Flame imaging on the ALSTOM EV-burner: thermoacoustic pulsations and CFD-validation

Transcription

1 Flame imaging on the ALSTOM EV-burner: thermoacoustic pulsations and CFD-validation F. Güthe *, R. Lachner, B. Schuermans, F. Biagioli, W. Geng, ALSTOM Switzerland, Baden, CH-5401, Switzerland and A. Inauen, S. Schenker, R. Bombach, N. Tylli, W. Hubschmid Paul Scherrer Institut, CH-5232 Villigen, Switzerland The ALSTOM low emission swirl-induced premix EV-burner is investigated by OH-planar laser induced fluorescence (PLIF) and OH*-chemiluminescence (CL) imaging on a full-scale industrial burner test rig. Three different burner variants have been compared by their flame shape and position as well as emission and pulsation behavior. The flame images have been used to enable comparison and validation of thermoacoustic and computational fluid dynamics (CFD) models. The flame movement upstream inside the burner can be related to emissions and pulsation. Depending on the burner two different mechanisms dominate the acoustic pulsations: One is based on equivalence ratio fluctuations coupled to a sudden displacement of flame anchoring point into the burner. Another mechanism seems to be related to turbulence intensity fluctuations. The experimental images were compared with the results of Reynolds-averaged Navier-Stokes (RANS) CFD simulations for varying parameters for validation. The turbulence treatment in time-averaged RANS models is not sufficient to describe the flame movement properly and encourages to apply a more sophisticated treatment like LES, which is capable of describing bi stable behavior. * Corresponding Author, TGNTB, Brown-Boveri Str. 7, Felix.guethe@power.alstom.com, TGNTB, Brown-Boveri Str. 7. Combustion Diagnostics Group, Paul Scherrer Institut. 1

2 OH*-CL = OH- excited state chemiluminescence OH-PLIF = OH planar laser induced fluorescence OH-PLIF-FF = flame front averaged OH-PLIF = adiabatic flame temperature T flame T ref u burner u 0 Nomenclature = reference temperature for T flame operating point = average flow velocity in the burner exit plane = u burner extrapolation to flame off = mass flow = mass flow φ = = equivalence ratio T LBO,, _LBO, LBO f D Sr = φ =lean blow out (~LBO) quantities: T flame, φ, = frequency = burner exit diameter f D u burner (Strouhal number) u t = turbulence burning rate u = turbulence intensity, l t = turbulence lengths scale λ = molecular diffusion coefficient = laminar flame speed s l Q & = heat release or thermal power ( ) = ( )(t) - ( ) (mean) = fluctuating part of a quantity δ, δ = phase shift, m & = m(t & + δ) p = ρ u 2 ζ burner 2 = pressure drop coefficient burner ζ ζ rel = = relative pressure drop coefficient ζ ref k = heat release rate used in CFD 2

3 I. Introduction The swirl induced ALSTOM environmental burner (EV) has been developed for low emissions, particularly addressing nitric oxides (NOx), relying fully on lean premix combustion technology in heavy-duty gas turbines. Low emissions are achieved without compromising the thermoacoustic behavior, i.e. keeping the amplitudes of combustion instabilities as low as necessary for ensuring that required lifetime targets are met. An overview of ALSTOM combustion technologies is given in a general frame in Ref 1 and concentrating on the EV burner in Ref 2 and 3. Gas turbine development is largely aided by the use of numerical simulations like computational fluid dynamics (CFD), chemical kinetics, and other applications. To gain confidence and to validate these tools, a comparison to experimental data is required. For this purpose, the EV burner has been have been investigated using imaging techniques with or without sophisticated laser diagnostics 4 differentiating different burner variants. From the study of steady operating points and comparison to phase locked images the pulsating flame is interpreted. The work documents a number of these efforts and describes post-processing methods of the recorded images as well as the CFD results. With these methods a comparison of model simulations for two different flame models and experimental data can be performed. The comparison has been done for OH-PLIF and OH*-CL data. The study is extended to thermoacoustic phenomena and emission modeling and discusses the use of flame images for burner development. Thermoacoustic studies using flame visualization techniques have been described for labscale burners 5, 6, 7 and industrial scale burners before 8. A number of experimental methods have been reviewed 9. The method can be used as a tool to guide the burner development at ALSTOM and has been applied to three different burner variants and the results are used to derive guidance in the design of new burner hardware. II. Experimental set-up The tests have been performed in the atmospheric single burner combustion test rig at a full-scale industrial burner as described earlier 10, 11. Combustor is electrically preheated, fed into a plenum chamber, and flows from there through the burner and the combustion chamber. The combustion chamber has a rectangular cross-section with the dimensions of 283mm x 381mm x 1250mm. Cooling of the combustor section is achieved by a combination of radiation and water-cooling. 3

4 The burner is based on the ALSTOM EV-Burner 1-3 but only results from generic EV-type burners are presented here. The design idea of the burner is to hold the flame on the vortex breakdown point induced by the swirl of the double cone burner. In the standard configuration the gaseous is injected through a row of holes in a cross flow direction into the entering the tangential burner slots. The location of the stagnation point shows only small sensitivity to operating conditions yielding a stable flame as long as the flame is anchored at the stagnation point. The burner was operated in three variants: pre-premix (as reference), standard and with staged injection. In the latter is injected through the burner slots as well as through the lance in the center. The three variants differ in mixing pattern and flame shapes. The staged injection can be seen as a result of an optimization for pulsations and emissions from the standard case. For a reference case with perfect mixing the test facility can be operated in pre-premixing mode. In this mode, the is mixed with the in the supply tube. This is taken as a reference without unmixedness and without temporal equivalence ratio fluctuations. An exhaust probe located about 550 mm downstream of the burner exit extracts flue gas, which is analysed for contents of oxygen O 2, carbon dioxide CO 2, carbon monoxide CO, unburned hydrocarbons UHC, and nitric oxides NO x (NO and NO 2 ). The gas analysis data (and the temperatures of combustor inlet and ) as well as the measured mass flows allow a determination of the adiabatic flame temperature T flame within 10 K relative from one measurement to another. The accuracy of the absolute T flame measurement is estimated to be ~ 25 K. The Pressure drop coefficient ζ is obtained from measurement of the pressure difference ( p) upstream and down stream of the burner. Flames inside the burner having a higher flow resistance lead to higher ζ -values. A microphone is placed in the combustion chamber around 180 mm downstream of the burner exit to measure combustion-driven pressure fluctuations. Two -cooled quartz windows of ca. 200mm x 200mm realize optical access to the combustion chamber and are placed 90 to each other and to the mean flow direction. The upstream window edge is located 20 mm downstream of the burner exit in the combustion chamber. To enable a visualization of the pulsations the test rig is excited with loud speakers mounted upstream of the combustor 8, 11 forcing the system to follow the naturally occurring frequency. The laser and camera were locked on this forcing signal and the phase shift adjusted by manually changing the delay constant. The laser set-up has been described before 12, 13. The view angle of the intensified camera was perpendicular to the flow. The laser light sheet was parallel to direction of mean flow. The frequency doubled output of a pulsed Nd:YAG/dye laser system at approximately 286 nm was used for excitation of the OH radicals. By appropriate combination of cylindrical and spherical lenses the laser beam was 4

5 shaped into a thin divergent light sheet, passing from below through the combustion chamber tube. The sheet plane was arranged parallel to the main flow direction. Using an intensified CCD camera equipped with a band pass filter centered at 308nm, the LIF signal was detected perpendicular to the plane of the light sheet. The camera was locked on the microphone signal (filtered by specially designed phase locked loop PLL filter) this forcing signal and the phase shift adjusted by manually changing the delay constant. The trigger signals for the laser and the camera were delayed using a digital delay/pulse generator as to resolve the phase length of the dominant thermoacoustic oscillation in steps of 45. Presented are images of chemiluminescence (CL) and laser induced fluorescence (LIF) intensity. CL images are actually the projection of a 3D flame onto the 2D image of the camera, indicating heat release from regions where the OH* radical is formed by chemical energy converted into heat and light integrated along the line of sight for each image. The LIF intensity is sensitive to the concentration of [OH] radicals in their electronic ground states, which are found in regions of heat release, but also in the exhaust gas. It results from resonant laser excitation in the focus of the beam or light sheet and subsequent fluorescence from that excited state of the OH radical. The use of a planar light sheet results in 2 dimensional concentration profiles of ground state OH (OH-PLIF). At least for atmospheric conditions their interpretation as temperature or heat release can be misleading. However the high concentration and quantum yields of the OH-LIF process enable the visualization of instantaneous and narrow flame structures. The LIF-images are 2D cuts through the combustor along the illuminating laser sheet, which is located in the center of the combustor. To derive an estimate of the 2D heat release profile in that plane a flame front tracking procedure has been developed, which uses the gradient of each individual single laser shot within the sample to be analyzed to detect regions of fast change and then applying a threshold to binarise for flame front detection. In the LIF-flame front (LIF-FF) the probability to detect such a flame front at that location is plotted. The procedure is similar but not identical to the one recently described by Balachandran et al. (14, 7). It is believed that the OH-PLIF-FF better resembles heat release than the OH-PLIF intensity. III. Results A. Steady operating points 5

6 The OH*-CL images in Figure 1 reveal the change of the flame shape and position with increasing T flame for the three burner variants. While the flame at lower adiabatic flame temperatures (T flame ) is located outside of the burner in the combustion chamber, for higher T flame the flame sits inside the burner attached to the stagnation point. The images seem to indicate that the flame can be stabilized at two different locations: inside or outside of the burner. Both positions seem to be stable over a large range of temperatures once the transition has occurred. This bi-stable behavior has been pointed out before 15. The flame anchored inside the burner is relatively stable as the original design idea had stated 3 and is not attached to the material of the burner shell, as this would result in a harmful flashback event. Figure 1: Chemiluminescence images for three burner variants for different T flame. The flow direction is from left to right. The observation window starts about 20 mm downstream of the combustor and is ca.180 x 120 mm. Each image is scaled from minimum to maximum value. The overall intensity changes as shown in Figure 4. This change in flame position is clearly detectable for the standard burner around T ref and ca. 50 K higher for pre-premix with an unsteady transition while for the staged case the transition is smoother and at higher T flame. The same trend is seen from two-dimensional OH-PLIF images in Figure 2 and flame front (OH-PLIF-FF) averaged data of Figure 3. The OH-PLIF images are difficult to interpret in a straightforward manner. The intensity is not only related to heat release or temperature. The trend for LIF intensities is not as clear since the LIF technique is sampling also regions where no heat is released but OH-radicals are still present in the post reaction zone in super equilibrium concentrations. 6

7 Figure 2: As Figure 1 OH-PLIF images for the same operating points. Each image is scaled from minimum to maximum value. Figure 3: As Figure 1 flame front averaged from PLIF images for standard burner and pre-premix. Each image is scaled from minimum to maximum value. The leading edge is defined as the point along the central part (symmetry line+- 15% of the image) of the flame where the measured intensity is 30% of its maximum As shown in (Figure 4) it gives a good measure of flame position if it can be resolved from the image. This definition has been applied to OH*-CL (Figure 1) and OH-PLIF (Figure 2) images. To compare to 2 dimensional heat release plots a flame front averaging method has been used for the OH-PLIF technique. A quantitative analysis of the flame movement is shown in Figure 4, where the leading edge of the flame is plotted over T flame.. For flames clearly anchored inside the burner an arrow is drawn towards the inside of the burner. While the standard and pre-premix burner shows a stronger dependance, the staged flame moves relatively little. A similar result is obtained from the behaviour of the reduced pressure drop ratio ζ rel (Figure 4). The flame 7

8 inside the burner causes a higher flow resistance and a higher pressure drop at given mass flow in the single burner test rig. The measured pressure drops correlate well with the observed flame positions. Average Intensity T LBO Average Intensity vs. T flame staged LIF staged CL Prepremix LIF Prepremix CL Standard LIF Standard CL T flame -T ref (K) Leading Edge (mm) Leading Edge vs. T flame window edge staged LIF staged CL pre-premix LIF pre-premix CL standardlif standard CL T flame -T ref (K) 8

9 ζ relative ζ burner vs. T flame staged CL prepremix CL standard CL T flame -T ref (K) Figure 4: Summed intensities and leading edge of LIF and CL images as well as ζ rel values plotted vs. T flame for the three burners. The black line indicates the edge of the observation window. The arrows indicate an observation by visual inspection of the flame image. To link the OH*-CL-intensities I CL to variations of and mass flow both variables have been changed in Figure 5. The results have been plotted over heat release (Q & ~ thermal power). For lean flames, assuming complete combustion, the thermal power can be related directly to the mass flow: Q& T (T - T ) flame inlet Equation 1 which also includes the dependence of T flame on. For lean flames this relation can be approximated to be linear. 9

10 Average Intensity Average Intensity vs. Heat release u o staged T varied Prepremix T varied Standard T varied Standard ub varied prepremix ub varied u burner varied T flame =T ref T flame varied u burner= u ref T LBO relative heat release ~ m Figure 5: OH*-CL-intensity I CL (arbitrary units) vs. heat release (relative to reference point) for varied u burner velocities and T flame for the different burners. The dashed lines extrapolate linearly for constant u burner and the solid lines for constant T flame. The T flame variation is shown in Figure 1 to Figure 3. The u burner variations are not shown here. The extrapolation for constant T flame (corresponding to const ) and constant burner velocity u burner (corresponding to const. ) reveal the different dependencies on and flow. Intensity increases linearly with both quantities but the offset differs significantly. A slight deviation from this linearity is observed if part of the flame disappears inside the burner as can be seen on the flame images. Images for the u burner variation are analyzed but not presented here. While the T flame -extrapolation leads to a temperature close to the lean blow off of the burner T LBO, the u burner extrapolation leads to u 0 values close to 0. The values do not change for the different burner variants. From these fits the following empirical I CL (as summed on the displayed flame image) formula can be derived from Figure 5 to describe all operating points in one equation: I u ) (T T ) Equation 2 CL (u burner 0 flame LBO which for small u 0 and u burner ~ (at constant T flame ) assuming a linear dependence of T flame on (for constant u burner ) and φ = (for lean flames) can be estimated to: 10

11 I CL 2 ( ) ( φ φ ) LBO Equation 3 _LBO This equation will be valid at least locally close to the reference point (at T ref and u ref ) of measurement and can be transferred later to processes occurring instantaneously in the flame front. The linearity of the intensity I CL on (or T flame ) reflects the dependency of the [OH*]-concentration on the formation pathway 16, which seems to be controlled by concentration of its precursors (CH, CHO, C 2 H,...) rather than a strong temperature-dependency of a formation rate of some limiting steps, which occur for reaction with high barriers 17. This observation contradicts other findings 9, but seems be confirmed also by our data at least for the observed range of conditions. The different slope of the linear curves in Figure 5 results in a much higher sensitivity to variations of T flame (T flame ) than u burner. For a given T ref the ratio of slopes is approximately constant C (C ~ 5 in our example) di d = const di C d ~. at reference Equation 4 = const B. Emission and pulsations The corresponding relative emission and pulsation values are plotted vs. T flame in Figure 6. The flames move out of the burner with decreasing T flame as pointed out earlier. For NO x -emissions the absolute lower limit for a premix burner is represented by the pre-premix case, which is not affected by residual unmixedness of the injection or flame position and rises exponentially with T flame, as can be seen in the logarithmic-plot. The burner with standard injection deviates from this exponential behavior with a clear step slightly above T ref, when the flame moves upstream and inside the burner towards regions of worse mixing in vicinity of the injection holes. The move of the flame upstream in the burner towards the injection results in different mixing quality and emissions lying on a higher curve corresponding to higher unmixedness parallel to the pre-premix curve. This level of unmixedness is than maintained as the curve continues parallel to the ideal curve. The staged burner shows remarkably low NO x close to the pre-premix curve corresponding to a well-mixed flame located outside the burner. 11

12 pulsation / arb. units flame off Pulsations STD Emissions and Pulsations Pulsations pre-premix Pulsation staged lg Nox: STD lg Nox: pre-premix lg Nox staged flame outside burner flame inside burner T flame /K unmixedness log (rel NOx) Figure 6: Relative NO x - emission and pulsation levels for three burners. The flames move out of the burner with decreasing T flame while NOx decrease and pulsation go through maxima. The pulsations for the standard burner seem to have a maximum slightly lower than the T flame, where the flame moves inside the burner. The pulsation maxima for the staged and pre-premix burner lie at lower T flame and much below that of flame movement. C. Pulsating flames The nature of the mechanism for the pulsation will become clearer in Figure 7 for the standard burner at T ref. The images have been recorded by phase locking to a frequency around Sr ~ For a number of operating points and burner configurations phase locked images have been recorded. Note that the pulsation amplitude is of the same order as the pressure difference of the p corresponding to the ζ-values as shown in Figure 4 linking the observed unsteady flame to observations of steady operating points for varying conditions. The acoustic fluctuations occurring naturally in this experiment are forced by loudspeakers upstream of the combustor. The full forcing power of the speakers resulted in an almost constant ~30% increase of pulsation amplitude for all variants affecting all naturally occurring levels very similar. The weakest enhancement through forcing was given for the weakest instability - the pre-premix at T ref. We are therefore confident that forcing does not create artifacts but rather enhances only phenomena that are naturally present. 12

13 Figure 7: Phase locked OH*-CL and OH-PLIF images sorted clockwise in a circle according to phase angle for Sr=0.57. Resolving the phase angle shows that the flame is moving between the two positions where the flame can be stable. The images of the opposing phase angles are comparable in shape and positions as well as in intensity to the difference in T flame. The maximum of the OH-PLIF seems to be slightly shifted in phase with respect to the OH*- CL-images. The flame position is in agreement with such a movement in and out of the burner compared to the steady state results. The standard case has the largest amplitude in intensity and pressure. At the same T flame the prepremix does not show any instability but about 50 K lower a weaker instability occurs. For the standard burner, resolving the phase angle shows that the flame is moving between the two positions where the flame can be stable (Figure 7) similar to a variation in T flame. The flame movement can be seen clearly 13

14 from the contour lines and from the plot of the leading edges (Figure 8). There is a small phase shift between the intensity maximum and the minimum of the leading edge with the latter lacking ca. one eights of a phase behind. Comparing the images of the OH*-CL in Figure 7 with the T flame variation for stationary points of Figure 1 the intensity fluctuations can be interpreted as fluctuation of and ( and consumed in the surface of the flame). From the relation derived above the fluctuating quantities (indicated by ) can be expressed as (the time dependency actually refers to phase shifts δ and δ.): m & = m(t & + δ) I I CL CL ( = = + ( _LBO )) ( ) ( _LBO ) + ( ) ( ) ( ) _LBO _LBO = = (t + δ ) + _LBO (t + δ ) _LBO or using equation 3: I I CL CL ( = 2 = 2 ' ) (φ φlbo ) + φ (φ φ ) 2 (t + δ LBO ) φ( t + δ + ( φ φ m LBO 2 ) ) = 2 ' φ + ( φ φ LBO ) Equation 5 and generally Q& = Q& and u u burner burner = Equation 6 From this relation it follows that in general intensity fluctuations are not proportional to fluctuations of heat release or mass flow. Q& Q& u u burner burner I I CL CL φ φ Equation 7 An exception is the pre-premix case where φ (and T flame ) are constant ( u burner Q& ) giving: 14

15 I I CL CL m & uburner Q& = 2 = 2 = 2 for = 0 u Q& burner φ Equation 8 This justifies the interpretation of the I CL fluctuation as u burner fluctuations for the pre-premix in Figure 8, while for the standard injection both quantities are fluctuating. The fluctuation for the pre-premix case is also clearly distinguishable but an order of magnitude smaller (3-4% in intensity) and phase shifted by 180 (δa-δf=180 ) hence counteracting the fluctuation in the standard case. It is due to its different nature as will be pointed out later. Therefore the interpretation of intensity fluctuation as temperature fluctuation for the standard burner refers to a lower limit for T flame as done on the left y-axis in Figure 8. I I CL CL < Q& = Q& T flame = T flame Equation 9 For the standard burner an I CL fluctuation of 28% results from heat release fluctuations of ca. 5-6% or T flame variation of ca. 45K while for the pre-premix 3-4% result from 1-2% fluctuation in u burner. 15

16 Fluctuating Quantities:T flame / U burner Standard Tflame /K I CL : ±28% p'=2.9*p' REF T flame : ±45K Tref standard; Sr= 0.59 Tref-50K pre-premix; Sr= 0.55 Tref-80K staged; Sr= 0.58 I CL : ±4 % p'=0.57*p' REF 103% 102% "pre-premix"/ staged l U 101% 100% 99% I CL : ±3.5 % p'=p' REF 98% u burner :±3-4% 97% phase angle/ 40 Leading edge 35 mm Standard Tref pre-premix Tref staged Tref-80K phase angle/ Figure 8: Fluctuation of I CL and leading edge (OH*-CL) for standard, pre-premixed and staged burner at different T flame. Note that the phase angles refer to slightly different Strouhal numbers on the same burner and the same time reference, which is determined by the forcing signal. For the standard burner the fluctuation lead to the flame clearly inside the burner for some of the phases. Values for given. I CL, p and T flame respectively uburner are D. Discussion The pulsations for the case of standard injection can be rationalized by the following reasoning assuming constant mass flow and varying mass flow. The occurrence of two stable positions for the flame anchoring point plays an important role. The instability for the standard burner occurs slightly below the temperature of the flame movement and can be explained by a flame, which starts to move upstream of the burner for the rich phases (0-90 ) (Figure 7). In this phase the is consumed faster resulting in increasing the heat release rate, temperature 16

17 and CL-intensity. The upstream flame (90 ) causes an increase of flow resistance (and ζ burner ) reducing the mass flow. This leads to conditions depleted of combustible mixture and slows down the heat release rate, which causes a displacement of the flame back outside of the burner ( ). This decreases ζ burner and leads to higher mass flow and leaner mixture burning slower ( ). The reduced heat release rate causes the region upstream of the flame to become slowly richer to be burned in the next phase upstream of the burner ( ). This periodic movement is an aerodynamic effect enhanced by the modulation of and equivalence ratio 11. Part of the pressure fluctuation can then also be explained by the change in p between the two stable flame positions in / out. It seems clear that the possible flame bi-stability, if coinciding with a feedback mechanism, can greatly enhance the amplitude of the pulsation. In order to accept the flame inside the burner, given aerodynamic conditions must be fulfilled involving an interaction between heat release and velocity flow field, in particular close to the burner axis. The nature of such interaction seems to be a baroclinic effect 18, 15, i.e. the production of azimuthal vorticity due to non alignment of pressure (due to centrifugal forces) and density gradients. The flame out / flame in could in principle be explained in terms of this interaction. The equivalence ratio (or whatever else) fluctuations play the role of modulating such interaction. More deep analysis of this mechanism is in progress. The I CL -fluctuation for the pre-premix case is also clearly distinguishable but smaller and phase shifted. The standard case has the largest amplitude in intensity and pressure. At the same T flame the pre-premix shows almost no instability but about 50 K lower a weaker instability occurs. The very weak peak at T ref seems to an intermediate point with two overlapping effects canceling each other. Since the equivalence ratio must be constant for all phases the pulsation has to be explained by a different mechanism. The laminar burning rate increases with T flame and can be related to the increased consumption but is not coupled to the aerodynamics. Therefore another mechanism must be invoked for the feedback of heat release fluctuation. The turbulent burning rate (u ~ heat release rate) can be described by 19 : u t u' 0.75 l 0.25 t λ sl Equation 10 Since l t and λ remain constant for standard injection and pre-premix the fluctuation of consumption must arise from the fluctuation of turbulence intensity as is described in more detail in ref 11. The increased heat 17

18 release causes acoustic energy to be released, which can be converted into turbulence intensity, which in turn speeds up the consumption rate. If a feedback allows such instability to grow it might cause a relevant pulsation. The pulsation amplitude for the pressure signal and the phase locked CL intensity of the pre-premix and staged case are lower than for the standard case. Since for the pre-premix case the equivalence ratio is constant for all phases the temperatures must also be constant and therefore the fluctuation in intensity is attributed to fluctuations of burner velocity alone, which give rise to much smaller amplitude in intensity fluctuations. For the staged case no instability seemed to be related to the bi-stability, which occurs at the highest T flame of all variants. The pulsation maximum appeared rather at a lower T flame not far from the pre-premix maximum. In this case the mechanisms of instability might be due to two reasons: fluctuations and turbulence intensity fluctuations. A coupling to the aerodynamic change of flame position has not been observed. The nature of the instability appears to be more similar to the pre-premix case but could be supported by equivalence ratio fluctuations as well. The staged burner is the outcome of a development for optimized emissions and pulsations (among other parameters) resulting in a flame burning outside of the burner up to high T flames and avoiding a sudden jump of flame position as well as an improved mixing. At least two different types of instabilities can be separated: pulsations related to flame sudden displacement of flame anchoring point and pulsations, which only fluctuate in intensity but not much in flame position. The staged burner is a realistic burner configuration with real injectors, that can be used in heavy duty gas turbines while being close in performance to the idealized pre-premix case. The further CFD analysis is focused on that burner. IV. CFD validation The experimental images are compared to CFD-simulations for their validation and comparison. The simulations were performed with steady-state k-ε realizable turbulence modeling on an unstructured half burner grid with over 1 million cells. The images were recorded and processed in the plane that lies parallel to burner slots. An example of unsteady LES simulation is given as outlook. Two different flame models were used in comparison: The finite rate eddy dissipation model (Further called EDC ) 20, 21, and The turbulent flame speed closure (further called TFC ) 22 18

19 Figure 9: 3D -CFD CL-intensity integrated compared to OH*-CL images for different T flame for the staged burner. To enable comparison of the CFD results with the experimental 2D and 3D images) a line-of-sight 3Dintegration program has been developed. To estimate the CL intensity the temperature and the heat release rate must be known. From this for each cell in the CFD grid a luminescence strength (~ intensity) has been obtained by using a linear approximation of the experimental intensity behavior as shown in Figure 4. This assures that the CLintensity from reactions in hot regions is contributing more than the intensity from reactions in cold regions. The intensity is taken to be from the CFD output T flame and reaction rate k, which corresponds to the heat release with a given T LBO : I CFD ( T T ) k Equation 11 flame LBO Although there is no strict proof for this formula it is used as a semi-empirical rule. The CFD simulations are carried out on unstructured grids. Therefore the results are first interpolated to a structured grid and then summed along the line of sight to match the camera view. The results obtained (Figure 9) were cross-checked with summing the rate k alone and do not alter the conclusions.the OH_PLIF images are directly compared to the 2D-cuts for the reaction rate as is shown in Figure 10. Only the standard and pre-premix cases were recorded in single shots and flame fronts could not be analyzed. The OH-PLIF-I images are not showing regions of heat release but are strongly influenced by the regions of recirculating gas and hot temperatures. A change in flame anchoring point over the T flame range however can clearly be deduced. Out of many CFD-simulations only some examples comparing two different flame models are shown. The dependance on post-processing method, treatment of wall temperatures has been tested but is not shown here. 19

20 However some trends shall be mentioned: The difference of the 2 dimensional cuts to the 3 dimensional projections show quite different pictures revealing the fact that using 2D cuts only can be misleading for a judgment of the flame properties. The EDC simulations were done with measured wall temperatures. For comparison a case with adiabatic walls was also simulated. The adiabatic simulation leads to unrealistically high temperatures near the wall for some spots and consequently a major part of the integrated heat release occurs at the wall. This spot is not on any of the central planes and can therefore only be seen only in the integrated image. The TFC simulations were performed with adiabatic walls. As in the EDC simulations the flame is always located inside the burner for all temperatures in contradiction to the experimental findings. For both flame models the shape remains similar over all temperatures. Figure 10: CFD heat release rate as 2D cut compared to OH*-PLIF images in the same plane for different T flame for the staged burner. A flame outside the burner has never been seen in the RANS CFD simulations using different theoretical flame model models under all conditions even for low T flame, where the experimental images show clearly a different flame position. This might be due to the treatment of flow field and turbulence, which is common in both versions discussed here. For the RANS simulations all flames are predicted to be anchored in the inner part of the burner which is not confirmed for lower temperatures and other staging ratios. The flame position seems to be mainly governed by flow field instead of reactivity or injection. Both flame models seem to have the same shortcomings suggesting that the turbulence modeling of the flow field itself is not sufficient. 20

21 The use of more sophisticated (and expansive) models like LES can mitigate this 15. This is supported by promising results, which are briefly presented here in Figure 11. The LES were not carried out as systematically as the RANS simulations and should be viewed as an outlook. The LES are significantly more expensive and the results have been averaged over only several hundred time steps, but the trend becomes already clear: The lower graph shows the flame anchored inside the burner at a temperature slightly above T ref as also the RANS models predicts. The upper graph however shows a stable flame located outside the burner at only slightly lower T flame. Such a flame position could not be obtained by the RANS method. Figure 11: Top Outside burner flame stabilization mode as 2 D cuts using LES with an TFC model. Bottom: inside burner flame stabilization mode. Colours represent axial velocity, black lines the reaction progress variable isolines and the white line the zero axial velocity isoline. A theoretical and CFD analysis of the flame behavior in the EV burner has been already presented in 10. The main results of this analysis are summarized as follows: a) The use of RANS (Reynolds Averaged Navier-Stokes equations) yields too thick recirculation region and lower turbulence intensity in relation to water rig experiments. It is shown that, at the burner exit, these 21

22 two factors lead to an inflow of reactants into the flame, which is much lower than the turbulent burning rate. In RANS, therefore the flame, quite independently from the combustion model used and from the magnitude of the turbulent burning rate, adjusts always deep inside the burner where the inflow of reactants is shown to match the turbulent burning rate. b) A simple boundary layer type 1D model, which determines the flame anchoring position at the axis, was able to give the correct flame position below the critical flame temperature at which the flame moves inside the burner (so called outside burner flame stabilization mode) when the water rig flow field was used. The 1D model, was however unable to give the sudden displacement of the flame inside the burner because the water rig flow field gives, inside the burner, a rate of reactants fed into the flame which is much higher than the turbulent burning rate (the flame is predicted always outside the burner by this simple 1D model where inflow if reactants and turbulent burning rate match together. c) Based on the 1D model results, it was speculated that until the critical flame temperature at which the flame moves inside the burner, the flow field in the burner must be as observed in the water rig. However, at the critical flame temperature, a modification of the approaching flow must necessarily take place in order for the flame to adjust deep in the burner. Such idea was fully confirmed by the LES results as shown in figure 11. It is speculated that at the critical flame temperature, flame and velocity flow field start interacting. The nature of such interaction is presently under investigation. The thickness of the averaged heat release zones is also underestimated in the CFD simulation. While the experimental images show relatively smoothly distributed images the CFD usually displays two lobes of higher reaction (luminescence) density. For the OH-PLIF the findings are similar. This might be due to the relatively high [OH]-concentration in the post flame zone under atmospheric conditions. The steady Reynolds-averaged (RANS) CFD fails to predict the flame positions and shape in dependence of T flame for the EV burner. The flame appears to be always inside the burner contradicting the observation. This is the case for the EDC and the TFC model if used with standard realizable k-e turbulence treatment while LES methods might yield better results. The former are not recommended for the determination of flame position or CO burnout (at least for EV type burners). The LES model has not been optimized to predict the occurrence of such flame 22

23 movements, but it is shown that it is capable of explaining such bi stabilities, while the RANS-methods fail to shows such behavior. V. Conclusions OH-planar laser induced fluorescence (OH-PLIF) and OH* -chemiluminescence (OH*-CL) images were recorded for several operating conditions and phase angles for a pulsating flame. To compare to 2 dimensional heat release plots a flame front averaging method has been used for the OH-PLIF technique. For comparison with CL flame images a CFD integration tool has been used. The comparison reveals several shortcomings of the RANS-CFD predicting flame positions and shapes, which may be attributed to the use of the half burner grid and the way the turbulence chemistry interaction is treated. The results are similar for both flame models (TFC and EDC). The view angle and the method of post processing of the CFD results have a significant influence on the obtained images. The use of LES model is capable of describing such bi-stable behavior. In general the CL intensity is good measure for T flame but not for change in staging ratio. The flame is shown to move inside of the burner for increasing T flame depending on the hardware. The flame movement could not be reproduced by the CFD, which predicted the flame to be anchored inside for all cases. For the standard injection the flame movement coincides with NOx increase and the occurrence of the instability. Phase locked images for pulsating cases of the three configurations have been recorded and show fluctuations of the flame intensity for all cases and a change in flame position (in and out of the burner) in axial direction for the standard but not for pre-premix and staged. For the standard case this can be explained with an equivalence ratio feedback mechanism. The pre-premix instability has lower amplitudes and needs to be explained by a different mechanism. Fluctuating turbulence intensity has been proposed. For the staged burner the pulsation is weak and seems to be more similar in nature to the pre-premix case. Two different mechanisms for thermoacoustic instabilities could be found on a swirl stabilized burner. Phase locked images for pulsating cases of the three configurations have been recorded and show fluctuations of the flame intensity for all cases and a change in flame position (in and out of the burner) in axial direction for the standard but not for pre-premix and staged. For the standard case this can be explained with an equivalence ratio feedback mechanism enhanced by the movement of the flame. This movement with T flame coincides with NOx increase and the occurrence of the instability. The pre-premix instability has lower amplitudes and needs to be explained by a 23

24 different mechanism. Fluctuating turbulence intensity has been proposed. For the staged burner the pulsation is weak and seems to be more similar in nature to the pre-premix case. Three different burner variants have been compared: The pre-premix experiment as an idealized burner allowed separating effects of burner velocity (u burner ) and equivalence ratio fluctuations. Further a standard burner as a counter example with high pulsations and emissions and a burner of improved design using two stages for injection were studied. The staged burner has been optimized for emissions and pulsations using its increased flexibility. The flame images reveal that this behavior is related to the movement of the flame within the combustor with T flame. This burner has been developed optimizing for emissions and pulsations using its increased flexibility in a result of the combined effort of analytical and experimental tools. This example shows one of the trends of the development of swirl stabilized burners, where the flame is kept out of the inner part of the burner by rearranging the injection minimizing NO x and stabilizing the flame at the same time. Staged injection can be used to control flame position and mixing quality to optimize emissions and pulsations. This burner development has been achieved by a combined effort of analytical, numerical and experimental tools. Acknowledgments The experimental data are the outcome of a joint project financed by the KTI (Swiss Kommission für Technologie und Innovation -KTI) Projekt-Nr.: EBS, and partners ALSTOM and PSI Villigen. Note The images are given as black and white for this journal publication but can be obtained from the author on request as colour scaled images. 24

25 References 1 Döbbeling K., Hellat J., Koch H., 25 Years of BBC/ABB/ALSTOM Lean Premix Combustion Technologies, ASME Turbo Expo 2005, GT Sattelmayer T., Felchlin M. P., Haumann J., Hellat J.: Second Generation Low-Emission Combustors for ABB Gas Turbines: Burner Development and Tests at Atmospheric Pressure, Transactions of the ASME, Journal of Engineering for Gas Turbines and Power, Jan. 1992, Vol Aigner M., Mayer A., Schiessel P., Strittmatter W.: Second Generation Low-Emission Combustors for ABB Gas Turbines: Test under full engine conditions, ASME Turbo Expo 1990, 90-GT Guethe F., Lachner R., Schuermans B., Biagioli F., Geng W., Inauen A., Schenker S., Bombach R., Hubschmidt W., 44th AIAA Aerospace Sciences Meeting and Exhibit, 2006, Reno, Nevada 5 Weigand P., Duan X. R., Meier W., Meier U., Aigner M., Bérat C., Experimental investigations of an oscillating lean premixed CH4/ swirl flame, Proceedings of the European Combustion Meeting, Külsheimer C., Büchner H., Combustion Dynamics of Turbulent Swirling Flames, Combustion and Flame 131: (2002), p Balachandran R., Ayoola B. O., Kaminski C. F., Dowling A. P., and Mastorakos E., Experimental Investigation of the Non-linear Response of Turbulent Premixed Flames to Imposed Inlet Velocity Oscillations," Combustion and Flame, 143, 2005, pp Paschereit C. O., Gutmark E., Weisenstein W., Flow-Acoustic Interactions as a Driving Mechanism for Thermoacoustic Instability, Physics of Fluids 11, 1999, pp Lee J. G., Santavicca D. A., AIAA Journal of Propulsion and Power, Vol. 19, No. 5, 2003, pp 735. Paschereit C. O., and Gutmark E.,: The Effectiveness of Passive Combustion Control Methods, ASME Turbo Expo 2004, GT Schuermans B., Bellucci V., Guethe F., Meili F., Flohr P., Paschereit C. O., A detailed analysis of thermoacoustic interaction mechanisms in a turbulent premixed flame, ASME Turbo Expo 2004, GT Reinke M., Mantzaras J., Schaeren R., Bombach R., Kreutner W., Inauen A., Homogeneous ignition in highpressure combustion of Methane/ over platinum: comparison of measurements and Detailed numerical predictions, Proceedings of the Combustion Institute, Volume 29, 2002, pp ,. 13 Schenker S., Bombach R., Hubschmid W., Inauen A., Kreutner W., Flohr P., Haffner K., Motz C., Paschereit C. O., B. Schuermans, and Zajadatz M., 2-D LIF measurements of thermo-acoustic phenomena in lean premixed flames of a gas turbine combustor, Inst. Phys. Conf. Ser. 177, 2003, pp Ayoola B.O., Balachandran R., Frank J. H., Mastorakos E., Kaminski C.F., Spatially resolved heat release rate measurements in turbulent premixed flames, Combustion and Flame 144, 2006, p, Biagioli F., Stabilization mechanism of turbulent premixed flames in strongly swirled flows, Combustion Theory and Modelling, vol. 10, 2006, pp Higgins B., McQuay M. Q., Lacas F., Rolon J. C., Darabiha N., Candel S., Systematic measurements of OH chemiluminescence for -lean, high-pressure, premixed, laminar flames, Fuel 80, 2001, pp Gaydon A. G., Spectroscopy of flames Chapman and Hall, London, Hasegawa T., Nakamichi R., Nishiki S., Mechanism of flame evolution along a fine vortex, Combustion Theory and Modelling 6, 2002, pp V. L. Zimont, The theory of turbulent combustion at high Reynolds numbers Combust. Expl. Shock Waves 15, 305 (1979). 20 W. Polifke, W. Geng, K. Döbbeling, Optimisation of Rate coefficient for simplified reactions with genetic algorithms, Combustion & Flame, 113, ,

26 21, B. F. Magnussen, B. H. Hjertager, On Mathematcal Modeling of Turbulent Combustion with Special Emphasis on Soot Formation and Combustion, 16th Symp. (Int.) on Combustion (1976). Comb. Inst., Pittsburg, Pennsylvania, , V. L. Zimont, "Theory of Turbulent Combustion of a Homogeneous Fuel Mixture at High Reynolds Numbers" Combustion, Explosions and Shock Waves, vol. 15, n.3, p ,

27 Colour pictures Figure 1 Figure 2 Figure 3 27

28 Average Intensity vs. T flame Average Intensity T LBO staged LIF staged CL Prepremix LIF Prepremix CL Standard LIF Standard CL T flame -T ref (K) Leading Edge (mm) Leading Edge vs. T flame window edge staged LIF staged CL pre-premix LIF pre-premix CL standardlif standard CL T flame -T ref (K) ζ burner vs. T flame Figure 4 ζ relative staged CL prepremix CL standard CL T flame -T ref (K) 28

29 Average Intensity Average CL-Intensity vs. Heat Release u o staged T varied Prepremix T varied Standard T varied Standard ub varied prepremix ub varied u burner varied T flame =T ref T flame varied u burner= u ref T LBO 500 Figure 5 Figure pulsation / arb. units flame off relative heat release ~ m Emissions and Pulsations Pulsations STD Pulsations pre-premix Pulsation staged lg Nox: STD lg Nox: pre-premix lg Nox staged flame outside burner 1.0 unmixedness flame inside burner T flame /K log (rel NOx)

30 , Figure 7 30

31 Fluctuating Quantities:T flame / U burner Standard Tflame /K I CL : ±28% p'=2.9*p' REF T flame : ±45K I CL : ±3.5 % p'=p' REF u burner :±3-4% Tref standard; Sr= 0.59 Tref-50K pre-premix; Sr= 0.55 Tref-80K staged; Sr= phase angle/ Leading edge I CL : ±4 % p'=0.57*p' REF 103% 102% "pre-premix"/ staged rel. Uburner 101% 100% 99% 98% 97% 35 mm Standard Tref pre-premix Tref staged Tref-80K 20 Figure phase angle/ Figure 9 31

32 Figure 10 Figure 11 32