is likely to be due to: 99

100 2.9 SUMMARY OF EXPERIMENTAL UNCERTAINTIES Large particles passing off-centre of the detector, thus covering a full half-length of the array and resulting in rejection. Beam wandering that yields voltage output levels from the extreme channels below 75% of the value corresponding to maximum local irradiance (as determined during referencing ), which is misinterpreted by the signal processor as a shadow. Provided that particle flow is not omni-directional as, for example, in the region of stagnation points in recirculating flows, reduction of the magnification ratio would result in smaller shadow images on the detector and therefore reduction of occurrence of image clipping along the detector. In omni-directional flows there is always a high probability that the velocity vector of some particles will be near-parallel to the photodiode array resulting in rejection of the measurement. Reduction of the magnification ratio is realisable in particle suspensions where the size dynamic range does not exceed 10:1, the limit of the current SDV detector; in flows like pulverised-coal flames, where the usual particle distribution contains sizes between a couple of microns up to 100 µm or more, or equivalently a dynamic range in excess of 30:1, a compromise has to be made as to which sizes it is desirable to measure. In the present thesis, the magnification ratio was adjusted to measure down to about 10 µm, because on the grounds of Stokes numbers defined in Chapter 4 it is likely that all particles smaller than 10 µm will follow the flow details, a result confirmed by the work of Orfanoudakis (1994). Thus, there is little to gain by such measurements as far as particle velocity is concerned. In addition, larger particles carry most of flux and, thus, they are of higher importance to burner stability and emissions. Beam wandering is a consequence of refractive index gradients along the axis of propagation of the laser beams; the latter occur and are noticeable with the SDV in reacting flows such as the swirl burners of this work. Because the photodiode array is shorter than the magnified and projected LDV probe volume, the effect of beam wandering results in data rejection due the Gaussian distribution of the illuminating laser beams. A method to reduce the influence of beam wandering without increasing the magnification ratio (because that would affect the measurable size range) is to use a cylindrical transmitting lens, as in the case of the water tunnel flow of this chapter, to increase the width of the LDV probe volume along the detector, and thus, reduce spatial irradiance gradients due to Gaussian distribution that result in data rejection because of beam wandering. 2.9 Summary of Experimental Uncertainties The uncertainties on the measurement of particle size and flux based on the previous work are shown in Table 2.4 and are briefly summarised under the next three subheadings. 101

101 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY Size Uncertainties in Open Isothermal Flows The random uncertainty in the measurement of the size of a single particle was smaller than 6% and the random uncertainty in the mean size became insignificant with increasing sample size ( 2.6.4). Morikita et al. (1995) have confirmed the accuracy of size measurement of irregular particles in isothermal flows who showed that the maximum difference between the measured mean diameter and microscopic measurement was 10%. Size Uncertainties in Reacting or Confined Flows Temperature gradient effects in reacting flows result in beam wandering and Hishida et al. (1995) have measured a maximum 4% size underestimation in Bunsen burner flames. Extrapolating their results they concluded that this error would not exceed -15% with increasing Table 2.4 Summary of Experimental Uncertainties with SDV open flows Effect of beam wandering and of windows Source Systematic Random Size uncertainty in open isothermal flows Size of single particle 6% Mean size of sample 10% Mean Size uncertainty due to Beam wondering (Hishida et al. 1995): in Bunsen burner flame -4% in large flames, extrapolated -15% Size uncertainty of 15 mm-wide optical windows (Hishida et al. 1995): of a 25 µm particle (pinhole) -12% of a particle larger than 25 µm -5% These uncertainties depend on the size of window relative to the particle size under consideration and are due to deterioration of image quality because of the presence of the window. See Hishida et al. (1995) for details. flame size. The influence of windows is to block part of the scattered light by diffraction thus resulting in less sharp shadow images on the detector plane. Hishida et al. (1995) also measured the effect of a 15 mm wide window behind a 25 µm pinhole on the sizing uncertainty and found a maximum systematic underestimation of 12%. This decreased to less than -5% for scatterers larger than 25 µm and wider windows, which is also the case in the measurements presented in Chapter 4. Uncertainty of Estimation of Sampling-Space Area and of Flux Measurement The uncertainty in determining the area of the sampling space included random contributions of 7% from estimation of the defocus distance and of 3% of the width of the sampling space due to the discretisation of measurement of the particle position along the detector. The total 101

102 2.10 SUMMARY OF CHAPTER 2 uncertainty in the measurement of flux was 14% which included 8% systematic uncertainty, whilst random uncertainties did not contribute because their significance decreases with increasing size of samples used in the measurement of average quantities, according to the central limit theorem Summary of Chapter 2 The principle and operation of a shadow Doppler velocimeter instrument, which is an imaging technique, for measuring the size, velocity and the volume (mass) flux of particles of arbitrary shape simultaneously were described in this chapter. The text focused on the description of the associated hardware and the method of processing the raw images in software, the method and accuracy of measuring particle volume flux, and the effect of biasing on the measured particle velocity distribution due to limitations of the present photodetector. The SDV signal processor was combined with a zero-crossings counter which processed particle velocity signals conditionally on shadow images being detected by the photodiode array. The main findings from work presented in this chapter are summarised below. The cross-sectional area of the sampling-space of SDV measurement is geometrically defined unlike, for example, phase Doppler anemometry and amplitude-based techniques, in which the effective cross-sectional area for flux measurement depends on the amplitude of scattered light. This difference renders flux measurement using SDV more robust than the equivalent measurement with phase Doppler anemometry. The accuracy of particle flux and volume fraction was assessed from measurement in a constant-velocity turbulent water-channel flow and it was found that the maximum departure from the actual particle flux was smaller than 20% and typically 10%, for particulate volume fractions in the flow of up to 0.05%. Of this, random uncertainties of 12% were attributable to the integration procedure used in the estimation of the flux, and a systematic 8% and a further 6% to the inaccuracy in the determination of the cross-sectional area of the sampling-space and in the measurement of the particle volume respectively. A Monte-Carlo simulation was performed in order to estimate the influence of velocity- and trajectory angle-limitations of the present photodetector on the measured velocity distributions. Two Gaussian pdfs for the axial and the radial velocity were sampled and only velocities which were within the limitations were included in the calculation of the velocity pdfs. Parametric studies for the mean and the rms axial velocity showed that the additional uncertainty in the estimation of the mean axial velocity was typically 10% whilst for the mean radial, rms radial and rms axial velocity these were about 5%. 103

103 CHAPTER 2 MEASUREMENT OF PARTICLE VELOCITY, SIZE AND FLUX BY SHADOW DOPPLER VELOCIMETRY Comparison between size- and velocity-validated data (included bias effects) and only velocityvalidated data (excluded bias effects) obtained in the swirl burner of chapter 4 for swirl number 0.41, at r/d=0 and z/d=2.67, showed that the uncertainty attributed to the bias effects was about 5%. Statistical analysis of rejected measurements made across radial profiles at z/d=2.67 and z/d=4 in the burner of chapter 4, for swirl number 0.41, indicated that the largest contribution to the total number of rejected measurements from the SDV was from those images which were vignetted by the edges of the detector and amounted up to 20-30% of the total rejected data. It was suggested that the rejections were due to large particles or due to beam wandering caused by refractive index gradients in the reacting flow. 103

104

105 CHAPTER 3 Non-Intrusive Simultaneous Measurement of Velocity, Size, Temperature and Flux of Burning Particles

106 3.1 INTRODUCTION 3.1 Introduction Background Two-colour pyrometers have been extensively used in the past to study the combustion characteristics of single or clouds of burning droplets and particles. In most cases, the experiments were conducted in drop-tube laminar-flow furnaces, in which experimental conditions were easily controlled and the flow was strictly one dimensional. For example, Gurevich and Shteinberg (1958) used a pyrometer to measure the temperature of free-falling monodisperse burning droplets in a laminar furnace (using, however, the incorrect assumption of grey emission). Numerous studies of single coal particle combustion in drop-tube furnaces have been reported using arrangements of two- or three-colour pyrometers. Studies include use of visible pyrometers (e.g. Habib and Vervisch 1987; Timothy et al. 1982), infrared two-colour pyrometers [which also measured particle size simultaneously using coded apertures] (e.g. Tichenor et al. 1984; Fletcher 1989), studies of coal water fuels using three-colour nearinfrared pyrometers (e.g. Atal and Levendis 1993 and 1994) and three-colour near-infrared pyrometers to study the combustion characteristics of plastic particles under high heating rates (e.g. Panagiotou and Levendis 1994). The number of studies using optical pyrometry in turbulent, 2-D or 3-D flows is smaller, they mainly involve measurements of an ensemble, rather than single particles and range from measurements in flat-flame burners (e.g. Bradley et al. 1984; Mackowski et al. 1983, diesel engines (e.g. Beatrice et al. 1995), laboratory-scale kerosene-fuelled swirl burners (Israel et al. 1995), to full-scale utility boilers (e.g. Butler et al. 1992) and fluidised bed furnaces (e.g. Hernberg et al. 1993). For example, the studies by Butler et al. (1992) in full-scale utility boilers report the temperature of a stream of particles along a line of sight and the measurements, therefore, did not represent the temperature of a single particle, but an arbitrary temperature which was an average over many particles with low spatial resolution. One of the phenomena most ensemble measurements ignore is that burning char and volatile flames have different optical characteristics, i.e. emissivity, and measurements obtained by two-colour pyrometry can be erroneous if no proper account is made of the two types of the emission characteristics of the radiant body. For example, because of the non-grey character of soot clouds, measured particle temperatures by two-colour pyrometry can overestimate actual char temperatures by up to 100%, if volatile flames are present (Grosshandler 1984). The main advantage of infrared pyrometry is that one can measure the temperature of char, because the soot is transparent to infrared radiation 1, whilst in the visible spectrum (in the absence of a criterion for discrimination between char and soot emission, such as that presented in this chapter) a pyrometer measures an average between the temperatures of the soot and the particle (Grosshandler 1984; Habib and Vervisch 1987). The main advantages of a visible pyrometer is the fact that readily available optics can be used, alignment is simple and the instrument can 106

107 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES be combined with a sizing anemometer (such as the SDV) for simultaneous velocity, size and temperature measurements. In addition, the sensitivity of Planck s law (and, hence, of the output signal of the pyrometer) to temperature variations in the visible range is higher than in the infrared (Habib and Vervisch 1987). Three colour pyrometers enable cross-checking of the measured temperatures between each two wavelengths 2 and thus enable a more robust measurement validation procedure than twocolour pyrometers, but the additional optical components, including colour filters, required for their operation makes three-colour pyrometers more complex and difficult to align than two-colour ones Present contribution Temperature measurement is required in order to identify individual burning particles and the effect of changes in the operational parameters of the burner, such as the swirl number, on the number -and volume- of burning particles. Simultaneous size, velocity and temperature measurements are required for the investigation of particle combustion as a function of particle size and also determination of the burning mass flux (and thus, rate) of particles. As concluded in Chapter 1, knowledge of the latter is important for identifying the influence of flow parameters on NO x emissions. With the exception of the experiments of Israel et al. (1995) with spherical droplets, simultaneous coal particle size and temperature measurements are limited to those performed at SANDIA national laboratories in USA (Tichenor et al. 1984; Fletcher 1989) where sizing was achieved by use of coded apertures and the residence time of the image of particles over the apertures. This sizing principle is only feasible if the velocity vector is always normal to the side of the aperture used for sizing, and renders the method unusable in turbulent flow such as, for example, a swirl burner, where at any point and particularly near stagnation points particles can travel in different directions. The purpose of the present contribution is to develop an optical instrument to make simultaneous non-intrusive measurements, with high spatial resolution, of individual coal particles velocity, size and temperature, as well as of the volume flux of a particle ensemble, in turbulent flows. In addition, a criterion which was devised by Israel (1997) in order to distinguish between pyrometer signals from char and volatile mantle-flames, is described in detail. The optics, data acquisition electronics and control software of both the SDV (Chapter 2) and the two-colour pyrometer (described later on) were combined. This chapter begins by outlining the principle of two-colour, or ratio, pyrometry ( 3.2), the optical and electronic arrangement 1 Although this is a claim by Habib and Vervisch (1987), experimental evidence, presented below, suggests that the soot cloud surrounding a single particle is optically thin. See Hence three independent temperature measurements are made simultaneously with a three-colour pyrometer. 107

108 3.2 FUNDAMENTALS OF TWO-COLOUR PYROMETRY of the pyrometer operating at visible wavelengths and the combined SDV and pyrometer instrument ( 3.3). It continues with formulation of a criterion for the identification of the origin of temperature signals into those which originated from char and those from volatile mantle-flames ( 3.4). The latter relies on the work of Israel (1997) which is unpublished to date and is thus extensively described here for completeness. 3.2 Fundamentals of Two-colour Pyrometry Two colour, or ratio, pyrometry is a method for determination of the temperature of a radiant body from measurement of the intensity of the emitted thermal radiation at two distinct wavelengths, l 1 and l 2. In the following text of this section, the equations from which the temperature can be determined will be derived and the associated assumptions clearly stated. Assume that a black body 3 of temperature T is emitting thermal radiation at wavelength l. The spectral hemispherical emissive power of the radiant energy 4, i.e. the integrated emissive power over all solid angles is expressed from Planck s spectral distribution of emissive power (Siegel and Howell 1992): 2πC 1 2 eλ b( λ, T) = π i λb( λ, T) = [Js m µ m 5 λ 1 C / T ( e 2 λ 1) 1 ] (3.1) where i λ b( λ, T) is the spectral distribution of radiant intensity, and has the same units as the emissive power and C 1 and C 2 are Planck s first and second constant respectively. Planck s distribution law is usually approximated by Wien s law (Siegel and Howell 1992): e λb ( λ,t) = π i λb 2πC ( λ, T) = 5 C λ e 2 1 / λt (3.2) Wien s law is accurate to within 1% for lt less than 3000 µm K, implying that measurements using Wien s law are accurate to within 1% if the temperature of the source is smaller than about 4500 K for the range of wavelengths used in the current pyrometer and presented in Below we shall need Wien s law for a non-black body: 3 By definition, a body which absorbs all incident radiation and hence also a perfect emitter. 4 The energy leaving the surface of the black body per unit time per unit area and per unit wavelength interval around l. Alternatively called spectral flux of the radiant energy. 108

109 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES e ( λ,t) = π i ( λ,t) = ε λ λ λ 2πC 1 5 C / λt λ e 2 (3.3) where e l is the spectral emissivity of the radiant body. Note that all of the equations presented above implicitly hold for emission into vacuum and they are used in the same form for emission in air or combustion exhaust gases because these gases are non-absorbing in the region of wavelengths considered in this work, and refraction due to temperature gradients can be ignored. The spectral distribution of emissive power in equation (3.3) can be integrated over a small bandwidth Dl i to yield the emissive power of radiation e λ at wavelength l i i where i denotes each of two distinct wavelengths: e e λ λ 1 2 (T) = λ ε 1 2 λ (T) = λ ε 1 λ 2 λ 2πC 1 5 C / λ T 1 λ e 2 2πC 1 5 C / λ 2 e T (3.4) in which it was assumed that the integrand is independent of the wavelength within each bandwidth l i. The ratio of the emissive powers at the two wavelengths of equation (3.4) is: R C2 1 1 e λ (T) ε 5 1 λ λ1 λ λ λ 1 2 = = 1 2 T e eλ (T) λ ε 5 λ λ (3.5) Equation (3.5) expresses the ratio of the emitted thermal fluxes as a function of the temperature of the source and this exponential relationship is expected to characterise the behaviour of the two-colour pyrometer. In practice, the emitted radiation is collected by an optical system, such as that described in 3.3, and is converted into current by photodetectors. Hence, the term in front of the exponential of the functional relationship between the measured ratio of thermal fluxes and the temperature of the emitter is expected to be modified from that of equation (3.5) since it will also account for the gain of the amplifiers in the electronic circuitry of the photodetectors. Let G i be a factor which accounts for the gain of the circuitry of the photodetector converting radiation at wavelength l i into voltage. This voltage, V λ, can be then written with aid of i equation (3.4) as a function of the power of incident thermal radiative energy as: 109

110 FUNDAMENTALS OF TWO-COLOUR PYROMETRY T / C 5 i 1 i i i i 2 i i i e C 2 A G (T) Ae G V λ λ λ λ λ π ε λ = = (3.6) where 2 2 d A = ω (3.7) is the surface area of the radiative particle of size d (assume a spherical particle for simplicity) which contributes to the signal from the photodetector 5 when the light is collected through solid angle ω. Introduction of equation (3.7) into (3.6) and rearrangement yield: T / C 2 5 i i i 1 i 2 i i e d 2 G C V λ λ ε λ λ ω λ π = (3.8) in which the term in brackets is independent of the size or the emissivity of the radiative body and is a function only of the design characteristics of the optical arrangement of the twocolour pyrometer and will be henceforth called the gain of the pyrometer 6 : 5 i i i 1 2 C G X i λ λ ω π = λ (3.9) The pyrometer gain is determined by calibration, as explained in Appendix II. Combining equations (3.8) and (3.9) one can obtain by division of the measured voltages at the two wavelengths (as was done in equation 3.5) that: λ λ λ λ λ λ λ λ ε ε = T C e X X V V (3.10) From which one can solve for the temperature: ln X X ln V V ln 1 1 C T λ λ λ λ λ λ ε ε λ λ = (3.11)

111 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES This is the fundamental equation for the determination of the temperature of a non-black non-grey body using two-colour pyrometry. A common simplification in equation (3.11) is that the emitter is grey, i.e. its spectral emissivity is independent of l and, hence, equation (3.11) reads: T grey 1 = C2 λ1 1 V ln λ 2 V λ1 λ 2 X ln X λ1 λ 2 1 (3.12) Equation (3.12) gives the functional relationship between the temperature of a grey emitter, T grey, and the ratio of the measured voltages of the photodetectors, V λ 1 / V λ 2, as already explained. It can be observed in the last equation that the inverse of the temperature is a linear function of the ratio of voltages and can be therefore generally written as: V ln V λ1 + λ2 = A B 1 T grey (3.13) where constants A and B can be calculated from comparison with (3.12): X A = ln X B = C 2 λ 2 λ1 1 λ 2 1 λ 1 (3.14) In practice constants A and B are directly estimated from calibration of the two-colour pyrometer, as explained in Appendix II, from which the gain of the pyrometer in each channel is also computed. Although knowledge of the gain is not necessary for the measurement of the temperature since equation (3.13) provides a direct measure, the gain of the pyrometer is used for the interpretation of the measured voltages and hence temperature, i.e. whether they 5 Although most equations are presented for hemispherical properties, i.e. those resulting from integration over the solid angle of a hemisphere, only a fraction of that energy can be collected by a lens of finite size, and the differential surface area introduced in equation (3.6) accounts for that fraction. This differential surface area is related to the solid angle (equation 3.7) through which radiant energy is collected by a lens and the latter will be defined in section 3.3 where the optical configuration shall be presented. The validity of equation (3.6) is general and independent of the size of the collection optics and, thus, use of a different optical configuration would result only in different magnitude of the term in front of the exponential in equation (3.6). 6 The reader should not confuse the gain of the pyrometer with the gain of the amplifier of the electronic circuitry, G, introduced earlier, which it incorporates. 111

112 3.3 COMBINED SDV/TWO-COLOUR PYROMETER INSTRUMENT originated from char particles or volatile flames 7, as shown in 3.4. Such interpretation allows exclusion of signals for which grey body assumption does not hold and, therefore, yields minimisation of the measurement uncertainty. 3.3 Combined SDV/Two-colour Pyrometer Instrument Optical arrangement of the two-colour pyrometer In this section the optical arrangement of the two-colour pyrometer is described. 8 Chapters 4 and 5 present results obtained with the combined SDV (Chapter 2) and two-colour pyrometer instrument, but the optics of the two-pyrometer in the absence of those which corresponded to SDV are described first, so that the reader can understand how the sampled signal voltages substituted in equation (3.13) were measured. Description of the combined SDV and twocolour pyrometer instrument is deferred to Figure 3.1 shows an isometric schematic view of the transmission and receiving optics. The transmitting optics were identical to those of the SDV instrument and have been already presented in detail in Chapter 2 (see also COLOUR PLATE 1 in Appendix VI). The receiving optics comprised a pair of f/300 achromats which collected thermally-emitted and laser-scattered light from a particle crossing the probe volume formed from the intersection of the two laser beams (Chapter 2). Use of a pair of lenses ensured that light was collected from only a small area in space because, owing to the depth-of-focus of the lenses, light which originated substantially from defocused locations 9 was attenuated and did not therefore contribute to the temperature signal. 10 It was experimentally found by traversing a pinhole and measuring the output signal of the photomultipliers that the 1/e 2 length of the probe volume 11 for the current design is about 8 mm (Israel 1997). Because, as explained earlier, the pyrometer was used in this work only in combination with SDV, the probe volume was ultimately coincident with that of the latter as shown in For which the assumption of a grey emitter does not hold and, if used, can result in errors in the determination of temperature of the order of hundreds of Kelvin. 8 Although strictly a two-colour pyrometer measures only temperature, the optical instrument called here two-colour pyrometer was actually a combined two-colour pyrometer and laser Doppler velocimeter instrument. The combined two-colour pyrometer and laser Doppler velocimeter was the basis on which the combined SDV/Two-colour pyrometer was developed ( 3.3.2) and, thus, the optical arrangement of the former is explained here for completeness. 9 From a particle, for example, which was passing at a defocus distance (Chapter 2) through the probe volume simultaneously with another passing through its centre. 10 The case of multiple particles emitting comparable amounts of radiation and which could result in erroneous temperature measurement is treated later on where the validation algorithm is explained. 11 Literally it is a viewing area as it is not strictly defined as a probe volume as the case of the intersection of laser beams, or as the sampling space of the SDV which is rigorously defined. The term probe volume is used to preserve the consistency with the rest of the instrumentation used here. 112

113 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES Figure 3.1 Isometric schematic view of the transmission and receiving optics of the two-colour pyrometer. Light collected from the pair of achromats was imaged onto a 100 µm pinhole which acted as a spatial filter permitting light collected only, theoretically 12, from within a 100 µm diameter region in the probe volume. Although it is common in practice to use water-cooled fibre probes to collect thermally-emitted radiation along a line-of-sight, particularly in the case of measurement in furnaces (e.g. Hernberg et al. 1993; Levendis et al. 1992), pairs of lenses have been routinely used to increase spatial resolution (e.g. Bradley et al. 1984) and even incorporated slits to define the probe volume (Habib and Vervisch 1987). Another optical setup reported in the literature which achieved high spatial resolution and small aberrations, particularly 12 In practice, lens aberrations result in larger probe volumes. This fact does not affect the accuracy of measurement but it is desirable that the probe volume is kept small to allow point measurements of single particles with high spatial accuracy. 113

114 3.3 COMBINED SDV/TWO-COLOUR PYROMETER INSTRUMENT chromatic, was that of Fletcher (1989) who used a pair of spherical mirrors instead of lenses to collect emitted light from burning particles in a drop-tube furnace. The light passing through the pinhole was collected by an f/2 achromat and passed through a dichroic mirror, which transmitted the 488 nm laser line and reflected wavelengths above 510 nm. This dichroic mirror acted as a filter to separate the scattered laser light for the velocity measurement from the thermally emitted radiation for the temperature measurement. 13 Finally, a f/2.5 lens focused the light transmitted through the dichroic mirror, onto a photomultiplier (Hamamatsu R1477) for the velocity measurement. Radiation above 488 nm was reflected by the dichroic mirror and passed through an edge filter which transmitted light above 505 nm and further removed unwanted residual laser light. The collimated transmitted light was launched into an 250 µm-diameter multimode optical fibre (Fujikura GI:200/250) by means of a f/5 achromat. All optics described in this paragraph were enclosed, as shown in figure 3.1, and were protected from ambient light which would have resulted in noise at the output of the photomultipliers. common pyrometer velocimeter Table 3.1 Principal characteristics of the two-colour pyrometer Achromat lens f/3.6 (f/300) Achromat lens f/3.6 (f/300) Pinhole 100 µm Achromat lens f/2 Dichroic Mirror Transmission at 488 nm 70% Reflection above 510 nm 95% Edge Filter Transmission at 488 nm 10-5 % Reflection above 505 nm 80% Plano-convex lens f/200 Photomultiplier (Hamamatsu) R1477 Optical fibre (Fujikura) GI:200/250 Achromat lens f/2.5 Dichroic Mirror Transmission at nm 91% Reflection at 633 nm 98% Interference Filter Transmission at % Bandwidth (nominal) 6 nm Achromat lens f/4.7 Photomultiplier (Hamamatsu) R1477 Interference Filter Transmission at 633 nm 92% Bandwidth (nominal) 6 nm Achromat lens f/4.7 Photomultiplier (Hamamatsu) R The flame, of course, emitted light at around 488 nm, but its irradiance was orders-of-magnitude smaller than the laser light.

115 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES The fibre conveyed radiation above 505 nm into a second enclosure in which the light emerging from the fibre was collimated by means of a f/4.7 achromat and split by a second dichroic mirror. The transmitted and reflected parts passed through interference filters centred at 514 and 633 nm respectively which had 6 nm bandwidth and were focused onto photomultipliers (Hamamatsu R1477). The optical characteristics of the two-colour pyrometer are summarised in Table 3.1. The sampling and processing of the signals are described next Optical arrangement of the combined SDV/pyrometer Figure 3.2 shows the modifications to the receiving optics of the pyrometer (transmitting optics remained identical throughout the work) for the accommodation of the receiving optics of SDV. The reader should also refer to figure 2.1 at this point, which shows the optical arrangement of the SDV instrument alone. The f/100 lens and the photomultiplier in figure 2.1 were replaced by a rectangular aluminium-coated mirror and the front f/300 lens of the pyrometer in figure 3.1 was removed. As shown in figure 3.2, the f/500 lens which collected the laser beams for the SDV acted also as collector of scattered laser light and thermallyemitted radiation. The added aluminium-coated mirror directed the collimated radiation onto the f/300 lens which focused it onto the pinhole of the first enclosure of the pyrometer (see COLOUR PLATES 2 AND 3). The arrangement inside the enclosures remained as described in The photomultiplier which was used in the SDV instrument (figure 2.1) for the measurement of the velocity was redundant as the velocity signal was detected by the photomultiplier in the first enclosure of the pyrometer ( 3.3.1). Placement of all of the photodetectors behind the pinhole of the first enclosure (see COLOUR PLATE 4), with the exception for obvious reasons of the linear photodiode array, allowed observation of the same region in space by all photodetectors 14 resulting in virtually assured simultaneity of the measurement of particle velocity, size and temperature. This would have been otherwise difficult to achieve had the velocity and pyrometer photomultipliers been placed behind independent collection lenses. Another change to the collection optics of the SDV in order to accommodate the pyrometer optics was that the elliptical aluminium-coated mirror behind the f/500 lens (figure 2.1) was replaced by a 25 x 95 mm rectangular mirror in order to maximise the reflection area 15 and thus the output signal of the pyrometer photomultipliers. It should be noted that the obvious remedy was to increase the cathode 14 observation of the same region in space by the pyrometer and by the linear array was achieved by comparison of the output signal of the central channel of the array with that of the photomultiplier for velocity measurement, in a manner similar to that described for SDV in Chapter 2. If the output signals were not synchronous the receiving optics of the pyrometer were traversed until simultaneity was achieved. 15 within the limits permitted by the presence of the collimated laser beams along which the diffracted light, carrying the size information, is propagated. Alternative designs, such as a circular mirror with holes to permit passage of the beams were considered, but the high cost, long manufacture times and the potentially small gain in increasing pyrometer signals found experimentally using mirrors of various sizes along the vertical axis rendered the current mirror size acceptable for the present optical configuration. 115

116 3.3 COMBINED SDV/TWO-COLOUR PYROMETER INSTRUMENT voltage in the photomultipliers to levels above 500 V, but this did not work, because beyond this limit the noise increased with increasing voltage and thus the net gain in signal output was insignificant. Table 3.2 Principal characteristics of the transmitting optics of the combined SDV and two-colour pyrometer instrument Wavelength 488 nm Laser Power Open flows 100 mw/beam Confined flows no pyrometer 300 mw/beam Exit Laser Beam Diameter 1.2 mm Transmitting Lens focal length 600 mm LDV Half Angle of Intersection 2.86 º Dimensions of the LDV Probe Volume (1/e 2 ) Diameter 300 µm Length 3200 µm Frequency Shift 4 MHz The transmitting optics of the combined instrument (also common to the two-colour pyrometer, should it be utilised as a stand-alone instrument) comprised a commercial single Bragg cell unit (DANTEC DISA 55X system), which split in two the laser beam produced by an Ar + laser operating at wavelength 488 nm (blue), chosen rather than the more common Ar + line at nm, because the latter wavelength is being used by the pyrometer for the temperature measurement. One of the two beams was optically shifted by 40 MHz. Because of use of a single Bragg cell, variable frequency shifting was achieved by use of a downmixer (DANTEC DISA 55N10) electronic device which mixed the raw photomultiplier output with a generated signal of frequency 40±f sh MHz, where f sh was the required frequency shifting. A f/600 singlet focused the beams to form a probe volume containing interference fringes for the velocity measurement. The optical characteristics of the transmitting optics are summarised in Table Data acquisition system In Chapter 2 the electronic arrangement with which SDV signal were processed was extensively described. To briefly summarise here, the output of the linear array was processed on-line by the custom-built SDV processor and, upon detection of shadows, a gating pulse was fed to the zero-crossings counter to initiate the simultaneous velocity measurement. The signal processors were arranged in such a way to ensure that measurements of velocity and size of particle were simultaneous. 116

117 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES Figure 3.2 Isometric schematic view of the transmission and receiving optics of the combined SDV and two-colour pyrometer instrument. For the acquisition of the pyrometer signals a two-channel, 8 bit resolution, 20 MHz 8 kbytes/channel buffer memory transient recorder (Datalab DL 922) was used. The outputs of the photomultipliers for the temperature measurement were amplified by a 0.3 V/µA current-to-voltage converter (Hamamatsu DA-C ), low-pass filtered at 1.5 khz 16 by active filters (Krohn-Hite) and digitised at µs (1-2 MHz), depending on the location, so that signals from different particle sizes and velocities would fit in the 8 kbytes record window. Figure 3.3 shows a block diagram of the electronic equipment used for data acquisition. The 16 This setting was not found to affect the maximum amplitude of the signal and, thus, the temperature measurement. 117

118 3.3 COMBINED SDV/TWO-COLOUR PYROMETER INSTRUMENT arrangement is similar to that of Chapter 2 (figure 2.7) with the addition of the transient recorder which digitised the filtered analogue signal from the pyrometer photomultipliers and transferred it to the host computer via an IEEE-488 interface. The transient recorder sampled the analogue input continuously until the SDV signal processor detected the presence of a particle shadow on the linear array, and this processor issued a TTL pulse which provided the external trigger event to the transient recorder. The recorder, which was operating in pretrigger mode, i.e. it was digitising signals before occurrence of the trigger event, stopped sampling upon receipt of the TTL pulse and stored the temperature trace which corresponded to the detected shadow into its memory, until this trace was requested by the host computer. Using this arrangement the SDV signal processor (as in the case of the SDV instrument of Chapter 2) provided the external request signals which triggered the transient recorder and gated the zero-crossings counter. For consistency the software validation schemes also followed a similar sequence. In practice, the software criteria designed to ensure valid temperature signals ( 3.3.4) were applied only after a valid particle velocity was measured. This was because, according to Chapter 2, the image of a particle could not be reconstructed if the particle velocity was unknown. If a measured shadow was invalidated because of noise or multiple occupancy in the probe volume, the pyrometer signals were still processed and if valid, only pairs of valid velocity and temperature were stored. This method not only allowed triplecorrelated measurements (velocity, size and temperature) to be accepted as valid, but also double-correlated velocity and temperature ones. All valid data were stored, but only triplecorrelated results were used in this thesis to examine the size-dependent combustion characteristics of particles Amplitude validation of pyrometer signals As explained in the previous section, upon occurrence of the triggering event the transient recorder contained the digitised pyrometer signals from the two channels. The digital data was then transferred to the memory of the host computer via the IEEE-488 interface for processing. The algorithm employed for processing was identical to that described by Israel et al. (1995) and shall be described below. It was designed to distinguish signals from multiple particle occupancy in the probe volume, ensure that signals were distinguished from noise and above a fixed level set at 25 mv in this work which was about 8 times the noise level. This voltage was a compromise between rejecting noisy signals but still accepting those that originated from small particles. A signal was accepted when all of the following four conditions, illustrated in figure 3.4, were met: (i) A signal from a flame was deemed processable when the signal amplitude at the time of triggering point, denoted A t in figure 3.4, was above the noise level (i.e. the noise bandwidth): 118

119 . CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES DATALAB TRANSIENT RECORDER PHOTOMULTIPLIER WAVELENGTH 1 PHOTOMULTIPLIER WAVELENGTH 2 KROHN-HITE ACTIVE FILTER Synchronisation IN IN External Trigger Via IEEE-488 6'96,*1$/352&(6625 TTL Out *DWHÃSXOVH *$7(2XW 6KDGRZ,1 no shadow "bright" shadow 6KDGRZ "dark" shadow 'RSSOHU =(52&5266,1*6&2817(5,1 ([WHUQDO5HTXHVW ISA BUS Analog signals DOPPLER FROM PHOTOMULTIPLIER SIGNAL FROM ARRAY +267&20387(5 '$7$$&48,6,7,2162)7:$5( Scattered Light Shadow Light Figure 3.3 Arrangement of the electronics for data acquisition using the combined SDV and twocolour pyrometer instrument. A t > 6 σ n (3.15) 119

120 3.3 COMBINED SDV/TWO-COLOUR PYROMETER INSTRUMENT where σ n was the noise rms in the channel, measured in the absence of light at the beginning of the experiment. 17 Value A t was measured relative to a reference, which was defined as being 3σ n below the baseline amplitude A b, which was in turn defined as: A b = Amin + 3 σ n (3.16) where A min was the minimum measured signal amplitude in the absence of light. Level A b was taken to represent the mean amplitude recorded by the transient recorder in the absence of particles, following the suggestion of Israel et al. (1995), because it is in the middle of the 6σ n band within which noise lies with probability 95%. As noted by Israel et al. (1995) this criterion was sufficient to judge whether a measurement originated from a burning particle or noise and in the case of the latter, application of further criteria was redundant and the measurement was aborted and re-started with the next particle. (ii) The entire temperature signal record from an isolated burning particle had to be captured in both channels within the window length of 8192 samples. The latter corresponded, in most cases, to 8192 µs, because the sampling rate was typically 1 MHz, although it had to be adjusted after inspection at certain locations, depending on the particle residence time in the pyrometer probe volume. In conjunction with definitions in condition (i), a signal consisting of a series of samples, A i, was considered to exist and wholly lie within the record window provided that: A A i i > 6σ 6σ n n + A + A min min for for i t 1 i < i i i t 1 t 2 and and i > i t 2 (3.17) where it and 1 it are the samples corresponding to times t 2 1 and t 2 respectively. Such a signal is shown in figure 3.4 which begins from time t 1 and lasts up to time t 2, whilst the rest lies within the noise band. (iii) A condition was employed to identify signals which were the result of multiple occupancy in the probe volume. Figure 3.4b shows such a signal having three peaks, which were recognised as being such (rather than noise) when the amplitude of the signal at a local peak A p relative to the amplitude of the signal at each valley A v,i before and after the local peak was: A p A v,i > 6 σ n (3.18) 17 The mean value of the measured signal in the absence of light was not used as a signal baseline. The latter was defined on the basis of the minimum measured voltage level and the noise levels, as explained in the main text. 120

121 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES A max (a) &KDQQHOÃ A b σ n A p,max A t external trigger digital range 'LJLWDO A min &KDQQHOÃ t 1 t 2 5HFRUGÃWLPHÃZLQGRZ A p &KDQQHOÃ (b) A v,1 A v,2 &KDQQHOÃ QRRIVDPSOHV Figure 3.4 Typical signal records as digitised by the transient recorder corresponding to the two operating wavelengths of the pyrometer for two cases explained in the main text: (a) signals exhibiting a single peak and (b) signals exhibiting multiple peaks. The figure also defines characteristic quantities used in the text for the description of the signal validation criteria. When condition (3.18) was not satisfied the existence of multiple peaks was due to noise and they were ignored. Although signals with three, or more, peaks were rejected due to multiple occupancy, signals with two peaks were considered valid. The latter was not a concession, but rather a provision to avoid erroneous rejection of measurements, during which the collection optics captured the likely variation of temperature and soot concentration along the depth-offocus. (iv) In order to minimise the random uncertainty due to presence of noise in the signals, a threshold in the signal to noise ratio (SNR) was placed. Pyrometer signals with amplitude A p,max (see figure 3.4, where A max is the maximum signal value, and A p,max is the corresponding 121

122 3.3 COMBINED SDV/TWO-COLOUR PYROMETER INSTRUMENT amplitude) were considered of SNR and were accepted only if : SNR = A p,max σ n > 8 (3.19) In this thesis the minimum acceptable value for the SNR was relaxed relative to previous work (Israel et al. 1995), because in practice the current pyrometer had reduced sensitivity owing to the use of the rectangular mirror which only reflected part of the thermally-emitted light collected by the front lens. As explained earlier, the necessity of collecting light using a common lens for both the pyrometer and the SDV did not permit within the time constraints of this work further improvement of the detectability limits (see Appendix II). For this reason, where temperatures are reported in this work are limited to data that originated from char particles: relative to the emission from (sooty) volatile flames, char has high emissivity (close to unity, according to Field et al. 1967), hence higher signal amplitudes and, thus, equation (3.19) was easier to satisfy Uncertainties Because of the way the combined instrument was designed, the uncertainties associated with the temperature measurement were independent of those associated with the size measurement. In the next few lines the uncertainty in the temperature measurement will be analysed (the reader should refer to Chapter 2 for the uncertainty of size measurement). Recall equation (3.11) which gives the temperature of a radiant body in terms of of the ratio of the amplitude measured signals V R = V λ V 1 λ, of the ratio of the gain of each of the channels 2 of the pyrometer ΧR = Χ λ Χ 1 λ and the ratio of the spectral emissivities 18 ε 2 R = ε λ ε 1 λ. 2 Calculation of the propagation of uncertainty from each of the aforementioned gives the total temperature uncertainty: T T = C λ λ T ε R V + Χ + ( λ λ ) ε Χ 2 1 R VR R R R (3.20) where the following properties of combining uncertainties have been taken into account (see, for example, Kirkup 1994): 122

123 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES Table 3.3 The characteristic uncertainties of the temperature of a single particle measured by two-colour pyrometry Systematic uncertainty Random uncertainty Source T T/T Spectral emissivities ratio +80 K (temperature of volatile flame) Signal digitisation ±3% Electronic noise ±5% Calibration +40 K Total +90 K ±6% x f(x) = Aln(x) f = A x f(x, y) = A(x + y) f = A( x + y) f x f(x) = 1/ x = f x and (3.21) where f describes a function of independent x or y, A is a constant, the differential operator (...) denotes the uncertainty of the quantity enclosed in brackets, whilst operator... denotes the absolute value. All uncertainties explained in the rest of this section are tabulated in Table 3.3. Equation (3.20) was evaluated at temperature of 2000 K, which was appropriate for the present experimental studies. A typical pyrometer signal of 50 mv (although for char particles it approached 100 mv) had a noise rms of 3 mv which gave an uncertainty of ±10% in the measurement of R/R. The latter contributed about ±5% in the measurement of temperature according to equation (3.20). Another random source of uncertainty in the estimation of R/R was the digital resolution of the 8-bit transient recorder which, for a full-scale deflection of 500 mv 19 gives a resolution of ±1.9 mv which corresponded to less than ±8% uncertainty in the estimation of R/R and resulted in a ±3% uncertainty in T. The contribution of calibration to the temperature uncertainty through determination of X R was estimated from Israel et al. (1995), who used an identical method and equipment for calibration. They estimated that the uncertainty of X R /X R, which was systematic, was no more than +5% which corresponded to a +2% in the determination of temperature. 18 The latter applies only in the case of volatile flames (and, hence, is related with the accuracy of the discrimination criterion of 3.4.2), because for char particles the spectral emissivity is constant and not included in the calculation of the temperature. 19 commonly used in most experiments in channel 2 (channel 1 was usually set to 200 mv full-scale deflection owing to lower signal amplitudes). 123

124 3.3 COMBINED SDV/TWO-COLOUR PYROMETER INSTRUMENT The last source of uncertainty to be examined here is the term ε R /ε R which includes departures from the thin-limit approximation 20 (see 3.4) used to estimate the temperature of volatile flames. 21 The resultant uncertainty is systematic resulting in a lower temperature measurement for volatile flames. For its estimation, the maximum measured value for ε l of the soot cloud, of which a volatile flame consists, of about 0.01 (Kadota and Hiroyasu 1984; Gurevich and Shteinberg 1958) was used. This value of ε l corresponds to a relative uncertainty of +10% in ε R and, hence, to +4% in T. The random uncertainties quoted in this table, as all random uncertainties, are smeared out with increasing sample size, when ensemble properties are extracted from the measurements. The reason was mentioned in Chapter 2 and is explained again here as a reminder: according to the central limit theorem of probability theory, the uncertainty, σ x, in the mean, x, of an ensemble is 22 (Papoulis 1991; Kirkup 1994): σ x = s N (3.22) where N is the sample size and s the random uncertainty of a single measurement, as described in Table 3.3. If an ensemble mean or rms temperature is quoted, it will then be dominated by the statistical uncertainty rather than instrumentation-related random uncertainties Estimation of burning fraction and flux of particles Section 3.3 has been concerned, up to this point, with description of the combined SDV/ two-colour pyrometer with emphasis on the latter, aiming to explain how simultaneous measurements of particle velocity, size and temperature were made. In addition to these quantities, which are single particle variables, one can measure ensemble particle properties, such as the particle burning fraction and volume flux. The latter has been extensively described in Chapter 2 and in this chapter, therefore, only a brief outline of the additional flux information that can be obtained using the combined instrument will be given here. 20 According to the thin limit approximation ε λ = α λ l, where a l is the soot absorption coefficient and l the geometrical depth of the soot cloud. 21 As mentioned before, temperature measurements are reported in this thesis for char particles rather than volatile flames and, thus, the ratio of emissivities was constant and omitted from the calculation of temperature. The temperature of volatile flames is nevertheless used in the criterion described in 3.4 in order to distinguish between signals originating from char or volatile flames. 22 Also known as standard error of the mean. 124

125 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES The fraction of burning particles, or burning fraction, is defined as the ratio of the number of particles in an ensemble which were detected as burning 23 and the number in the ensemble. The burning fraction is the statistical frequency of occurrence of the event a measured particle was burning. In mathematical terms it is written: N f b = N b t (3.23) where N b and N t were the number of particle counts detected as burning and the total valid counts respectively. A minimum of N b =400 signals from burning particles were accumulated, the number being limited by the practical limit on the duration of the experiments, i.e. below two hours per measurement location. A measured particle could be either burning or nonburning and therefore determination of the burning fraction constituted a binomial experiment. Using the probability density function equation of a binomial distribution (e.g. Papoulis 1991), it was estimated that the maximum deviation between the measured and the real burning fraction due to finite sample size did not exceed ±5% and was typically ±3%. Note that this uncertainty was not included in Table 3.3 because it refers to an ensemble property rather than to uncertainties per single measurement of Table 3.3. Because the fraction of burning particles quantifies only the statistical occurrence of burning particles rather than the arguably more useful from the engineering point of view variable volume flux, the latter was also calculated from the measured data, for two particle categories: those which were burning and the ensemble, which included particles regardless of their burning state. Particle volume flux is a measure of the distribution of the incoming supply of pulverised fuel in the burner. The method of calculation in the case of burning particles is identical to that of the non-burning. The particle burning state was used only to separate particles into two categories (mentioned above) and, hence, the calculation of volume flux of burning particles is bound to the analysis of It is worth emphasising here is that, unlike the flux correction to account for size data-validation-rates less than 100% described in Chapter 2, no attempt was made to correct the flux for temperature data-validation-rates below 100%, because apart from measurements which corresponded to multiple occupancy, also data below the threshold levels of the pyrometer were invalidated ( 3.3.4). In any case, correction of the flux measurement with the validation rate of size measurement has been applied to preserve consistency with Chapter 2 and allow comparison with the flux results obtained using SDV. 23 i.e. the temperature signal was valid according to the conditions set in and the signal amplitude was above a pre-determined threshold (condition iv in 3.3.4). 125

126 3.4 INTERPRETATION OF MEASUREMENTS OBTAINED WITH TWO-COLOUR PYROMETRY 3.4 Interpretation of Measurements Obtained with Two-colour Pyrometry Background This subsection briefly describes findings of previous experimental work, on which the interpretation of the measurements using the two-colour pyrometer is based. The difficulty arises because of the way coal particles burn. Ignition can be either homogeneous (i.e. sooty volatile flames) or heterogeneous (depending on particle size, or joint heterohomogeneous ignition, which is the dominant mode for all particle sizes at sufficiently high data rates (>10 3 K/s; Essenhigh et al. 1989), as those which obtain in practice. Burning char and volatile flames, which occur in the case of hetero- and homogeneous ignition respectively, have different optical characteristics, i.e. emissivity, and measurements obtained by two-colour pyrometry can be erroneous if no proper account is made of the two types of the emission characteristics of the radiant body. In addition, char and volatile flames can radiate simultaneously and one has to identify the origin of the radiation and hence to what temperatures the measurements correspond. For example, because of the non-grey character of soot clouds, measured particle temperatures by two-colour pyrometry can overestimate actual char temperatures by up to 100%, if volatile flames are present (Grosshandler 1984). Previous experimental investigations concluded that char surrounded by a volatile flame did not contribute to the pyrometer signals because char temperature was much lower than that of the flame (Atal and Levendis 1993). In their work Atal and Levendis (1993) inferred that the temperature differences between char and volatile flames were of the order of 1000 K. They also calculated adiabatic flame temperatures of typical hydrocarbons released as products of pyrolysis and compared them with the measured temperatures. The close agreement between the calculated and the measured flame temperatures strengthened their conclusion that they were measuring the temperature of the volatile flames. Similar results were reported by Shaw and Essenhigh (1991) and Timothy et al. (1982, 1986) who concluded that the temperatures measured by their pyrometers represented that of the flame because char did not contribute to the pyrometer signals. Photographic observations supported this conclusion, because char appeared as black spots in the centre of luminous flames (Timothy et al. 1986). Independent temperature measurements of 3 mm devolatilising particles using a thermocouple, on which these were supported, resulted in minimum difference of 700 K between the pyrometric and thermocouple-based temperatures (Shaw and Essenhigh 1991). 126

127 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES Although the experimental findings indicate that the temperature of char is much lower than of the volatile flame and, thus, the pyrometer overwhelmingly detects radiation from the soot cloud of the flame, an assumption is necessary for the determination of the temperature of the soot cloud because, unlike char, soot clouds are not grey emitters and temperatures calculated in this manner are erroneous. Such an assumption was used by Gurevich and Shteinberg (1958) to measure the temperature of free-falling burning droplets by a two-colour pyrometer. On the basis of measurement of flame absorptivity, which was between and 0.06 thereby implying that the flame surrounding the droplets was optically-thin, and using error analysis the authors concluded that the assumption of grey emission could overestimate the flame temperature by as much as 500 K. The finding that the soot cloud surrounding single droplets is opticallythin is supported by other works, in which optical thicknesses of the order of 0.01 was measured (Gollahalli and Brzustowski 1973; Kadota and Hiroyasu 1984) and therefore lie in the region of the optically thin limit. This is the also case for devolatilising single coal particles, since photographic studies of the volatile flames showed that they were transparent (Shaw and Essenhigh 1991) and, hence, the flame was optically-thin. The two experimental findings that char has temperature several hundred degrees lower than the surrounding volatile flame and that soot clouds around isolated particles (or droplets) are optically thin form the basis for the definition of a criterion for discrimination between char and volatile-flame signals, which is developed in the rest of this chapter A criterion for discrimination between char and volatile flame signals One of the conclusions from the literature survey in was that the assumption of grey emitter does not hold for volatile flames surrounding char particles and temperatures of flames calculated using this assumption can be erroneous by several hundred degrees. The following sections describe the criterion devised by Israel (1997) which distinguishes signals which originated from volatile flames from those corresponding to char and calculates the temperature of the former using the optically-thin flame approximation. Before formulating the criterion, the spectral emissivity of a char particle surrounded by volatile flame is calculated. Discrimination of char from volatile flames As equation (III.14) of Appendix III, repeated below for convenience, shows, the compound emissivity of the system of char particle surrounded by volatile flame is a complex function of the emissivities of char and soot, their temperatures, the spectral absorption coefficient of soot and the size of the soot cloud: 127

128 3.4 INTERPRETATION OF MEASUREMENTS OBTAINED WITH TWO-COLOUR PYROMETRY ε c λ = 2 s C2 Ts p min(d ph,2r s) λ e 1 ε + ε λ λτ s p C2 λt d ε p λτs e (III.14) It is unlikely that all of them can be simultaneously known, measured or calculated with sufficient accuracy to permit determination of the temperature (either T s or T p or both) of a coal particle. One has then to resort elsewhere and an alternative solution emerges if one takes into account the results of previous investigations, presented in The idea is that instead of trying to use estimates of the compound emissivity of a coal particle to find T s and T p, one should recognise that coal particles (at least isolated ones, as in the current experiments) burn in one of two predominant modes with radiative emissions derived primary from either char or volatile mode burning. This simplification is useful because one can then apply different and tractable equations for emissivity to estimate the temperature T p or T s, depending on the mode of burning. 24 The purpose of this subsection is to derive equations for the voltage output of a pyrometer channel, as a function of temperature T s or T p, for the two modes of burning. The related but separate problem of identification of the mode, given the voltage output, is deferred to the next subsection. Assume that a char particle is surrounded by a volatile flame and the spectral emissivity is given by equation (III.14). According to equations (3.8) and (3.9) the signal amplitude of the pyrometer at wavelength l i due to radiation emitted by the system of char and flame is: V λ i = X λ i d 2 ε c λ i e C2 λ itp (3.24) where c λi ε is the compound spectral emissivity of a char particle and its surrounding soot cloud. By replacing the compound spectral emissivity from (III.14) and rearranging equation (3.24) becomes: V λ i char contribution 6 soot contribution C2 λ itp 2 s C2 λ it = X d e + X min ε e s (3.25) λ i λ i [ (d,2r )] ph s λ i 24 More on this follows later on. 128

129 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES where d is the diameter of char, min takes the minimum of the arguments in brackets and T p and T s are the temperature of char and soot respectively. It has been assumed in equation (3.25) that the emissivity of char, as mentioned in 3.3.4, is unity. Equation (3.25) gives the signal amplitude of the pyrometer as the sum of two contributions: those from char (first term of (3.25) and from soot (second term). The second term, unlike the first, does not include a fixed diameter, since as is explained in Appendix III it is likely that soot clouds around some particles will be larger than the image of the pinhole of the receiving optics and, hence, will be masked by the latter. Because of the assumption of the optically-thin flame that surrounds the char particle, radiation originating from the char at the centre of the cloud is not attenuated through the cloud. It has been found that char particles are usually at much lower temperature than the volatile flames and, thus, their contribution to the total radiation is small. 25 This is the basis for the discrimination criterion, which is explained later on, after the spectral emissivity of the soot cloud is estimated using the assumption of optically-thin flame and a ceiling is placed to its value. For the calculation of the spectral emissivity of a soot cloud the model is identical to that presented in preceding paragraphs with the exception that the particle at the centre of the soot cloud does not radiate and only contributions from the soot cloud are taken into account. Equation (III.1) in Appendix III: i ( κ, ω) = i (0, ω)e λ s λ κs + κs 0 I ( κ, ω)e λ * s * ( κs κs ) dκ * s (III.1) is solved again with the difference that the boundary condition is everywhere the one corresponding to angles larger than the critical in equation (III.2). Only the part of equation (III.5) which corresponds to angles θ larger than the critical applies here. If one performs the integration in (III.8), where the optical depth of each ray, S, is now: S( θ) = 2R s cosθ (3.26) where again R s is the radius of the soot cloud and θ was defined in the analysis of the system of char and volatile flame. The spectral emissivity of the soot cloud can be then given in the closed form 26 (subscript i is dropped for simplicity from the expressions of emissivity): 25 This is not, of course, the case after the volatile flame has consumed all the volatiles and combustion proceeds through char burning. 26 The assumption of optically-thin flame has not been introduced yet in order to demonstrate that, when contributions from char are ignored, the spectral emissivity assumes a much simpler mathematical formula. 129

130 3.4 INTERPRETATION OF MEASUREMENTS OBTAINED WITH TWO-COLOUR PYROMETRY ε s λ e = 2α λ R s (2α 2α λ 2 λ R R s 2 s + 1) (3.27) where the absorption coefficient is given by (Siegel and Howell 1992): α λ C 0 f = α λ v g(n, k) (3.28) in which C 0 is a constant and equal to 7 according to Grosshandler (1984), α=1 for most hydrocarbons 27 (Siegel and Howell 1992) and g is a function of the complex refractive index of the soot cloud. Since experimental studies of flames surrounding droplets have shown than the optical thickness of the soot cloud, defined in equation (III.3), is of the order of 0.01 (Gollahalli and Brzustowski 1973, Kadota and Hiroyasu 1984), the assumption of opticallythin flame applies. If this assumption is introduced before integration of equation (III.8), equation (3.27) reads: ε 4 3 s λ = α λ ot R s (3.29) where ot refers to optically-thin limit, and is dropped from equations in the rest of this chapter. Equation (3.29) in conjunction with (3.28) shows the functional dependence of emissivity with 1/l. The experimental values of the optical thickness imply a maximum value to the soot volume fraction f v of about 10-4, which in turn implies a maximum for the spectral emissivity of soot. The latter, as shown by equation (3.29), is also dependent on the size of the soot cloud. As already explained, the existence of the pinhole in the receiving optics of the present pyrometer limits the size of the probe volume and, thus, the maximum size of soot cloud from which emitted radiation is collected. To get an estimate of the (maximum) emissivity 28, the experimental value for the optical thickness of k l =0.01 was substituted into (Siegel and Howell 1992, p. 658): ε s λ = 1 e κ λ (3.30) Exponent a also depends on wavelength. Hottel (1954, quoted by Siegel and Howell 1992, p. 658) suggested the value 0.95 for l>0.8 µm, whilst Hottel and Broughton (1932) suggested the value 1.39 for visible wavelengths (see also Siegel and Howell 1992). Because no soot temperatures are reported in this thesis, it is not important which value for l one chooses; the maximum difference between soot temperatures calculated by exponents of 1 and 1.39 is about 7%. The value of 1 has nevertheless been selected because is has been extensively used in the literature (e.g. Atal and Levendis 1993; Flower 1983; Israel et al. 1995). 28 According to Siegel and Howell (1992) the word emittance should be used to describe the amount of radiation a given thickness of an isothermal material emits relative to a black body, whilst emissivity is reserved for describing the same property for a body. In this thesis, however, the work emissivity is used indiscriminately to avoid confusion.

131 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES to give a value of the order of This estimate agrees with the experimental values reported by Gurevich and Shteinberg (1958) for soot flames surrounding kerosene droplets, which were in the range of Israel (1997) concluded from scatter plots of the measured amplitude versus coal particle temperature that the maximum emissivity of volatile flames surrounding char using the present pyrometer was about 0.01, in agreement with experiments with burning droplets. Because of the two upper limits for the size of the soot cloud over which radiation is collected by the pyrometer and the spectral emissivity, an upper bound on the signal amplitude due to the soot cloud exists and is (cf. equation 3.25): s ph 2 s C λt V X d e s λ = λ ελ 2 ( λ = λ1 or λ 2); max max soot (3.31) in which T s is the temperature of the soot cloud. No volatile flame assuming that there are no contributions from char can produce larger signal for a given temperature, because of the spatial limitation imposed by the pinhole in the receiving optics, and the fact that the flame around the char is optically-thin. Equation (3.31), which corresponds to curve A in figure 3.5, constitutes the first part of the discrimination criterion and states that measured signals which are of higher amplitude can be a result of contribution from char, but cannot have originated from volatile flames. Separate condition is necessary to identify signals that originated from char. Equation (3.8), for example, shows that as particle size decreases, so does the signal as a function of the square of particle size. This implies that, at a given char temperature, the signal due to char, which is generally higher than that from a volatile flame due to the higher emissivity of char (by about two orders of magnitude), approaches the maximum signal due to soot (see figure 3.5). The minimum size of char which would give the same signal amplitude as the soot cloud at the same temperature is: d = min d ph ε s λ max (3.32) which can be easily deduced if both parts of the right-hand side of equation (3.25) are equated to each other. Particle size calculated from (3.32) corresponds to the smallest incandescent char particle that can be discriminated from a volatile flame and was about 20 µm. This minimum is actually a limitation of the pyrometer but does not harm the generality of the results presented in Chapters 4 and 5, because 20 µm do not contribute substantially to the total heat release and, as shown in Chapter 4, faithfully follow the gas flow. 131

132 3.4 INTERPRETATION OF MEASUREMENTS OBTAINED WITH TWO-COLOUR PYROMETRY By taking the first part of the right-hand side of equation (3.25) and substituting the particle size from (3.32), the minimum signal amplitude due to radiation that can only have originated from char becomes: p λ min ph 2 s C λt p V = Xλ d ελ e 2 ( λ = λ1 or λ 2); char particle max (3.33) where T p is the temperature of the char particle. Equation (3.33) corresponds to curve B in figure 3.5. The core equations of the criterion for the determination of the origin of temperature signals and, hence, accurate temperature measurement are (3.31) and (3.33) which place a ceiling and a floor on identifiable signals that originated from volatile flames and char respectively. From calculation of the signal from a coal particle (i.e. char surrounded by flame) and its compound emissivity, using an analysis similar to Grosshandler s (1984) that was extended to include the presence of the pinhole in the receiving optics by Israel (1997), it was found that the pinhole is the important design parameter of the present pyrometer, given the emission characteristics of soot clouds, known from the published work. The pinhole imposes an upper limit on the signal amplitude from volatile flames and this allows discrimination between signals from char and volatile flames. A char particle that is surrounded by a volatile flame has char T p B A T s V λ [µv] soot T p, T s [K] Figure 3.5 Typical plot of signal amplitude in channel 2 versus measured particle temperature calculated on the assumption of grey emission. Curves A and B define, for a given temperature, the ceiling and floor values of amplitude of signals that originated from soot cloud and char particles respectively. 132

133 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES temperature several hundred degrees lower than that of the flame and, hence, char not contribute to the received radiation and, consequently, to the output signal. Use of the criterion and determination of temperature The purpose of this subsection is to show how measurements of the output voltages of the pyrometer can be used to identify which of the two modes of burning, namely volatile and char, is taking place. This is not a straightforward matter because, even given that these are indeed only two modes, it is not known a priori which emissivity to use. All equations used for the discrimination of the signals have been mentioned in the preceding sections, and the procedure by which the discrimination is performed will be described here, by means of the flow diagram in figure 3.6. The pyrometer measures the output voltage of each of the two pyrometer photomultipliers, namely V λ and V. 1 λ 2 A temperature, denoted T grey, is calculated using equation (3.12), assuming that the emitter is grey. After this first step, each pair of output voltage of photomultiplier 2 (which corresponds to red wavelength) and grey temperature T grey is compared with curves A and B, defined according to equations (3.31) and (3.33) respectively if these are solved for the temperature, on a x-y plot such as that of figure 3.7 ( V λ is used instead of V 2 λ, because signal voltages were higher at l 1 2 compared with those at l 1 ) 29. Using the graph of figure 3.7, the criterion for discrimination between signals from char or volatile flames can be expressed as follows: V λ 2 A signal with output voltage and corresponding grey temperature T grey originated from char, if the pair (, T grey ) lies below curve B in figure 3.7, whilst it originated from volatile flame if V λ the pair ( V λ 2 2, T grey ) lies above curve A. Pairs ( V λ 2, T grey ) lying between curves A and B are deemed indiscriminable and are therefore rejected. If a signal is identified as char, then the original assumption of grey emission for calculation of the temperature T grey was correct for that particular signal and, thus, T grey is the temperature of the particle T p. If a signal was identified as soot (from volatile flame) then temperature T grey is an incorrect measurement of the temperature of the volatile flame, because of the non-grey emission characteristics of soot, as already explained. In such a case, a temperature T s, which accounts for the optically-thin emission character of soot, is estimated from: 1 C2 λ2 T 1 λ1 λ = ln λ grey C2 λ2 λ1 + Ts (3.34) 29 Curves A and B are the inverse of curves A and B respectively. 133

134 3.4 INTERPRETATION OF MEASUREMENTS OBTAINED WITH TWO-COLOUR PYROMETRY V λ1, V λ2 from photomultipliers HTXDWLRQ T grey 3ORW9 λ2 and T grey on figure 3.7 (V λ2,t grey) below curve B? Yes signal from char T p =T grey No (V λ2,t grey) above curve A? Yes signal from soot T s from equation (3.34) No indiscriminable signal Figure 3.6 Flow chart of the discrimination criterion. which was a result of calculation of T grey from (3.12), T s from (3.11) and combination of the two. The natural logarithm of the ratio of the two wavelengths of the pyrometer accounts for the ratio of spectral emissivities in (3.11), and was calculated from equation (3.35) below and corresponds to the optically-thin limit of emission: 134

135 CHAPTER 3 SIMULTANEOUS MEASUREMENT OF VELOCITY, SIZE, TEMPERATURE AND FLUX OF BURNING PARTICLES 9 λ2 A B T grey Figure 3.7 Replot of curves A and B of figure 3.5 as a function of the voltage output of pyrometer channel 2. ε ε λ1 λ 2 λ = λ 2 1 (3.35) Equation (3.35) was deduced by taking into account equations (3.28) and (3.29) with the further assumption that function g in (3.28) is constant, which indeed the case in the visible spectrum, as the measurements of Dalzell and Sarofim (1969) showed. 3.5 Summary of Chapter 3 An optical instrument, based on a shadow Doppler velocimeter (Chapter 2) and a two-colour pyrometer, was developed for non-intrusive, simultaneous measurements of velocity, size and temperature of individual coal particles, as well as volume flux of an ensemble of particles in complex flows. In addition, a criterion devised by Israel (1997) for interpretation of the pyrometer signals and, thus, accurate measurement of particle temperature was extensively described. There are two main points from this work, summarised below. 135

136 3.5 SUMMARY OF CHAPTER 3 The combined SDV and two-colour pyrometer was designed to ensure that all photodetectors observed the same point in space so that the measurement of velocity, size and temperature was simultaneous. Common collection optics and pinhole ensured that the observed area by the pyrometer had maximum width of the order of 150 µm and, thus, only one particle was measured at a time with high spatial accuracy. The pyrometer was designed to operate in the visible spectrum of radiation so that visible laser light could be simultaneously collected for the sizing part of the instrument, using common readily-available collection optics. A criterion for interpretation of the pyrometer signals, based on the work of Israel (1997), was developed to ensure that only temperatures that corresponded to char particles were estimated from the response curves of the pyrometer using the grey-body assumption. According to the criterion, signals originating from volatile flames were distinguished from those which originated from incandescent char, and the assumption of optically-thin flame was used for the calculation of the temperature of volatile flames. The criterion was based on experimental evidence that char particles have lower temperatures than the envelope flame by hundreds of degrees and the fact that the optics pinhole restricted the size of the probe volume (the area observed by the pyrometer) and, hence, the maximum signal amplitude due to emission from the soot cloud. 136

137 CHAPTER 4 Pulverised Coal Combustion in Open and Confined Piloted Swirl Stabilised Flames

138 4.1 INTRODUCTION 4.1 Introduction Background As concluded in Chapter 1 from review of the literature, energy generation from coal combustion can continue to be a major contributor to the total production of electricity, provided that the associated combustion technologies can comply with the increasingly stringent regulations regarding emissions and demands for higher generation efficiency. This involves collaboration between industry and laboratory and, accordingly, development of suitable instrumentation which allows to obtain the necessary experimental data to assist in the design low-emission burners. It has already been mentioned that this thesis focuses on understanding of the phenomena associated with NO x emissions, and extends the investigations to understanding of aerodynamic centrifuging, which is demonstrated here to be also relevant to particulate emissions from a furnace. Minimisation of the latter is essential in situations where it is desirable to combine gas turbines in the energy-production cycles that increase the total efficiency of the process. The common denominator between NO x and particulate emissions is that they both depend to a certain degree on the near-burner aerodynamics: the near-burner region is where NO x is mainly generated, whilst for particulate emissions, the near-burner region constitutes the upstream boundary conditions of the flow, since it is there that particles enter the flow 1. It has been recognised that the way particles mix with the combustion air, heat-up and ignite in the flame is important for the minimisation of NO x emissions (Soud and Fukasawa 1996). Both productive and destructive chemical reactions for NO x have been identified (Miller and Bowman 1989) that generate or reduce NO x emissions respectively, through control of particle motion and, hence, particle combustion near the coal injection region (e.g. Abbas et al. 1993; Smart and Weber 1987 and 1989). This, and the fact that spatially-precise in-flame measurements of the aerodynamics and combustion characteristics of pulverised coal can aid the development and evaluation of sub-models in computer codes that assist in the design of low-no x burners, were incentives for the development of the optical instrumentation presented in Chapters 2 and 3 and its application in the flows of this chapter. Although it is desirable that the velocity, size (and temperature) of burning particles are simultaneously measured in full-scale facilities, there are only few publications reporting any measurements in full-scale facilities. They are limited either to only velocity measurement of burning coal particles using LDV as, for example, in a 37 MW burner (Ereaut and Gover 1991) or the size distribution of an ensemble, rather than of individual, particles using an amplitude-based sizing instrument in a 85 MW furnace (Bonin and Queiroz 1991). Regarding particle temperature, the averaged temperature of a stream of particles along a line-of-sight in Of course, regarding particulate emissions, investigation farther downstream of the flame stabilisation zone is necessary.

139 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES full-scale utility boilers with a two-colour pyrometer has been reported, for example, by Butler et al. (1992) in an 80 MWe installation, who treated particles effectively as a continuum. However, measurements in full-scale facilities have many disadvantages, amongst which are the inflexibility of control over the boundary conditions. The measurements reported by Ereaut and Gover (1991) were obtained in a 2000 MWe multi-burner installation and the parameters one could vary were limited by the operational parameters of the power station. Other difficulties, equally important, include the cost of operation of full-scale facilities and the complexity in using the necessary optical instrumentation demonstrated by the limited published work. As a consequence, the measurements in the cited references are all limited to a few isolated points (whose location is dictated by maintenance access ports rather than chosen) with little, if any, parametric study. Thus, despite the effort expended, it can be argued that the knowledge generated is comparatively little. 2 The development and application of advanced optical instrumentation which permits nonintrusive measurement of velocity, size and temperature as well as flux- of coal particles is, by itself, a formidable task and most published work is concerned with instrumentation which can be applied only in uni-directional flows as, for example, the measurements in a drop-tube furnace of Tichenor et al. (1984). It is not surprising therefore that the only simultaneous particle velocity, size and temperature measurements in a 10 kw swirl flame were those of Israel et al. (1995), using a combined phase Doppler anemometer and two-colour pyrometer, and their measurements were thus related to burning spherical droplets of atomised kerosene. Simultaneous size and temperature measurements were reported in a laboratory-scale fluidised bed by Hernberg et al. (1993) but the size was inferred from the temperature signals rather than measured directly by a suitable sizing instrument. Such a sizing instrument was developed by Orfanoudakis (1994) and was applied to make simultaneous size, velocity as well as flux measurements of burning coal particles in a 10 kw swirl burner with gas support, as a function of a wealth of parameters, including the equivalence ratio of the pilot flame, the swirl number and the primary to secondary air momentum ratio The Present Contribution NO x emissions are dependent on local gas composition and fuel stoichiometry and are thus a function of local particle velocity, temperature and flux characteristics. Operational parameters such as the swirl number and the primary to secondary air momentum ratio influence particle motion and, hence, the history of particle temperature and of the local gas and gas composition along the particle trajectory 3. In more detail, response of particles to the mean flow and/or to turbulent characteristics decides the trajectories that the particles will follow and whether 2 This is best illustrated by the results of Butler et al. (1992) who measured the temperature of a stream of particles and subsequently constructed temperature spectra at different locations in a boiler, a measurement which was difficult to accomplish and offered little understanding of either the flow inside the boiler or the physics of combustion of pulverised coal. 139

140 4.1 INTRODUCTION particles will experience unfavourable or favourable conditions, for example, for NO x emissions. In particular (as described in Chapter 1) availability of oxygen and short-living radicals are two of the most important variables that determine favourable conditions for NO x generation and destruction respectively. Measurement of the correlation between particle size and velocity vector is difficult in practice using existing instrumentation. Previous investigations, such as those of Orfanoudakis and Taylor (1995), reported only the axial velocity component, which corresponded to flow direction along the axis of symmetry of the burner. This component shows whether particles enter the swirl-induced recirculation zone and reverse their direction of motion but knowledge of the radial (i.e. normal to the axial) velocity component would provide indication on whether particles are centrifuged away from the recirculation zone. Early escape from the recirculation zone due to centrifuging is detrimental for NO x generation, as it results in reduced residence times (see Chapter 1) inside the recirculation zone. As concluded in Chapter 1, the recirculation zone provides conditions favourable for low NO x emissions and it is thus desirable to maximise particle residence time inside it. Centrifuging could also be the mechanism for separating particles from the mean gas flow and, hence, allow diversion of a clean hot exhaust gas stream from the coal furnace into a gas turbine for increased cycle efficiency. This chapter addresses these points by presenting measurements of particle volume flux, velocity (two components), size, temperature and burning flux as a function of swirl number, momentum ratio and equivalence ratio of the pilot flame necessary due to the small scale of the burners as explained in detail later in 8 kw swirl burners, one of which was identical to that of Orfanoudakis (1994). The sequence of results starts from measurements in the near-burner region of the burner of the size and velocity characteristics of particles in open flames, with emphasis on centrifuging and show the effect of confining the flow ( 4.3.1). The presentation continues with simultaneous measurements of the temperature, the size and the velocity of burning particles to address the effect of swirl number, momentum ratio and gas equivalence ratio on the particle burning pattern and to extrapolate the results to infer the potential effect of these parameters on NO x generation ( 4.3.1). The section on experimental results concludes with the flow in the far-burner region of a confined burner ( 4.3.2). The latter flow regime, although it depends on the near-burner region, is thought to be important for particle retention in the furnace and thus, minimisation of particulate emissions. Before describing the conclusions of the work of this chapter ( 4.5), particle centrifuging is further discussed ( 4.4) concerning its implications on particle motion in the near burner region (regarding NO x emissions) and in the far-burner region (regarding particulate emissions from the present furnace design). 140

141 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES 4.2 Flow Configurations and Experimental Method Burner Geometries In all experiments described in this thesis two stainless steel swirl burners were used and were both capable of firing pulverised-coal flames with natural gas pilot support. The first burner, denoted burner I, was identical to that of Milosavljevic (1993) and Orfanoudakis (1994) and had 16 mm throat diameter whilst the second, denoted burner II, was a modified version of burner I with 18 mm throat diameter and longer flow development section, described in detail later. The modifications were necessary as burner II was designed for operation in a ceramic furnace ( 4.2.2) at pressures higher than atmospheric 4 and therefore some modifications to the coal injector were necessary to ensure that it would survive at pressurised furnace. Burner I shall be described in detail and the differences between the two burners will be subsequently presented. Burner I is shown in figure 4.1 along with the z-r co-ordinate system used on all graphs. It consisted of a main cylindrical body, a contraction downstream of the cylindrical body and a straight flow-development section with D=16 mm internal diameter after the contraction, at the end of which there was a quarl, with half-angle of φ=20 and non-dimensional length L/D=2. As shown in figure 4.1, four axial and four tangential inlets of 10 mm diameter were symmetrically placed on the periphery of the main cylindrical body to generate variable swirl levels in the secondary air. A baffle was placed concentrically with the main body of the burner, 20 mm downstream of the entry of the four axial jets, to help achieve a uniform flow inside the main body. Two fuel injectors designs were available, namely a multihole radialand an axial injector 5 and each was coaxial with the burner and was centred by means of a spider to precision better than ±0.2 mm. Both injectors had the same external diameter of 6 mm, chosen from readily available stainless-steel tubing, resulting in an area blockage ratio of 14%, similar to that of Milosavljevic (1993) and Orfanoudakis (1994), and kept as small as possible in order to minimise the bluff-body effect and increase the maximum obtainable swirl number at a particular oxidant flow rate (Hagiwara and Bortz 1984). Figure 4.2 shows detail of the tip of the multihole radial injector. Coal was pneumatically transported through 3 Although it is common in the literature to express the phenomena occurring between pulverised fuel and air as mixing it is the author s belief that this stems from treatment of a particle stream as a continuum, since the term is better suited for describing molecular mixing of fluids. Mixing with gases is a property that solids do not possess: indeed, even after coming into contact with a gas they remain solid. One would similarly refer to liquids such as water and olive oil as non-mixing when they come into contact with each other. However the use of the term is preserved when the author refers to mixing of volatiles with the surrounding gases and whenever it refers to published work, because this is the way the authors of those publications described the phenomena. 4 in this thesis only experiments under atmospheric pressure are described. No pressurisation was performed due to lack of time. 5 In this Chapter only the multihole radial injector was used. This injector was designed for firing pulverised coal flames with gas support as described later. The second, axial injector, suitable for firing natural gas flames was used for the experiments of Chapter

142 4.2 FLOW CONFIGURATIONS AND EXPERIMENTAL METHOD r D z φ / )ORZ 'HYHORSPHQW 6HFWLRQ Four-arm centering spider A A 7DQJHQWLDO 6HFRQGDU\ $LU Baffle $[LDO 6HFRQGDU\ $LU Natural Gas Coal P ÃP SECTION ON A-A Figure 4.1 Drawing of the 10 kw swirl burner I indicating the method of swirl generation. The throat diameter of this burner is 16 mm. 142

143 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES PP PP PP PP PP PP PP PP Figure 4.2 Detail of the tip of the gas and pulverised coal injector. Gas was transported through the annulus and was injected via 6 holes of 1 mm radially placed whilst pulverised coal was fed through the central 3 mm tube. the central pipe by primary air whilst natural gas was fed through six holes of 1 mm each in the annulus. All pulverised coal flames were fired with natural gas pilot support because the residence times in the region of the swirl induced recirculation zone were small compared to coal devolatilisation times and it is not possible to stabilise flames fired solely by pulverised coal. In addition, owing to the small scale of the burner, the heat release due to coal combustion was low and the surface-to-volume ratio of the flame was high and, hence, heat losses to the surroundings were comparatively large and detrimental to coal flame stability. The second burner geometry, burner II, is shown in figure 4.3 along with detail of the coal gun used in co-operation with this burner. As mentioned earlier, this geometry was based on burner I and therefore not only the method of swirl generation but also the dimensions of the body of the burner were the same. The differences were concentrated on the straight flow development section. Burner II had a 18 mm throat diameter instead of the 16 mm of burner I. The enlargement was necessary to accommodate the larger coal injector which had 8 mm external diameter (figure 4.3) rather than 6 mm of burner I. The use of thicker material was due to the need to construct a more robust injector for use with burner II in pressurised operation and the burner throat diameter was accordingly increased so that the width of the annular gap was kept constant. Another difference between the two burners was that burner II had a flow development section three times as long as burner I to accommodate a ceramic plate leading to the quarl, described later. The consequence of the larger throat diameter was that the air bulk velocities through burner II were about 15% smaller than those through burner I for the same air mass flow rate. 143

144 4.2 FLOW CONFIGURATIONS AND EXPERIMENTAL METHOD Figure 4.3 Drawing of Burner II a modified version of the burner depicted in figure 4.1 along with detail of the tip of the fuel injector used with this burner. 144

145 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES Metal and Ceramic Furnaces Both metal and ceramic furnaces were one-sixth scale models of an industrial-scale furnace capable of operating at pressures up to 20 bar, described by Reichert et al. (1990). The metal furnace, figure 4.4 (see COLOUR PLATE 5 in Appendix VI) was made at an early phase of the work of this thesis and the purpose behind its design was to provide a light construction with optical access in the far-burner region near the lateral exhaust pipe and which could be easily traversed to allow measurement of profiles of particle velocity and size by SDV. In the design of the ceramic furnace, particular care was taken to make a construction capable of surviving at pressures up to 5 bar gauge and wall temperatures of the order of 1700 K. The down-fired 1 MW full-scale furnace (Reichert et al. 1990) was cylindrical with a slag drip ring, equipped with a swirl burner which could burn either natural gas or coal particles. The hot gases exhausted through a lateral exit near the bottom of the furnace and above a slag granulating bath. The laboratory model, shown in figure 4.4, was fired upwards at atmospheric pressures. This orientation was chosen as more convenient for optical access without any compromise of the fluid mechanics of the particulate phase, as the purpose was to investigate the effect of inertial centrifuging which is independent of the direction of gravity, as will be discussed later. The model had a 172 mm diameter 585 mm long cylindrical body giving an overall aspect ratio of 3.4, similar to that of the full-scale furnace. The lateral exit was 80 mm long and 40 mm diameter, arbitrarily selected to give an area ratio between the furnace and the exit of about A conical head was fitted to the top of the cylindrical body to emulate the ash collection hopper of the full-scale furnace and its dimensions were identical to the conical head used by Orfanoudakis (1994). Provision for optical access at two regions in the furnace was made, as shown in the plan view of figure 4.4, through 25x95 mm windows. The location of window 2 was selected to provide optical access as close to the near-exit region of the furnace as possible in order to study particle motion near the exhaust, where centrifuging effects would manifest themselves. 7 The burner was connected to a 172 mm diameter base plate with a quarl embedded in it at the bottom of the furnace, figure 4.4. The embedded quarl was identical to the quarl attached on the burner, figure 4.1; the latter was removed before connecting the burner to the furnace. The ceramic furnace was built to operate up to 5 bar and is shown in figure 4.5 (see COLOUR PLATE 6). The inner and outer diameters were 150 and 190 mm, and the length was around 600 mm. The refractory wall was constructed from silicon carbide which could resist the potential operating pressures and continuous temperatures up to 1500 C, according to the manufacturer (Morganite Thermal Ceramics, UK). Silicon carbide has high thermal conductivity, which allowed the furnace to heat up within one hour and resulted in 6 Reichert et al. (1990) did not provide details on the dimensions of their full-scale furnace. 7 An additional window (window 1) was placed upstream of window 2, but in this thesis only results from locations accessed through window 2 are reported. 145

146 4.2 FLOW CONFIGURATIONS AND EXPERIMENTAL METHOD granulating bath lateral exhaust pipe Exhaust z v u Window Window swirl burner Air & Pulverised Coal frame for window 2 frame for window 1 80 y x 4 0 Figure 4.4 Drawing of the measured metal laboratory furnace in elevation and in plan view showing the cartesian velocity components u,v. The windows which provided optical access for the measurements are also indicated in the figure. 146

147 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES Figure 4.5 Elevation of the ceramic furnace designed to operate in pressures up to 5 bar. The figure shows the various components of the furnace, i.e. the base and end wall plates, the two pieces of the metal casing surrounding the ceramic the access ports for optical instrumentation and the exit ports which were symmetrically placed. Only one exit port was used at a time. 147

148 4.2 FLOW CONFIGURATIONS AND EXPERIMENTAL METHOD comparatively short startup time of the experiments. An external water-cooled stainless steel casing consisting of four segments protected the nearby lenses from the heat radiated by the ceramic furnace. The end furnace walls were also constructed from silicon carbide. The burner end wall, here termed as base plate, incorporated an igniter port and the burner quarl was cast into the refractory material. The quarl had a length to exit diameter ratio of around 1 with entrance diameter of 18 mm and exit diameter of 38 mm. The exhaust end wall simulated an ash collection port, as in the case of the metal furnace. The refractory wall, figure 4.5, had four optical rectangular 70 x 40 mm access windows. Two of the windows were on opposite sides of the refractory wall and provided the forward, and nearly-forward, scatter access required by the SDV instrument s receiving optics and the third window was at 90 relative to the direction of the two previous windows to allow photographic studies of the flow field in the near burner region. The fourth window was located at either a 30 or 60 off-axis angle, depending on the window used to provide the incident light source, and could also provide alternative optical access for any future instrumentation. Ports for temperature or species concentration measurements were available along the wall of the furnace. 8 Two exhausts were placed at 520 mm from the exit of the quarl, on opposite sides of the furnace, to facilitate the positioning of the furnace relative to the optical instrumentation and convenient placement of the laboratory exhaust ducts. Note that figure 4.5 shows the ceramic furnace as manufactured, but there were modifications in the metal casing during its construction which were not fed back into the drawing because of the complexity of the latter and lack of time. The changes were minor and they were concentrated in blocking the access ports to thermocouples, which were not used in the experiments, as well as constructing the base plate where the burner was mounted from a rigid stainless steel piece, rather than the water-cooled shell depicted in figure The ceramic furnace was constructed to allow measurements of particle velocity and size, concentration of stable gas species and gas and wall temperatures under pressures between 1 and 5 bar. In the progress of the work the actual construction time exceeded by far the quoted schedule and only limited time was eventually available for test firing and measurement of the particulate phase inside the furnace at atmospheric pressures. It was also necessary to apply last-minute modifications on the metal casing for the measurements, as described in the next paragraph, in order to make measurement feasible. Although the furnace can be potentially fired at design conditions, the set of measurements obtained during the course of this thesis served to investigate the effect of the confinement on the velocity and size of the particles as described in

149 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES Table 4.1 Proximate and Ultimate analyses of Bentinck bituminous coal (UK). The table also shows details on the particle size distribution of the coal batch used in the experiments (except that of 4.3.2), measured by a Malvern instrument (Abbas 1994). Proximate Analysis Weight (%) as received Moisture 3.9 Volatile matter 34 Fixed carbon 56.6 Ash 5.5 Ultimate Analysis Weight (%) as received Carbon 75 Hydrogen 4.8 Nitrogen 1.5 Sulphur 1.6 Oxygen (by difference) 11.6 Coal Particle Size % under 85 µm 99 under 11 µm 25 Average size measured by SDV AMD=35 µm SMD=45 µm Heating Value (Hu) in MJ/kg Note that the proximate analysis was performed before 1990 so that at the time of measurements smaller quantities of volatile matter than the quoted are expected to remain in the coal, due to the finite volatile release rate at atmospheric temperature Pulverised Coal Analysis The coal used in this burner was bituminous Bentinck (UK) coal and the batch used in most measurements had particles 99% by volume smaller than 85 µm and 25% smaller than 11 µm (Abbas 1994). 9 Table 4.1 presents the proximate and ultimate analyses of the coal (Abbas et al. 1993) Experimental Conditions and Bulk Flow Quantities The amount of air fed through the burners was kept constant for all conditions (except where otherwise indicated) to 300 lt/min. The air bulk velocities, U 0, at this flow rate through the burner annulus were 29 and 24.5 m/s in burner I and II respectively, corresponding to Reynolds numbers of and calculated at 20 ºC using the diameter of the annulus of the burner. These Reynolds numbers ensured fully-turbulent jet flow from the exit of the burner throat in combustion. 10 Burner I has been previously calibrated for swirl by Milosavljevic (1993) for a total of 250 lt/min air throughput and Orfanoudakis (1994) used the calibration curves provided by the former to extrapolate the swirl numbers at a total 300 lt/min flow rate. 149

150 4.2 FLOW CONFIGURATIONS AND EXPERIMENTAL METHOD In this work all swirl numbers referring to burner I are those reported by Orfanoudakis (1994). Burner II was a new design, as explained earlier, and was therefore calibrated for a flow rate of 300 lt/min. In this chapter results are also presented for 400 lt/min but the corresponding swirl number was extrapolated from calibration curves at 300 lt/min following Orfanoudakis (1994). The swirl number is defined as: & ϑ 2G S = G& D z (4.1) where D was the diameter of the burner throat and the axial fluxes of the angular, G & axial momentum, G &, are defined as: z ϑ, and G& G& ϑ z = 2πr = 2πr r r o r r i o i WrUrdr UUrdr (4.2) where integration is performed as a function of burner radius, r i and r o correspond to the external radius of the coal injector and the internal radius of the burner throat respectively and U, W are the mean axial and tangential components of flow velocity. Details on the measurement of these velocity components in burner II and the calculation of the swirl number are presented in Appendix IV. Natural gas was drawn from the laboratory compressed supply and was metered by calibrated rotameters whilst pulverised coal was supplied by means of a vibrating feeder at mass flow rates between 0.24 and 1.5 gr/min, and was transported to the burner through a 3 m-long copper pipe with 8 mm internal diameter. The fact that the primary air pipe was long and the coal particles were fine (see 4.2.3) resulted in frequent clogging of the pipe and in surges of coal particles which extinguished the pilot flame, although air was purged through the feeder and the primary air pipe before starting measurements at every new location of a profile. Pipe clogging did not have any influence on the measurements but the frequent extinctions due to particle surges delayed the experiments, because a flow rate of around 60 lt/min had to be fed for a few minutes through the clogged primary air pipe to ensure that the pipe was clean with the exception of the measurements of for which a coal batch identical to that of Orfanoudakis (1994) was used. 10 The reader is reminded that kinematic viscosity increases with temperature faster than bulk velocity (due to density decrease) and thus the corresponding local Reynolds number decreases. As a consequence the local Reynolds number under combustion can be between one-third and half of that calculated at room temperature. Provision of flows through the burners with Reynolds numbers of the order of at room temperature ensure that under combustion are above 10000, necessary for fully-turbulent flow.

151 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES For most conditions 10 lt/min of primary air was used which corresponded to a particle bulk exit velocity of 23.6 m/s, assuming no slip between particles and transporting air. All measurements reported here (with the exception of those of 4.3.2) were made at a particle loading of 1.5 gr/min which corresponded to an average particle concentration at the exit of the primary air jet of about particles/µm 3 (=10 6 particles/m 3 ), calculated from the bulk air flow rate and coal mass flow rate, assuming that all particles were 20 µm, a diameter representative of the smaller sizes present in the batch. This result is an indication that the event of multiple occupancy in the SDV probe volume is unlikely, also confirmed by estimation of the group combustion number (Chiu and Liu 1977), which was of the order of << The coal feed rate of 1.5 gr/min was the highest which could be supplied through the current primary air pipe before feeding problems resulted in unworkably frequent clogging of the pipe and interruption of the experiment. Two values for the primary/secondary momentum flux ratio were used, namely 1/50 and 1/40. The momentum flux ratio was calculated from: (m& air,p + m& MR = m& U air,s coal 0 )U 0,p (4.3) where m &, m&, m& were the mass flow rates of primary and secondary air and pulverised air,p air,s coal coal respectively, and U 0,p and U 0 the bulk velocities of the primary and secondary streams respectively. It must be noted that the momentum of the solid phase at a feed rate of 1.5 gr/min was smaller than 0.1% of that of the primary air and had, effectively, insignificant contribution to equation (4.3). 11 As mentioned earlier the flow rate of pulverised coal in the flame was small and, thus, so was the corresponding coal based equivalence ratio 12 and potential heat release. Most pulverised coal flames were stabilised using 22 lt/min natural gas at a (gas) equivalence ratio of and potential heat release of about 8.2 kw. The maximum gas supply was 32 lt/min, and corresponded to stoichiometry and a potential heat release of 12 kw. The coal mass flow rate of 1.5 gr/min corresponded to equivalence ratio of and a potential heat release of 0.7 kw. As can be seen from the above figures, the contributions of coal to the total equivalence ratio and potential heat release were less than 10% that of the gas. 11 It must be noted here that in a full-scale burner mass loading in the primary air is 1-1, i.e. one kg of particulate matter is fed per kg of air. 12 The term equivalence ratio is borrowed from premixed flames and is probably less suitable for quantifying the amount of supplied air and fuel in coal burners than the term air to fuel ratio, most commonly used in internal combustion engines. The term equivalence ratio is nevertheless used throughout this thesis to enable comparisons between the present results and previous work on the subject, particularly with the work of Orfanoudakis (1994). 13 This was calculated using equation 1 of Chapter 5 in which the stoichiometric fuel/air ratio has been already included. As explained in Chapter 5, in this thesis all calculations of the gas equivalence ratio were performed assuming that the fuel was pure Methane. 151

152 4.2 FLOW CONFIGURATIONS AND EXPERIMENTAL METHOD Instrumentation and Uncertainties This chapter presents measurements in the near-burner region of open and confined flames obtained by the SDV instrument described in Chapter 2 and the combined SDV and twocolour pyrometer instrument of Chapter 3. The measurements in the far-burner region of the confined flame in were obtained with an older version of the SDV instrument which used analogue-to-digital (A/D) converters instead of D/A. Detailed description of the A/D version of the SDV instrument is beyond the scope of this thesis, particularly because that instrument was the first prototype and was only used in the aforementioned experiment. The optical and electronic characteristics of the A/D-based SDV instrument are nevertheless described below. 14 Instrumentation Figure 4.6 shows a block diagram of the SDV arrangement relative to the furnace. An Ar + laser operating at nm and output power of 300 mw/beam was used. The control volume was generated by a commercial transmitting laser Doppler velocimeter (LDV) optical unit (DISA 55X, Dantec A/S), with shifting provided by a single Bragg cell operated at 40 MHz, and the control volume was imaged through the windows onto the photodiode array by the receiving optics, along with the images of any particles passing through the volume. Table 4.2 summarises the optical parameters of the transmitting optics of the SDV. The principle and the details of operation of the SDV instrument have already been given in Chapter 2. The purpose of the next paragraph is to outline the difference of the version of the SDV instrument used throughout this thesis (as described in Chapter 2) and the older version used only for the experiments of this section, dubbed the prototype SDV. The reader should recall that the size of the irregular particle is here quoted as a single number, corresponding to the diameter of a sphere, d p, which has the same area as that measured from the projected area, S proj of the particle, dp( µ m) = S proj 4 π 1 2 (4.4) 14 The reader should recall at this point that regardless of the method of processing implemented in the signal processor (i.e. A/D or D/A) the principle of particle sizing and, hence, the photodetector was the same in both versions. The later use of D/A converters, in the place of A/D, allowed faster measurements to be made since much of the processing could be performed by hardware instead of by software, as required in the A/D version. The benefits with D/A over A/D have already been described in Chapter 2; it worth mentioning that the A/D version was incapable of measuring particle velocities higher than about 15 m/s and also flux, owing to the slow response of the electronic circuitry. 152

153 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES Burner & Confinement LDV Transmitting Optics Optical Windows Exit Receiving Optics Photomultiplier Photodiode Array Unit Swirl Burner Host Computer Fuel+Air Transient Recorder Multi-channel Transient Recorder Low Pass Filter Pre-amplifier Figure 4.6 Block diagram of the flow configuration and the SDV. Table 4.2 Parameters of optical configuration Ar-Ion laser wavelength (nm) Beam interaction angle (deg) 6.87 Focal length of transmitting lens (mm) 500 Focal length of first receiving lens (mm) 200 Total magnification ratio (-) 213 Laser intensity at measuring volume 300 (mw/beam) Size of measuring volume 250x500x4000 (µm x µm x µm) In this prototype instrument, it was convenient for the velocity measurement to be made by collecting scattered light with a separate set of receiving optics, imaged onto a photomultiplier, and processing the signal by an FFT-based processor. In principle, the velocity information could be extracted from the linear array. The FFT-processor was triggered by the sizing signal to avoid capturing strongly out-of-focus images of particles (see Hardalupas et al for a discussion) which nevertheless result in valid Doppler signals. More details about the FFT signal processing, in Maeda et al. (1988). The sampling rate per photodiode was f c = khz which limited the maximum particle velocity which would provide 10% size accuracy to about 11m/s. The limitation will be relaxed in future circuits by arranging for faster sampling 153

154 4.3 RESULTS of each photodiode. The measurable size range for the current optical configuration extended from 10µm to about a hundred microns. The statistical uncertainties were as follows. In the measurements of the aerodynamic characteristics of particles of a minimum of 1000 samples were taken per point, a number which was limited by the time required to validate the measurements. Statistics of the radial velocities were based on 30-40% fewer measurements than for the axial, because only measurements of particles which were defocused by a minimum distance were included in the calculation of the mean and the rms radial velocities. The measured sizes were sorted into three classes, namely µm, µm and µm, from now on referred to as 18, 36 and 60 µm, to allow statistical uncertainties in mean velocity lower than 10% and characterise the response of particles to the turbulence of the flow according to the Stokes number, defined as the ratio between an appropriate flow time scale and the particle response time (e.g. Hardalupas et al. 1992). The minimum size of 12 µm was a limitation due to the particular magnification ratio of the SDV receiving optics. 15 Thus, 200 measurements per size class were typically used to calculate the axial velocity statistics resulting in 5% and 10% statistical uncertainties in the mean and the rms respectively (Yanta and Smith 1978); the number decreased to as low as 50 measurements for the largest size class at the outermost radial locations, which accordingly corresponded to uncertainties of 10% and 20%. Typical statistical uncertainties for the radial velocity were 8% and 12% respectively. In the vicinity of the stagnation point of the recirculation zone the uncertainty for the radial component is not expected to exceed 10% and 14% in the mean and rms respectively Results Section 4.3 presents the experimental results from the investigations in the open and confined flames, in the following sequence: first, the size and velocity characteristics of coal particles in the near-burner region of the open and confined flames ( 4.3.1). The presentation is then extended to the size- and velocity-dependent temperature characteristics of particles in open flames ( 4.3.1) and concludes with the results from the downstream region of the metal furnace ( 4.3.2). Because each section focuses on a different property of the flow, the experimental conditions are described in each section to avoid confusion between results from different geometries and boundary conditions Although that was the minimum size which would be validated by the instrument for the present magnification, the minimum measurable size, at any magnification ratio, cannot be smaller than about 5 µm, provided that the resolving power of the collection optics is sufficient to discriminate images of this size. The reason is that the laser wavelength was 488 nm and if a particle were smaller than about 5 µm, less than approximately 10 light waves of 488 nm wavelength would have been diffracted by the particle and, hence, would not allow resolution of particle shape. 16 According to the theoretical analysis of Chapter 2, this location corresponds to the worst-case owing to zero mean axial velocity.

155 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES The Near-Burner Region Particle Aerodynamics in Open Flames Table 4.3 summarises the operating conditions of the burner, which included two swirl numbers, 0.41 and 0.57, and two secondary air flow rates of 300 and 400 l/min. The swirl number was calculated from the radial profiles of axial and tangential velocity of the non-reacting flow. The velocity measurements and the method of calculation of the swirl number are presented in Appendix IV. For flows 1 and 2 the conditions were those described in Table 4.3 Summary of conditions of parametric study. Flow Swirl number Gas ratio equivalence Coal equivalence ratio Secondary air flow rate [lt/min] (Bulk velocity m/s) (24.5) (24.5) (32.6) (24.5) (24.5) 17.7 Particle bulk velocity at exit of jet [m/s] Examples of flames corresponding to conditions 1, 2, 3 and 4 are shown in COLOUR PLATES 7, 8, 9, and 10 respectively. There is no representative photograph of flame of condition 5, because measurement of this condition was decided after completion of most of experiments, but visualisation of the influence of reducing the momentum ratio (quantified by condition 5) is shown in COLOUR PLATE 11, which corresponds to condition similar to 5, except for the swirl number which was The results of this condition are described in subsequent section and this condition is included in Table 4.2 for completeness only. For flow 3, the gas and coal flow rates were increased by 30% to maintain the overall equivalence ratio equal to that of flows 1 and 2. Results are presented in the form of radial profiles of particle volume flux, mean and rms of fluctuations of axial and radial velocities, U, u and V, v respectively, mean and rms trajectory angle relative to the centreline of the burner, φ, as well as the correlation between trajectory angle and particle size. The spatial distribution of the particle volume flux in the near-burner region is an indication of the amount of particles entering the recirculation zone close to the axis of injection, which is a favourable condition for NO x emissions. Particle centrifuging is expected to increase the volume flux at radial distances far from the axis, and the mean and the rms velocity characteristics and in the following text it will be shown that the correlation between trajectory angle and particle size will identify the phenomenon. Figure 4.7(a)-(c) presents comparison between radial profiles of absolute particle net and positive volume flux corrected with the validation rate of data acquisition for flows 1-3 respectively, at axial location z/d=2.67, selected to be as close as possible to the quarl exit, where a gas flow recirculation zone is present. Flux is generally defined as (see also Chapter 2): 155

156 4.3 RESULTS G = 1 T sm V A i i i ij i n j= 1 u u ij (4.5) where T sm is the total sampling time, u ij is the axial velocity component normal to the fringes of the SDV probe volume of particle j in size class i, V i is the volume of a particle of size class i, n i is the total number of particles in size class i and A i is the size (m 2 ) of the sampling space for size class i through which particles flow with unit vector normal to the fringes. Positive flux is defined for equation (4.5) taking into account only velocities larger than zero, whilst net includes all contributions regardless of direction of the velocity vector. Two characteristic patterns are observed: one in flow 1, figure 4.7(a), where penetration of the centreline by all size classes results in a flux maximum on the centreline, and a second in flows 2 and 3, figure 4.7(b) and (c), where a second local maximum occurs at radial distances r/d=0.7 and 1 respectively. This maximum implies that particles enter the recirculation zone close to the axis of injection, reverse their direction, as illustrated by the negative fluxes, figure 4.7(b) and (c) and disperse away from centreline at the shear layer between the secondary air stream and the recirculation zone where they acquire positive velocities. The flux distribution for flow 1 suggests that a large amount of the mass of coal particles entering the recirculation zone penetrated through it without reversing its motion, which is unfavourable for the reduction of NO x emissions. On the contrary, reduction of the swirl number (flow 2) and increase of the secondary air flow (flow 3) led to increased number of particles reversing their motion and increasing their residence time inside the gas recirculation zone, which is favourable for the reduction of the NO x emissions. However, the resulting second maximum at large radial distances of the flux distribution may limit the potential improvement on NO x emissions and the mechanism responsible for the second maximum is therefore important and will be identified below. Figures 4.8 and 4.9 show radial profiles at z/d=2.67 of the mean and rms of fluctuations of axial and radial velocity 17 of particles as compared with those of the gas flow for flows 1 and 2 respectively. The mean axial velocity profile of figure 4.8 shows no recirculation zone in the central region of the flow and this is in agreement with the observation of the flux profile of figure 4.7(a) for flow 1; all particle size classes had positive axial velocities on the centreline which increased with particle size, probably due to the initial condition at the exit from the injector. The mean axial and radial velocity profiles for flows 1 and 2 (figures 4.8 and 4.9) for 18, 36 and 60 µm particles almost collapse onto a single curve suggesting that particles respond to the mean flow. By defining a mean flow timescale on the ratio of the maximum velocity in the region of the forward flow (r/d>0.6 in figure 4.8) and on the quarl exit diameter, mean Stokes numbers of 10, 3 and 1 are calculated for the 18, 36 and 60 µm classes respectively, which are all greater than one and hence response of the particles to the mean flow is indeed 156

157 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES G [arbitrary units] µm, positive µm, net µm, positive µm, net µm, positive µm, net (a) r/d G [arbitrary units] (b) r/d G [arbitrary units] 10 0 (c) r/d Figure 4.7 Radial profiles of net and positive particle volume flux corrected with the measurement validation rate, for (a) S=0.57 (flow 1 in Table 4.3), (b) S=0.41 (flow 2) and (c) S=0.57 and 400 lt/min secondary air (flow 3), for three particle size classes indicated in the legend. 17 Radial velocity components have been obtained from the axial velocity, measured by LDV and the trajectory angle, measured by SDV, as explained in Chapter

158 4.3 RESULTS µm µm µm gas (a) (b) U/U u /U r/d r/d 0.3 (c) 0.3 (d) V/U v /U r/d r/d Figure 4.8 Radial profiles of the (a) mean and the (b) rms axial velocity, as well as the (c) mean and the (d) rms radial particle velocity as compared to those of the gaseous phase, for S=0.57 (flow 1 in Table 4.3) at z/d=2.67, for three size classes. Symbols: (ç) µm, (á) µm and (ó) µm and (ò) gas phase. expected. Except for local differences beyond r/d=0.9 for flow 2 (figure 4.8), a region at which high volume flux values are observed, the particle mean radial velocity is similar to that of the gas flow, which is also true for flow 1. The profiles which present the most interesting feature are those of the rms fluctuations of radial velocity, especially in the case of flow 2 (figure 4.8) which, in the region r/d>0.5, increase with increasing size. A similar phenomenon has been previously observed by Hardalupas et al. (1989) for the rms fluctuations of particle axial velocity in a jet flow and a fan spreading mechanism, based on the existence of quasideterministic particle trajectories for Stokes numbers less than unity 18, was proposed to explain the measured fact that the rms velocities for large particles were higher than smaller. In the present case, the manifestation of large rms fluctuations of the radial velocities for the 60 µm size class was not accompanied by similar observations in the rms fluctuations of axial velocity profile, which was also the dominant direction of the flow and the fan spreading mechanism 158

159 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES 1.0 (a) (b) U/U u /U r/d r/d 0.3 (c) 0.3 (d) V/U v /U r/d r/d Figure 4.9 Radial profiles of the (a) mean and the (b) rms axial velocity, as well as the (c) mean and the (d) rms radial particle velocity, for S=0.41 (flow 2 in Table 4.3) at z/d=2.67, for three size classes. Symbols: (ç) µm, (á) µm and (ó) µm and (ò) gas phase. need not be the explanation for our observations, particularly since the Stokes numbers were greater than unity. A mechanism for the explanation of the observed trends will be suggested later in this chapter. The effect of increasing the total secondary air flow rate, while keeping the swirl number constant, on the mean axial and radial velocities is presented in figure In accordance with the flux profile, figure 4.7(c), all particle sizes penetrated the recirculation zone on the centreline, but they attained velocities almost 50% lower than for flow 1, figure 4.9. The major difference between the mean radial velocity of flows 1 and 3 was that the 25% increase of the total secondary air resulted in a maximum mean radial velocity which was almost doubled, although the shape of the radial profile remained the same. In addition, in the region of r/d=1.2, a correlation of the size with the radial velocity is expected which can be translated 18 i.e. particles which remain unresponsive to the flow time scales. 159

160 4.3 RESULTS 1.0 (a) U/U r/d 0.3 (b) 0.2 V/U Figure 4.10 Radial profiles of the (a) mean axial velocity, and the (b) mean radial particle velocity, for S=0.57 and 400 lt/min at z/d=2.67, for three size classes. Symbols: (ç) µm, (á) µm and (ó) µm. r/d into an equivalent correlation between size and trajectory angle given the observation that the axial velocity does not change with size. A similar correlation for the mean radial velocity was observed in flow 2 (figure 4.8) at r/d=0.9 which, as in flow 3, lies in the region where a maximum for the volume flux was measured off the centreline. In order to emphasise the latter observation we present in figure 4.11(a) and (b) radial profiles of the mean trajectory angle relative to the centreline of the burner 19 multiplied with the sign of the mean axial velocity as function of the size, measured at z/d=2.67 and for flows 2 and 3 respectively. By multiplying with the sign of the axial velocity, positive plotted values indicate trajectories which deviate away from the centreline. Both cases show that there is a mean net increase of the trajectory angle in the location between the centreline and large radial distances. However, between these two extreme locations, both profiles show a non-monotonic increase, which is size-independent, except at r/d=0.9 for flow 2 and at r/d=1.2 for flow 3 where the measured trajectory angle increases with the size. These two locations have been mentioned earlier as (i) being in the region where respective secondary maxima occur in the flux and (ii) where a similar trend in the mean radial velocity profiles was observed. 19 In all experiments of this thesis the trajectory angle was measured relative to the (centreline) axis of the burner, a fact which is implied in all further references to this quantity. 160

161 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES Φ ( 0 ) sign(u) 25 (a) Φ ( 0 ) sign(u) 25 (b) r/d r/d 50 (c) 50 (d) φ [ 0 ] 20 φ [ 0 ] d [mm] d [mm] Figure 4.11 Radial profiles of the mean particle trajectory angle relative to the centreline of the burner, multiplied with the sign of the particle axial velocity for (a) S=0.41 and (b) S=0.57 and 400 lt/min, for three size classes. Positive abscissa denotes trajectories away from the centreline of the burner. Also, sizetrajectory angle correlation for (c) S=0.41 at r/d=0.9 and (d) S=0.57 and 400 lt/min at r/d=1.2. Vertical bars indicate the rms values. Some evidence on the origin of the observed trend in the mean angles is provided by the correlation between the size and the mean trajectory angle, plotted for flow 2 at r/d=0.9, figure 4.11(c), and for flow 3 at r/d=1.2, figure 4.11(d). The vertical bars indicate the rms values, which in turn show the range of angles of particle trajectories. In both cases the measured mean trajectory angle increases with the size up to about µm at which point a large scatter in the correlation is observed which is partly due to the statistical uncertainties (see 4.2.5). This size-angle correlation implies that acquisition of different radial velocities as a consequence of different trajectory angles is not fortuitous; the increasing trajectory angle as a function of particle size suggests that particles larger than 40 µm deviate away from the centreline more than smaller particles, possibly under the effect of centrifugal forces from mean gas flow 161

162 4.3 RESULTS (a) 0.3 (b) U/U u /U r/d r/d (c) 0.2 (d) V/U v /U r/d r/d (e) φ ( 0 ) r/d Figure 4.12 Radial profiles of particle (a) mean and (b) rms axial velocity, (c) mean and (d) rms radial velocity and (e) rms value of the trajectory angle for S=0.57 at z/d=4. Symbols: (ç) µm, (á) µm and (ó) µm. 162

163 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES rotation, under the influence of the mean tangential velocity component, or from response to a large-scale turbulent structure as identified below. In flow 1 not presented here contrary to the case of lower swirl number in flow 2, no particular mean correlation between the size and the trajectory angle was found thus indicating an absence of particle centrifuging due to the mean flow. The penetration of the recirculation zone by the incoming primary jet resulted in the formation of a very weak torroidal recirculation zone which was unable to decelerate the particles so that they could follow the mean tangential flow, at least near the exit of the quarl at z/d=2.67. In flow 1 at z/d=4, 60 µm particles have higher mean radial velocity at r/d>1 than 18 and 36 µm, figure 4.12(a), probably due to mean flow centrifuging. A more prominent trend is observed in the radial profile of the rms fluctuations of the radial velocity, figure 4.12(d), where in the region of r/d>0.6 the values for 60 µm increase almost linearly with increasing radial distance, unlike those for 18 and 36 µm, in accordance with the variation of the range of particle trajectory angles in figure 4.12(c). If we take into account that the radial profiles of the axial mean and rms velocity for 18, 36 and 60 µm (not plotted) collapse on a single curve, which renders the explanation of Hardalupas et al. (1989) unlikely in the present case, we suggest that the measured values of the radial rms velocity (a consequence of the measured range of particle trajectory angles) are a result of particle centrifuging by the large eddies of the flow turbulence. Further discussion on centrifuging effects on coal particles will be presented in 4.4. Particle Aerodynamics In Confined Flames The mean and rms axial velocities, the mean and the rms tangential velocities and the arithmetic and Sauter mean diameters of coal particles in the flame for a 0.57 swirl number were measured in the ceramic furnace at atmospheric pressure. The tangential, rather than the radial, velocities of the particles were measured in this case because the optical access through the windows allowed profiles to be taken only when traverses were performed at right angles to those reported in the previous section. Hence, the trajectory angle measured by the SDV from which the second velocity component was inferred- was that between the axial, U, and the tangential, W, velocity components samples were taken on average at each measurement location and the data rate varied greatly, between a few minutes for locations with high particle flux to a couple of hours for the rest. Measurements at locations with low particle flux were not only time consuming because of the data rate, but also due to the fact that particle accretion on the windows caused the total laser power to reduce and, therefore, frequent referencing 20 measurement was required for the operation of the SDV. 20 For explanation of referencing refer to Chapter

164 4.3 RESULTS An observation on the radial profile of the mean axial velocity at z/d=3.22, figure 4.13, is the asymmetry in the peak velocities on either side of the geometric centreline, one of which reached values of 0.8U 0 and the other reaches 0.6 U 0. The two peaks occurred at r/d=0.83 and at r/d=-1.11, respectively. Except for the centreline, where 60 µm particles had velocities of the order of 0.2 U 0, which are higher than those of the 18 and 36 µm that had 0.1U 0, everywhere else along the profile the axial velocities for all particle classes collapse on a single curve, an observation which is more pronounced than was in the case of unconfined flames already presented. It is also observed that at r/d=-2.0 particles entered the external recirculation zone, formed between the swirling flame and the wall of the ceramic liner of the furnace. The profile could not be radially extended to locations r/d=2.0 because of limitations in the traversing mechanism of the furnace and not because of absence of particles. No important differences in the velocity characteristics of particles as a function of their size were observed in the rms axial velocity profiles. The mean tangential velocity for all size classes, figure 4.13, increased almost linearly with radial distance up to about 0.8D where it reached its maximum of about 0.2U 0. Again, there is a departure from symmetry in the form of different peak values for the mean tangential velocity, by about 20% between the peak at r<0 and the peak at r>0. Also the almost total coincidence of the profiles for the three size classes observed in the axial velocity profiles is absent for the tangential velocities, where the 60 µm particles have consistently smaller tangential velocities than the smaller particles, especially in the region r>0 where the differences are as high as 50%. This can be attributed to particle centrifuging which was apparent in the open flow, where measurements of the radial rather than tangential velocity component were made (see 4.4). The tangential rms velocity profile exhibited two pronounced peaks at about 1.0D from the centreline which are probably due to either pressure redistribution or direct production. 21 The possibility that the peaks had substantial contribution from the fan-spreading mechanism (Hardalupas et al. 1989) is small, because the tangential rms velocities did not increase with increasing particle size. Figure 4.13 also shows radial profiles of the arithmetic mean (AMD) and Sauter mean (SMD) diameters, defined as: AMD = i i n d i n i i, SMD = i i n d i n d i 3 i 2 i (4.6) 164

165 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES U/U u /U r/d 0.3 r/d µm µm µm W/U w /U r/d 80 AMD SMD r/d 60 d (µm) r/d Figure 4.13 Radial profiles of the mean and rms axial and tangential velocity components as well as the AMD and SMD diameters inside the ceramic furnace fired at atmospheric pressure for S=0.57, at z/ D=3.22. Symbols for velocity profiles: (ç) µm, (á) µm and (ó) µm. Symbols for diameters: (ç) AMD, (á) SMD. 21 For the gas flow, but it is assumed that particles follow the flow faithfully. 165

166 4.3 RESULTS where n i and d i were the volume fraction and the diameter of size class i. The AMD and SMD diameters remained almost constant at 30 and 40 µm respectively, with the exception of a narrow region around r/d=-0.75 and r/d=1 where the SMD exhibits a small drop to about 35 µm. 22 The almost constant value of the SMD along a radial profile confirmed the earlier observation that the presence of the furnace did not result in size-dependent velocities and therefore justifies the decision to make most of the measurements in open flames, particularly in view of the difficulties which stem from window fouling due to the particles. The latter, although it does not affect significantly the accuracy of size measurement using SDV (Hishida et al. 1995), makes the use of the pyrometer under these conditions almost impossible, because signals through windows become weaker compared with the no-windows case due to extinction and, thus, the detectability limits of the pyrometer become unacceptably low. 23 Combustion Characteristics of Particles in Open Flames Effect of Swirl The effect of the swirl number on the burning pattern of coal particles in the near-burner region of the swirl burner has been quantified from radial profiles of the velocity, size and temperature 24 of particles at axial distances z/d=2.67 and z/d=4, where z was measured from the inlet of the quarl. The upstream stations were selected as being the closest at the quarl exit which was accessible to the optical instrumentation and the latter as being just downstream of the recirculation zone. Two swirl numbers were investigated, namely 0.41 and 0.57, the former resulting in mean particle recirculation on centreline (type II flow), and the latter which was the highest achievable by this burner design and the available flow rates of air. In contrast to the lower swirl number, the higher resulted in recirculation zone penetrated by the incoming primary air (type I flow), as shown in the results in the first subsection of 4.3.1, with no onaxis reversed particle axial velocities. The conditions were those of flows 1 and 2 in Table 4.3. Results are presented below for the effect of the swirl number on particle burning fraction and its correlation with the temperature of the surrounding gas due to combustion of the gaseous pilot fuel. In addition, comparison between the velocity, size and temperature probability functions (pfs) of burning particles and the ensemble, which includes both burning or not burning particles, is presented for selected locations. Figure 4.14 presents the mean gas flow temperature for S=0.57 and S=0.41, at z/d=2.67, measured by a 250 µm platinum-platinum/13%-rhodium thermocouple in the absence of particles. Note that the mass flow rate of coal particles was so low that the presence of burning Note that this location does not correspond to a zero axial velocity location, around which the particle velocity pf measured by the SDV is clipped as explained in Chapter In fact I did not manage to get any pyrometer signals during measurement through the windows in my test experiments. 24 The temperature of char particles, estimated after application of the criterion described in detail in Chapter 3.

167 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES T [K] r/d Figure 4.14 Radial profiles of the mean gas temperature, measured in the absence of particles at z/ D=2.67, for (í) S=0.41 and (ó) S=0.57. particles would not have contributed materially to the local gas temperature. No correction for radiation losses was applied to the measurements. Although the maximum temperatures were about 1800 K and 1850 K for S=0.41 and S=0.57 respectively, for S=0.41 the presence of the recirculation zone on the centreline defined a zone up to about r/d=0.80 that behaved as a stirred reactor and had insignificant temperature variation across a radial profile. On the contrary, the penetration by the primary jet of the gas flow recirculation on the centreline for S=0.57 yielded a region extending between r/d=0 and r/d=0.80 which had temperatures lower than 1800 K and down to about 1000 K on the centreline. The combination of gases cooler than 1300K between r/d=0 and r/d=0.4 and the local positive particle velocities is likely to have resulted in ignition delay times which were long compared to the particle residence times, and this is reflected in the low values of the fraction of burning particles of figure Figure 4.15 shows the burning fraction of particles calculated from equation (4.7) at z/d=2.67 for the two swirl numbers S=0.41 and S=0.57. The symbols in equation (4.7) have been explained in Chapter 3 (equation 3.23), where it was first presented. f = b N N b t (4.7) There is a clear distinction between the behaviour of the burning fraction in type II flames occurring at S=0.41 and type I for S=0.57. The maximum values occurred at around r/d=0.44 for S=0.41 whilst for S=0.57 this was near r/d=0.9, although both maxima were comparable 167

168 4.3 RESULTS 100 burning fraction [%] Burning Not burning r/d 100 burning fraction [%] r/d Figure 4.15 Radial profiles of the fraction of burning particles, measured at z/d=2.67, for (a) S=0.41 and (b) S=0.57. at 95% and 90% respectively. For the non-penetrated S=0.41 flow, the centreline value for the burning fraction was about 85% but less than 5% (depicted as 5%) burning particles were found near the centreline for S=0.57. In the latter case burning of the particulate phase was confined in the region between r/d=0.44 and r/d=0.90, where a weak torroidal recirculation zone existed and the absence of burning particles on the centreline was correlated with the location of positive particle axial velocities presented earlier. 168

169 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES 0.4 Burning S= Ensemble (a) r/d= (b) r/d= S= (c) Probability r/d= (d) r/d= (e) r/d= velocity [m/s] velocity [m/s] Figure 4.16 Comparison between velocity pfs for burning particles and the ensemble, measured at z/d=2.67 for (a, b) S=0.57 and (c-e) S=0.41 at the following locations: (c) r/d=0, (d)=0.22, (a, e) r/d=0.44 and (b) r/d=0.67. The velocity pf of burning particles compared to that of the ensemble is presented in figure 4.16 as a function of the swirl number. The velocity pfs are not presented for S=0.57 at r/d<0.44, because only few burning particles were found there, whilst for the radial locations which are 169

170 4.3 RESULTS Burning Ensemble r/d= r/d= Probability r/d= r/d= Size [µm] r/d= Size [µm] Figure 4.17 Radial profiles of the measured size distribution of burning particles and the ensemble, measured for S=0.41 at z/d=2.67. not presented in figure 4.16 there was only minor difference between burning particle pf and the ensemble, implying that the probability of encountering a burning particle for those locations was independent of particle velocity. The question of whether the local burning fraction for S=0.41 at r/d>0.44, and S=0.57 at r/d>0.67, depended on particle size is deferred to figure In both flows the velocity pf of burning particles that corresponded to the measurement location closest to the centreline (i.e. r/d=0 for S=0.41 and r/d=0.44 for S=0.57) was bimodal, 170

171 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES with one peak corresponding to reversed particle motion and the other to forward motion: the probability of encountering a burning particle with reverse velocity was the higher. It must be mentioned here that, due to the use of a linear instead of a two-dimensional photodetector array in the SDV instrument, particles travelling with velocities close to zero were rejected during measurement, thus resulting in the pronounced bimodality in the velocity pf of figure 4.16 (see Chapter 2). The absence of burning particles with positive velocities in, for example, figures 4.16(a) and (c), is nevertheless independent of this bimodality and is attributed to the flow rather than to the shortcoming of the particular detector arrangement. The differences between burning particles and the ensemble were smoothened out closer to the shear layer between the recirculation zone and the main forward flow, as indicated by the increasingly similar velocity pfs between burning particles and the ensemble with increasing radial distance. The radial profile of the measured size distribution is presented for swirl number S=0.41 in figure 4.17 for burning particles and the ensemble, at z/d=2.67. Two observations can be made. The size pf for the ensemble at r/d=0 and 0.22 includes a substantial number of particles larger than 60 µm which were not burning. For locations beyond r/d=0.44 the differences between burning particles and the ensemble became increasingly smaller and were confined to the small sizes of about 10 µm, which was the smallest particle size that could be resolved owing to the fixed magnification ratio of the collection optics. The existence of non-burning particles larger than 60 µm near the centreline was due to insufficiently large heat-up rates as a consequence of, first, the local positive mean velocities which resulted in smaller residence times within the high-temperature recirculation zone than for smaller particles and, secondly, their larger thermal inertia with accordingly longer heat-up times. The smaller number of small (<25 µm) burning particles compared with the number of small particles in the ensemble is not a feature of the flow and is attributed to the detectability limits of the present pyrometer. As shown in figure 4.17, with the exception of particles larger than 60 µm near the centreline of the burner for S=0.41, the size distribution of burning particles across a radial profile remains approximately constant. Radial profiles of the char temperature pf of burning particles averaged over all size classes is presented in figure 4.18, for S=0.41 and at z/d=2.67 and The measurements also showed that there was only weak size-temperature correlation (inferred from scatter plots, see, for example, figure 4.24) and it is therefore reasonable to construct temperature pfs from the entire batch of burning char particles. As figure 4.18 shows, the temperature pfs are quasi-gaussianly distributed with an approximately constant mean temperature of about 1900 K across a radial profile. The calculations of Lau and Niksa (1992) have shown that char particles of a given size burn at almost a constant temperature throughout their burnout and, hence, it is likely that the width of the pfs of figure 4.18 were a result of size-dependent particle trajectories and accordingly, variation in particle heat-up history and local oxygen concentration during particle flight in the turbulent recirculating flow. This explanation is plausible in view of published measurements of particle ignition and burn 171

172 4.3 RESULTS 0.2 z/d= z/d= r/d= r/d= Probability r/d= r/d= r/d= Temperature [K] Temperature [K] Figure 4.18 Radial profiles of the measured temperature distribution of burning char particles, measured for S=0.41 at (a) z/d=2.67 and (b) z/d=3.05. temperature in drop-tube furnaces (e.g. Tichenor et al. 1984) were a clear correlation between particle size and temperature was found. Moreover, of the heat losses from a particle in the flow, which are due to radiation and convection, the latter is more important in a confined flow for the particle size ranges investigated here (Wall 1987). According to the calculations of Wall (1987) convection losses are about 6 times larger than radiation losses, for a 30µm particle. The relative importance of radiation is expected to be higher than of convection in the current 25 Assuming, of course, that in no case the trajectory results in high local particle concentration and, thus, radiative heat transfer between particles, which is a reasonable assumption on the basis of the typical particle concentrations measured in the current flows. Another assumption is that the flame is transparent, which in not necessarily true, because there are species in the flow, such as CO, CO 2 and H 2 O, which absorb radiation. 172

173 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES open-flow system, but nevertheless constant for any particle trajectory. 25 On the contrary, convective heat transfer depends on the local velocity, i.e. the Nusselt number, as well as temperature gradient between particles and carrier gas, and thus, it is probably the one which controls heat-up history, with the contribution, of course, of the local oxygen partial pressure which affects the reaction rate. There are however some differences between measurements at z/d=2.67 and z/d=3.05. At z/d=3.05, the median temperature on the centreline was higher than at r/d=0.89 by about 200K and the pf was narrower than the respective at r/d=0.89. The lower temperatures at r/d=0.89 were probably due to higher convection losses in the shear layer, formed between the recirculation zone and the forward flow, caused by entrainment of lower-temperature air from the secondary stream into the shear layer. Despite the fact that the median particle temperatures in the shear layer were lower than on the centreline, it is likely that NO x generation is higher at the shear layer, owing to the influence of higher local oxygen concentration there (Miller and Bowman 1989). Effects of Momentum and of Gas Equivalence Ratio The effect of the gas equivalence ratio and the primary to secondary air momentum ratio on the burning pattern of coal particles in the near-burner region of the swirl burner has been quantified from radial profiles of the velocity, size and temperature of particles at axial distances z/d=2.67 and z/d=4. Gas flames with overall equivalence ratios of 0.69 and unity were selected to result in recirculation zones with different oxygen mole fractions and thus simulate coal burner firing with different degrees of staging. The momentum ratio, which is a burner design parameter, was varied from 1/30 to 1/40, conditions similar to previous investigations of Orfanoudakis and Taylor (1995). These two flows are referred to as flows 4 and 5 in Table 4.3. Results are presented below for the effect of the gas equivalence ratio and the momentum ratio on the particle burning fraction. The volume flux of burning particles and the ensemble 26, the mean and rms char temperature and size-velocity correlations as a function of the equivalence ratio for axial stations of z/d=2.67 and 4 are also presented. Before describing the results for the particulate phase, the variation of the mean gas temperature as a function of the gas equivalence ratio will be presented. Figure 4.19 shows the mean gas flow temperature for S=0.41 and equivalence ratios φ=0.69 and 1.0, at z/d=2.67, measured by a 250 µm platinum-platinum/13%-rhodium thermocouple in the absence of particles. No correction for radiation losses was applied to the measurements. As shown in figure 4.19, the result of the 45% increase of the fuel equivalence ratio was a small increase of the maximum mean gas temperature from 1800 to 1900 K, which occurred at about r/d=0.8. The centreline mean temperature was the same in both cases and both flows exhibited the same well-stirred reactor-like behaviour where the temperature was almost constant across the radial profile and up to about r/d=0.9. Although no measurements of the mean gas temperature were measured 173

174 4.3 RESULTS T [K] φ=0.69 φ= r/d Figure 4.19 Radial profiles of the mean temperature of the gaseous flame in the absence of particles as a function of the overall gas equivalence ratio f. Symbols: (í) f=0.69 and (ó) f=1.0. for axial distances greater than z/d=2.67, air entrainment due to the swirling flow is expected to reduce the mean gas temperature, but this reduction is likely to be insignificant (smaller than 100 K) for 2.67<z/D<4 where measurements for the particulate phase are reported below. In addition, the change of the momentum ratio from 1/30 to 1/40 is not expected to have any effect on the mean gas temperatures through alteration of the overall equivalence ratio 27 because the mass of primary air is about 3% of the secondary. For the swirl number of 0.41, even for the higher measured momentum ratio (namely 1/30) no transition from flame type II (nonpenetrated) to type I (penetrated) was observed, clearly shown in COLOUR PLATE 8, where the brightest region appears to occur inside the quarl, compared with PLATE 7, where it manifests downstream of the quarl, a characteristic of type I flame. This means that the results presented in the next paragraph cannot be attributed to flame type transition, contrary to the results of the previous section on the influence of the swirl number. Figure 4.20 shows radial profiles of the burning fraction of particles calculated from equation (4.7) as a function of the equivalence ratio, at z/d=2.67 and z/d=4. In the case of φ=0.69 the maximum occurred around r/d=0.5 and was about 95% whilst the centreline value was slightly lower. From radial locations larger than r/d=0.6 the number of burning particles dropped abruptly and beyond r/d=1 no burning particles were found. Under the same conditions which includes the entire particle sample at each point of measurement, whether they were measured as burning or not. 27 The reader is reminded that the discussion concerns non-premixed flames and, thus, the equivalence ratio does not affect flame temperature, as in premixed flames. It nevertheless affects the mean gas temperature, because for decreasing overall equivalence ratio the same amount of released heat is distributed over larger gas mass.

175 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES 100 z/d= z/d=2.67 burning fraction [%] burning fraction [%] r/d z/d=4 100 r/d z/d=4 burning fraction [%] burning fraction [%] r/d r/d Figure 4.20 Radial profiles of the fraction of burning particles, measured at z/d=2.67 and 4, for gas equivalence ratios f=0.69 (left) and 1.0 (right) and momentum ratio MR=1/30. measurements further downstream, at z/d=4, showed a broader distribution of high proportion of burning particles and, for example, 80% of those around r/d=1 were already burning. Moreover, burning fractions of almost 100% were measured near the centreline and it can be safely assumed that all particles which were not transported from the gas flow away from the recirculation zone, and at z/d=4 were measured close to the centreline were already burning. Comparison shows that the burning fraction generally increased with downstream distance and this is explained by the fact that there was sufficient residence time in high temperature regions to allow most particles to start burning by z/d=4.0. A similar trend was observed for φ=1.0, although the burning fractions were generally lower as compared with the lower equivalence ratio, and, for example, the increase of φ from 0.69 to 1.0 resulted in reduction of the near-centreline values from 90-95% to 60-70%. This is initially surprising but is an effect of reduced O 2 concentration. 28 Farther downstream, the proportion of burning particles in 175

176 4.3 RESULTS the case of φ=1.0 increased and reached 100% on the centreline, but the region of burning fraction higher than, say, 50% was narrower than for φ=0.69. It is likely that the decrease was a consequence of smaller oxygen concentration in the recirculation zone owing to the use of stoichiometric combustion. Figure 4.21 shows the effect on the burning fraction of decreased primary to secondary air momentum ratio from 1/30 to 1/40, achieved by reduction of the primary air supply, which did not have any influence on the gas equivalence ratio because the amount of primary air was smaller than 3% of that of the secondary, by use of which the equivalence ratio was calculated. A 75-80% reduction was measured at z/d=2.67 but the profile was measured only at one axial station because the small burning fraction made the measurement difficult as more than one hour per location in profile was necessary to collect sufficient data for its calculation. This initially surprising result was confirmed by visual observation of the brightness of the flame, which decreased substantially relative to the momentum ratio of 1/30 (flow condition 2 of Table 4.3). 29 As a result of this, measurement at z/d=4 was abandoned, particularly after test measurement on the centreline showed no increase in the number of burning particles relative to the results of This result exemplifies the importance of the high-temperature recirculation zone on particle ignition. As shown in the previous section, reduction of the momentum ratio results in increased particle volume flux at radial distances r/d larger than about 0.6 due to particle interaction with the tangential fluid motion which yields decreased particle residence times near high temperature regions and, hence, reduced probability for ignition up to z/d=2.67. The fact that most particles escape to the shear layer implies that there is a higher possibility that these particles will yield higher NO x emissions than with particles travelling through the recirculation bubble. In view of the low particle burning fraction in this case, a consequence of their motion away from the hot regions (an effect which is not expected to be as pronounced in full-scale burners where mean flow time scales are an orderof-magnitude larger), particles react with the secondary air in the shear layer between the recirculation zone and main forward flow, and thus nitrogen-bound species are likely to be converted to NO x. The burning fraction gives the number of burning particles amongst all counts regardless of the particle size and, hence, does not take into account the burning mass on which the total volatile production and, thus, heat release depends. The burning mass can be represented by The influence of O 2 concentration in the recirculation zone on char particle ignition is theoretically addressed in the discussion of Chapter As explained in the first footnote of Table 4.3, there is no photograph of condition 5. If one compares COLOUR PLATE 11, which is photograph of a flame of swirl number 0.57 and all other conditions as in flow 5, and PLATE 7, which is photograph of flame of condition 1, one observes that the brightest region of the flame in PLATE 11, moved upstream towards the quarl. The latter was a result of smaller penetration of the recirculation zone by the incoming particle-laden jet, which is a phenomenon similar to the one measured here in flow condition 5, i.e. the reduction of the burning fraction due to smaller penetration of the recirculation zone by the incoming jet.

177 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES 100 z/d=2.67 burning fraction [%] Burning Not burning burning fraction [%] r/d r/d Burning Not burning Figure 4.21 Radial profiles of the fraction of burning particles, measured at z/d=2.67, for gas equivalence ratio f=0.69 and momentum ratios MR=1/30 (top) and 1/40 (bottom). the mass flux or here volume flux if we omit particle density- of burning particles, estimated according to equation (4.5). Figure 4.22 presents radial profiles of the (axial) volume flux of burning pyrometer-detected particles and of the ensemble, which was defined as the total of burning and non-burning particles, for φ=0.69 and 1.0, momentum ratio 1/30 and at z/d=2.67 and 4. Four different curves are presented in each graph that correspond to the ensemble of the particles, burning particles, positive and net flux respectively. Positive flux was defined as flux of particles with positive axial velocities only whereas net flux contained all particles, regardless of the sign of the axial velocity component. At z/d=2.67, in the region 0<r/D<0.5, the net particle flux corresponded to reverse flow for both equivalence ratios and the difference between burningand ensemble-flux was small, which is the result of figure 4.20 where burning fractions close to 100% were observed in the region. In the case of φ=1 at z/d=4 the difference between burning- and ensemble-flux increased with increasing radial distance for r/d>0.5, thus implying that the bulk of particulate mass was carried downstream by the forward flow at r/d>

178 4.3 RESULTS Flux [Arbitrary units] Flux [Arbitrary units] positive, ensemble net, ensemble positive, burning net, burning No particles r/d r/d z/d= Flux [Arbitrary units] Flux [Arbitrary units] z/d=4 r/d r/d Figure 4.22 Radial profiles of the particle volume flux, measured at z/d=2.67 and 4, for gaseous equivalence ratios 0.69 (left) and unity (right). Positive flux takes particles with positive velocities into account only. The ensemble consisted of all particles, whether pyrometer-detectable burning or not. Symbols: (à) positive flux of the ensemble, (ò) net, (á) positive flux of burning particles only, (ó) net. without burning. Similar observations can be made for the rest of the profiles, with the exception of φ=0.69 at z/d=4 where this observation is not so pronounced. The discrimination criterion for particle temperature described in Chapter 3 was applied to those measurements corresponding to burning particles and counts which qualified as incandescent char were used to construct mean and rms temperature profiles for φ=0.69 and 1.0. Figure 4.23 presents the axial development of the radial profiles of char temperature. The temperatures were conditionally averaged according to their measured diameter into 12-24, and µm classes to investigate whether there was a correlation between particle diameter and char temperature. Despite the fact that the mean temperature radial profiles are slightly noisy, particularly for the µm sizes because of the smaller number of particles in this class in the parent fuel, no particular size-char temperature correlation exists in figure 178

179 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES z/d=2.67 z/d=3.05 <T> [K] <T> [K] µm µm µm φ= T [K] 100 T [K] z/d= z/d=4.00 <T> [K] <T> [K] φ= T [K] 100 T [K] r/d r/d Figure 4.23 Radial profiles of the mean and the rms temperature of incandescent char particles, measured at z/d=2.67 and 3.05 for gas equivalence ratio f=0.69, and at z/d=2.67 and 4 for f=1. The temperatures are plotted for three particle size classes, namely 12-24, and mm. Symbols: (ç) µm, (á) µm and (ó) µm. 179

180 4.3 RESULTS r/d= Temperature [K] r/d=0.22 r/d=0.44 Temperature [K] r/d= r/d= Size [µm] Size [µm] φ=0.69 φ=1.0 Figure 4.24 Scatter plots of the size of the incandescent char particle as a function of its instantaneous temperature, measured across a radial profile at z/d=2.67, for gas equivalence ratios of 0.69 (left) and 1.0 (right). 180

181 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES 4.23; this observation will be emphasised in figure The mean profiles nevertheless show that char temperature decreased with increasing radial distance although the mean variation across the profile was only about 100 K in all cases, except for φ=1.0 at z/d=4 where it was about 200 K. Lower char temperatures are expected as we move radially away from the highgas temperature regions and the small variation across the profile was a result of the detectability limits of the pyrometer which measured only particles already burning and not those in the early stages, i.e. inside the quarl, of heat-up. The rms char temperature profiles indicate an increase with increasing radial distance although this is not necessarily a feature of the gaseous flow; it rather reflects the fact that particles away from the recirculation zone are likely to have larger temperature variation owing to their different heat-up histories. Although turbulent dispersion is expected to be the source of variation in the heat-up histories, size-char temperature correlations, presented in figure 4.24 in the form of scatter plots as a function of the equivalence ratio, indicate that heat-up is a size-independent phenomenon in this burner for the chosen size-ranges of the coal batch. The detailed size-velocity results of (subsection particle aerodynamics in open flames) have already shown that with the exception of the near centreline region, where smaller particles tended to have velocities closer to that of the gaseous phase, particle velocity did not vary with diameter for the present swirl number. This result, which can be justified on the grounds of the Stokes numbers of the particles, is also representative of a larger scale burner where all particles are expected to follow the gas flow. In figure 4.24, for φ=0.69, for example, the calculated correlation coefficient was only of the order of 0.05 on average and, on the centreline, larger particles tend to have smaller temperatures than smaller particles. This is expected in view of the results of the aforementioned section where the mean axial velocity of the 60 µm particles indicated that those particles moved in the positive axial direction, in contrast to the smaller sizes. From the observations in figures it is inferred that in a full-scale burner, where particles would be expected to follow the instantaneous gas flow, we would expect no sizetemperature correlation or small char temperature variation across a radial profile - taking into account the detectability limits of the present pyrometer. Hence, the most important variable which would modify NO x production and, therefore, result in a more environmental-friendly combustion system, is the spatial distribution of the volume (i.e. mass) flux distribution The Far-Burner Region Experimental Method In this part of Chapter 4 the flow in the downstream region of the metal furnace is investigated. The two-phase flow in this region is of particular interest in pressurised furnaces designed to operate in a combined cycle where the hot, particle-laden gases are directed to the inlet of a 181

182 4.3 RESULTS gas-turbine to increase cycle efficiency. As explained in the introduction, for such a cycle to be feasible, the particulate matter must be minimal, to avoid damage of the blades. Although such facilities are to-date mostly in the experimental stage, understanding of the flow near the exit of such furnaces can provide evidence of the mechanisms which can potentially yield clean hot gases from the furnace. From the point of view of the end-user of the hot gases, the most common methods for particulate removal from gases, i.e. electrostatic precipitators and ceramic filters, have a ceiling in the maximum operable temperature which is lower than the maximum permissible gas-turbine inlet temperature. The feasibility of combined cycles that incorporate pressurised combustion of pulverised fuel (including perhaps exhaust gas recirculation, see Chapter 5) and gas-turbine topping cycle thus depends on the ability of the furnace to detain the particles before they enter the turbine and cause serious corrosion on the blades. The flow in the downstream region of the furnace, where the exit port is situated, certainly depends on the upstream region, already investigated, and it was found that the size-dependent particle velocity is a function of the upstream flow boundary conditions and not of the presence of absence of the surrounding walls for the flows presented earlier in this chapter. In the following text the particulars of the flow will be presented because, as already explained at the beginning of the chapter, there were slight differences between the burner and the furnace results which have already been described in In the rest of section 4.3.2, the experimental findings will be presented. Discussion of the mechanisms thought to be responsible for particle retention, based on particle centrifuging, is offered in 4.4. Flow Details The measurements of this section were taken in the metal furnace, described in The hot gases exhausted through a lateral exit near the bottom of the furnace and above the slag granulating bath. Our laboratory model, shown in figure 4.4, was fired upwards at atmospheric pressures. This orientation was chosen as more convenient for the optical setup without any compromise of the fluid mechanics of the particulate phase, as we investigate the effect of inertial centrifuging which is independent of the direction of gravity, as will be discussed in 4.4. Provision for optical windows at two different areas was made to allow optical measurements close to the exit to study the mean streamline curvature effects on particle dispersion and also within the main body of the furnace. The air flow in the burner was turbulent, based on the Reynolds number of in the annulus. The swirl number, calculated for cold flow according to Milosavljevic (1993), was the smallest for which a stable flame was possible and was chosen to minimise the mean centrifuging effects in the vicinity of the burner. The natural gas flow rate corresponded to an equivalence ratio of 0.62 which was close to the lean limit of this burner; this flow rate minimised 182

183 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES high furnace wall temperatures and hence reduced the thermal radiation to the adjacent optical equipment. The coal was high bituminous Bentinck coal 30 and contained particles 99% smaller than 70µm by mass. The coal mass flow rate of 2.5 g/min was about 1% of the total air mass flow rate and corresponded to a coal equivalence ratio of 0.07, and the effect of the dispersed phase on air momentum was therefore negligible. The experiment was reproducible within the 5% accuracy of the calibrated rotameters. The regime of operation was convenient both for experimental reasons, explained above, and because it permits assessment of many fundamental aspects of CFD calculations without the complications of two way mass, momentum and energy interaction between the two phases. Measurement Strategy The prototype SDV ( 4.2.5) used in this experiment made use A/D converters and hence had slow response to velocity vector changes compared with the final D/A version (Chapter 2). Particles generated no useful sizing signal when their trajectory lay parallel to the axis of the linear array and sizing accuracy increases as the trajectory angle approaches the normal to the axis. The results of Hardalupas et al. (1994) showed that the error for trajectory angles about ±30 to the normal to the array axis is about 7% for velocities of the order of 1 m/s and it is not expected to exceed 10% for the velocities measured in this work, provided that particles at large (see below) out-of-focus distances are rejected. For smaller trajectory angles this error falls below 5%. In cases of simple, one-dimensional flows, the implied necessary orientation between the linear array and the (mean) flow direction is easily achieved. In the present experiment, existing LDV data on the gaseous phase measured at window 2 on the meridional symmetry plane of the furnace, and shown in figure 4.29, indicated that the velocity field was strongly two-dimensional. Thus, for the use of the prototype SDV with the particulate phase, the measurement region was divided into two sub-regions, for each of which a predominant flow direction could be defined and which would reduce the resulting errors. Hence, the photodiode array axis was aligned at 45 relative to the horizontal plane for the lower subregion (Geometry B in figure 4.25a) and at 90 (Geometry A in the same figure) for the upper sub-region. 31 Provided that the trajectory lies within the approximate angle range stated above, the SDV signal can also yield the angle of the trajectory relative to the axis of the photodiode array (Morikita et al. 1995). By analysing the measured velocity onto a local cartesian system, it was 30 although the same coal type was used for all experiments in this thesis, see Orfanoudakis 1994 for details on the size distribution of the batch used in this section. 31 It is reminded that due to A/D hardware implementation in the prototype instead of the D/A of the final version the response of the instrument was slower which necessitated use of the measurement strategy described here. 183

184 4.3 RESULTS possible to derive two-dimensional velocity information from an effectively single channel one-dimensional velocimeter with a tolerance not exceeding 10% in the mean, and 15% in the rms, velocity and less than 10% in the angle estimation. Following figure 4.25b, the instantaneous velocity components u, v in the local cartesian co-ordinate system were obtained from: u = V v = V m m sinψ + V m cosψ V m tanφcosψ tanφsinψ (4.8) where V m is the component of particle velocity measured by the SDV system, y is the inclination angle of the photodiode array relative to the axis of v velocity in figure 4.4, and φ is the trajectory angle in the x-y system, where x was aligned with the horizontal orientation of figure 4.25, obtained from processing the SDV image data. In order to minimise the error from sources other than the large angle, i.e. φ, of the velocity vector relative to the photodiode array only particles with out-of-focus distances between 400µm and 100µm from the centre of the LDV probe volume were accepted. During the course of the measurement, the windows became soiled owing to particle accretion. Despite the low coal loading used, the transmitted laser beam light intensity had fallen to 10% of its original value, and the size validation rate from 70% to 20%, within 15 minutes of continuous operation of the furnace. To avoid sizing errors because of this effect, the data acquisition software stopped automatically when the reference level, which corresponded to measurement of the intensity of the image of the LDV control volume in the absence of particles, had fallen below 75% of its original value: new reference levels were established by suspending the coal and natural gas flows. The procedure of taking new reference levels was continued until the transmission had fallen to 10% of its original value. At this point, it was necessary to interrupt the experiment and to remove the windows for cleaning. The slow measurement rate was also due, in part, to the low coal particle data rate, being about 10 particles/min. Because window fouling caused the mean data rate to fall with time, we limited the measurement time at each point to around 1 hour. Typically, this allowed a total of three batches to be measured, providing a sample size of 300 velocity validated data. The statistical error was 8% in the mean and 10% in the rms particle velocity for the entire range of particles, but the maximum error for only the µm size class of particles reached 22% and 30% respectively (Yanta and Smith 1978) because of the small number of samples of these particles. It should be noted that less than 30% of the 300 data were due to out-of-focus particles and these were rejected for size, and size-velocity, statistics and scatter plots presented below and only data which was both velocity- and out-of-focus-validated has been taken into account in the calculation of the measurement uncertainties mentioned earlier. 184

185 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES Exhaust Pipe Window 2 z 90 (geometry A) ψ 45 (geometry B) (a) normal to axis of array particle velocity vector Vm ψ φ v vertical u Axis of photodiode array Vm tanφ horizontal (b) Figure 4.25 (a) Alignment of the SDV photodiode array in the two adjacent sub-regions. Note that the arrows indicate alignment of the normal to the axis of the photodiode array; (b) Cartesian components u, v of measured particle velocity when the normal to the photodiode array is aligned at y and the velocity vector is at angle f to the normal. 185

186 4.3 RESULTS Results First, figure 4.26 shows correlations of V m with diameter, diameter with φ, V m with φ and u with v by means of scatter plots: note that these plots contain less than 300 data points because many measurements were rejected in the post-processing owing to the out-of-focus limits mentioned earlier. This, though, does not affect the velocity measurements presented below in figure Figures 4.26a and 4.26b are, respectively, at the extreme locations of the measured vertical profile, namely z/r o = -1 and z/r o =+1, where R o is the exhaust pipe radius and z is measured along the vertical direction normal to the exit duct axis with origin in the horizontal plane containing the centreline of the exit duct, as shown in figure 4.25a. After taking account of the small sample size, the results do not indicate any strong correlation between velocity magnitude and particle size, V m and trajectory angle or diameter with trajectory angle at either location. This would have to occur if centrifuging had taken place, in which case the trajectory angle would increase with increasing particle size, indicating larger inclination of the velocity vector away from the axis with increasing size. The u-v correlation plots are different for each location, for reasons which become apparent below. The calculated cartesian ensemble mean and rms u and v components of the measured particle velocity, normalised by the bulk velocity in the main body of the furnace U 1b and using for convenience air density at 300 K, are presented in figures 4.27(a,b) and 4.27(c,d), respectively. 32 Three size classes, namely µm, µm and µm were chosen, see 4.2.5, to characterise the range of sizes available in the coal, taking into account that the total number of measurements is only 300 and that more size classes would lead to unacceptably large statistical errors. In the same plots, air velocities are also plotted for reference which were measured by using aluminium oxide particles as seed introduced into the secondary air and based on 5000 samples so that the statistical accuracy is greater than that of the coal. The ensemble mean profiles for both u and v components and all size classes almost collapse onto a single curve, taking into account the experimental uncertainty, although the particle velocities are systematically slightly smaller that those of the air. This indicates that, in the mean at least, all three particle size classes follow the air motion closely and that there is no evidence of mean centrifuging. The scatter on the particle rms velocity profiles is due to the limited number of measurements in each size class but show a weak trend for the rms velocity to be lower than that of the air and to decrease with increasing size. Figure 4.28 shows the particle size distribution, or probability function, at three z locations. The diameter probability function of each diameter, p(d i ) was calculated as: 32 U 2 i = V & / d /4, where V & was 300 lt/min and d i the diameter of the furnace. 1b π 186

187 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES Diameter [µm] Trajectory Angle [deg] Trajectory Angle [deg] v [m/s] Diameter [µm] Trajectory Angle [deg] Trajectory Angle [deg] v [m/s] (a) (b) Figure 4.26 (a) Scatter plots of V m vs. diameter, diameter vs. f, V m vs. f and u vs. v at z/r o =-1 and 34.5 mm upstream of the exit, and (b), as in (a), but at z/r o =1. 187

188 4.3 RESULTS air µm µm µm u v z/ro 0 air µm µm µm z/ro (a) (c) air µm µm µm z/ro air µm µm µm z/ro (b) d) Figure 4.27 (a) Mean, and (b) rms axial velocity component, u, of particle as compared with that of air for three different particle diameter classes 12-24, and µm at 34.5 mm upstream of the exit; (c) and (d) are mean and rms, respectively, for the radial velocity component, v, at the same location and for the same particle diameter classes. Symbols: (ç) µm, (æ) µm and (õ) µm and (ì) gas phase. p(d ) = i ni A i ni A i (4.9) where n i was the number of measurements in size class i and A i is the effective cross-section of the SDV probe volume for size class i, and d i the mean diameter of size class i. 188

189 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES z/ro=1 z/ro=0 z/ro= Diameter [µm] Figure 4.28 Particle size distribution at three vertical locations at 34.5 mm upstream of the exit. Symbols: (ç) z/r 0 =1, (æ) z/r 0 =0 and (õ) z/r 0 =-1. Further evidence for the absence of mean centrifuging is that the calculated arithmetic and Sauter mean diameters for z/ro = -1, 0, 1 are all equal to 33 and 40µm, respectively, whereas if centrifuging had occurred, it would be reasonable to suppose that the mean diameter should increase with increasing z. The velocity profiles, presented earlier, are rearranged as mean velocity vector plots for both air and the ensemble of the particles in figure It is evident that, apart from minor, local differences in the velocities, the particles follow the gas flow closely. Once again, had centrifuging occurred, it would be expected that there would be deviation away from the air flow and that particles would be directed away from the exit, and particularly towards the upper part of the furnace. Whether this is reasonable on theoretical grounds is discussed in Discussion: Particle Centrifuging In Chapter 1 it was concluded from analysis of the literature that it is desired that particles remain as long as possible inside the swirl-induced recirculation zone. The latter provides an oxygen-deficient high-temperature environment where particles can devolatilise with reduced possibility of oxidation of the nitrogen-containing volatiles into NO x. On the other hand, in pressurised combustion, particles which are centrifuged away from the flame zone are likely to hit the furnace walls and form a slag layer. This process is, depending on furnace design, beneficial as it forms a layer which protects against particle erosion and can result in clean-up of the gas flow from particles and thus, low particulate emissions. In this section an attempt is 189

190 4.4 DISCUSSION PARTICLE CENTRIFUGING 1.0 [m/s] 1.0 [m/s] (a) Air (b) Particle Figure 4.29 Vertical profile of (a) air and (b) particle ensemble velocity vectors at 34.5 mm upstream of the exit as measured by LDV and SDV, respectively. made to justify theoretically the measurements presented so far for both the near- and the farburner regions and identify the mechanisms which can explain particle centrifuging in the region the swirl-induced recirculation zone condition which is unfavourable for lower NO x emissions- and also help increase the particle retention efficiency of a pressurised furnace, where pulverised coal is used as fuel Near Burner Region It was earlier ( 4.3.1) assumed that in the near-burner region the increasing rms velocity fluctuations and the increasing mean trajectory angle as a function of particle size were a result of particle centrifuging due to the gas flow. In the region of the recirculation zone, the length scale of the large eddies is comparable to the size of the recirculation zone and for burner II is, therefore, around l =1.2 D» 22 mm. According to the velocity profiles of figure 4.8, the convection velocity of these eddies can be approximated by the mean gas velocity inside the recirculation zone, which is U=0.2 U 0» 5 m/s. Therefore, the time scale of the large eddies is 4.5 ms. The response of a 60 µm particle to these eddies is then characterised by a 190

191 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES corresponding large eddies Stokes number (see equation 4.14 below) which is approximately unity. It is well documented by Crowe et al. (1993) in plane shear layers, for example, that particles with response time comparable to the time scale of the vortices formed in the mixing region of a shear layer can be centrifuged by those vortices. The consequence of mean, and turbulent, centrifuging due to the recirculation zone is that particle escape from it and, thus, their residence time inside the recirculation zone is reduced which is detrimental for NO x emissions, according to calculation results presented by Abbas et al. (1994) and Smart et al. (1988) Far Burner Region In this section a theoretical analysis based on Stokes numbers, and extended to include the effects of gravity (via the drift parameter), is offered in order to explain the absence of particle centrifuging in the far-burner region of the confined flame. In addition to the use of the large eddies Stokes number for the description of particle response to turbulence (source of centrifuging according to Hardalupas et al. 1992), particle centrifuging can be characterised by a centrifuge Stokes number which describes the fidelity of the response of the particles to the mean curvature of the streamlines as these approach, for example, the exit of the furnace (in this case the metal furnace of figure 4.4). We use the centrifuge Stokes number defined by Hardalupas et al. (1992), where we adopt the definition of the Stokes number of Clift et al. (1978) which has the particle response time, τ p, in the denominator (see also 4.14): St ω 18µ = ρ ωd p g 2 p (4.10) where µ g, ρ p and d p are the exhaust gas dynamic viscosity, coal particle density (taken from Orfanoudakis, 1994) and particle diameter, and ω is a suitably-defined angular velocity: ω = R U ω ω (4.11) where R ω is the characteristic radius of the streamline and is here taken as the radius of the main body of the metal furnace. The magnitude of velocity U ω is taken from the vector at z/r o =0 in figure 4.29, namely 4 m/s. By applying the above relations we estimate centrifuge Stokes numbers of 38, 10 and 3 for particles of the average diameter of the µm,

192 4.4 DISCUSSION PARTICLE CENTRIFUGING µm and µm size classes, respectively. For all three size classes this Stokes number is greater than unity, which implies that it is expected that particles up to 70 µm will follow the streamline, without centrifuging. Gravity is not expected to affect particle motion near the exit port at this scale, as can be assessed by means of the drift parameter, defined as (Siu 1996) U γ = U T f (4.12) where U T is the particle terminal velocity, defined as: U = τ g p T (4.13) τ p was the particle response time and U f was taken (see above) as 4 m/s. The estimated drift parameters for the 12-24, and µm particles are 0.001, and respectively near the exit port. The small values of the drift parameter imply that it is unlikely that forces due to gravity will affect particle motion implying that it is irrelevant whether the furnace is fired vertically upwards or downwards or, hence, particle centrifuging as particles approach the exit. However, the retention of particles to form a continuous slag layer on the furnace walls is a primary requirement in some designs. This layer subsequently drains as a Couette flow under the action of gravity. Reichert et al. (1990) have measured the slag retention capacity of their scale model (see Flow Details in 4.3.2) and found that it is up to 95% for an air throughput of about 500 m 3 /h. It is then necessary to argue what mechanism might promote this, in the light of our measurements. There are two potential mechanisms for slag retention. The first, mechanism A, is by particle turbulent dispersion across the external recirculation zone, formed between the flame and the furnace liner, towards the furnace wall and subsequent formation of a slag layer on the wall in the upper part of the furnace. This mechanism is effective when particles can respond to the large eddies which are responsible for particle dispersion in the recirculation zone, provided that the particle time-of-flight along the recirculation is long enough to allow interaction between particles and large eddies and that the crossing trajectories effect (Yudine 1959) is insignificant. Whether particles will disperse into the recirculation zone can be assessed by the large eddies- and the transit- Stokes numbers respectively. Both the large eddies- and the transit- Stokes number are defined as (Hardalupas et al. 1992) 192

193 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES St e Tf = τ p (4.14) In the case of the large eddies Stokes number, St e, T f is the time scale of the large eddies, estimated following Milosavljevic (1993) from: T f l = 0.3U 0q (4.15) where l is the length scale of the large eddies and U 0q is the air bulk velocity at the exit of the burner upstream of the quarl diffuser, whereas in the case of the transit Stokes number, St tr, T f is the particle time-of-flight along the recirculation zone. Reichert et al. (1990) mention that fine pulverisation of the fuel is necessary in order to reduce the unburned fraction owing to the smaller residence times of a particle in the furnace compared with a conventional atmospheric-pressure combust (a consequence of the smaller size of a Pressurised Pulverised Coal Combustion PPCC system) hence we use 50 µm as the representative particle diameter for the present calculations. From data presented in Reichert et al. (1990) one can estimate that the internal diameter of the furnace is of the order of 1 m, the gas bulk velocity in the upper half of the furnace (i.e. near the ash hopper) is 0.05 m/s whereas in the lowest it is 0.03 m/s owing to the lower temperatures, U 0 is 7.6 m/s assuming that the furnace to burner diameter ratio in practice is 10:1 (Orfanoudakis 1994) and the length scale of the large eddies l is taken as m following Milosavljevic (1993). Using this data, we estimate that the time scale of the large eddies is 16 ms and the time-of-flight along the recirculation zone is 220 ms respectively, assuming that particles are injected with velocities similar to the bulk velocity U 0, the recirculation zone extends down to the slag drip ring which is located at about the mid-height of the furnace, therefore the length of the recirculation zone is 1.7 m and the particle response time is 3.3 ms. Hence, St e»5 and St tr»70. Because they are both larger than one, particles will therefore disperse into the external recirculation zone and eventually form a slag layer at the wall. Gravity does not affect particle dispersion in this region, and therefore does not interfere with slag formation mechanism A, because the drift parameter g, calculated on the basis of the bulk velocity of the flow in the upper part of the furnace is 0.6. If the drift parameter were approximately unity, or greater, then the effects of gravity would be large and, specifically, would hinder the turbulent transfer of particles across the recirculation zone to the furnace walls (see Siu 1996). For comparison we mention that, in the laboratory scale, where the air bulk velocity at the inlet of the burner was 28 m/s and the diameter of the quarl entry was m, the time scale of the large eddies in the region was 0.67 ms. Thus, the large eddies Stokes number is St e =0.26 which implies that it is unlikely that in the laboratory-scale 193

194 4.5 SUMMARY OF CHAPTER 4 furnace a particle of this size can respond to the large eddies and disperse into the recirculation zone. The second mechanism for slag retention at the pilot scale, mechanism B, is by the settling of the particles under the action of gravity in the lower half of the furnace -ultimately into the ash hopper. Thus, gravity is beneficial to slag retention, in contrast to the first mechanism of the preceding paragraph. The importance of the action of gravity is, once again, assessed through the drift parameter and in the lower part of the pilot furnace (the part which includes the exit port), especially near the ash collection port, the drift parameter is 1.1. This is higher than that estimated at the upper part of the furnace owing to the smaller flow bulk velocity, and therefore gravity effects do become significant at pilot scale and therefore promote ash retention. Estimation of the centrifuge Stokes number in the furnace of Reichert et al. (1990) suggests that it will be about one order of magnitude larger than in the present experiment and therefore aerodynamic centrifuging will be even less important than in our laboratory-scale furnace: hence, particles can settle due to gravity in the ash collection port. Increasing the particle size is unlikely to change the effect of mechanisms A and B, and fine ash particles which did not disperse into the recirculation zone due to mechanism A are likely to escape because of the decreasing effectiveness of gravity at the lower part of the furnace with decreasing particle diameter, say for sizes smaller than 50 mm. It is interesting to examine whether the above arguments explain the observation of Reichert et al. (1990) that particle retention became worse with decreasing air throughput in the furnace, which was as low as 40% for a throughput of 250 m 3 /h. When the air throughput is reduced from 500 m 3 /h to 250 m 3 /h, the large eddies- and the transit Stokes numbers almost double in magnitude and, although at 9.6 and 140 respectively these remain much larger than one, at the same time the drift parameter doubles and reaches the value of 1.2, just larger than unity. Hence, gravity effects become important at the upper part of the furnace and therefore hinder particle dispersion across the recirculation zone to the furnace walls, thus reducing the efficiency of mechanism A. In the lower part of the furnace the same reduction of the air throughput means that the drift parameter becomes 2.2 implying that capture mechanism B becomes more effective. Because of the reduction of particle retention in the pilot scale when the air throughput is reduced, it is inferred that mechanism A contributes the most to the total retention efficiency of the furnace: indeed, if it were mechanism B, retention efficiency would have increased with decreasing air throughput, due to the increased drift parameter at the lowest part of the furnace which would have promoted particle settling in the ash collection hopper. 194

195 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES 4.5 Summary of Chapter 4 This chapter presented simultaneous measurements of velocity, size and temperature of burning particles as well as particle volume flux in the near-burner region of a 8 kw swirl burner with natural gas support. Measurements were made for swirl numbers S=0.41 and 0.57, overall equivalence ratios of the support flame of 0.69 and unity and primary to secondary air momentum ratios of 1/30 and 1/40. The effect of confining the flame was quantified for S=0.57 and measurements were made in the far-burner region of confined flames for S=0.5. The main conclusions of this work regarding the influence of the boundary conditions on particle motion and burning characteristics and their potential consequences on NOx generation and particulate emissions are summarised in the following three subsections Particle Aerodynamic Characteristics in the Near-burner Region 1. The volume flux distribution had a single maximum on the centreline due to axial penetration of coal particles through the recirculation zone for swirl number A second local maximum existed in the flux profile at the shear layer between the secondary air stream and the recirculation zone for reduced swirl number of 0.41, or increased secondary air flow rate to 400 lt/min for swirl number 0.57, which was combined with a reduced flux on the centreline. The latter volume flux distribution is unfavourable for the reduction of the NO x emissions, because it implies increased particle residence time inside the gas recirculation zone, although the second local maximum in the shear layer may limit the improvement. 2. The mean particle trajectory angle increased with particle size at z/d = 2.67 and r/ D=0.9 and 1.2, coinciding with local maxima in the flux distribution existed for swirl 0.41 and 300 lt/min secondary air and swirl 0.57 and 400 lt/min secondary air respectively. The rms fluctuations of the radial velocity increased with size for swirl 0.57 at r/d>0.6 and downstream of the recirculation zone, and those of the 60 µm were more than twofold that of the 36 µm near r/d=1. Both these phenomena were explained by particle centrifuging due to mean swirling motion and large scale structures of the flow turbulence respectively. The large eddies were characterised by a length scale similar to the length of the recirculation zone and velocity scale similar to the mean gas recirculating velocity and had time scale 4.5 ms. The Stokes number based on the timescale of the large eddies and the response time of the 60 µm particles was around unity, confirming that particles of this size can be centrifuged, thereby reducing their residence time in the recirculation zone, a condition which is unfavourable for NO x emissions. 195

196 4.5 SUMMARY OF CHAPTER Particle Burning Characteristics in the Near-burner Region 1. Increase of the swirl number from S=0.41 to 0.57 resulted in creation of a forward flow along the centreline of the burner and hence the local gas temperature at z/d=2.67 near r/d=0 was reduced from 1800 K to 1000 K. As a consequence, the local burning fraction decreased from 85% to less than 5% which meant that, effectively, almost no burning particles were found on the centreline for S=0.57. Comparison between the velocity and size distributions of burning particles and the ensemble, which consisted of both burning and non-burning particles, showed that, for S=0.41, although particles which had reversed their direction of motion on the centreline were burning, only few of the forward-moving particles were burning. In addition, most burning particles were smaller than 60 µm and those which were larger had not ignited, implying that their residence time near high-temperature regions was insufficient for ignition. With the exception of the centreline, there was no correlation between particle size or velocity and the burning probability. 2. Radial temperature distributions of burning char particles for S=0.41 showed that at z/d=3.05 the peak of the distribution and, therefore, the median particle temperature, at r/d=0.89 was lower by about 100 K than on the centreline, implying that hotter particles are more likely to be found inside the swirl-induced recirculation zone rather than near the shear layer, between the recirculation zone and the forward flow, probably due to the higher local gas temperatures and hence, smaller heat losses from the particles. Despite the higher particle temperatures on the centreline, the relatively high burning fraction close to the oxygen-rich shear layer implies that those particles are more likely to yield NO x emissions from fuel-bound nitrogen. 3. The effect of increasing the gas equivalence ratio by 45%, from 0.69 to 1.0, and reducing the primary to secondary air momentum ratio by 25%, from 1/30 to 1/40, resulted in reduction of the maximum fraction of burning particles across a radial profile at z/d=2.67 from the inlet of the quarl, chosen to be the closest possible location to the exit of the quarl which was optically accessed. The reduction was of the order of 25% when the equivalence ratio was increased, but reduction of the momentum ratio resulted in a marked reduction of the burning fraction of about 80%. This result can be attributed to insufficient residence times near hightemperature regions owing to changes in the flow pattern, because of the undesirable effect of the swirling flow to transport particles away from the recirculation zone. The larger time of flight at z/d=4 relative to z/d=2.67 is also the reason that the burning fraction increased at the former axial station for both equivalence ratios and reached almost 100% near the centreline. 4. After application of an amplitude-based criterion on the measured temperatures to distinguish between signals from volatile flames or incandescent char, presented in Chapter 3, radial profiles of the mean and rms char temperature were constructed. These showed that, in 196

197 CHAPTER 4 PULVERISED COAL COMBUSTION IN OPEN AND CONFINED PILOTED SWIRL STABILISED FLAMES all cases, the temperature decreased with increasing radial distance and was size-independent, a result confirmed from size-temperature scatter plots for equivalence ratios 0.69 and 1 at z/d=2.67. The decrease was about 100 K except for unity equivalence ratio at z/d=4 in which case it was about 200 K. In contrast, rms temperatures showed a slight increase towards larger radii which was a consequence of the different heat-up histories of particles measured at radial locations far from the centreline. 5. A consequence of conclusions 3 and 4, in conjunction with the results for the volume flux of burning particles and the ensemble, defined here as the total of measured particles burning or not, which showed increasing unburned particulate volume flux with increasing radial distance, is that it is likely that the most important parameter for the design of low NO x burners is the spatial distribution of particle volume flux. In full-scale burners this lack of any size-temperature (and size-velocity, as measured in this work) correlation is likely to be preserved, because particles will be able to respond to the gas flow and therefore flux profiles would indicate directly whether particles devolatilise and burn near the oxygen-lean recirculation zone which is considered favourable to low NO x production Particle Aerodynamic Characteristics in the Far-Burner Region of Confined Flow 1. The scatter plots of velocity, trajectory angle and diameter and the mean velocity of three chosen size classes, namely µm, 24-48µm and 48-72µm, show that there is no evidence of inertial centrifuging of any size class away from the exhaust pipe. This is in accordance with expectations based on the magnitude of the Stokes number of the flow. The particle rms velocities, in spite of the scatter in the graphs due to the limited number of measurements, are generally smaller than air rms velocities due to the inertia of the particles. 2. Scale-up considerations show that in the full-scale furnace, operating at 20 bar pressure, the centrifuge Stokes number is expected to increase by more than one order of magnitude, for the classes of particles considered here. This implies that centrifuging will be weaker at the full as compared to the laboratory scale. At full scale two mechanisms are proposed as responsible for particle retention, on the basis of the calculated large eddies- and transit Stokes number and the drift parameter. The first involves the formation of a slag layer on the walls of the near-burner region due to particle dispersion in the recirculation zone formed between the flame and the furnace liner. The second mechanism is particle settling due to gravity near the exit port. 197

198

199 CHAPTER 5 Gaseous and Pulverised Coal Reacting Flows in Vitiated Air

200 5.1 INTRODUCTION 5.1 Introduction Background The incorporation of a gas turbine topping cycle in a conventional coal-fired utility, in which the high-temperature exhaust gases from the turbine containing less than 21% oxygen 1 are used as oxidant in the furnace, results in increase of the coal combustion cycle efficiency from to (Floris 1981; van de Kamp et al. 1992; Robertson et al. 1993) together with potential reduction of NO x emissions by up to about 50% depending on coal rank (e.g. Smart and van de Kamp 1994). According to IEA Coal Research (Soud and Fukasawa 1996) there are 49 pulverised coal fired units with total capacity of >15GW e which utilise vitiated air ( flue gas recirculation ) for NO x control. However, there is an alternative arrangement for incorporation of a gas-turbine cycle in a coal-fired utility, proposed by Westinghouse Electric Power Corp., in which pulverised-coal-derived synthetic gas from a gasifier is burned in swirl burners, rather than the pulverised coal itself, using vitiated air from pressurised fluidised bed combustion of the remnant coal char from the gasification, with expected efficiencies up to 0.45 (Domeracki et al. 1995; Robertson et al. 1993). Because both technologies, i.e. combustion of coal-derived gas or pulverised coal, can be incorporated in a combined cycle which uses exhaust gas as oxidant, previous work on both gaseous and pulverised coal flames in vitiated air is reviewed here, with emphasis on the influence of vitiation on flame stability. Gaseous flames Most published work on gaseous flames using vitiated air is concentrated on the influence of parameters such as the oxygen mole fraction, preheating and the strain rate on flame stability in non-recirculating flows. For example, in opposed jet laminar diffusion flames the effect of dilution (i.e. use of vitiated air instead of air) without simultaneous preheating was experimentally quantified and it was found that there was a minimum oxygen mole fraction, about 0.15, below which no flame could exist, due to insufficient heat release to sustain chemical reactions (e.g. Puri and Seshadri 1986; Tsuji 1982). Also, for oxygen mass fractions in vitiated air in the range , the strain rate for extinction of laminar counterflow diffusion flames was experimentally determined to decrease by about 40% for every 10% reduction of the oxygen mass fraction (Puri and Seshadri 1986). For comparison, in a turbulent counterflow arrangement, it was recently found that the effect of a 0.02 reduction of oxygen mole fraction in vitiated air required 100K temperature increase of the vitiated air to maintain stability, a result which accords closely with theoretical results from laminar flames (Mastorakos et al. 1995). Apart from stability, radiative heat transfer from the flame is also affected by air vitiation, with consequences on the operation of boilers where heat is principally transferred from the flame to the walls by radiation. Previous work on pool flames showed that the total heat transfer 200

201 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR from flames in vitiated air was not affected (Santo and Delichatsios 1984), although the radiative part decreased by about 25-30% when vitiated air with 0.17 oxygen mole fraction was used instead of atmospheric air, owing to reduced soot production (Santo and Delichatsios 1984). It is important to understand how air vitiation and preheating affect the stability of gaseous swirl-stabilised flames, because the latter are commonly found in power generation installations. Unfortunately, the application of experimental findings from investigations on laminar or turbulent counterflow flames, regarding the influence of vitiation on flame stability, to turbulent swirling flames is not straightforward, owing to the presence of the swirl-induced recirculation zone. Even in non-vitiated flames, this zone modifies the mixing between the fuel and the oxidant relative to the counterflow flame and hence flame location and flame length (Rawe and Kremer 1981), causing continuous broadening of the lean 2 stability limits with decreasing swirl (Milosavljevic et al. 1990; Tangirala et al. 1987). The combined influence of dilution and swirl number on gaseous flame extinction is not currently well understood, mainly because of lack of experimental data. In addition, it is reasonable to assume that the pulverised coal flame in vitiated air will behave differently compared with the gaseous flame. Pulverised coal flames At semi-industrial scales (i.e. of the order of a MW) the effects of vitiated air on flame stability and on the potential reductions of the NO x emissions from combustion of pulverised coal using swirl burners have been investigated and the findings of such work is reviewed in this section. In investigations in a 100 kw furnace it was found that a minimum of 30% excess air and a 17% oxygen content in the primary vitiated air was necessary in order to maintain efficient conversion of carbon, whereas coal ignition and combustion stability was insensitive to the oxygen concentration in the secondary stream and the latter could be maintain within 15-17% (Floris 1981). In a 0.5 MW furnace it was found that use of vitiated air containing 14% oxygen in the secondary oxidant stream resulted in reduction of particle burnout from 96% to 92% (O 2 from 21% to 14%) and from 97% to 90% % (O 2 from 21% to 14%) when an annular and a central orifice injector were utilised respectively, accompanied by respective reductions of 70% and 30% in the NOx emissions (Abbas et al. 1993). It must be noted that the latter reduction is quoted without correction for the increased volumetric flow rate when using vitiated air and that might be misleading in the interpretation of the emissions, as Smart and van de Kamp (1994) point out. Combustion of pulverised coal directly using vitiated air containing 11-21% oxygen as oxidant in a 1.3 MW furnace showed that, depending on coal rank, a potential 50% reduction in the NO x emissions can be achieved at the expense of 1 termed as vitiated air. 2 The designation lean refers to the overall air/fuel ratio derived from the flow rates of air and fuel (explained later in 5.2.3). The flame is of the non-premixed type. It is not premixed. 201

202 5.1 INTRODUCTION serious compromise in the fraction of unburned carbon which can be more than 5 times higher than the case of combustion in pure air (van de Kamp et al. 1992). Observations similar to gaseous flames regarding the reduction of radiative heat transfer in coal-fired flames were reported where weaker flame luminosity, implying smaller radiative heat transfer, was observed (van de Kamp et al. 1992). One of the main undesirable effects of vitiated air on pulverised coal flames, as compared with air, was that, the range of the ratio of primary air, transporting pulverised coal/secondary air momentum for which stable flames can be sustained becomes narrower with the decreasing oxygen content (Abbas et al. 1993), an effect which is probably associated with the devolatilisation processes in the near-burner zone of pulverised coal-fired burners. It is important that use of vitiated air does not result in deterioration of the operability of the burners of a furnace and, consequently, of a power plant The Present Contribution From the work reviewed in the previous paragraphs it is clear that further research on the effect of air vitiation and preheating on the stability of swirl-stabilised flames was necessary, given direct extrapolation of findings in opposed-jet flames to swirling flames, in the absence of any experimental results from the latter, is beyond current knowledge. Additionally, because of the possibility of using gas turbine exhaust gas to burn pulverised coal in swirl burners and thus increase the overall efficiency of power plants, particularly in view of the associated potential reductions in NO x emissions, a detailed study of the effect of vitiation on the combustion characteristics of isolated coal particles would be useful for understanding the combustion of pulverised fuel in swirling vitiated air. The latter understanding is particularly important given the results of Abbas et al. (1993) and Smart and van de Kamp (1994) who reported a deterioration in flame stability and particle burnout (hence, reduced efficiency) in vitiated air, but did not provide detailed in-flame measurements to study the effects of vitiation on individual particles. The purpose of the present contribution is to quantify the effects of the swirl number, the mean strain rate, the geometry of fuel injection, the oxygen concentration in- and the temperature of- the vitiated air on the stability of non-premixed natural gas flames, stabilised inside the quarl of a nominally 10 kw swirl burner, and quantify the effect of oxygen concentration on the combustion probability 3 of isolated coal particles in vitiated-air flames with natural gas support. This Chapter is divided in two parts: the first ( 5.2) presents results from investigations on gaseous flames and the second ( 5.3) those from simultaneous measurements of velocity, size and temperature of burning coal particles in natural-gas supported flames. 202

203 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR 5.2 Stability of Natural Gas Flames in Vitiated Air Flow Configuration and Experimental Methods The experimental facility comprised the swirl-stabilised natural gas burner and an vitiated air generation system. The 10 kw swirl burner was that described in Chapter 4 with 16 mm throat diameter and illustrated in figure 4.1. In the work presented here two fuel injectors were used. The first, figure 5.1a, was designated axial with 4 mm internal diameter and its tip tapered at 10, to reduce the bluff-body effect of the primary pipe. The second, figure 5.1b, was designated multihole radial and was similar to 0 the injector described in Chapter 4, although it had six lateral holes instead of ten for natural gas (a) injection. An outline of the vitiated air generation system is shown in figure 5.2 (se also COLOUR PLATES 12, 13 AND 14 in Appendix VI). It consisted of two identical commercial premix burners (model P34PE; Igniters Combustion Engineering Ltd, UK) which burned natural gas (94% CH 4 by volume; net calorific value 35 MJ/m 3 ) and air. The burners were fitted at the end of a premix unit (model CAITO; Igniters Combustion Engineering Ltd, UK) which produced a combustible mixture from air and methane and the equivalence ratio of the mixture was electronically controlled (model CU6; Igniters Combustion Engineering Ltd, UK). They were housed in two separate cylindrical steel chambers (fig. 5.2) of 3" (76.2 mm) external diameter, 10 mm thick and 250 mm long. The combustion chambers were designed to be about 30% longer than the flame stabilised at the exit of the premix burner as a compromise between avoiding flame/ wall interaction (i.e. quenching) with the downstream wall but, at the same time, keeping the chamber surface area and, hence heat losses ÃPP ÃPP ÃPP ÃPP ÃPP ÃPP ÃPP ÃÃPP Figure 5.1 Detail of the tip of the axial (a) and the multihole radial (b) injector, showing the dimensions of the holes through which natural gas was injected. (b) 3 the probability of measuring a burning particle at a given location in the flame. 203

204 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR small. A four outlet manifold was arranged at the downstream end of each combustion chamber, which directed the exhaust gases to the four axial and the four tangential inlets of the swirl burner, after these had been mixed with an additional amount of air drawn from the laboratory shop air supply to produce vitiated air of variable oxygen content, see figure 5.2. The outlets of the cylindrical combustion chambers and the inlets of the swirl burner were connected in stainless steel tubes of 8 mm external diameter. The vitiated air generation system was located close to the swirl burner, in order to minimise the total length of the metal tubes (to about 1 m) and the consequent heat losses, and all metal tube connections between the combustion chambers and the swirl burner were cut to similar lengths, to ensure equal pressure losses and thus to avoid flow imbalance between each of the four tubes to the axial and tangential inlets. The swirl burner and the tubes which connected it with the cylindrical combustion chambers were insulated with a layer of 20 mm thick fibre insulation and a layer of self-adhesive aluminium foil on the external surface 4. The exhaust gas generation system provided hot gases with oxygen content and temperature in the range of and C respectively, which according to van de Kamp (1992) correspond to typical gas turbine inlet conditions in combined-cycle power plants. During the execution of the experiment, the thermal inertia of the system required a heat-up interval of about one hour to reach steady state. The temperature of the exhaust gases was monitored at the inlet of the quarl with a 250 µm diameter R-type (Pt/Pt+13% Rh) thermocouple connected to a digital voltmeter and it was assumed that thermal equilibrium between the experimental facility and the surrounding air was reached when the vitiated air temperature did not change more than 5 K within 5 minutes. For extinction measurements which were arranged at burner exit temperatures higher than that of thermal equilibrium, constant (within 5%) vitiated air temperatures could not be sustained for periods longer than 20 seconds, owing to heat losses. Burner exit temperatures higher that those of thermal equilibrium were achieved by running the vitiated air facility for short period of time at heat releases higher than the desired for a given condition, before measuring the given condition. In a similar manner, heated unvitiated air was produced by heating up the facility to temperatures higher than the required for each condition using both premix burners, and monitoring the air temperature by thermocouple. Extinction limits were obtained by the procedure described in the next paragraph at the desired preheat temperatures Trials with an early design of the exhaust gas supply system, which used only one premix burner and the swirl burner was operable at only one swirl number, and fivefold longer tubes to direct the vitiated air to the swirl burner, showed that the length of the tubes connecting the premix burner with the inlets of the burner was critical for the heat losses, which were insensitive to the addition of even three insulation layers. The maximum exhaust gas temperature of 350 C at the exit of the swirl burner with this design was insufficient to obtain a wide range of swirl-stabilised flames, as will become apparent in the following pages. 5 This method sufficient for the determination of the extinction limits for a swirl number, but this way it was impossible to measure, at the same time, the gas species concentration and the temperatures of the reacting flow at those vitiated air temperatures, as temperatures higher than the equilibrium were sustained for a period not exceeding 20 seconds, much shorter than the time scale of species measurement (of the order of minutes).

205 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR 6ZLUO%XUQHU $[LDOVXSSO\ RIYLWLDWHGDLU 7DQJHQWLDOVXSSO\ RIYLWLDWHGDLU 6KRSDLUIRUGLOXWLRQ 6KRSDLUIRUSUHPL[ EXUQHU 0L[WXUH 1DWXUDO*DV 3UHPL[EXUQHU 1DWXUDOJDVIRUVZLUOIODPH 3UHPL[EXUQHU 1DWXUDO*DV 0L[WXUH 6KRSDLUIRUGLOXWLRQ 6KRSDLUIRUSUHPL[ EXUQHU Figure 5.2 Block diagram of the exhaust gas supply system, showing the tube connections between its various parts and the air and gas supplies. The extinction limits were determined by observation of the visible flame while the natural gas supply through the swirl burner was varied, for a constant vitiated air mass flow rate, swirl number and vitiated air temperature (COLOUR PLATE 16 shows an example of vitiated air flame, compared with a non-vitiated air flame in PLATE 15). The temperature of the vitiated air was measured before and after each extinction measurement, and the results were discarded if the difference was larger than 5%. Lean limits (see earlier footnote) of stable operation were determined in all cases but rich limits were attainable only in a few cases, due to limitations in the natural gas flow rate of the apparatus. Lifted flames were also observed with operation with the axial injector but not with the multihole radial injector. With the latter, the flame appeared 205

206 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR to detached from the wall of the quarl and it was assumed that this corresponded to a lifted flame. The mean mole fractions of stable gas species, namely CO 2 and CO, O 2 and unburned hydrocarbons were measured at the exit of the quarl of the swirl burner with, respectively, non-dispersive infrared analysers (Analytical Development Company), paramagnetic analyser (Servomex 540A) and Flame Ionisation Detector (FID, Rotork Analysis Ltd.). The hydrogen mole fraction was assumed to be 0.65 of that of the CO, following Masri and Bilger (1986). The water-cooled sampling probe had 4 mm external and 0.5 mm internal diameter and imposed an area blockage of 1% at the measurement region, calculated on the basis of the cross section of the probe and the cross sectional area at the exit of the quarl where the measurements were made. Following Mastorakos (1993), it was estimated that the spatial resolution of the measurements due to averaging varied from 2 mm near the shear layer around the recirculation zone and 4 mm inside it. A 50µm Pt/Pt+13%Rh thermocouple was used to take radial profiles of temperature. The signal was passed through an amplifier with gain of 100 and the amplified signal was sampled by an A/D converter (Analogic, model LSTA/16) with a rate of 5 khz. In order to measure temperature fluctuations as well, a total number of 4096 samples were taken and the signal was compensated for the finite response of the thermocouple wire using the method described by Mastorakos (1993). Its time constant was calculated from the expression of Collis and Williams (1959), using the transport properties of air at local temperature, which was reduced by an empirically-determined factor of 0.76 (Mastorakos 1993) to account for the local transport properties of the combustion products which are different from those of the air at the same temperature, assumed in the expression of Collis and Williams. In the compensation for the finite response of the wire, an approximate flow velocity was used, based on the measurements of Milosavljevic (1993) and Orfanoudakis (1994), as simultaneous temperature and local velocity measurements were unavailable in this work. The reading resolution of the rotameters was: 10 lt/min for the air, 1 lt/min for the natural gas in the premix burners and 0.5 lt/min for the gas in the swirl burner. The estimated uncertainties in the measured and calculated quantities were as follows: ±5% for the measured flow rates and the calculated bulk velocities and the bulk strain rates, ±0.03 for the overall equivalence ratio (based on the uncertainty of the flow rate measurement), ±10% for the reading of the measured species mole fractions and 150K systematic underestimation for the temperatures due to radiation losses Boundary Conditions 206

207 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR The boundary conditions of the flow were defined by the total oxidant mass flow rate through the burner; the swirl number, determined from velocity measurements at the throat diameter in the absence of the quarl (see below); and the temperature and the oxygen content of the oxidant, measured at the inlet of the quarl by means of instrumentation described in the previous section. The velocity boundary conditions were considered to be identical to those measured by Milosavljevic (1993) since the Reynolds number, defined by the bulk velocity at the throat diameter D, was always greater than 17000, ensuring that the flow was Reynolds number independent (Chen et al. 1990). Flow symmetry was established from temperature measurements as described in Appendix V, where also a detailed parametric study of the influence of misalignment of the fuel injector relative to the axis of the burner of flame symmetry is presented. The oxygen content of the vitiated air was measured for different combinations of exhaust gases produced by only one premix burner feeding tangential vitiated air (because it is the bulk of the flow) and additional air. Both burners were not operated at the same time except for preheating at the start of the experiment, or for achieving temperatures higher than those of thermal equilibrium. In the case of simultaneous operation of both burners the minimum flow rate of vitiated air containing oxygen mole fraction which could be produced exceeded the maximum of 300 lt/min investigated here (shown in Table 5.1) Bulk Quantities to Characterise the Flow Table 5.1 Cases examined, corresponding flow rates and oxygen concentration of exhaust gases. CASE Air flow rate through premix burner [lt/min] Gas flow rate through premix burner Potential heat release due to gas through premix burner [kw] Additional air flow rate [lt/min] Measured oxygen concentration at the exit of the swirl Measured air temperature at burner exit at thermal equilibrium [ C] [lt/min] burner [%] Ambient air Ambient air A B C D E F All flow rates were metered at 20 C. 6 Assumed rather than measured. 207

208 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR All flames were characterised by bulk quantities such as the overall (fuel) equivalence ratio f, the fuel/oxidant momentum ratio (MR), the thermal power due to preheating H & T and the bulk strain rate s. The overall equivalence ratio j, for the cases using unvitiated air, was calculated from: ϕ = Q&.52 Q& 9 f (5.1) where Q & f and Q & were the volumetric flow rates of methane and oxidant respectively, and the constant in equation (5.1) was determined from the stoichiometry of the one-step, global reaction of methane with air 7 : CH 4 + ( 0.21O N2 ) CO2 + 2H2O + N (5.2) For vitiated air, the constant in equation (5.1) had to be modified from the value used when the oxidant was air, i.e. 21% O 2 and 79% N 2. The overall equivalence ratio for vitiated air was: ϕ =.52 x /0.21 Q& Q& 9 f (5.3) where x was the (measured) oxygen mole fraction in the oxidant stream. Along with the overall equivalence ratio, the stoichiometric mixture fraction changes with vitiation. When no partial premixing takes place, it can be defined as (Spalding 1979): Y / w f stoich = 1+ Y / w (5.4) where w is the fuel/oxidant mass stoichiometric ratio in equation (5.2) and it is 4 kg/kg, whereas the oxygen mass fraction in air Y was calculated from the measured oxygen mole fraction x: 7 For all calculations of the equivalence ratio and the mixture fraction, defined below, CH 4 was assumed as fuel instead of natural gas, following Mastorakos (1993) who estimated that this assumption results in a 3% difference in the calculated equivalence ratios. 208

209 Y ( 32/ MW ) air x = 1 x + ( 32/ MW )x air CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR (5.5) MW air, the molecular weight of air, was assumed the same for both vitiated air and ambient air. A small percentage, about 2%, CO 2 existed in the exhaust gas along with water vapour as a result of the combustion in the premix burners, but the error in the calculation of the molecular weight of the vitiated air using this assumption is only about 0.1%. The stoichiometric mixture fraction for ambient air was The fuel jet/oxidant momentum ratio where oxidant was either vitiated or non-vitiated air was defined as: MR = ρ f / ρ 0 ( U / U ) 2 f 0 (5.6) where ρ i denotes the density of fluid stream, U i the corresponding velocity and the subscripts f and o correspond to the fuel and the oxidant respectively. In the case of the multihole injector, U f was calculated for one hole 8. Both the overall equivalence ratio and the momentum ratio were used as independent variables, as a function of which the stability limits of the burner were plotted. A consequence of using vitiated air generated by hot combustion products was that thermal energy was added to the swirl burner because of the higher temperature of the vitiated air. If T is the temperature of the vitiated air, then the enthalpy flow rate H & of an oxidant stream of mass flow rate m& was calculated from: T H& T = mc & T p (5.7) where the specific heat was taken as 1.1 kj/kg K (Incropera and DeWitt, 1985). With increasing temperature of the vitiated air its density decreased and, accordingly, its bulk velocity 9 U 0 increased as well due to the increased volumetric flow rate through the swirl burner. Consequently, the flow bulk straining at the base of the quarl increased with increasing temperature. To estimate bulk straining a bulk strain rate s was defined as: 8 ρ f =0.664 kg/m 3, ρ air =1.2 kg/m 3, ν f = m 2 /s, ν air = m 2 /s, at 20 C. 9 Throughout this work, the term bulk, where applied, refers to quantities calculated at the quarl inlet corresponding to diameter D. 209

210 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR U0 s = D (5.8) where the velocity U 0 of the vitiated air was estimated from mass conservation considerations, assuming oxidant density changes which followed the ideal gas law. Table 5.2 presents the variation of bulk strain rate with temperature for 250 lt/min and 300 lt/min total oxidant flow rates through the swirl burner. It is noted that preheating to 600 C yields increase of the bulk strain rate of a factor of 3 relative to 20 C. For comparison, a CH 4 turbulent diffusion counterflow flame extinguishes at a strain rate of 350 s -1 (Mastorakos et al. 1992). Table 5.2 Bulk strain rate as a function of vitiated air temperature Total vitiated air flow rate (lt/min) Bulk strain rate s (1/s) containing 0.21 O 2 mole fraction at temperature 20 C 400 C 500 C 600 C Temperature and Stable Chemical Species Figures 5.3 and 5.4 present the effect of increasing the equivalence ratio from 0.7 to on the radial profiles of the at z/d=2.2 mean and the rms temperature for a unvitiated air and x=16.5% O 2 (case C) respectively, at a swirl number of The measurement location was the nearest location to the quarl exit which was physically accessed and where measurement of both temperature and species concentration could be obtained. In all cases presented in this section the multihole radial injector was used. The vitiated air mean temperature profile was flatter, with no clear temperature peak, in contrast to the case of unvitiated air, and the maximum mean temperature occurred at a location closer to the centreline of the burner. The mean T profile was about 200 C greater for the vitiated air case. In both cases the mean temperature increased as the overall equivalence ratio increased; the normalised rms of temperature fluctuations at the location of maximum temperature for the case of unvitiated air decreased from 10.3% to 7.4% and for x=16.5 O 2 (case C) it increased from 5.8% to 6.3%. The decrease of the rms temperatures with increasing equivalence ratio is expected, as it is likely that as the fuel supply is increased from the lean extinction limit of the burner, mixture fraction fluctuations and hence, temperature fluctuations decrease. Recall that, for a diffusion flame, combustion 210

211 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR (a) T [K] φ=0.7 φ=0.8 z/d=2.2 r/d 400 (b) 300 φ=0.7 φ=0.8 T [K] z/d= Figure 5.3 Radial profiles of mean (a) and rms (b) temperatures measured for ambient air at 300 lt/min, for a swirl number of 0.68 and equivalence ratio 0.7 and 0.8. The profiles were measured at z/d=2.2. r/d always occurs in the region of stoichiometric mixture fraction (e.g for CH 4 ) and an overall leaner mixture would result in smaller probability of encountering a locally stoichiometric mixture. For the vitiated air case, the influence of increased equivalence ratio on the normalised rms temperatures is small, and although to the opposite direction compared with the case of unvitiated air, it could be attributed to overcompensation, because the velocities used in the expression of Collis and Williams (1959) for the calculation of the thermocouple time constant were assumed to be the same as in the case of air at 20 C. Figure 5.5 compares the mean temperatures for unvitiated and vitiated air at comparable heat release, which was 8.5 and 8.2 kw respectively and equivalence ratios 0.8 and 0.85 respectively (see also footnote 10). As expected, for the vitiated air case the mean temperatures were higher, which, in principle, should be higher by amount equal to the vitiated air temperature; in practice heat losses might reduce this difference. 10 The numbers quoted in figure 5.4 are slightly higher than in figure 5.3, because the equivalence ratio in 5.4 was calculated using equation (5.3), whilst for the results in figure 5.3 equation (5.1) was used; the gas flow rate was the same in both cases. 211

212 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR (a) T [K] φ=0.73 φ=0.85 z/d= φ=0.73 φ=0.85 z/d=2.2 r/d (b) Figure 5.4 Radial profiles of mean (a) and rms (b) temperatures measured for vitiated air of 400 C and 16.5% oxygen mole fraction (case C) at 300 Normal lt/min, for a swirl number of 0.68 and equivalence ratios 0.73 and The profiles were measured at z/d=2.2. T [K] r/d Figures 5.6 and 5.7 present radial profiles of CO 2, CO, O 2 and unburned hydrocarbons (UHC) and the mixture fraction for case C and a swirl number of 0.68, for equivalence ratios of 0.81 and 0.93, respectively, measured 5 mm downstream of the quarl exit, which corresponded to z/d=2.2. The equivalence ratio of 0.81 for the vitiated (and the 0.73 for the unvitiated case used below) corresponded to the minimum at which a stable flame could be sustained in the presence of the probe, in order to measure the species at conditions close to extinction 11. The concentrations plotted in these graphs are the corrected ones on wet basis, using the method of Milosavljevic (1993). Figures 5.8 and 5.9 also present axial profiles 11 The reader will observe that the overall equivalence ratios quoted in this paragraph are slightly different to those of the previous paragraph; for example, 0.81 instead of 0.85 of the temperature measurement - whilst 0.93 was intentionally selected as a higher equivalence ratio for the parametric investigation. The reason for this variation was the fact that small increase (0.5-1 lt/min) in the fuel supply was necessary when measuring with the sampling probe a condition close to the lean limit, to ensure stable flame formation during the experiment; pressure of time and luck of equipment did not allow the author to repeat the temperature measurements to ensure consistency in the used equivalence ratios. In addition, from inspection of the results it was considered that this small difference between the used equivalence ratios does not influence the interpretation of the measurements. 212

213 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR measured along the centreline of the burner, for the same conditions respectively. The background CO 2 concentration in the exhaust gases due to previous combustion was 2% while no CO and UHC was present. For comparison with unvitiated (ambient) air, figures 5.10 and 5.11 present the radial and axial profiles respectively for an equivalence ratio of 0.73, measured for the same swirl number and at z/d=2.2 and correspond to figures 5.6 and 5.8 respectively. If we consider the calculated stoichiometric mixture fractions of and for case C and unvitiated air respectively, it is inferred from figures that the unvitiated flame is stabilised inside the quarl as the maximum mixture fraction at the quarl exit corresponded to lean local conditions and it is decreasing further downstream of the quarl exit (figure 5.11). In contrast, it is likely that the vitiated air flame did bot extend inside the quarl as much as the unvitiated flame, was possibly surrounding the recirculation zone and moved even farther as the equivalence ratio increased (figure 5.7), a fact attributed to the smaller oxygen concentration in the vitiated air compared with ambient air. The flame location can D 7>.@ 4 N: 4 N: ]' $PELHQW FDVH& Figure 5.5 Comparison between radial profiles of mean temperatures corresponding to a swirl number of 0.68 and heat releases of 8.5 and 8.2 kw. The profiles were measured at z/d=2.2. U' E 7>.@ 4 N: 4 N: ]' $PELHQW FDVH& U' 213

214 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR "Wet Mole Fraction" r/d CO 2 O 2 Mixture Fraction (b) (a) Figure 5.6 Radial profiles of mean concentrations of CO 2, O 2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured at z/d=2.2, for the case of vitiated air of 400 C and 16.5% oxygen mole fraction (case C), at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used. "Wet Mole Fraction" CO UHC r/d (a) 0.08 "Wet Mole Fraction" "Wet Mole Fraction" r/d 0.6 r/d CO 2 O 2 Mixture Fraction CO UHC (b) Figure 5.7 Radial profiles of mean concentrations of CO 2, O 2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured at z/d=2.2, for the case of vitiated air of 400 C and 16.5% oxygen mole fraction (case C), at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used. 214

215 ... CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR Figure 5.8 Axial profiles of mean concentrations of CO 2, O 2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured along the centreline starting from the quarl exit, for the case of vitiated air of 400 C and 16.5% oxygen mole fraction (case C), at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used. "Wet Mole Fraction" CO 2 O 2 Mixture Fraction z/d (a) (b) "Wet Mole Fraction" CO UHC z/d (a) "Wet Mole Fraction" CO 2 O 2 Mixture Fraction Figure 5.9 Axial profiles of mean concentrations of CO 2, O 2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured along the centreline starting from the quarl exit, for the case of vitiated air of 400 C and 16.5% oxygen mole fraction (case C), at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used. "Wet Mole Fraction" CO UHC z/d z/d (b) 215

216 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR "Wet Mole Fraction" r/d CO 2 O 2 Mixture Fraction (a) Figure 5.10 Radial profiles of mean concentrations of CO 2, O 2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured at z/d=2.2, for the case of ambient air at 300 lt/min, at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used (b) "Wet Mole Fraction" 0.01 CO UHC r/d (a) "Wet Mole Fraction" "Wet Mole Fraction" CO 2 O 2 Mixture Fraction 1.0 CO UHC z/d (b) Figure 5.11 Axial profiles of mean concentrations of CO 2, O 2, CO and unburned hydrocarbons (UHC), and mean mixture fraction, as calculated from the measured mole fractions, measured along the centreline starting from the quarl exit, for the case of ambient air, at a swirl number of 0.68 and equivalence ratio of D is the quarl inlet diameter. The multihole radial injector was used z/d

217 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR not be estimated accurately from the present measurements, because of the spatial averaging effects of the probe. The oxygen mole fraction inside the recirculation zone was at least double when using ambient air as oxidant, and it is unlikely that this is only due to the equivalence ratio difference, and decreased as the equivalence ratio increased, an indication that the recirculation zone mostly entrained hot exhaust gases. Another observation on the CO mole fraction curves (compare figures 5.7 and 5.10) is that a clear peak appeared for the vitiated air case and the equivalence ratio of 0.91 which was shifted towards the centreline which could be due to a similar shift of the mean temperature as the equivalence ratio increases, shown in figure 5.4. It is likely that this peak has been smoothed, due to probe spatial averaging (Mastorakos 1993) Lean Extinction Limits Results are presented in the form of shaded areas within which stable flames could be sustained (as, for example, in figure 5.12) the boundaries of which were determined by the measurements, with the swirl number as well as the overall equivalence ratio as an independent variable. The arrows, where applicable, indicate that no rich limits were found, within the apparatus limit of gas, or air, flow rates, and that the rich extinction limits can be expected to exist in the direction of the arrows. Swirl burner operational limits ( i.e. limits within which the burner could sustain stable flames) were plotted with the abscissa representing the reciprocal of the equivalence ratio, following the practice of Tangirala et al. (1987) and Milosavljevic (1993), and also with an abscissa representing the momentum ratio of the gas and the oxidant stream. The reason for the presentation of the results in two ways was the observation that overall equivalence ratio, although it is a parameter of engineering importance and useful to the designer and user of swirl burners, was found inadequate to correlate the behaviour of vitiated flames. Therefore we have chosen the fuel/oxidant momentum ratio (MR), defined as: ρ MR = ρ f o U U f o 2 K (5.9) Where ρ i and U i were the density and the velocity of the fluid 12, K was the number of fuel nozzles (1 for axial and 6 for multihole radial injector) and subscripts f and o corresponded to fuel and oxidant respectively. The momentum ratio is a convenient variable for elucidating the effects of rate of strain due to the fuel jet as it mixes with the surrounding co- or cross-flowing oxidant (for axial and multihole radial injection, respectively) in the region of injection. It has been assumed in the calculation of the momentum ratio that the fuel heats up as it flows through the injector by heat transfer from the vitiated air and thus assumes the temperature of the vitiated air. 217

218 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR Lifted flames 20 C Sw C 500 C lean 3.5 1/φ Sw lean MR Figure 5.12 Extinction limits for the ambient air case and 300 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as an independent variable, for the multihole radial injector. Shaded areas correspond to stable operation of the burner at secondary air temperatures indicated on the graph. Lean limits of operation correspond to the right-hand side in (a) and the left-hand side in (b). Symbols indicate the measured conditions and arrows the extension of the stability limits. Dotted lines indicate the boundary of the measured shaded area where it is not obvious in the figure. 12 In axial injection the momentum ratio is a measure of fuel jet penetration into the oxidant stream whilst in radial injection it represents the momentum ratio of a jet-in-crossflow -type flow. It has been found by McGuirk and Rodi (1978) from numerical studies of jet discharge into an open channel that the square of the velocity ratio (not significantly different from the momentum ratio used here) controls the centreline trajectory of the jet (and, hence, mixing farther downstream in the present recirculating flow). The momentum ratio has nevertheless little influence on the dilution of the fuel jet due to oxidant entrainment (McGuirk and Rodi 1978). 218

219 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR C Lifted flames 20 C 500 C Sw C 0.5 unstable lean /φ Sw lean MR Figure 5.13 Extinction limits for the ambient air case and 250 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as an independent variable, for the multihole radial injector. See also figure Lean Limit as a Function of Unvitiated Air Temperature Figures 5.12 and 5.13 present the extinction limits of the burner for the multihole radial injector, for ambient air as oxidant and flow rates of 300 and 250 lt/min respectively, at temperatures of 20 C and 500 C (see for a description on how to obtain unvitiated air of desired temperature), and figure 5.14 presents similar results for the axial injector and the 300 lt/min flow rate case. The oxidant flow rate, as measured at 20 C, is quoted instead of its bulk velocity U 0 to distinguish between the flow conditions, because of the dependence of the oxidant bulk velocity on temperature. The oxidant bulk velocity also varied between results 219

220 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR 0.8 AXIAL INJECTOR Lift-off C 20 C 500 C Sw C lean /φ Sw lean MR Figure 5.14 Extinction limits for the ambient air case and 300 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as an independent variable, for the axial injector. See also figure obtained at different temperatures, for the same oxidant flow rate, due to density changes, as explained in 5.2.3, a feature which has been taken into account in the presentation of the lean extinction limits as a function of the momentum ratio (MR). In figures 5.12a, 5.13a and 5.14a it is observed that increasing the temperature of the oxidant resulted in leaner extinction limits - flames could be stabilised at smaller equivalence ratio φ (i.e. larger 1/φ). The momentum ratios (MR) which corresponded to the lean limits decreased with increasing temperature for the multihole radial injector, figures 5.12b and 5.13b, but remained constant for the axial injector, figure 5.14b. In all cases, decreasing swirl numbers resulted in broadening the extinction limits, particularly for the 500 C cases (figures 5.13 and 5.14), but a minimum swirl number 220

221 . CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR C Lifted flames 400 C 500 C Sw C lean /φ Sw lean MR Figure 5.15 Extinction limits for case C, with a 16.5% oxygen mole fraction in vitiated air and 300 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as an independent variable, for the multihole radial injector. See also figure was necessary to obtain a stable flame and depended on temperature. This minimum was found to be 0.55 for the 500 C temperature, although flames were stabilised for 0.49 swirl number at 20 C. Given the low swirl numbers obtained by this burner of the order of 0.5 the 10% difference in the swirl number mentioned above is important. The lean limits for the axial injector, which were leaner than those for the multihole injector (figures 5.13a and 5.14a) became broader with decreasing swirl numbers although the broadening became less pronounced as the swirl number was decreased and for a swirl number 221

222 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR C Lifted flames 410 C 500 C Sw C 0.5 unstable lean /φ Sw lean MR Figure 5.16 Extinction limits for case D, with a 15.5% oxygen mole fraction in vitiated air and 250 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as a parameter, for the multihole radial injector. See also figure of 0.5, the temperature rise had no effect on the lean stability limit, figure 5.14a. In terms of momentum ratio (MR) this means that MR never fell below 0.2 for the axial injector, figure 5.14b. Lean Limits as a Function of Vitiated Air Temperature Figures 5.15 and 5.16 present the lean extinction limits of the swirl burner obtained with the multihole radial injector, for cases C and D with conditions detailed in Table 5.1 and figure 5.17 presents the lean extinction limits for the axial injector and case C. Case C corresponded 222

223 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR C Lift-off 400 C 500 C Sw C 0.5 lean /φ Sw lean MR Figure 5.17 Extinction limits for case C, with a 16.5% oxygen mole fraction in vitiated air and 300 lt/min, as a function of the reciprocal of the overall equivalence ratio (a) and the fuel/oxidant momentum ratio MR (b), with the swirl number as a parameter, for the axial injector. See also figure to 300 lt/min while D to 250 lt/min and the respective oxygen concentrations in the vitiated air were 16.5% and 15.5%. Results for cases B, D, E and F were similar to cases C and D and are not therefore presented. For case A which also corresponded to 250 lt/min, but the oxygen in the vitiated air amounted only to 12.6%, no flame could be stabilised at any swirl number, within the operational temperature range of the experimental apparatus, the upper bound of which was about 650 C. Again, as for the case of hot unvitiated air, the influence of increasing temperature while keeping the oxygen mole fraction in the vitiated air constant was to extend the lean limits of operation as well as the lift-off limits towards the rich side, figures 5.15, 5.16 and 5.17, for both injector geometries. In figure 5.15 (case C, 300 lt/min) it can be also 223

224 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR seen, that a 100 C temperature rise to 500 C enabled the burner to operate at a lower swirl number of 0.52, whereas the lowest at 400 C was Recall that a similar result was described in the previous section for the unvitiated air. The same trend also exists in figure 5.16 for case D (250 lt/min). It is interesting to observe that the use of the momentum ratio rather than 1/φ as abscissa improved the correlation of the lean limits as a function of temperature for both injector geometries, figures 5.15b, 5.16b and 5.17b. For the axial injector (figure 5.17b) the correlation was also better and the lean limits collapse on a single curve. The minimum MR which corresponded to the lean limit for the swirl number of 0.5 was 0.9 for the multihole radial injector and 0.2 for the axial, as is better shown in figure 5.18b, where the lean limits for case C and 500 C are plotted for both injection geometries. The fact that the axial injector could sustain flames for lower MR compared with the multihole radial is attributed to the different flow patterns which generate the mixing between the oxidant and the fuel: the mixing in the case of the multihole radial injector/nozzle is achieved by a quasi-bluff-body effect produced by each of the six laminar 13 fuel jets emerging in the cross-flowing oxidant, whereas with axial injection the laminar 13 fuel jet mixes partly through a mixing layer and partly through counterflow against the hot gases from the recirculation zone, in an opposed-jet-like pattern. Despite the smaller momentum ratios MR at which flames can exist with axial injection, overall leaner flames could be sustained with radial injection in case C, figure 5.18a, which seems to imply that a swirl burner with multihole radial fuel injection can be designed to operate at lower thermal loadings than with axial injection, possibly due to the better mixing achieved from the jet-in-crossflow-like fuel injection. Nevertheless, the fact that the value of 0.2 of the MR for axial injection was consistent for the various flames studied suggests that the use of an axial injector with smaller internal diameter would result in higher MR for the same amount of fuel and, therefore, flames that could be stabilised at MR of 0.2 would be leaner than the present. The process of reducing the nozzle diameter and stabilising flames at lower equivalence ratios cannot continue indefinitely, of course, because a minimum amount of fuel flow will be required to achieve high temperatures of the recirculation zone which are responsible for flame stabilization, as explained in Table 5.3 compares the amount of thermal energy input to the swirl burner due to vitiated air with the potential heat release due to natural gas combustion at the lean limit of operation for the swirl number of 0.5. It can be observed that the potential heat release in the burner for stable operation at the lean limit for the multihole injector and for case C at 500 C was 4.3 kw, more than 30% higher than for the 3.2 kw for the case of 300 lt/min air preheated at the same temperature. Also, the 13 The Reynolds number for each of the 1 mm jets emerging from the multihole radial injector, when the burner operated near the lean limit and evaluated at 20 C is 3000 (and lower for 500 C). For the axial injector, the Reynolds number is For turbulent jet flow is required, although Reynolds numbers of above are used throughout this thesis to ensure fully-turbulent flow at the exit of the burner(s). 224

225 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR Axial Lift-off Axial Sw Multihole Multihole lean 1/φ Sw lean MR Figure 5.18 Comparison between the axial and the multihole radial injectors, for 16.5% oxygen mole fraction (case C) and 300 lt/min and 500 C. See also figure effect of 100 C preheat increase in case D reduced the potential heat release from 4.6 kw to 3.9 kw, a reduction which was balanced by the additional thermal energy added to the system due to vitiated air enthalpy flow rate. The calculated potential heat releases summarised in Table 5.3 indicate that there are fuel savings to be made by using vitiated air, at least as far as changes in the flow and, hence, in the operation of the burner at partial load, are concerned, which result in leaner combustion. It is known from the literature ( 5.1.1) that there are overall savings in energy because of the increased efficiency of power plants incorporating exhaust gas recirculation. The only concern that arises is whether the stable operation of the burner) is influenced by vitiated air at partial load operation (say, at turn-down of a power 225

226 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR Table 5.3 Thermal power due to vitiated air H & T and potential heat release from combustion in the swirl burner at lean limit of operation for the multihole radial injector, except where otherwise indicated. Calculations are for swirl number of 0.5, at which the lower lean limits of operation of this burner were achieved. CASE ENTHALPY FLOW Vitiated Air 400 C 500 C C (300 lt/min) 2.5 kw 3.2 kw D (250 lt/min) 2.1 kw 2.6 kw HEAT RELEASE IN SWIRL BURNER AT LEAN EXTINCTION Ambient 20 C 400 C 500 C Air 300 lt/min 6.2 kw kw Air 250 lt/min 4.8 kw kw Vitiated Air C - no flame possible 4.3 kw Vitiated Air D kw 3.9 kw AXIAL INJECTOR Ambient 5.0 kw kw Air 300 lt/min Vitiated Air C kw 5.0 kw plant). This might be the case in some vitiated air conditions, as the present results show (e.g. case C and 400 C). Lean Limit as a Function of Oxygen Mole Fraction in Vitiated Air Figure 5.19 presents the effect of oxygen mole fraction in the oxidant stream: for this comparison, the extinction limits for case D (0.155 O 2 mole fraction) and the ambient air of 250 lt/min flow rate at 500 C are replotted. The results show that lean stable operation of the burner is improved as the oxygen mole fraction in the oxidant increases. The improvement in the leanest measured flames at swirl number of 0.55, when the oxygen mole fraction was increased by 0.56, from to 0.21 was 46% in terms of equivalence ratio. The improvement was 30% on the basis of the potential heat release, Table 5.3, in order to have the same lean limits at the same swirl number, while the thermal energy added due to vitiated air enthalpy flow rate was identical for both cases. It is useful to calculate the improvement in the lean limit due to a 100 C increase in temperature by choosing case C as reference, because the oxygen mole fraction of the oxidant in this case (i.e ) is in the mid-range of full scale installations (e.g. Domeracki et al. 1995). It is estimated that the 100 C increase extended the lean limit for that case at 0.56 swirl number by about 17%, in terms of equivalence ratio, which is the consequence of a similar reduction in fuel consumption, equation (5.3). If we combine these results we estimate that, in terms of equivalence ratio, a 100 C increase extends the lean extinction limit by as much as a 0.02 change in the oxygen mole fraction. These values are estimated for the particular temperature and oxygen mole fraction variation examined in this experiment, and they are likely to differ for different oxidant conditions because of the non-linear dependence of chemical reaction rate on temperature. It is also likely that it will vary non-linearly with the 226

227 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR % 21% Lifted flames 15.5% 21% Sw unstable flames lean /φ Sw lean MR Figure 5.19 Comparison between extinction limits for 21% and 15.5% (case D) oxygen mole fraction, at 250 lt/min and 500 C temperature for the multihole radial injector. See also figure swirl number; this estimate nevertheless compares well with that of Mastorakos et al. (1995) measured in a counterflow geometry. Effect of Bulk Straining on the Lean Limit Figure 5.20 presents the effect of a 20% increase of bulk straining from /s on flame stability, by comparing the extinction limits for cases C and F, for 400 C. It is clear that the lean limits are the same, within the experimental uncertainty. The lean extinction limits, as a function of the equivalence ratio (figure 5.20a), for the case of the higher bulk strain rate are slightly leaner, a fact which is attributed to the higher heat input in the burner of 2.5 kw for 227

228 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR the higher bulk strain rate as compared with the 2.1 kw for the lower. The momentum ratios which correspond to the lean limits almost collapse on a single curve for both strain rates, suggesting that flame stabilization in such a flow is insensitive to changes of bulk straining of this magnitude. This initially unexpected finding can be justified if one considers that the bulk strain rate for case C is already increased by a factor of about 2 compared with the unheated, unvitiated air case, see Table 5.2. As will become apparent in the analysis below, the bulk strain is already much higher than those for extinction calculated using laminar diffusion flamelet theories and measurements in turbulent counterflow flames (e.g. Kostiuk, 1991; Mastorakos, 1993). Thus a change of 20% in bulk straining by increasing the oxidant flow rate is trivial compared with the % increase of the bulk strain rate due to vitiated air temperature, or ambient air preheating Stability of Gaseous flames: Discussion In the previous sections we investigated the effect of temperature increase and oxygen content decrease on primarily the lean extinction limits of swirl-stabilised vitiated air flames fired by natural gas. Both for axial and multihole radial injection and for all swirl numbers for which stable flames were observed, the lean limits became broader with increasing oxidant temperature or increasing oxidant oxygen mole fraction. Moreover, by combining the effects of temperature and oxygen content variation, it was found that 100 C temperature increase extended the lean limits by as much as 0.02 oxygen mole fraction increase did, for swirl numbers in the midrange of the investigated. In previous studies, Mullins (1954) reported a 0.016/100 C relation for a T-scheme aircraft gas turbine combustion chamber and Mastorakos et al. (1995) measured a 0.023/100 C relation for turbulent counterflow flames. The aforementioned values are comparable, in spite of the different experimental configurations on which these were measured. But of fundamental importance to the operation of swirl burners using vitiated air is whether the combined effects of the temperature increase and the oxygen decrease result in lean extinction limits which are broader than those of the unheated, unvitiated air. Comparison between figures 5.12a and 5.15a shows that, although a flame stabilised in oxidant with and 500 C mole fraction and temperature, respectively, is stronger by about 15% in terms of equivalence ratio than that in ambient air of 20 C for all swirl numbers, a flame stabilised in oxidant environment with oxygen mole fraction but 400 C temperature, extinguishes at comparable equivalence ratios to ambient air at 20 C, implying that flames in hot vitiated air will experience broader lean limits only for certain combinations of temperature and oxygen. The similarity between the measured effects of temperature and oxygen variation on lean flame extinction in this work and in the counterflow flames of Mastorakos et al. (1995) suggests that, theoretical results from laminar diffusion 14 flames would provide insight on how hot vitiated flames extinguish under straining as compared with air, therefore calculations for laminar counterflow diffusion flames (i.e. diffusion flamelets) were carried out in mixture fraction 228

229 . CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR Lifted flames /s /s Sw /s /s 0.5 lean /φ Sw lean MR Figure 5.20 Comparison between extinction limits for bulk strain rates corresponding to 250 (case F) and 300 lt/min (case C), at 400 C and 16.5% oxygen mole fractions for the multihole radial injector. See also figure space with detailed chemistry using the numerical code RUN1DL (Rogg, 1993; Rogg, 1994). The form of the unsteady conservation equations for the energy and the species mass fraction in mixture fraction space, Z, assuming that the flamelet is thin in Z direction and which RUN1DL employs are (Mauss et al., 1990, Peters 1989): 14 It is arguable whether diffusion flame analysis can be applied here for this purpose; it is certain that all flames in this thesis are non-premixed and, thus, diffusion flamelets were selected for the analysis. 229

230 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR Y ρ t i χ = ρ 2 T χ ρ = ρ t 2 st st 1 Le i 2 Z T 2 Z 2 Yi 2 n i= 1 + w& h C i p i w& i (5.10) where ρ denotes the density, t the time, T the temperature, Y i the mass fraction, w& i the (Arrhenius) reaction rate, h i the enthalpy of species i and χ st the scalar dissipation evaluated at stoichiometry (i.e. Z=Z st ). The scalar dissipation is defined as (Peters 1989): 2 Z D y χ = 2 (5.11) where D is the diffusion coefficient. Throughout the calculations was assumed that the Lewis number was unity. For a particular diffusion flamelet (i.e. for given temperature and reactants concentration) the system of equations (5.10) was solved by RUN1DL, using the boundary conditions: Oxidant side T = T Y Y i Fuel i o = Y T = T f = Y o,i side f (Z = 0) (Z = 1) (5.12) The oxidant was considered to consist of a mixture of nitrogen, oxygen and carbon dioxide (depending on the condition) of mass fraction Y 0,i and had temperature T 0 and the fuel was pure methane of temperature T f. The scalar dissipation at stoichiometry was a variable externally supplied by the user of RUN1DL. A detailed CH 4 reaction mechanism using 16 reactive species and 116 reactions was used, supplied with RUN1DL. According to Rogg (1994) equations (5.10) are solved in RUN1DL by use of a modified Newton method and adaptive grids. The solution proceeded by initially considering a weakly strained laminar flamelet (χ st =1) which was gradually exposed to straining by increasing χ st, and solving until the flamelet reached the steady state (for straining less than that for extinction), or extinguished. The stoichiometric scalar dissipation for extinction χ st was assumed to have been determined when a further increase to the scalar dissipation by 0.5 1/s caused extinction. 230

231 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR In order to compare with the bulk quantities used to characterise the flow in this work, the stoichiometric scalar dissipation for extinction output value from RUN1DL was used to estimate an equivalent strain rate assuming constant density and diffusivity (Darabiha and Candel 1992): s = πχ α = Y w st exp 2 erf α 1+ α (5.13) where s is the strain rate, χ st the scalar dissipation for extinction, calculated from RUN1DL, Y the oxygen mass fraction in the oxidant, erf the error function, w is the stoichiometric fuel to oxygen mass ratio 15 and the quantity α is directly related to the stoichiometric mixture fraction, see equation (5.4). Figure 5.21 presents the results from the calculations from RUN1DL plotted in the form of maximum temperature as a function of the logarithm of the reciprocal of the scalar dissipation at the stoichiometric mixture fraction; the curves in figure 5.21 correspond to the upper branch of the S-shaped curve. The curves which correspond to the 20 C/0.21 and 400 C/0.165 flamelets almost collapse onto each other, whereas the 500 C/0.165 shows higher resistance to straining. The strain rates which correspond to the scalar dissipation for extinction calculated with RUN1DL, as estimated from equation (5.13) are 750, 1100 and /s, for the 20 C/0.21, 400 C/0.165 and 500 C/0.165 respectively. The scalar dissipation at extinction for the case of unheated unvitiated air, as calculated from RUN1DL, is higher than for 400 C/0.165; the estimated strain rates from equation (5.13) show the opposite. This, though, does not affect the generality of the results, as the estimated strain rates for extinction of the laminar flamelets are only used here for order-of-magnitude comparison with the bulk strain rates experienced by the turbulent flames of the experiments. Although the general trends for the effect of temperature and oxygen content in the oxidant stream on the extinction of hot vitiated air flames are captured by the theory, the scalar dissipation at extinction and, hence, the strain rates leading to extinction estimated from laminar flamelet calculations are smaller by a factor of 4 than the estimates of the bulk strain rates in the present experiment. Indeed, the bulk (i.e. average) strain rate is likely to be much smaller that the total (i.e. instantaneous) strain rates the flame experiences, if we also considered the turbulent component, raising questions as to how is it possible for a flame to exist. It has been proposed 15 It corresponds to w in equation (5.4). 231

232 5.2 STABILITY OF NATURAL GAS FLAMES IN VITIATED AIR Extinction &2 T [K] &2 &2 ORJχ st ) Figure 5.21 Maximum calculated flame temperature of a laminar diffusion flamelet plotted as a function of the logarithm of the inverse of the scalar dissipation at the stoichiometric mixture fraction for three different oxidant conditions: Blocked squares: air at 20 C, open circles: vitiated air with oxygen mole fraction and 400 C and open squares: vitiated air with oxygen mole fraction and 500 C. Vectors indicate the point of extinction. that the effect of the swirl is to reduce the local velocity magnitude and thus the local strain rates in the region of the flame and, hence, improve stability (Feikema et al 1991). In the context of the recent theories for flame extinction (e.g. Peters and Williams 1983), which treat the scalar dissipation as the important parameter for extinction and which have received confirmation from turbulent opposed jet diffusion flames (Mastorakos 1993), it is unlikely that this can explain why flames can exist in regions of low local velocity magnitude, but high velocity gradients, which imply high local mixture fraction gradients and therefore, high local scalar dissipation rates. By the same token, theories which correlate flame extinction with high local velocities and which argue that extinction occurs because, at the lean limit, the flame moves towards the shear layer of the recirculation zone, like that of Rawe and Kremer (1981), ignore the important fact that the recirculation zone becomes cooler as the equivalence ratio decreases and this has a detrimental effect on the resistance of the flame to straining. Preliminary calculations with RUN1DL showed that for temperatures of the order of 1850 K, which is in the range of temperatures in the recirculation zone for the vitiated air flames of this experiment, the scalar dissipation at extinction for a vitiated air flamelet containing a mere 0.05 oxygen mole fraction, again typical of the oxygen content of the recirculation zone, is greater than 50 1/s as compared to the value of 30 1/s for the 500 C/0.165 flamelet calculated in the previous paragraphs. In the case of the experiments of Rawe and Kremer (1981), the reduction of the maximum temperature of the recirculation zone as we approach the lean limit is about 150 K. 232

233 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR Recall that a 100 C decrease from 500 C (773 K) to 400 C (673 K) of a vitiated laminar flamelet with oxygen mole fraction resulted in a 38% decrease of the scalar dissipation for extinction from /s to 19 1/s, an effect which is expected to be much stronger at temperatures higher than 1500 K, due to the exponential dependence of the chemical reaction rate on temperature. In view of experimental findings, such as those of Rawe and Kremer (1981), and of the present calculations it is likely that small increases of the temperature of the oxidant 16 in the region of flame stabilization can yield flamelets which can survive the local strain rates. Increasing the bulk straining has been found to have weak influence on the lean extinction limit. Indeed, the flame did not respond to an imposed 20% increase on top of that due to preheat, which has already increased bulk straining by a factor of about relative to the unheated unvitiated air case. Mixing (i.e. the macroscopic flow structure in terms of the mean and the fluctuations of the mixture fraction) is not expected to change by the change in the bulk straining if the swirl number is kept constant, since the flow is Reynolds number independent. Assuming that the flow structure remains unchanged from 20% changes of the bulk velocity (i.e. bulk strain rate), the resistance of the flame to extinction can be attributed to the fact that the temperature in the stabilization region is sufficiently high and the flamelets survive the local straining. Another important fact that is likely to contribute to increasing flame resistance to straining is the decrease of the stoichiometric mixture fraction due to vitiation. In the context of the flamelet model of a flame, this means that a vitiated-air flame stabilises even farther away from the stagnation plane of a diffusion flamelet, where the mean stoichiometric scalar dissipation is smaller than in the case of ambient air as oxidant. The gravity of this fact is unlikely to be estimated in the present flow, because of the complexity of the structure of the latter. In addition, the influence of the swirl number on lean extinction is not well explained by the assumption that the mean stoichiometric scalar dissipation decreases with vitiation, because it is hard to see how the swirl number influences the scalar dissipation. Swirl number is likely to influence mixing (in terms of mixture fraction) as well as the local velocity gradients (and, thus, rate of strain), and such a hypothesis is described in the next paragraph. Contrary to small changes in the bulk strain rate, small changes in the swirl number influenced the lean limit, and it was found that the lean limit was extended with decreasing swirl number. This observation is consistent with measurements of Tangirala et al. (1987), who found similar behaviour for a swirl burner with 22 mm throat diameter (compared with the 16 mm of the burner of the present experiments) and with Milosavljevic (1993) who also reported similar findings in a 100 kw swirl burner. The explanation for this phenomenon cannot be traced to 16 The temperature of the fuel, for the present hydrocarbon fuel, has very little effect on the value of the scalar dissipation of extinction. This is due to the small amount of fuel required (Z st =0.055) for combustion and, hence, to the small contribution of the fuel stream in supplying the combustion zone with heat. 233

234 5.3 COMBUSTION OF PULVERISED COAL IN VITIATED AIR turbulence level changes; Milosavljevic (1993) reported an unexpected reduction in turbulence intensities as the swirl number increased from 0.30 to which is not very different to the range of swirl numbers of the present burner - and Tangirala et al. (1987) found that the turbulence intensities increased with increasing swirl up to 1.0. It is likely that mixing which changes with swirl number is responsible for this observation. The large scale laboratory burner of Milosavljevic (1993) enabled the measurements of species concentration inside the quarl which showed that the recirculation zone in the region of the separation streamline becomes progressively leaner with increasing swirl, implying that the stoichiometric mixture fraction contour moves away from the separation streamline. In the context of the previous discussion about the beneficial effect of high temperatures in the recirculation zone on flame stability, this means that with decreasing swirl number the stoichiometric mixture fraction contour (which corresponds to stoichiometric fuel/oxidant mixture, but not necessarily to ignited one) moves closer to hotter regions, and the flame thereby becomes more resistant to straining. Similar shifting of the stoichiometric mixture fraction contour was also observed by Rawe and Kremer (1981) with increasing equivalence ratio for a constant swirl number, and this partly explains why a flame becomes more stable as we move away from the lean limit. This view of the flow 17 provides a plausible explanation for the extension of the lean limits of the flame with decreasing swirl number, in agreement with the theoretical results of the previous paragraph. 5.3 Combustion of Pulverised Coal in Vitiated Air Experimental Method The measurements were made in the near-burner region of the same swirl burner used in the experiments of the previous half of this chapter and described in Chapter 4, using pulverised coal with natural gas support. The oxidant was either pure or vitiated air, the latter produced by mixing air with exhaust gases from combustion of natural gas using the facility of For the present investigations three flow conditions were arbitrarily selected from the range mentioned in 5.1 which were of practical interest, to span oxygen concentrations in the oxidant between 16.5 and 21%, and these are summarised in Table The coal batch was the same as in the investigations of Chapter 4 and contained particles 99% smaller than 85 µm and 25% under 11 µm by volume. 234

235 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR Table 5.4 Flow conditions of the present investigations with pulverised coal Flow Oxygen Swirl number Gas/Coal Temperature of Bulk velocity mole Equivalence oxidant of secondary fraction ratio [ C] air [m/s] A / B / C / o Measured by Milosavljevic (1993) at 20 C In all flow conditions the natural gas flow rate was kept constant at 22 lt/min and corresponded to 8 kw. This though did not correspond to a constant equivalence ratio owing to the different oxygen concentration in the oxidant ( 5.2.3). The bulk velocities were calculated from the flow rate of the secondary air taking into account the effect of the preheat temperature by means of the perfect-gas law. The preheat temperature quoted in Table 5.4 is the thermal equilibrium temperature for particular oxygen mole fraction in the oxidant and was measured during the experiments with the gaseous phase of 5.2. Simultaneous measurements of velocity, size and temperature of single coal particles were made with the combined shadow Doppler velocimeter and two-colour pyrometer instrument presented in Chapter 3. Both the optical magnification of the SDV receiving optics and the calibration characteristics of the pyrometer were as described in Chapter 3 and the reader should refer there for details Uncertainties In the velocity results presented below for flow A, a minimum of 1000 measurements were made and they were arranged in three size classes, namely µm, µm and µm, referred to in the following text by the mean class diameters, namely 18, 36 and 60 µm, to identify the behaviour of the small, medium and larger particles present in the coal batch, resulting in statistical uncertainties no worse that 5% and 10% in the mean and the rms respectively according to Yanta and Smith (1978). The minimum of 12 µm was due to the fixed magnification ratio of the collection optics and was experimentally determined. 17 although not experimentally confirmed in this burner because small scale of the latter did not allow detailed in-quarl species concentration measurements. 18 Note that although the conditions investigated in 5.3 and summarised in Table 5.4 were a subset of the conditions of Table 5.1 in 5.2.2, those conditions referred to with the same letter in Table 5.1 and Table 5.4 were not the same. 235

236 5.3 COMBUSTION OF PULVERISED COAL IN VITIATED AIR Results In the following paragraphs, results from the investigations in the flow conditions mentioned in will be presented in the sequence: first, profiles of the mean axial velocity of particles in all flow conditions, followed by the mean axial velocity profiles and the size as well as the velocity probability functions (pf) at two locations for the ensemble and for the burning particles only for the unvitiated flow. The next heading will present the distribution of the fraction of burning particles along a profile for all three measured flows and this section will conclude with comparison between the measured volume flux of burning particles and the ensemble, which consisted of both burning and non-burning particles as explained in Chapters 3 and 4, for all three flow conditions. Figure 5.22 presents radial profiles of the mean axial particle velocity normalised with the bulk velocity of the secondary air, U 0, calculated at the measured temperature of the oxidant from the perfect gas law for every flow condition. All profiles have been measured at axial location z/d=2.6 which was as close as possible to the exit of the quarl. In all flow conditions, the maximum in the axial velocity profile occurred at radial location r/d=1.25 and was of the order of 0.6 U 0, thus confirming that the flow conditions were Reynolds number independent. The velocity profiles had comparable location of mean zero axial velocity of the gaseous phase, which it can be reasonably assumed is represented in all flow conditions by the velocity of the smallest size class, that of µm particles. In flow conditions A and B, this location was r/d=0.5, whereas in flow C it extended a bit farther radially to r/d=0.7. The behaviour of the 36 and 60 µm particles is also similar in all flow conditions and those of the larger size class present in the recirculation zone on the centreline and have larger mean axial velocity (up to r/d=0.9) than the smaller size classes. The similarity between the mean particle velocity profiles provides confidence that the aerodynamics of the dispersed phase is similar for all reported flow conditions and, hence, the differences in the burning patterns which will be described below are not due to discrepancies in the structure of the flow in the near-burner region. In addition, the similarity in the flow fields allows use of the particle transit time, defined on the basis of the particle bulk injection velocity and the centreline distance between the measurement and the injection locations, to parametrically characterise the influence of the flow conditions on the velocity field. In the case of unvitiated air (flow A), mean profiles of axial velocity were constructed from raw data after they were discriminated into those which were burning and to the ensemble and are presented in figure 5.23(a), where velocity- and size probability functions are also presented. The mean axial velocity included all measured sizes between 10 and 130 µm; that profile is compared in figure 5.23 with velocity profiles for the 18, 36 and 60 µm particles constructed from measurements which were SDV-valid regardless of the measured temperature and hence, include measurements of both burning and non-burning particles, hereafter referred to as the 236

237 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR X O2 =0.21, T=20 0 C (U 0 =29 m/s) µm µm µm U/U r/d X O2 =0.177, T=350 0 C (U 0 =61 m/s) U/U r/d X O2 =0.165, T=400 0 C (U 0 =65 m/s) U/U r/d Figure 5.22 Radial profiles of the mean axial velocities for flow conditions A-C, measured at z/d=2.6 for three particle size classes. ensemble. The characteristic of the axial velocity profile of burning particles is that it is similar to the profile of the 18 µm particles rather than that of the 60 µm. Also the axial velocity of burning particles is consistently smaller than that of the smallest size class regardless of the 237

238 5.3 COMBUSTION OF PULVERISED COAL IN VITIATED AIR burning state of the latter, particularly in the region r/d=1.2 where particles travel in the gaseous flow faster than they do in the region r/d<1. The lack of a velocity measurement of burning particles for r/d=1.5 is because no such particles were found in this location. It can be inferred from figure 5.23(a) that on the centreline the axial velocity pf of the ensemble has positive and negative parts, which correspond to the larger and the smaller sizes respectively. This is confirmed in figure 5.23(b) and it can be seen that on the centreline the bimodal velocity pf is transformed into one which resembles a Gaussian distribution when only burning particles are taken into account. In the case of burning particles, the pronounced maximum in the positive side of the pf is missing, implying that it is more likely that particles which reverse their direction of flow will be burning. Whether the probability to find a burning particle is size-dependent, will be examined below in this section. It was shown in figure 5.23(a) that the mean axial velocity of the 60 µm particles was positive on the centreline, in contrast to that of the smaller sizes. In the size pf of burning particles, figure 5.23(d), the secondary maximum near 80 µm which is observed in the case of the size pf of the ensemble has disappeared, and particles smaller than 50 µm are predominantly present in the size distribution of burning particles on the centreline. It is reasonable, therefore, to assume that those particles which enter the recirculation zone on the centreline due to their initial momentum 19 do not burn on the centreline at z/d=2.6. Whether this is because of the reduced residence time, relative to recirculating particles, in the hot regions due to the penetration, the slower heat-up rate owing to their size, compared with smaller particles, or insufficient oxygen concentration for ignition will be considered in In figure 5.23(c) and (e) the velocity and the size pf of burning particles and the ensemble are plotted for comparison at r/d=0.75. At this location the differences between burning and ensemble are smaller than for r/d=0; there is a tendency for burning particles to have smaller velocities than the ensemble, as already seen in the mean velocity profile, figure 5.23(a), an indication that those particles with higher velocities experienced smaller residence times in hot regions inside the quarl and were not burning therefore at the measurement location. The differences between the size pfs on the other hand are concentrated in the smaller sizes, in which the probability of measuring a burning particle of size smaller than 20 µm was at least 30% smaller than of a 30 µm, although the measured probability of detecting particles of 10 and 30 µm at that location is the same, a result which can be attributed to the detectability limits of the pyrometer. 19 rather than dispersing into it, as in the case of a sudden-expansion flow (Hardalupas et al. 1992). 238

239 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR (a) µm, Burning µm, Ensemble µm, Ens µm, Ens. U/U (b) r/d=0, Ensemble r/d=0, Burning r/d (c) r/d=0.75, Ensemble r/d=0.75, Burning pf 0.15 pf U [m/s] U [m/s] (d) (e) r/d=0, Ensemble r/d=0, Burning r/d=0.75, Ensemble r/d=0.75, Burning pf 0.2 pf d [µm] d [µm] Figure 5.23 (a) Radial profiles of the mean axial velocity of burning particles and the ensemble in flow A; Comparison between axial velocity probability density functions of burning particles and the ensemble in flow A and at (b) r/d=0.0 and (c) r/d=0.75; Comparison between particle size probability density functions of burning particles and the ensemble in flow A and at (d) r/d=0.0 and (e) r/d=

240 5.3 COMBUSTION OF PULVERISED COAL IN VITIATED AIR It is already evident from the preceding paragraphs that a fraction of the ensemble of measured particles is not burning. The fraction of those which are burning has been quantified through the ratio of the number of size-, velocity- and temperature-valid measurements of burning particles and the number of size-, velocity- and temperature-valid measurements of the ensemble (equation 3.23), provided that at least 1000 size-, velocity- and temperature-validated measurements were made. The local burning fraction, as defined above, gives the probability of a encountering a burning particle at a particular location. Figure 5.24 presents radial profiles of the burning fraction for flows A-C. In the case of unvitiated air (flow A) the maximum in the burning fraction occurs at r/d=0.5, and not on the centreline and is about 95%. The burning fraction on the centreline is less than half the maximum at r/d=0.5, about 45%, and reflects the fact that particles which eventually penetrate the recirculation zone and have positive axial velocity at z/d=2.6 near the centreline 20, do not burn, as was inferred from comparison between the velocity and size pfs of burning particles and the ensemble in figure 5.23(b) and (d). For r/d>0.5 the burning fraction rapidly decreases to values close to zero at r/d=1.25, beyond which no single burning particle was found. Although it is likely that the rapid decrease of the burning fraction for r/d>0.5 in flow A is due to the effect of low temperatures. The effect of reduction of the oxygen concentration in the oxidant was quantified by measurement of the burning fraction in flow conditions B and C. The smaller oxygen fraction in the oxidant is not the only difference between flow conditions A and B, C; as has been mentioned above, since the latter two conditions are preheated the bulk velocity is about 2.5 higher and, accordingly, the residence time of particles is smaller than in flow A. Reduction of oxygen concentration in the oxidant has detrimental effect on the burning fraction of measured particles. On the centreline, the reduction was about 40% when the oxygen mole fraction was reduced from 0.21 to and a further reduction by about 60% was observed when the oxygen mole fraction was further reduced from to The preheat temperature and the bulk velocities (61 and 65 m/s respectively) were comparable for both flows B and C and, as we have already seen in figure 5.22, the flow fields were similar. The reduction of the burning fraction is therefore likely to be, in large part, owing to the reduction of the oxygen concentration in the oxidant. Along the radial profile more dramatic reductions of the burning fraction were observed and, for example, at r/d=0.25 and r/d=0.50 for flows B and C it was one order of magnitude smaller than for flow A. 20 which are mostly those larger than about 50 µm (figure 5.23). 240

241 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR X O2 =0.21, T=400 0 C X O2 =0.21, T=20 0 C X O2 =0.177, T=350 0 C X O2 =0.165, T=400 0 C Burning fraction [%] r/d Figure 5.24 Radial profiles of the burning fraction for flows A-C. In order to confirm that the reduction was mainly due to the decrease of the oxygen mole fraction in the oxidant, the radial profile of the burning fraction was measured under the following conditions: once the facility was at thermal equilibrium, following heat-up at conditions which corresponded to flow C, the preheat burner was switched off and pure air of the same mass flow rate as flows A-C was immediately passed through the burner to provide the oxidant to burn the same amount of natural gas as all previous flow conditions. Since the thermal source for preheat was switched off, simultaneous measurements of the velocity, size and temperature of coal particles were made for a minute, then the facility was further heated up for another 5 minutes and the process was repeated about 5 times, until we had more than 100 size- and velocity- validated data. The burning ratio was then calculated and the result is plotted on figure The measured local velocity in this case was within 20% of that measured in flow C and is therefore reasonable to assume that the particle residence time in the flow was not substantially affected. 241

242 5.3 COMBUSTION OF PULVERISED COAL IN VITIATED AIR Under these conditions which corresponded to heated, unvitiated air the measured particle burning fraction was almost 100% in the region 0<r/D<0.6 and dropped for locations r/ D>0.6 to a minimum of about 7% at r/d=1.25, which was the outermost radial location where burning particles were found in this case. On the centreline the burning fraction was almost fourfold higher than that of case B and even higher compared to case C. Compared to the unheated, unvitiated case A, the burning fraction on the centreline became two times higher by preheating the oxidant to 400 C. The effects of heated, unvitiated air on the local burning fraction were more prominent with increasing radial distance, compared to cases A- C. In cases B and C the number of burning particles dropped sharply for off-centreline locations and fell to values of about 10%. The use of heated, unvitiated air kept the burning fraction to levels higher than 80% for locations as distant from the centreline as r/d=0.8. Hence, the increase of oxygen concentration from to 0.21 at the preheat temperature of 400 C resulted in augmentation of the local burning fraction from about 5-10% to around 90% in the region 0<r/D<0.7. The beneficial influence of using heated, unvitiated air as oxidant for the combustion of pulverised coal in increasing the percentage of burning particles in the near-burner region justifies the assumption made earlier (and theoretically investigated in below) that the reduction of oxygen concentration in the region of the recirculation zone by using vitiated air has a potentially detrimental effect on the combustion of pulverised fuel in swirl burners. The results of this thesis for the burning fraction are also in qualitative agreement with the reductions of the unburned carbon when using vitiated air reported by Abbas et al. (1993) and van de Kamp et al. (1992), in their 0.5 and 1.3 MW furnaces, respectively. Reductions of the NO x emissions are likely to occur in vitiated air flames since devolatilisation takes place in an environment containing less oxygen available to form pollutants than in the case of unvitiated flames. As described in Chapter 1, Abbas et al. (1997) measured 20-30% (and up to 70%) reductions in the NO x emissions in their 0.5 MW furnace, but Smart and van de Kamp (1994), who accounted for the increased volumetric flow rate due to the temperature of the vitiated air, even measured 15% increase in NO x in a 1.3 MW furnace for some flame types. Although the burning fraction defined above represents the statistical occurrence of burning particles at a given location, it does not take into account the size of the particles and, therefore, the mass of burning coal relative to the mass of the ensemble. This extra information can be supplied knowing the volume flux 21 of burning particles, defined as (Chapter 2): G = 1 T s V A i i ni i j= 1 u u ij ij (5.14) 21 or mass if we multiply with the density of a particle. 242

243 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR where T s is the total sampling time, V i is the volume of a particle, u ij is the axial velocity of particle j in size class i, n i is the total number of particles in size class i and A i is the size (m 2 ) of the sampling space (through which particles flow) with unit vector normal to the fringes. The volume flux expresses the burning volume of particles per unit time and area, rather than the number and therefore correctly accounts for the larger burning particulate volume transported by larger particles. The latter is of great significance in swirl burners because the heat and volatile matter released during combustion as well as the subsequent burnout of the remaining carbon in the char depends on the total burning mass rather than number. Figure 5.25(a)-(c) presents radial profiles of the positive and net volume flux of burning particles and the ensemble for flow conditions A, B and C respectively. The positive flux has been calculated from equation (5.14) taking particles with positive velocities only into account. In flow A the positive and net volume flux of burning particles throughout the width of the recirculation zone and up to about r/d=0.7, with the exception of the centreline, is almost identical to the volume flux of the ensemble as a consequence of over 90% burning fraction in the same locations. In the shear layer between the recirculation zone and the main forward flow, at r/d>0.7, the volume flux of burning particles decreases with increasing radial distance while the flux of the ensemble increases. As a consequence, an increasing amount of particulate mass is lost unburned at the outmost locations of the flame, where the velocities are higher than 15 m/s implying that the particles will be quickly swept downstream before they ignite. Although the shape of the radial profile of the flux of the ensemble has not changed substantially in flows B and C, the amount of flux of burning particles reflects even more prominently the reduction of the burning fraction in flows B and C compared with flow A, and only a fraction of the ensemble was burning and was confined to the region 0<r/D<0.6, where the net burning flux at all locations was negative. It would also be reasonable to estimate that the integrated volume flux of the ensemble along a profile is an order of magnitude larger than that of the burning particles Ignition of char particles: Discussion In the previous section we saw that the use of vitiated air resulted in dramatic decrease of the volume flux of burning particles implying an equivalent reduction in the combustion efficiency in this region of the flow. Similar observations, although not of the same magnitude, have been made in semi-industrial scale furnaces. In their 1.3 MW furnace van de Kamp et al. (1992) reported reductions in the carbon burnout which depended on the coal- and the flametype and, for example, for Goettelborn coal the burnout decreased form 99.1% to 91.2% when the oxygen concentration in the oxidant was reduced from 21% to 11%. In a 0.5 MW 243

244 5.3 COMBUSTION OF PULVERISED COAL IN VITIATED AIR furnace Abbas et al. (1993) reported a reduction from 97% to 90% for a reduction of the oxygen concentration from 21% to 14%, which depended on the injector type. In order to provide quantified understanding of the effects of decreasing the oxygen concentration in the oxidant and the residence time of particles in the high-temperature region of the recirculation zone due to the increased volumetric flow a theoretical study of the heat-up and ignition of a char particle was carried out. In what follows, the influence of the temperature of and of the oxygen concentration in the oxidant on the heat-up and ignition of char particles was investigated. The equation which describes the transient heating and combustion of a char particle 22 and includes heat transfer by means of convection and radiation is (Wall 1987): mc p dt =πd dt 2 Nuk q( H) + d g 4 4 ( T T) +εσ( T T ) g a (5.15) where m is the mass of a particle, d its diameter and C p its heat capacity, T the particle temperature and T g the temperature of the surrounding gases, H the heating value of coal, Nu the Nusselt number assigned the constant value of 2 23, k g the gas thermal conductivity, ε the emissivity of the particle, σ the Stefan-Boltzmann constant and T a the temperature of the environment to which thermal radiation losses were calculated. The char combustion rate, q, has been estimated from (Field et al. 1967): q = 1 K s P K d (5.16) where P 0 was the oxygen partial pressure. The surface reaction rate K s (kg m -2 s -1 [atmo 2 ] -1 ) in the equation above was calculated from an Arrhenius-type expression (Field et al. 1967): More complex models exist which take gas-phase ignition into account (e.g. Annamalai and Durbetaki 1977; Lau and Niksa 1992), but since the purpose of this section is to estimate time scales of ignition of small coal particles which contain little volatiles, a simple char ignition model was preferred. In any case, small particles can ignite on the particle surface rather than in the gaseous phase during pyrolysis (Annamalai and Durbetaki 1977). 23 This value corresponds to heat transfer between a particle and quiescent fluid. The Nusselt number depends on the slip velocity between the gaseous and particulate flows and its detailed calculation in the present flow requires knowledge of the particle and gaseous velocities along a particle trajectory. Accurate simultaneous measurement of these two variables requires suitable instrumentation for Lagrangian particle tracking (such as of Siu 1996) and is beyond the scope of the present instrumentation, which measures in Eulerian frames of reference. The assignment of the value 2 is not expected to harm the generality of the conclusions of this section.

245 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR Flux [Arbitrary units] Ensemble, positive Ensemble, net Burning particles, positive Burning particles, net r/d Flux [Arbitrary units] r/d Flux [Arbitrary units] r/d Figure 5.25 Radial profiles of the positive and net volume flux of burning particles and the ensemble (a)-(c) for flow conditions A-C respectively. 245

246 5.3 COMBUSTION OF PULVERISED COAL IN VITIATED AIR K s = A E0 RT 0e (5.17) where the activation energy E 0 and the surface reaction rate constant A 0 are coal-specific; they were taken from Abbas et al. (1994) and are presented along with other constants in Table 5.5. The diffusion reaction rate K d was estimated from the following equation derived by Field et al. (1967) and quoted by Wall (1987), on the assumption of an isolated coal particle surrounded by stagnant fluid, where oxygen diffuses to the particle via the boundary layer on its surface and the sole combustion product is carbon monoxide: K d = T + T 2 g 0.75 / d (5.18) In equation (5.18), dependence of the gas conductivity on temperature has been taken into account. The mean of the particle- and the free-stream gas temperatures has been assumed as a representative temperature for the gas in the boundary layer surrounding the particle. Table 5.5 Constants used in the model described in and their values Constant Value E (kj kmol -1 ) A (kg m -2 s -1 [atmo 2 ] -1 ) C p 1200 (kj kg -1 K -1 ) Nu 2 ( ) ρ coal 1500 (kg m -3 ) ε 1 ( ) σ (W m 2 K 4 ) The system of equation (5.15)-(5.18) can be solved to yield the temperature of a particle as function of time, assuming that the particle had initially low temperature and was suddenly immersed into a hot gas. The solution of equation (5.15) gives the temperature of a char particle as a function of time; however, a criterion must be used to estimate the time of char ignition from the instant the particle was exposed to the hot gas. In the present work we followed previous investigations on the subject (reviewed, for example, by Essenhigh et al. 1989) and used Semenov s thermal explosion theory, which states that ignition will occur at an instant when the temporal particle temperature gradient as well as the second derivative of the particle temperature vanish 246

247 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR simultaneously. These conditions, mathematically described in (5.19) correspond to the van t Hoff and Taffanel and le Floch conditions, respectively. 24 For the particle size range used in the experiments it is likely that heterogeneous ignition occurs (Essenhigh et al. 1989), thereby justifying the selection of this ignition criterion. In addition, as mentioned earlier in this thesis the coal used here was pulverised about 10 years ago and, hence, the amount of volatiles still present in the coal was a fraction of the volatile content immediately after pulverisation. As Essenhigh et al. (1989) pointed out, this criterion is restricted to those particle heat up histories for which the net heat (right-hand side of equation 5.15) becomes zero. In practice, and this was the case in our calculations, it is possible that the second temporal derivative of particle temperature vanishes, while the gradient is always positive. In order to overcome this problem, Chen et al. (1984) suggested that the ignition criterion be extended to assume that ignition occurs when the right-hand side of equation (5.15) experiences a minimum and the second temporal derivative of particle temperature is zero. In mathematical notation this is: dt d T net dt 2 dt ( ) Q 0 0 and = 0 2 (5.19) The system of equation (5.15)-(5.18) was numerically solved using a Runge-Kutta method of fourth order and the ignition criterion of equation (5.19) was applied at each time step. The initial conditions were of the Dirichlet type; the particle assumed the temperature of the vitiated air and the fluid assumed a fixed temperature quoted in the results. The calculation was interrupted when the particle reached thermal equilibrium with the surrounded gas. Figure 5.26 presents the time until a particle ignited or heated up to 1700 K, on the assumption that it entered the recirculation zone and stayed inside for time t RES (which is a guessed value in order to investigate the effect of this parameter) before it escaped the recirculation zone and continued its journey to the axial measurement station. The idea behind this simulation is that a particle injected axially against in the flow will enter the recirculation zone before it disperses away from it. The actual trajectory it will follow cannot be simulated by this model and requires use of computational fluid dynamics tools. 25 Here, it is assumed that it stays in the hot recirculation bubble for a predefined time and then it escapes to the main forward flow formed outside the recirculation zone. The temperature of the gas and the oxygen partial pressure in the recirculation zone as well as the main forward zone have been extrapolated 24 Given the heat release rate R and the heat losses due to convection L, it is not only required that R=L, but also that the curve corresponding L is tangent to the R-curve, and that point is called critical condition for ignition. This is a, of course, the basic idea of the criterion, but there are more conditions to be satisfied, mathematically described by equation (5.19). For an analysis of thermal ignition the reader should refer to a standard combustion textbook (e.g. Griffiths and Barnard 1995). It should be noted here that thermal ignition is an inevitability only in adiabatic conditions (Griffiths and Barnard 1995). 25 where, of course, other uncertainties such as the turbulence model would introduce departures of the solution from the actual process taking place in the flow. 247

248 5.3 COMBUSTION OF PULVERISED COAL IN VITIATED AIR from measurements by Milosavljevic (1993) and of this thesis. For the case of ambient air as oxidant the oxygen partial pressure in the recirculation zone was 0.07 atm and the temperature 1700 K, whilst in the case of the vitiated air these numbers were 0.03 atm and 1850 K. The initial particle temperature was 300 K and 600 K respectively, i.e. it assumed the temperature of the carrying fluid. 26 These conditions used in the simulations and detailed in figure 5.26(a) and (b), which show the results from the simulation of the case of the ambient and the vitiated air respectively. The temperature of 1700 K was selected as a plausible threshold level based on the detectability limits of the pyrometer mentioned in Chapter 3. The horizontal line on the graphs represents the transit time of a particle from point of injection to the measurement location assuming rectilinear motion. Using the simple char heat-up and ignition model described in the previous paragraphs it was estimated that in the case of flow A, represented by the simulation corresponding to figure 5.26(a), particles up to about 30 µm are likely to be detected by the combined instrument - and therefore be interpreted as burning- at the measurement location whilst even larger particles of the order of 40 µm are likely to have ignited. In the case of flow C, figure 5.26(b) the largest particle which would had sufficient time to heat up and be detected by the pyrometer was 30% smaller than in case A. However, not all particles ignited in this case, mainly due to the low oxygen concentration in the recirculation zone which was measured to as low as 3% (e.g. figure 5.6). It must be noted that the actual transit time is likely to be larger than what is assumed in figure 5.26(a) and (b) due to the different trajectories particles can follow until they reach the plane of measurement. In both figure 5.26(a) and (b) calculations showed that it is beneficial that particles stay inside the recirculation zone as long as possible to accomplish rapid heating up. Particles, however, would not ignite unless sufficient oxygen is brought in contact and this is the case in flow C (and presumably B). The oxygen concentration near the high-temperature region of the recirculation zone can be increased by using larger amounts of excess air, as has already been noted by Floris (1981), whose work suggested a minimum of 30% for stable combustion. In the present work, however, use of higher amount of excess air was limited by the stability of the gaseous pilot flame. It is inferred from the results from the simple model of this section that in the case of vitiated air the reduction of the oxygen mole fraction in the recirculation zone counterbalances the increase of the temperature. Although the reaction rate employed in the model is of the Arrhenius type, and as such, it has stronger dependence on temperature than on reactant concentration, In the case of vitiated air the coal injector acted as a heat exchanger between the vitiated air stream and the primary air, and it was assumed that within the length of the burner the primary air assumed the temperature of the vitiated air stream. This assumption which is not expected to be far from reality, given the length of the tube (of the order of 100 internal injector diameters) and the mass flow rate of the secondary (vitiated) air relative to the primary air (30:1). 248

249 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR Initially T=300 K; For 0<t<t RES (P O2 =0.07, T=1700 K) and for t>t RES (P O2 =0.12, T=1400 K) τ [ms] 20 Ignition time 18 t =0.5 ms 16 RES 8 Ignition time 7 t =1 ms RES 6 Ignition time 5 t =2 ms RES 4 Time for T>1700 K 3 t =0.5 ms RES 2 Time for T>1700 K 1 t =1 ms RES 0 Time for T>1700 K t =2 ms RES d [µm] Initially T=600 K; For 0<t<t RES (P O2 =0.03, T=1850 K) and for t>t RES (P O2 =0.08 or 0.10, T=1500 K) τ [ms] 20 No ignition t =0.2 ms, P = RES O d [µm] No ignition t RES =0.5 ms, P O2 =0.08 Ignition time t RES =0.5 ms, P O2 =0.1 Time for T>1700 K t RES =0.2 ms, P O2 =0.08 Time for T>1700 K t RES =0.5 ms, P O2 =0.08 Time for T>1700 K t RES =0.5 ms, P O2 =0.1 Figure 5.26 Time until ignition or until a char particle reaches 1700 K as a function of particle size on the assumption that the char particle initially entered the recirculation zone and escaped it after time t RES for two different conditions simulating (a) flow conditions A and (b) flow conditions C. Horizontal lines indicate the transit time of a particle from point of injection to the measurement location assuming rectilinear motion with the bulk injection velocity. 249

250 5.4 SUMMARY OF CHAPTER 5 the vitiated air temperature of the experiments cannot account for the reduction of the reaction rate due to the low oxygen concentration in the recirculation zone. In the case of heated, unvitiated air the benefits of higher reactant temperature (and temperature in the recirculation bubble) compared with the unvitiated case and oxygen concentration in the recirculation zone compared with the vitiated air case are shown in the increased burning fraction in figure Summary of Chapter 5 In this chapter the effect of vitiated air as oxidant on the lean stability limits of swirl stabilised, natural gas flames in a 10 kw burner were investigated as a function of the swirl number and on the combustion characteristics of single coal particles, in gas-supported flames, measured by the combined instrument described in Chapter 3 as a function of oxygen concentration. In the experiments with gaseous flames the swirl number was varied between 0.70 which corresponded to the maximum obtainable and 0.50, the minimum swirl number at which flames could be stabilised. The range of the oxygen mole fraction in- and the temperature ofvitiated air examined were and K respectively. A multihole radial and an axial geometry of fuel injection were used at two volume flow rates of 250 lt/min and 300 lt/min at 20 ºC. In the experiments with pulverised coal the oxygen mole fraction was varied between and 0.21, the preheat temperature between 350 ºC and 400 ºC whilst the swirl number and the volume flow rate at 20 ºC were kept constant at 0.55 and 300 lt/min respectively. The main findings are summarised in and Gaseous Flames 1. Lean extinction limits became leaner with decreasing swirl number, for all cases examined here and for all swirl numbers at which flames could be stabilised. The improvement depended on the temperature and the oxygen mole fraction of the oxidant and was of the order of 10% in terms of equivalence ratio, between swirl numbers of 0.70 and The lean extinction limits of the burner became leaner with increasing temperature at constant oxygen concentration and swirl number and increasing oxygen concentration at constant temperature and swirl number, when a multihole radial injector was used. Quantitatively, a 100 C temperature increase from 400 to 500 C at oxygen mole fraction changed the lean limits by 17% at most, corresponding to the minimum swirl number at 27 char combustion is a diffusion-limited process and as such, depends on the oxygen diffusion coefficient (equation 5.18), which becomes more important with decreasing particle size. 250

251 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR which flames were stabilised, in terms of equivalence ratio; a mole fraction increase, from to 0.21, of the oxygen mole fraction resulted in a change by 47%, at 500 C at the same swirl. It was estimated that for the range of temperatures and oxygen mole fractions examined here, a temperature increase of 100 C is counterbalanced by a reduction in the oxygen mole fraction by Radial fuel injection could sustain about 20% leaner flames than axial injection at vitiated air conditions of oxygen and 500 C. On the contrary, axial injection of fuel resulted in about 10% leaner limits compared with multihole-radial injection, when the burner was fired in unvitiated, unheated air. The lean limits of operation with axial fuel injection for the smallest swirl numbers were always found to be insensitive to the temperature rise of the oxidant stream, although became leaner with oxygen content rise, and radial injection was preferred. It is suggested that this is due to insufficient mixing between the axial fuel jet and the coflowing vitiated air at the lean limit, due to the lower injection velocities compared with the multihole radial injector. 4. A minimum swirl number at which stable flames could exist has been found which depended on the temperature of the oxidant. For the case of oxygen mole fraction, that minimum corresponded to 0.56 for 400 C and it was extended to 0.5 for 500 C, a change which is larger than the 1% uncertainty of the swirl number measurement (Milosavljevic 1993). For unvitiated air and swirl numbers below 0.5, although flames could be stabilised at oxidant temperatures of 20 C, a temperature rise by 480 C resulted in unstable flames, contrary to results at higher swirl numbers, probably due to insufficient mixing caused by the higher oxidant velocities and weak recirculation zone. 5. Both for the case of unvitiated air at 20 C and vitiated air at 400 C with oxygen content, the effect of increasing the equivalence ratio was to increase the maximum measured temperature on the centreline at the quarl exit by 100 C, contrary to expectations from the theory of turbulent jet flames, probably due to the balance between heat production and heat losses. The mean temperature profiles for the unvitiated and the vitiated air cases depicted qualitative differences, with no observable maximum in the case of vitiated air, in contrast to the unvitiated. Results for the calculated mixture fraction from species concentration measurements suggested that the flame surrounded the recirculation zone in the case of vitiated air, while in the case of unvitiated air the flame was present inside the quarl. 6. Comparison between measurements and theoretical results using the laminar flamelet formulation in mixture fraction space and detailed chemistry showed that, as the theory predicted, a 500 C/0.165 condition resulted in better lean limits than that for the 20 C/unvitiated air whereas a 400 C/0.165 in similar lean extinction limits. This suggests that bulk quantities, such as the strain rate, cannot be solely used to predict extinction without 251

252 5.4 SUMMARY OF CHAPTER 5 taking the temperature of the region where the flame is stabilised into account. It is further suggested that flames can be stabilised despite the high strain rates because of the high temperature of the recirculation zone, which results in a substantial increase of the scalar dissipation (and therefore, the strain rate) for extinction. This suggestion is also encouraged by the observation that with decreasing swirl number the lean limit becomes broader towards smaller equivalence ratios, a fact which is accompanied, according to the in-quarl measurements of Milosavljevic (1993), by a shifting of the stoichiometric mixture fraction surface towards hotter regions Pulverised Coal Flames 1. The particle ensemble mean axial velocity profiles normalised with the bulk velocity of the secondary flow, which was 29, 61 and 65 m/s for the case of 0.21, and oxygen mole fraction in the oxidant, were similar in all three flows for the 18, 36 and 60 µm particles implying that the structure of the flow was similar and hence, use of the transit time of particles between injection- and measurement location was sufficient to characterise the effect of increased volumetric flow rate due to preheating on particle residence time in the high-temperature region of the recirculation zone. Measurements at z/d=2.6 showed that the velocity distribution of the ensemble on the centreline included particles with positive and negative velocities, of which only those which reversed their direction of motion were burning whilst at radial distance r/d=0.75 this difference was only marginal. Both observations indicated that it is unlikely that particles which went through the recirculation zone did ignite. 2. The influence of reducing the oxygen concentration in the secondary stream by mixing with combustion products was to decrease the fraction of burning particles across a radial profile at z/d=2.6. The maximum burning fraction for oxygen mole fraction of 0.1 and no preheat was measured at r/d=0.5 and was 94%, whilst the centreline value was 44% and only 1% was burning at r/d=1.25. Reduction of the oxygen mole fraction to resulted in reduction of the maximum burning fraction by more than 60% to 30% and at an oxygen mole fraction of the maximum measured burning fraction was about 12%. It is likely that the reduction of the oxygen concentration has the largest influence on the reduction of the burning fraction, rather than the decrease of the residence time in the high-temperature regions, as confirmed by measurement with air preheated at about 400 o C as oxidant, in which case the burning fraction increased at all locations along a radial profile compared with both heated vitiated and unheated atmospheric air. The augmentation in the burning fraction varied between twofold on the centreline compared with unheated atmospheric air to about tenfold compared with vitiated air with oxygen mole fraction heated at 400 ºC at almost all measured radial locations. 252

253 CHAPTER 5 GASEOUS AND PULVERISED COAL REACTING FLOWS IN VITIATED AIR 3. Use of a theoretical model and experimental results to study the influence of the increased temperature of- and the decreased oxygen concentration in the oxidant when vitiated air was utilised indicated that, despite the higher temperature of the recirculation zone because of preheating the reduction of the oxygen mole fraction in the oxidant has detrimental effect on particle ignition, and it is unlikely that particles will ignite while inside the recirculation zone. 253

254

255 CHAPTER 6 Closure

256 6.1 CONCLUSIONS The contribution of this thesis is focused in the field of optical instrumentation applied to combustion. It presented a single particle counting optical instrument, based on shadow Doppler velocimetry and two-colour pyrometry, for the spatially-precise simultaneous measurement of the velocity, size and temperature of isolated irregular fuel particles (as well as volume and mass flux of an ensemble of particles). The instrument was applied in 10 kw gas-piloted swirl burners, open and confined by cylindrical furnaces with a lateral exit, using either atmospheric or vitiated air as oxidant, where the velocity, size and temperature of particles were measured, as well as the fraction and the volume flux of those particles which were burning. The measurements also included those of the temperature, species and extinction limits of naturalgas flames, stabilised using atmospheric or vitiated air as oxidant. This first part of this chapter summarises the most important conclusions of the work divided into sections concerning the instrumentation development and the results from the flows respectively, whilst the second includes recommendations for further research. 6.1 Conclusions Instrumentation A novel optical instrument was developed for the simultaneous measurement of the velocity, size and temperature of single particles of arbitrary shape with high spatial accuracy, as well as the volume flux of a particle ensemble. The instrument is based on a shadow Doppler velocimeter (SDV), which measures the velocity and size of isolated particles, and a twocolour pyrometer for the temperature of glowing particles. SDV is an imaging technique which measures the cross-sectional projected particle area, from which an area-based diameter was defined in this thesis accurate to within 10%. Because of its principle of operation SDV is not amplitude-based technique and, thus, no calibration is required, whilst the measurement is less sensitive to optical misalignment than common sizing techniques of irregular particles, based on the amplitude of diffracted light. The current design of the SDV allows the measurement of a second velocity component with accuracy better than 15% using singlecomponent transmitting optics. The temperature of a burning particle is estimated from the ratio of the photomultiplier signal amplitudes due to the thermally-emitted radiation, measured at two wavelengths, namely and nm. The use of wavelengths in the visible spectrum allows use of standard glass optics required for the simultaneous operation of the SDV. In order to minimise the uncertainty of temperature measurement which stems from the fact that coal particles can undergo homogeneous (volatile) or heterogeneous (char) combustion, which are characterised by different emission characteristics, a discrimination criterion, based on the work of Israel (1997), was proposed. The criterion relies on prior experimental findings, namely that the sooty volatile flames surrounding coal particles are optically-thin and that the char is several hundred degrees colder that the envelop volatile flame. The measurement of 256

257 CHAPTER 6 CLOSURE particle flux using the combined SDV/two-colour pyrometer does not suffer, as phase Doppler anemometry does, from the so-called trajectory ambiguity effect, caused by the Gaussian distribution of the irradiance of the laser beams and the accuracy of particle volume flux measurement is better than 20%, assessed from measurements of 100 µm spheres in a turbulent water flow through a vertical square channel with Reynolds number of Disadvantages of the SDV/Two-colour pyrometer include the probably lower particle concentration limit compared with sizing techniques for spherical particles, due to the forward scatter operation mode and the dynamic size range of the instrument which is currently limited to 10:1. The pyrometric part of the instrument has the disadvantage that its sensitivity is low at particle temperatures below approximately 1500 K (which is a size-dependent limit) Pulverised Coal Flames Combustion in vitiated air These experiments addressed the question of how the combustion of pulverised coal is affected by the use of vitiated, instead of atmospheric, air as oxidant. The investigations were carried out in a 10 kw swirl burner with 16 mm throat diameter. Simultaneous particle velocity, size and temperature measurements were made in the near-burner region of open flames using vitiated air at 400 C as oxidant. The main finding was that the number of coal particles which were found to be burning at the exit of the quarl decreased by about 60% with decreasing oxygen concentration in the oxidant from 17.7% to 16.5%, for constant oxidant temperature and the total number of burning particles decreased by about 70% relative to use of unheated unvitiated air. In an attempt to separate the influence of the oxygen mole fraction, oxidant temperature and particle residence time in the near-burner region as a result of the variation of volumetric flow rate with oxidant temperature the number of burning particles was also measured when atmospheric air heated at 400 C was used. The result was an increase in the number of burning particles by a factor of 10 relative to the case in which vitiated air containing 16.5% oxygen and 400 C temperature. These experimental findings suggest that although devolatilisation is likely to be more intense in the case of vitiated air due to the higher peak gas temperatures (see Chapter 1) the flame length will increase with vitiation. Calculation of the heating-up and ignition of an isolated char particle submerged in a hot-gas stream, using a simple one-step reaction scheme suggested that the reductions of the number of burning particles mentioned above could be due to the insufficient particle residence time in high temperature regions for the given local temperature and oxygen content of the continuous phase. 257

258 6.1 CONCLUSIONS Combustion in atmospheric air A strong correlation between particle burning fraction and temperature of the continuous flow has been found in the near-region of a 10 kw swirl burner with 18 mm throat diameter. The parametric studies included swirl numbers of 0.41 and 0.57, overall gas equivalence ratios of 0.69 and unity and primary to secondary air momentum ratios of 1/30 and 1/40. For a given equivalence ratio of 0.69 and momentum ratio of 1/30 the increase of the swirl number from 0.41 to 0.57 created a zone of forward flow on the axis of the burner with 700 K temperature, compared with the more than 1700K of the swirl number of The condition of swirl number of 0.57 simulated an internally-staged flame, such as that of Smart and Weber (1987), in which particles ignite farther downstream compared with an unstaged flame, and the delayed ignition results in low NO x emissions. The number of burning particles near the axis decreased from about 85% to less than 5% on changing the swirl number from 0.57 to That was due to the main forward flow on the axis of the burner for swirl number of 0.57 which resulted in slow particle heat-up, as a consequence of low local temperature of the continuous phase. Reductions in the fraction of burning particles (and the corresponding volume flux) of the order of 30-40% were observed when the overall gas equivalence ratio was increased from 0.69 to unity. This result was attributed to the fact that increased overall gas equivalence ratio decreased the available oxygen in the recirculation zone and, thus, delayed particle ignition. A similar trend is likely to be observed in a full-scale air-staged burner, where the stoichiometry in the primary combustion zone downstream of particle injection is rich. Although the measured size-velocity correlations were weak for most conditions and locations in the near-burner flow, the substantial reduction of about 80% of the maximum particle burning fraction, when the momentum ratio was reduced from 1/30 to 1/40, is likely to be an aerodynamic phenomenon (reduced residence time due to faster response of particles to the mean swirling flow). The smaller particle injection velocity, due to the smaller primary air flow rate in the case of 1/40 ratio, caused particles to be swept away from the recirculation zone towards the colder shear layer of the flow, where gas temperatures are smaller and, thus, particle ignition is less likely. However, this flow pattern is likely to yield high NO x emissions, since particles are inevitably going to ignite in the shear layer where oxygen availability is high. The influence of confining the flow by a cylindrical furnace with a lateral exit was also addressed in this thesis. Simultaneous measurements of particle velocity and size were made in both the near- and the far-burner region of the flow for swirl number of No size-velocity correlation was found in either flow region. Particularly in the far-burner region of the flow, the trajectory angle with which particles approach the lateral exit of the furnace was size-independent, implying that no inertial separation ( centrifuging ) effects were present in the particle trajectories. According to this result, two mechanisms were suggested to explain the aerodynamic separation of particles from the flow in a geometrically-similar full-scale pressurised furnace: slag-film formation on the furnace wall in the near-burner region which drips down due to gravity and particle settling in the downstream region due to gravity. 258

259 CHAPTER 6 CLOSURE Natural gas flames The question of whether vitiated air can be used as oxidant, such as that used in combined cycles with supplementary firing, or the advanced PFBC cycle, without deterioration of the flame extinction limits was addressed by parametric studies of mainly the lean extinction limits of natural-gas swirl flames in vitiated air with variable oxygen content and temperature, as a function of the swirl number. An axial and a multihole radial injector were used and measurements were made at two different oxidant flow rates which corresponded to different bulk strain rates. The lean extinction limits, in terms of the overall gas equivalence ratio, became leaner for both injectors by 10%, on average, with decreasing swirl number. Higher temperatures resulted in improvement of lean stability limits of the flames by about 20%, whilst lower oxygen mole fraction in the oxidant deteriorated flame stability by 50%, for example, from 21% to 16.5% oxygen concentration. The influence of these parameters was more prominent in the case of radial injection, which showed stronger improvement of flame stability with increasing temperature compared with axial injection. This behaviour is likely to be due to the more intense mixing between the radial fuel jet and the cross-flowing vitiated secondary air in the burner. An empirical correlation was found between the influences of the oxygen in and the temperature of the oxidant: 2% variation of the oxygen mole fraction had the same effect on the lean extinction limit as 100 K variation of the temperature in the opposite direction. This result implies that careful selection of the operational parameters of the burner is important to ensure that the operation of swirl burners is not deteriorated by the use of vitiated air. Discussion of the experimental results with the aid of calculation of the extinction limits of laminar flamelets using the RUN1DL code suggests that the vitiated air flame withstands the high local straining in the swirling flow because the high temperature of the recirculation zone compensates for the decreased oxygen mole fraction in the vitiated air. 6.2 Recommendations for Further Research Improvement of the Instrumentation The measurements of this thesis showed that the combined SDV/two-colour pyrometer can satisfactorily measure simultaneously the velocity, size and temperature of coal particles in omni-directional flows. There are, however, two improvements to be made which are concerned with the sizing and the pyrometric parts of the instrument. Immediate improvement in the sensitivity of the pyrometer can be achieved by use of a mirror (behind the pair of front lenses) with larger reflecting surface and/or Péltier coolers in the detectors, by which the photomultiplier voltage and, thus, the pyrometer sensitivity, could be further increased without reducing the signal to noise ratio. Improvement of the sensitivity of the pyrometer is desirable, so that smaller or colder particles can be detected. The obvious change would be to convert the 259

260 6.2 RECOMMENDATIONS FOR FURTHER RESEARCH present visible radiation pyrometer into one based on infrared (IR) light, because in the IR part of the spectrum more thermal radiation is emitted for given temperature. For the IR special lenses are required (Habib and Vervisch 1987) which are made from materials (such as ZnSe or Sapphire) which are opaque to visible radiation, whilst SDV relies on a continuouswave laser beam in the visible spectrum, most commonly from an Ar + or He-Ne laser. The increase of the sensitivity of the pyrometer, though, will be less than the case of IR region. This is a severe disadvantage, for although one could separate the SDV and pyrometer parts by use of different collection lenses albeit based on common synchronisation signals it will be almost impossible, in practice to ensure that both lenses observe the same point of space. In addition, aberrations will result in different sizes of the probe volume observed by the two instruments, which is, again, undesirable. An alternative wavelength range for the pyrometer is the near-ir region of the spectrum (up to 1000 µm), in which standard glass lenses can be used. An alternative detector in the SDV part of the instrument should be used in order to alleviate the problem caused by non-measurable particles with near-zero velocity components normal to the array. As explained in Chapter 2 this is a result of use of a linear photodiode array in the current instrument. Use of a two-dimensional detector would avoid this difficulty because it would be possible to obtain complete particle shadow images with one snapshot, rather than by continuous sampling of 1-D slices. This arrangement, though, would impose new difficulties, such as increase of the complexity of the electronics and the amount of data required for each measurement. A standard CCD camera cannot be used because of the 25 Hz refresh rate. Even if they were fast enough, CCD cameras have resolutions of the order of 512x512 pixels, which are unnecessarily high for this kind of measurement as the large amount of image data limits the maximum particle data rate because of computer-hardware limitations to date. Two dimensional detectors can be made out of, say, optical fibre-connected (at least) 32x32 matrices of avalanche photodiodes so that the sizing dynamic range of the instrument is no worse than the current arrangement. An additional advantage of a two-dimensional array, other than the instantaneous particle cross-sectional area measurement, is that an SDV with a two-dimensional detector could be used as a velocimeter, by taking two consecutive snapshots of a particle shadow. An instrument has been developed in the past which uses the passage of the image of seeding particles over a diode array for the measurement of flow velocity (the diode-array velocimeter, Devenport and Smith 1993). A simpler quasi 2-D detector arrangement for the SDV would make use of two linear arrays placed at right angles to each other, so that they form a Greek cross. Such a detector would increase the amount of data per measurement by a factor of only 2, compared with the linear arrangement (and unlike the 16-times increase of a 32x32 matrix) and it is possible that it would improve the response of the SDV to measurement of particles with near-zero velocity components normal to either the one or the other array. 260

261 CHAPTER 6 CLOSURE The data rates of the present SDV/Two-colour pyrometer can be immediately improved by replacement of the pyrometer transient recorder, which is limiting the overall acquisition rates of the instrument, by on-board A/D cards. The acquisition rates of the instrument would increase by more than one order of magnitude; it will be then the sizing part which will limit the maximum particle data rate to the present 300 Hz instead of the pyrometric part which is currently the limiting one Coal Combustion One of the most difficult aspects of interpretation of the measurements of the present work was the isolation of the influence of the turbulent flow on particle trajectories from those of the particle size and local temperature on particle heat-up and ignition. Drop-tube furnaces have been used in the past and are still used in order to obtain fundamental information about coal combustion processes, but the flow is laminar and not representative of the interaction between particles and a turbulent flow. Drop-tube furnaces are indispensable when it is required to study the heat-up/volatile structure/ignition and char burnout of various coal particles under perfectly repeatable conditions. They are, though, hardly representative of the real interaction between a flame and a particle and, of course, of the case of particle-particle interaction phenomena (such as group combustion). The latter was indeed one of the limitations of the present work and the current instrument, and is dealt with in the next paragraph. Flows between the extreme cases of a perfect one-dimensional laminar flow inside a drop-tube furnace and a turbulent swirling flow which can provide reasonable one-dimensionality are the cases of a turbulent counterflow and the flow from a flat-flame burner. As far as particle motion is concerned, the counterflow does not provide strictly one dimensional motion, but it is a flow easy to setup and control. In addition, it provides uninterrupted optical access (as the case of the flat flame) unlike a drop-tube furnace. A range of single or twin premixed flames can be stabilised in a counterflow, whilst particles can be injected on the centreline in a direction initially normal to the flame. Measurements of size and temperature can be made as a function of injection velocity to investigate size-temperature or velocity-size-temperature correlations of isolated particles without the smearing effect of the trajectory-dependent particle heat-up in a swirling flow. In the case of premixed flames stabilised in the counterflow, the mixture equivalence ratio controls the flame temperature and its composition providing the conditions for parametric studies of the interaction between a single particle and a flame. An additional counterflow is that of air versus high-temperature vitiated air, where one relies on particle heat up from the hot air to result in ignition and burning. Alternative fundamental reacting flow for studies of coal particles is a flat flame which can also provide a quasi-one-dimensional flow, the temperature, the oxygen content and chemical composition of which can be varied for parametric studies. This flow also provides the 261

262 6.2 RECOMMENDATIONS FOR FURTHER RESEARCH opportunity for investigation of the influence of interactions between burning particles by use of a CCD camera as a planar pyrometer, an arrangement which is common in engines (e.g. Kawamura et al. 1989). Combined measurements of the temperature (by collection of thermal radiation at two wavelengths) and scattered light from a laser sheet at a wavelength different from those of the pyrometer by use of separate CCD cameras can reveal group combustion effects and ignition patterns as a function of the local particle concentration, information unlikely to be obtained by an instrument such as the combined SDV/Two-colour pyrometer, which a single particle counter. It is also possible that with suitable optical magnification, laser sheet imaging can serve as a sizing instrument, by processing of particle images in CCD frames. In this way, the influence of local particle concentration on particle ignition can be investigated under well-controlled turbulent flow conditions. Volatile flame initiation from a cluster of particles is likely to be different from that surrounding a single particle: as a group of particles heat up they release volatiles less so in the (colder) centre of the cluster, but owing to the low temperature of the group centre, it is possible that a large proportion of the volatiles remain unignited until, at least, particle dispersion allows rapid heat-up of volatiles which leads to their ignition. The delay between cluster heat-up and flame initiation might be an important mechanism for low NO x emissions (because, for example, it allows volatiles to mix with oxygen and, thus, burn in partially-premixed mode). It is possible to reveal such phenomena using the method suggested earlier in this paragraph which not only visualises the particle flow, but also provides a means for quantification of their combustion via the temperature measurement. 262

263 APPENDIX I Derivation of turbulent stress tensor for Monte-Carlo simulation

264 In the Monte-Carlo simulation employed in 2.7 for the prediction of the influence of the velocity bias effects on the uncertainty of the measured velocity probability density functions, the model described in 2.7 accounted for the effect of anisotropy of the turbulent velocity field. In that model, the velocity fluctuations were calculated from sampling normal Gaussian pdfs in the manner of the model described by Gosman and Ioannides (1983) for the lagrangian simulation of two-phase flows. In contrast to Gosman and Ioannides (1983) who did not take turbulent cross-correlations (i.e. anisotropy) into account for the calculation of the fluctuations of the gaseous velocity, turbulent cross-correlations have been incorporated in the calculations of this work through modifications on the model of Gosman and Ioannides (1993) proposed by Bouris (1991) based on the work of Zhu and Leschziner (1991). Let V be the first order tensor of the instantaneous velocity fluctuations : V = ( u v w) T (I.1) where u, v, w are the three velocity fluctuations in a cartesian coordinate system and superscript T denotes the transpose operator. Assuming that each of the three velocity fluctuations is a random variable following a Gaussian pdf, the tensor of velocity fluctuations can be calculated by transformation with a tensor: P = ( P P ) T 1 2 P3 (I.2) which consists of random variables following Gaussian distribution. The transformation in mathematical notation can be written as: V = A P (I.3) where A is an appropriate second order transformation tensor to be subsequently calculated. It is obvious from equation (I.3) that the transformation tensor will take the properties of the turbulent flow field into account in order to have consistency between the units of the left and right hand sides of equation (I.3) which should be units of velocity. A simple way to express A is to take the covariance C of the tensor of the velocity fluctuations: C = E T ( V V ) 2 u = E vu wu uv v 2 wv 2 uw u vw C = vu 2 w wu uv v 2 wv uw vw 2 w (I.4) 264

265 where operator E( ) denotes the mean value. According to equation (I.4), is nothing other than the Reynolds stress tensor normalised by the density of the fluid. If we apply the same principle to the right-hand side of equation (I.3), the covariance of the velocity fluctuations reads: T T T ( V V ) = E( A P( A P) ) E( A A ) C = E = (I.5) the derivation of which is simple tensor algebra after taking into account that since (P i ) are independent random variables following Gaussian distribution their cross-correlation is zero and their autocorrelation is unity. If the second order transformation tensor A has elements A = ( aij), then by equating the right-hand sides of equations (I.4) and (I.5) and applying Choleski s method to the positive definite tensor C we calculate ( a ij ): a a a a = = = = a 2 ( u ) uv a 11 uw a = a a a = = = ( u v uv ) 1 ( vw a21a31) a33 = ( w a31 a32) a a (I.6) It is now simple algebra to combine equations (I.4) and (I.6) and produce the formulas for the calculation of the velocity fluctuations: u = a v = a w = a P 11 1 P + a P + a 32 P 2 P 2 + a 33 P 3 (I.7) In the case of isotropic turbulence, cross-correlation terms in equations (I.4-I.7) vanish and therefore tensor A becomes A = ( a ) no summation, in which case the velocity fluctuations ij as calculated from equation (I.7) will be exactly those used in the model of Gosman and Ioannides (1983). 265

266

267 APPENDIX II Calibration of the Two-colour Pyrometer

268 In order to determine the gain of the pyrometer (according to equation (3.9) with sufficient accuracy for temperature measurements, the response of the pyrometer, i.e. the relation between temperature and ratio of signal amplitudes, must be determined from calibration. Calibration was performed by experimental determination of the constants A, B required in equation (3.13) to establish the linear function of the inverse of temperature with the logarithm of the ratio of signal amplitudes. The experimental arrangement for calibration did only differ from that of figure 3.1 in the transmitting optics. Figure II.1 shows the modified (transmitting) optical arrangement. A tungsten-ribbon lamp (Gas-filled No. P132C) was used as source of known temperature. The lamp had been previously calibrated by the National Physical Laboratory (Reference: QT02/094/ December 1994) in terms of current versus filament colour temperature at nm. The filament of the lamp was imaged onto a 100 µm pinhole mounted on a rotating flat by means of a f/80 singlet. The flat was rotating in order to generate a transient signal which was subsequently digitised by the transient recorder. Lamp radiation diffracted behind the pinhole simulated that emitted by a thermal source and was collected by the pair of lenses described in The lamp was operated by means of a DC power supply (Farnell B30/20) in a current range between 9-19 A for which the lamp has been calibrated by the NPL. For a given power supply setting the maximum signal amplitude was measured during each rotation of the pinhole and this was repeated over 1000 cycles. The mean amplitude ratio, as well as the rms, was calculated and the corresponding filament colour temperature was obtained from the tables provided by NPL. The true filament temperature was estimated from its colour temperature at nm using de Vos (1954) data for the spectral emissivity of tungsten. Data collected for the current range mentioned above were plotted in terms of the logarithm of the signal amplitude ratio versus 1/T, where T is the temperature, corrected for a grey body as follows. Taking the logarithms of both sides of equation (3.11) and rearranging yield: Vλ λ λ Xλ ε ln R = ln = C2 + ln + ln V λλ T X ε λ2 1 2 λ2 λ1 λ2 (II.1) Had we taken the logarithm of both sides of (3.12), the ratio of the emissivities in the righthand side would be absent from the resultant equation. One can conclude therefore that the response of the pyrometer to radiation emitted from a grey body is related to that from a nongrey as calibrated with the tungsten lamp as: 268

269 Figure II.1 Optical arrangement for calibration of the two-colour pyrometer. ε ln Rgrey = ln Rtungsten ln ε λ1 λ2 (II.2) where ln R grey and ln R tungsten were the calculated signal amplitude ratio for a grey body, such as a char particle, and as measured respectively. Figure II.2 presents three different calibration curves as compared to that obtained by Israel (1993) for three different settings of the voltage of the photomultiplier. All curves correspond to the response of the pyrometer for char particles. Here, the voltage setting for the photomultipliers of 500V and 570V volts for channels 1 and 2 respectively was selected for the measurements as it provided enough sensitivity without excessive noise levels. The latter was evident when the voltage in channel 2 was increased to 600V (figure II.2) in which case noise levels at low signals (i.e. low temperatures) resulted in substantial departures from linearity of the calibration curve. 269

270 V, 600V 500V, 550V 500V, 570V Israel (1993) personal communication linear fit to set no 3 ln R e-4 4.0e-4 5.0e-4 6.0e-4 1/T (1/K) Figure II.2 Pyrometer calibration curves as a function of the voltage applied to the photomultipliers. The curves correspond to the response of the pyrometer to radiation emitted by a grey body. Determination of the gain of the pyrometer In this subsection the method for determination of the gain of the pyrometer Χ λi for each channel i from the calibration curve shall be described. Recall from equation (3.8) that the theoretical response of the pyrometer, i.e. the measured signal amplitude, is exponential: i Vλ = Ai B T e i (II.3) where Vλ is the amplitude of the signal in channel i and A, B i i i parameters to be determined. If we take the right-hand sides of (II.3) and (3.8) to be equal and also (3.9) into account, the following relation for the gain of the pyrometer yields: 270

271 Table II.1 Calibration parameters of the present two-colour pyrometer Channel 1 (514.5 nm) Channel 2 (632.8 nm) Calibration parameter A mv mv Calibration Parameter B K K Gain of pyrometer X l mv/µm mv/µm 2 Effective wavelength l eff nm nm Χ λ i = A i d ε λ 2 (II.4) A i and B i were determined from the calibration, by fitting a linear curve between 1/T and ln( Vλ i ) pairs of data and the gain of the pyrometer in each channel, Χ λi, was consequently calculated. Because the current pyrometer was used for the measurement of char particles I-1 the spectral emissivity ε l in (II.4) was assumed unity and the diameter d was substituted from: ε λ d 2 2 = ελ, tungstendpinhole (II.5) where ε λ,tungsten and d pinhole were the spectral emissivity of the tungsten lamp and the diameter of the pinhole (=100 µm) used in the calibration. The purpose of the substitution was to calculate the size of an incandescent char particle which would have given the same signal amplitude as the pinhole illuminated by the tungsten lamp. I-2 This information was used for the calculation of the discrimination band, explained in Comparison between equations (II.3) and (3.8) gives: C λ eff,i = 2 Bi (II.6) where i runs from 1 to 2 and λ eff,i is called the effective wavelength of the pyrometer as opposed to the central value of the interference filters λ i of the receiving optics. All the calibration characteristics of the present pyrometer are summarised in table II.1. I-1 Even if temperatures for volatile flames were reported in this thesis, The gain of the pyrometer would have been determined in the same manner. The temperature corresponding to measurements identified and categorised as volatile flames would have been calculated from the measured temperature calculated under the assumption of char particles by application of equation (3.34). I-2 The spectral emissivity of tungsten is about 0.5 therefore an incandescent char particle emits as much radiation as that scattered by a pinhole illuminated by the tungsten lamp, twice as big as the particle. 271

272 Detectability limits of the pyrometer Because the radiant heat from a glowing particle depends on its size (e.g. equation 3.11), there is a minimum temperature for a given particle size at which enough energy is emitted to produce current in the photomultipliers above the noise levels. In it was mentioned that threshold levels are also set during measurement and only signals above them are considered, in order to reduce the uncertainty due to noise. A particle of size d at temperature T was considered therefore detectable, if the radiant energy was sufficient to induce voltage output in the photomultipliers above the threshold level, set at 25 mv as explained in The least sensitive channel was that corresponding to interference filter of nm. In mathematical language, the minimum detectable size d of a char particle for a given temperature T was determined from (3.8), making use of (3.9) and assuming that the voltage was 25 mv, as follows: d 25mV = λ X e C 2 1 T 1 /λ (II.7) where the spectral emissivity of char has already been substituted with unity. Table II.2 presents the detectability limits for a range of sizes between 10 and 100 µm. Table II.2 Detectability limits of the pyrometer for sizes between 10 and 100 µm Char size (µm) Minimum detectable temperature (K)

273 APPENDIX III Calculation of the spectral emissivity of a coal particle

274 The spectral emissivity of a burning coal particle (i.e. burning char particle surrounded by a volatile flame) is calculated here in the general case where the entire radiation is collected by the receiving optics of the pyrometer, which is the basis of Grosshandler s (1984) analysis. The case where the pinhole restricts the pyrometer probe volume (e.g. presence of volatile flames with diameter larger than that of the image of the pinhole in the probe volume) is also treated subsequently. The calculation presented here is an extension of the method of Grosshandler (1984) and notation has been followed in this work. Assume a spherical char particle surrounded by a concentric and spherical volatile flame, shown in figure III.1. Radiation is emitted in all directions from the surface of the particle and rays travelling to the optics pass through various optical depths through the volatile flame. Provided that the entire emitted radiation (i.e. collected over 4π steradians) can be collected by the optics, the total collected energy is the energy emitted by every point on the particle integrated over all directions. According to figure III.1, R p is the char radius, R s the radius of the volatile flame, θ is the angle of a given ray passing through a point P on the outer surface of the soot mantle, measured from the normal to the surface and the spectral intensity i λ ( θ) is independent of the azimuthal direction due to symmetry. Assuming that the soot cloud is at local thermal equilibrium, the spectral intensity at point P in a given direction θ is found from the integrated form of the equation of radiative transfer (Siegel and Howell 1992, equation 14-13): i ( κ, ω) = i (0, ω)e λ s λ κs + κs 0 I ( κ, ω)e λ * s * ( κs κs ) dκ * s (III.1) * where κ s is the optical depth of the soot cloud, κ s is the dummy variable of integration and w the solid angle relative to the directional spectral intensity. It must be noted here that a critical value θ c of the angle θ in the direction PF (figure III.1) exists. If the angle is greater than the critical, radiation will have arrived along path EP without interaction with the solid surface. At the critical value, radiation will originate from point C and reach P along CP, whilst for smaller angles the contribution of the char surface will be substantial to the total radiation on P. As a consequence, the boundary condition i λ (0, ω) in equation (III.1) reads: p ε λb(tp ) λ λ i i (0, ω) = 0 θ θ θ > θ c c (III.2) where p i λ b is the directional spectral intensity of a black body, ε λ and T p correspond to the spectral emissivity and temperature of char, respectively. Scattering of radiation from soot particles can be neglected (Grosshandler 1984) because particles are small compared with the 274

275 A soot cloud (vignetted by image of pinhole) α λ char particle 5 V E Image of pinhole 5 S & θ c i λ (θ) θ 3 Radiation emitted along dotted line F D θ G Figure III.1 Geometry of a single char particle surrounded by a soot cloud. The receiving optics are located in the bottom part of the figure. wavelength of the radiation collected by the pyrometer (Kadota and Hiroyasu 1984) and measurements of the optical properties of soot from common hydrocarbons indicate that soot is highly absorbing (Dalzell and Sarofim 1969). If scattering is ignored, the optical thickness can be written as a function of the soot spectral absorption coefficient a l and the optical depth S as: κs = αλs (III.3) and, therefore, the integral of equation (III.1) reduces to: κs 0 I ( κ λ * s, ω) e * ( κ s κ s ) dκ * s = i λb (T ) s α S ( 1 e λ ) (III.4) 275

276 It follows from equations (III.1)-(III.5) that the spectral intensity of radiation at point P is: p i λb(tp ) ελ e i λ( κs, ω) = i λb(t s)1 α S( θ) λ + i α θ ( λs( ) e ) λb (T )1 s α θ ( λs( ) e ) θ θ θ > θ c c (III.5) Equation (III.5) can be further simplified by assuming that the soot cloud of the volatile flame is optically-thin, an assumption also made by Grosshandler (1984) in his analysis and which is justified by the experimental evidence presented in This assumption implies that the penetration depth of radiation into the soot cloud is smaller than its diameter (Siegel and Howell 1992). By expanding the exponential term in the right-hand side of (III.5) in a Taylor series, the latter can be approximated by the first two terms of the series as: e αλs 1 α λ S (III.6) Introducing the optically-thin limit approximation, equation (III.5) becomes: i i λ( θ) = i λb λb p (T )1 s p λ (T ) ε ( 1 α S( θ) ) λ α θ ( λs( ) e ) + i λb (T ) α S( θ) s λ θ θ θ > θ c c (III.7) which is only a function of the angle θ. The (hemispherical) spectral emissive power of radiation emitted from the system of the solid particle surrounded by the volatile flame can be found by integration over the hemispherical solid angle ω, which in detail can be written: e λb + = = λ Ω 2π θc 2π θd φ= 00 i λb φ= 0 θc i ( θ)cos( θ)dω p ( i (T ) ε α S( θ) + i (T ) α S( θ) ) λb λ λ λ λb (T ) α S( θ)cosθsin θdθdφ s p s λ cos θsin θdθdφ (III.8) 276

277 277 III-1 This equation accounts for the magnification of the receiving optics in the pyrometer side and is calculated from the focal lengths of the two lenses f/500 and f/300 as described, for example, by Hecht (1987). where Ω is the hemispherical space on which integration is done. With the exception of the last integral of (III.8), other terms can be easily recognised in (III.8) and correspond to contributions from the char particle and the soot cloud. The second double integral of (III.8) is a result of use of a pinhole (see and 3.3.2) as an aperture stop to limit the area of the pyrometer probe volume. If the diameter of the soot cloud, 2R s, is larger than the size of the image of the optics pinhole in the pyrometer probe volume, ) (500/300 d d ph ph = III-1, then integration is limited by angle θ d corresponding to the image of the optics pinhole. When the soot cloud is smaller than the image of the pinhole the upper limit of integration is θ d =π/2 since there is no geometrical limit, so integration is carried out for the hemispherical space. Equation (III.8) was integrated for this case by Grosshandler (1984). On the other hand, in the case mentioned earlier where the integration is limited by the size of the image of the pinhole, the upper limit can be deduced from simple geometrical considerations and is ) / 2R (d tan s ph 1 d = θ. The integrands in equation (III.8) do not depend on φ (i.e. the azimuthal angle) and therefore integration over that variable can be carried out immediately. The angular dependence of the optical depth S can be easily determined from figure III.1, and is: θ > θ θ θ θ θ θ = θ c s c p cos 2R sin R cos R ) S( (III.10) where R o =R p /R s. Integration of equation (III.8) using (III.10) yields: ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) s p o ph s 3 o 3 o 3 2 o ph p s b 3 o 3 2 o s 2 o p p b ph s 3 o 3 2 o s p s b 3 o 3 2 o s 2 o p p b R R 2 d,when R 2 R 1 R 1 3 d (T ) i R 1 R 1 R 3 2 R ) (T i i 2 d,when R R 1 R 1 R 3 2 (T ) i R 1 R 1 R 3 2 R ) (T i i = µ µ α ε + α + ε = < α ε + α + ε = λ λ λ λ λ λ λ λ λ λ λ λ λ λ (III.11)

278 278 Inspection of equation (III.11) reveals that the first term in each equation corresponds to the contribution of the char particle and the second to the contribution of the soot cloud. If it is considered that radiation emitted from the char passes through the soot cloud before collection by the pyrometer, the bracketed term of the first part of each equation can be interpreted as the transmissivity of the soot cloud: ( ) ( ) α + = τ λ 3 o 3 2 o s 2 o s R 1 R 1 R 3 2 R (III.12) whilst in a similar manner it can be deduced that the bracketed term of the second part of each equation corresponds to the emissivity of the soot cloud: ( ) ( ) ( ) ( ) µ α < α = ε λ λ λ 2 d R 2 R 1 R 1 3 d 2 d R R 1 R 1 R 3 2 ph s 3 o 3 o 3 2 o ph ph s 3 o 3 2 o s s (III.13) If the values of τ s and ε λ s are substituted into Wien s approximation of Planck s law of emissive power as used in equation (3.3), the following form of equation (III.11) emerges: τ ε ε τ + ε = λ λ λ λ λ λ λ p 2 s 2 T C T C s s s 2 s ph s p p b e e d ),2R (d 1 ) (T i i min (III.14) It is recalled that d is the size of the char particle. The term in square brackets corresponds to the compound emissivity of the system of char particle and surrounding volatile flame (see, for example, equation (III.2), denoted c λ ε : p 2 s 2 T C T C s p s 2 s ph s p c e e d ),2R (d 1 λ λ λ λ λ λ τ ε ε τ + ε = ε min (III.15)

279 APPENDIX IV Flow Boundary Conditions

280 This appendix briefly describes the method of determination of the swirl number from velocity measurements at the exit of the burner in the manner of Milosavljevic (1993) and presents the results of the velocity measurements, the calculated swirl number as a function of the flow rate through the tangential inlet of the burner and the velocity symmetry tests conducted before the measurements presented in Chapter 4 were initiated. Determination of swirl number The swirl number was determined as a function of the flow rate of the tangential supply for a 300 lt/min total air throughput in the burner. Measurements of the air velocity for isothermal conditions were obtained in the absence of the quarl at z/d=0.12, which was the location closest to the exit of the straight flow development section of the burner where optical access permitted velocity measurements. The mean and rms velocities presented below were obtained from 5000 samples and the statistical uncertainties were ±2% and ±1% in the mean and rms respectively. The laser Doppler system that was used for the measurement of the air velocity was slightly different to the one used throughout this thesis. A beam splitter and optical shifter based on a rotating diffraction grating was employed because of unavailability of the Braggcell based transmission optics at the time of these measurements. The principal characteristics of optical configuration of the laser Doppler system are summarised in table IV.1. Table IV.1 Principal characteristics of optical configuration of the LDV system used in the measurement of the swirl boundary conditions. Laser Wavelength 488 nm Exit Laser Beam Diameter 1.2 mm Transmitting Optics Focal Length of Focusing Lens on Diffraction Grating 80 mm Focal length of Collimating Lens 100 mm Focal Length of Beam Focusing Lens 300 mm LDV Half Angle of Intersection (measured) 4.48º Dimensions of LDV Probe Volume (calculated at 1/e 2 ) Diameter 60 µm Length 800 µm Figure IV.1 presents profiles of the mean and the rms axial and tangential velocity components for five different cases of partitioning of the secondary air through the burner. Details of the flow rates used for each case are given in the caption of figure IV.1 whilst the calculated swirl numbers, obtained by numerical integration of the measured profiles, which corresponded to the measurements are shown on the graph. It should be noted here that measurements were made every 1 mm starting from the centreline of the burner for the regions corresponding to r<0 and r>0 and, hence, two swirl numbers were calculated. The swirl number quoted in the figure is the arithmetic mean of the swirl number in each of the two sub-regions. The difference 280

281 1.2 S w =0.57 S w = u/u 0 or w/u u /U 0 or w /U u /U 0 or w /U r/d r/d 1.2 S w =0.41 S w = u/u 0 or w/u u /U 0 or w /U u /U 0 or w /U r/d 1.2 S w = r/d u/u 0 or w/u u /U 0 or w /U r/d Figure IV-1 Radial profiles of the mean and the rms axial and tangential velocity components of the isothermal flow, measured at 2 mm (z/d=0.02) downstream of the exit of the burner in the absence of the quarl, for a range of derived swirl numbers indicated on the graph. Symbols: (à) Mean axial velocity, (ò) Mean tangential velocity, (á) rms axial velocity and (ó) rms tangential velocity. 281

282 Cumulative swirl Fuel Injector Sw=0.57 Sw=0.40 Sw=0.20 Sw= r/d Swirl number, Sw Not Measured m tang /m tot Figure IV-2 (a) Cumulative swirl number along a radial profile at z/d=0.02 in the absence of a quarl for a range of swirl numbers between 0.16 and 0.57 as indicated on the graph. (b) Swirl number as a function of the fraction of the total mass flow rate which is supplied through the tangential inlets in the burner. Results correspond to measurements of figure IV-1. between the swirl numbers in the two sub-regions was smaller than 15% and was a result of the flow asymmetry between the two sub-regions. 282

283 As figure IV.1 shows, the flow is close to symmetric for all flow conditions although for the highest swirl number of 0.57 the asymmetry of the axial velocity in the region close to the fuel injector is more prominent than in the other cases. This swirl number was the highest obtainable with this burner and Milosavljevic attributed this limit to the high friction losses on the walls and particularly in the contraction. The swirl number of 0.57 was lower than the maximum value of 0.68 obtained by Milosavljevic for burner I, but that was expected because burner II has higher area blockage ratio due to the injector (Hagiwara and Bortz 1984). For all swirl numbers the typical rms of the velocity fluctuations was about 10% of the bulk velocity and only the rms axial velocity reached 20% near the walls due to shearing. Figure IV.2(a) presents the radial evolution of the swirl number (i.e. for r 0 replaced by r in equations 4.2) for swirl number spanning the range There is an almost linear increase of the calculated swirl with increasing radial distance until r/d 0 =0.4 where it reaches a plateau. A similar plateau can be observed in figure IV.2(b) which shows the evolution of the swirl number as a function of the tangential air supply for a total secondary air supply of 300 lt/min. Figure IV.2(b) shows that although the swirl increases almost linearly as a function of the tangential supply for small values (up to 0.7 of the total flow rate) a maximum is reached as almost all the air is supplied through the tangential inlets. This confirms that it is indeed unlikely that the swirl number can be further increased using this design of burner. Flow Symmetry Check The rest of this section presents results from symmetry checks of the flow using the LDV system described earlier. The purpose of the checks were to confirm that the flow was sufficiently symmetric in order to proceed with measurements of half profiles only and thus minimise the duration of the experiments. The latter was important in keeping the total duration of the measurements using the combined SDV and two-colour pyrometer instrument (Chapter 3) low. The measurements were presented earlier in this Chapter 4 (as well as chapter 5). The results of the flow symmetry checks will be presented systematically below for the following cases: (i) flow at the exit of the throat in the absence of the quarl, (ii) gaseous swirl-stabilised flames without primary air and (iii) pulverised coal flames with gas-support. Figure IV.3 presents profiles of the axial velocity traversed at right angles with each other for S=0.57, measured at z/d=0.02. As shown the flow is symmetric with the minor exception of the region close to the fuel injector. The small departure from symmetry was either due to a small (less than ±0.5 mm IV-1 ) misalignment of the fuel injector or of the LDV system relative to the injector or to structural asymmetries of the fuel injector. IV-1 As shown in Appendix I of chapter 5 this is the positional precision one should expect when aligning the fuel injector in the annulus or the optics relative to the fuel injector. 283

284 1.2 U/Uo U/Uo, X sweep U/Uo, Y sweep Injector boundaries r/d Figure IV-3 Radial profiles of the mean axial velocity component measured at the exit of the burner (z/d=0.02) in the absence of the quarl for isothermal conditions which corresponded to S=0.57. Symbols: (á) mean axial velocity measured along a profile denoted X and (à) mean axial velocity measured at right angles to X profiles. Figure IV.4 presents complete radial profiles of the axial velocity as a function of the swirl number and the overall fuel equivalence ratio of the gaseous swirl flame and radial profiles of the tangential velocity component as a function of the swirl number. As shown on the graph, both the effect of the swirl number and the fuel equivalence ratio on the symmetry of the flow is insignificant and all flows are adequately symmetrical. Therefore, one is justified in obtaining only half-profiles to save time in the execution of the experiments. It remains to be seen in the next paragraph whether particle velocity profiles exhibit symmetrical behaviour. Figure IV.5 presents radial profiles of particle Sauter mean diameter (SMD), axial velocity of two size classes, namely µm and µm and volume flux of the µm size class, all measured for S=0.41 at z/d=2.67. The purpose of the measurements was to confirm that symmetry which was already established for the turbulent gas flow- was also present in the particle flow. Figure IV.5 indicates that the flow was reasonably symmetric and therefore measurement of only half profiles was adequate to characterise the flow. 284

285 U/Uo, u /Uo throat boundaries S=0.57 S=0.41 S=0.57 S= r/d U/Uo, u /Uo throat boundaries r/d φ=1.0 φ=0.69 φ=1.0 φ= W/Uo, w /Uo S=0.57 S=0.41 S=0.57 S= throat boundaries r/d Figure IV-4 Radial profiles of (a) mean (á, à) and rms (ç, æ) axial velocity component for swirl numbers 0.57 and 0.41 respectively and f=0.69, (b) mean (á, à) and rms (ç, æ) axial velocity component for overall gas equivalence ratios of f=1.0 and 0.69 respectively, and (c) mean (á, à) and rms (ç, æ) tangential velocity component for swirl numbers 0.57 and 0.41 respectively and f=0.69, all measured at z/d=3.05 in the absence of primary air. 285

286 U/U SMD [µm] G [Arbitrary Units] r/d Figure IV-5 Radial profiles of particle Sauter mean diameter (ç) and volume flux G (ì), and radial profiles of the mean axial velocity for particles of the µm (á) and the µm (ó) size class. All quantities were obtained from a single measurement profile for S=0.41 and overall gas equivalence ratio f=0.69 at z/d=

287 APPENDIX V Influence of Orientation of Fuel Injector on Flame Symmetry

288 This appendix describes the method and results from investigations on flame asymmetries in the burner of Chapter 5. It focuses on the effect of thermocouple orientation, misalignment between the axes of the coal injector and the burner throat and fuel injector (coal gun) orientation relative to a fixed frame of reference on flame symmetry. The purpose of the experimental investigations was to identify sources of flame asymmetry due to the construction or misalignment of the coal injector, and due to the disturbance on the flow caused by the bluffbody effect of the thermocouple used in the measurements. The burner used in the investigations was the burner described in detail in Chapter 5 and the fuel injector was the multihole annular injector detailed in figure 5.1. The set of experimental data consisted of mean temperature measurements made by a 250 µm Platinum/Platinum+13% Rhodium thermocouple mounted within a 3 mm ceramic cladding. Results are presented in the form of radial profiles of mean temperature measured along traverses which were at right angles to each other. Some details on the notation used to identify the traverses and the orientation of the thermocouple (and the coal injector as described below) on the figures are explained using figure V.1 as example: each figure is accompanied by a schematic plan view of the burner where only the diameters corresponding to the throat (16 mm) and the exit of the quarl (38 mm) are indicated. The orientation of the circle corresponding to the exit of the quarl was described by arbitrarily designating the locations N(orth), S(outh), E(ast) and W(est) which are used on the graphs to describe the direction of the traverses and the orientation of the thermocouple and the fuel injector. It must be noted here that the orientation of the thermocouple was defined by the direction of the cladding, as shown on figure V.1, and the orientation of the fuel injector was defined relative to a fixed reference denoted coal gun entry on the figures. The fuel injector could be freely rotated about the axis of the burner and the measurements were performed in order to identify asymmetries in either the construction of the injector or a blockage of one of the six holes through which natural gas was injected and could result in flame asymmetries. All swirl flames were produced at swirl number S=0.41 by injection of 22 lt/min of natural gas, which corresponded to equivalence ratio of 0.69 and was typical of the support flames used in all experiments with pulverised coal presented in this thesis. Check of Secondary Air Flow Symmetry Before the parametric studies presented in this appendix vitiated air of 600 ºC, produced by the exhaust gas generation system described in figure 5.2, was fed through the burner and the flow was checked for symmetry. The mean temperature was measured across two radial profiles at right angles, at axial distance z/d=2.7 V-1 which was the position nearest to the exit of the V-1 This axial station was kept constant and all results presented henceforth correspond to the same axial distance. 288

289 Figure V.1 Radial profiles of the mean temperature of vitiated air preheated at 600 ºC in the absence of combustion, measured at z/d=2.7 along traverses at right angles explained in the adjacent drawing, and were (à) along S-N and (á) along E-W. quarl in which physical access of the thermocouple was possible. Figure V.1 presents two profiles of the mean temperature, measured along directions (S-N) and (E-W) as indicated in the adjacent drawing, which show that the temperature distribution is symmetric and both profiles collapse on a single curve. The temperature maxima correspond to maxima in the velocity distribution (Milosavljevic 1993). According to this result the flow is symmetric and, thus, it is unlikely that any temperature asymmetries presented below were due to the flow of the secondary air through the burner. Effect of Coal Gun Orientation Figures V.2 and V.3 present the effect of the orientation of the coal gun on the flame. The coal gun was rotated through 120º and the mean temperature was measured along (S-N) and (E- W) traverses. Figure V.2a shows that although in (S-N) direction the maxima in the mean 289

290 290 Figure V.2 Radial profiles of the mean temperature of a natural gas swirl flame with swirl number S=0.41 and equivalence ratio f=0.69, measured at z/d=2.7, along traverses at right angles, as explained in the figure, and corresponded to (à) S-N and (á) E-W directions; in (b) the measurements were made after the coal gun was rotated through 120º in reference to results of (a).

291 temperature are symmetrically placed and reach levels of about 1800 K, a peak higher by about 150 K was measure at one side of the burner along the (E-W) traverse. In order to ensure that the peak was a result of asymmetries introduced by the multihole annular injector, the coal gun was rotated through 120º (a complete 180º rotation was not possible due to obstructions) and the same temperature traverses were made. As shown in figure V.2b the asymmetrical temperature peak presented in figure V.2a rotated with the rotation of the coal gun and occurred along the (E-W) profile instead of the (S-N), because the rotation was only 120º. Figure V.3 presents a replot of the results of figure V.2, where profiles are shown as a function of the orientation of the coal gun, rather than the traverse (as in figure V.2). Figure V.3 shows that in this setup, direction (E-W) was more sensitive to the rotation of the injector and indicated a higher asymmetry. This was attributed not only to the 120º rotation (instead of a full 180º) but also to the fact that the initial orientation of the hole of the injector (from a total of six), which was likely to inject more fuel (hence the higher mean temperature maximum), was unknown. In any case, the measurements implied that this asymmetry in the value of the maximum mean temperature along one profile was probably inadequate to result in flow asymmetry, otherwise this would had reflected on (S-N) profile in figure V.2a. Effect of Coal Gun Misalignment A set of measurements were made with the coal gun misplaced by 0.3 mm along (S-N) direction to quantify the influence of positional asymmetries on the flame. However small, this misplacement of the coal gun corresponded to more than 10% of the external radius of the injector and was identifiable by naked eye. In this case, figure V.4, the flow was most probably asymmetrical, particularly at the direction of the positional asymmetry implied by the asymmetry on the mean temperature profile along a (S-N) traverse. Note that in this case the thermocouple was placed in the flow normal to the positional asymmetry so that the interference between the cladding of the thermocouple and the flow was constant along the profile. The result, however, was attributed to the flow rather than the presence of the thermocouple, especially in view of the results of figure V.5 presented below. Figure V.5 presents the effect of the orientation of the thermocouple on the measured temperature along a S-N traverse. The thermocouple was placed in the flow in three different orientations, one of which was along the axis of the injector, thought to cause minimal disturbance to the flow, and the same traverses were made and the results are compared in figure V.5. As shown in figure V.5b, the inevitable disturbance caused to the flow by the bluffbody effect of the physical probe, was insignificant compared with the effect of asymmetries in the construction of the coal injector. The temperature peak measured with all thermocouple 291

292 292 Figure V.3 Re-plot of data presented in figure V.2 as a function of coal gun orientation where the coal gun was aligned to the (à) north direction and (á) south direction; (a) presents measurements along a S-N traverse, and (b) along a E-W traverse.

293 Figure V.4 Radial profiles of the mean temperature of a natural gas swirl flame with swirl number S=0.41 and equivalence ratio f=0.69, measured at z/d=2.7, along traverses at right angles, as explained in the figure, and corresponded to (à) S-N and (á) E-W directions. The coal gun was intentionally misaligned by 0.3 mm along S-N direction. orientations near x/d=0.8, figure V.5a, is the same peak indicated in figure V.2b and was attributed to the construction of the coal gun. 293

294 294 Figure V.5 Radial profiles of the mean temperature of a natural gas swirl flame with swirl number S=0.41 and equivalence ratio f=0.69, measured at z/d=2.7, along a (a) S-N and (b) a E-W traverse, for three thermocouple orientations.

295 Colour Plates

296

297 PLATE 1 PHOTOGRAPH OF THE TRANSMITTING OPTICS OF THE COMBINED SDV AND TWO-COLOUR PYROMETER INSTRUMENT. THE AR + LASER AND THE DISA SINGLE BRAGG-CELL UNIT ARE SHOWN ON THE PHOTOGRAPH PLATE 2 PLAN VIEW OF THE RECEIVING OPTICS OF THE COMBINED SDV AND TWO-COLOUR PYROMETER INSTRUMENT 297

298

299 PLATE 3 DETAIL OF THE RECEIVING OPTICS OF THE COMBINED SDV AND TWO-COLOUR PYROMETER INSTRUMENT. THE PHOTOGRAPH SHOWS (FROM RIGHT TO LEFT) A F/500 ACHROMAT, THE RECTANGULAR ALUMINIUM-COATED MIRROR, THE F/300 ACHROMAT AND THE 40X MICROSCOPE OBJECTIVE LENS. AT THE BOTTOM OF THE PHOTOGRAPH THE SECOND COATED MIRROR IS ALSO SHOWN PLATE 4 FRONT VIEW OF THE MODULE CONTAINING THE PYROMETER OPTICS, AS WELL AS OF THE MODULE, THE LINEAR PHOTODIODE ARRAY (ON THE RIGHT OF THE PHOTOGRAPH). THE F/300 ACHROMAT, LOCATED IN FRONT OF THE 150 µm PINHOLE IS SHOWN. OBSERVATION ALONG THE AXIS OF THE OPTICS, FROM A LOCATION DOWNBEAM OF THE MICROSCOPE OBJECTIVE LENS 299

300

301 PLATE 5 PHOTOGRAPH OF THE METAL FURNACE SHOWING MAINTENANCE ACCESS PORTS 301

302

303 PLATE 6 PHOTOGRAPH OF THE CERAMIC FURNACE, SHOWING METAL CASING AND MAINTENANCE ACCESS PORTS 303

304

305 PLATE 7 PHOTOGRAPH OF GAS-SUPPORTED PULVERISED COAL FLAME FOR SWIRL NUMBER 0.57 AND GAS EQUIVALENCE RATIO OF PRIMARY AIR: 10 LT/MIN AND SECONDARY AIR: 300 LT/MIN. CAMERA SETTINGS: F/5.6 AND 1/

306

307 PLATE 8 PHOTOGRAPH OF GAS-SUPPORTED PULVERISED COAL FLAME FOR SWIRL NUMBER 0.41 AND GAS EQUIVALENCE RATIO OF PRIMARY AIR: 10 LT/MIN, SECONDARY AIR: 300 LT/MIN. CAMERA SETTINGS: F/5.6 AND 1/

308

309 PLATE 9 PHOTOGRAPH OF GAS-SUPPORTED PULVERISED COAL FLAME FOR SWIRL NUMBER 0.57 AND GAS EQUIVALENCE RATIO OF PRIMARY AIR 10 LT/MIN AND SECONDARY AIR: 400 LT/MIN. CAMERA SETTINGS: F/5.6 AND 1/

310

311 PLATE 10 PHOTOGRAPH OF GAS-SUPPORTED PULVERISED COAL FLAME FOR SWIRL NUMBER 0.41 AND GAS EQUIVALENCE RATIO OF PRIMARY AIR: 10 LT/MIN, SECONDARY AIR: 300 LT/MIN. CAMERA SETTINGS: F/5.6 AND 1/

312

313 PLATE 11 PHOTOGRAPH OF GAS-SUPPORTED PULVERISED COAL FLAME FOR SWIRL NUMBER 0.57 AND GAS EQUIVALENCE RATIO OF PRIMARY AIR: 7.5 LT/MIN, SECONDARY AIR: 300 LT/MIN. CAMERA SETTINGS: F/5.6 AND 1/

314

315 PLATE 12 PHOTOGRAPH OF THE VITIATED AIR FACILITY, SHOWING THE SYSTEM FOR GENERATION OF EXHAUST GAS, THE BURNER WRAPPED IN FIBRE INSULATION, AND A CYLIDRICAL FURNACE WITH ACCESS PORTS MOUNTED ON THE BURNER, USED FOR TEST EXPERIMENTS 315

316

317 PLATE 13 DETAIL OF COLOUR PLATE 12 SHOWING THE TWO FURNACES WHICH PRODUCE THE VITIATED AIR SUPPLY TO THE SWIRL BURNER. PLATE 14 PHOTOGRAPH OF THE UNITS FOR METERING OF AIR AND GAS SUPPLIES TO THE SYSTEM FOR GENERATION OF EXHAUST GASES, AND THE ELECTRONIC UNITS FOR THE CONTROL OF THE FLAMES IN THE PREMIX BURNERS 317

318

319 PLATE 15 PHOTOGRAPH OF GASEOUS FLAME OF 0.65 EQUIVALENCE RATIO, STABILISED USING AIR AS OXIDANT. PLATE 16 PHOTOGRAPH OF GASEOUS FLAME OF 0.80 EQUIVALENCE RATIO, STABILISED USING VITIATED AIR CONTAINING 17.6% OXYGEN AS OXIDANT. 319