DEVELOPMENT AND APPLICATION OF TRACER-BASED PLANAR LASER-INDUCED FLUORESCENCE IMAGING DIAGNOSTICS FOR HCCI ENGINES

Transcription

1 DEVELOPMENT AND APPLICATION OF TRACER-BASED PLANAR LASER-INDUCED FLUORESCENCE IMAGING DIAGNOSTICS FOR HCCI ENGINES A DISSERTATION SUBMITTED TO THE DEPARTMENT OF MECHANICAL ENGINEERING AND THE COMMITTEE ON GRADUATE STUDIES OF STANFORD UNIVERSITY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY Jordan A. Snyder March 2011

2 2011 by Jordan Andrew Snyder. All Rights Reserved. Re-distributed by Stanford University under license with the author. This dissertation is online at: ii

3 I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy. Ronald Hanson, Primary Adviser I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy. Christopher Edwards I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy. Mark Mungal Approved for the Stanford University Committee on Graduate Studies. Patricia J. Gumport, Vice Provost Graduate Education This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file in University Archives. iii

4 iv

5 Abstract Homogeneous charge compression ignition (HCCI) is an emerging engine strategy that can provide both high efficiency and low emissions, particularly in terms of NO x and soot. An important challenge of HCCI is the inherent narrow load range, bounded by combustion instability and misfires at low-load, and high pressure riserate (PRR) at high-load. In response, researchers have devised a number of strategies to expand the limits of HCCI operation. Negative valve overlap (NVO) with pilot injection can extend the low-load gasoline HCCI operating limit by increasing sensible energy during main compression through hot residual gas retention. Chemical effects due to reformation of the pilot injection may further impact combustion. Similarly, the high-load limit can be extended by increasing naturally occurring thermal stratification (TS) of the incylinder charge. These non-uniformities result in sequential auto-ignition that can effectively lower the PRR and thus expand the high-load limit. While demonstrations of these strategies have been successful and multiple engine studies have been completed, further characterization of key processes such as residual gas mixing and TS development is needed. This motivates the development of quantitative imaging diagnostics to improve the understanding of these complicated processes. In this study, tracer-based planar laser-induced fluorescence (PLIF) diagnostics for temperature and composition have been refined and optimized for application in HCCI engines at both load extremes. Acetone and 3-pentanone (both ketones) have been selected as seeded PLIF tracers as they provide good overall sensitivity and performance. Single-line and two-line diagnostic variations have been investigated, with an emphasis on optimizing overall diagnostic performance through v

6 excitation wavelength selection. Based on a detailed uncertainty analysis excitation wavelengths of 277 nm and 308 nm were selected for subsequent studies. Resulting single-shot temperature precisions were typically on the order 4 K and 12 K for the single-line and two-line techniques respectively. The corresponding mole fraction precision for the two-line technique was typically 4-5%. These results are consistent with the uncertainty analysis and demonstrate the utility of the optimization. HCCI studies were performed in two optically accessible engines, each configured for a specific load extreme. Residual mixing for low-load HCCI operation with NVO was first studied using the two-line technique to provide the simultaneous temperature and composition distribution. These measurements indicated rapid mixing of retained residuals during gas exchange and early compression, reaching a steadystate value midway through compression. Temperature stratification gradually increased throughout the remainder of compression while compositional stratification effectively remained constant. Variation of operating parameters such as main and NVO injection timing exhibited minimal differences in thermal or compositional stratification during main compression. Measurement during NVO recompression and re-expansion were also acquired to assess the in-cylinder temperatures stratification prior to chemical reaction and gas exchange. Next the development of thermal stratification for high-load HCCI with conventional valve timing was investigated using the single-line technique. These studies indicated a progressive increase in TS during compression, reaching a maximum standard deviation of 10 K at top dead center. Comparison of results for motored and fired operation exhibited similar trends. This finding indicates that the mechanism producing the TS is the same for both cases, although some differences in magnitude can occur. A subsequent parametric study proved that these differences can be attributed to the impact of both incomplete fuel mixing and cylinder-wall temperature variation, depending on the type of engine operation (DI skipfiring or premixed continuous firing). All measurements demonstrate the feasibility of quantitative tracer-based PLIF diagnostics in harsh engine environments and provide useful information for future HCCI engine development. vi

7 Acknowledgements First and foremost I would like to thank my advisor, Professor Ronald K. Hanson, for providing a wealth of opportunities at Stanford. His vast technical knowledge, research vision and relentless drive continue to inspire. It truly has been a privilege to be a part of the Hanson research group. I also wish to thank the members of my reading committee, Professor M. Godfrey Mungal and Professor Christopher F. Edwards for their continuing guidance and input. Thanks also to Professor Robert L. Byer for chairing my examination committee and to Professor Craig T. Bowman for serving as an examiner. I owe a debt of gratitude to Dr. David F. Davidson and Dr. Jay B. Jeffries for the countless discussions and the willingness to always answer questions or lend a hand when needed. A special thanks to all the members of the Hanson group, both current and former. It is a unique opportunity to work with such a large, talented group of students with overlapping but diverse interests. Dr. Dave Rothamer has been particularly helpful in shaping my research at Stanford, and remains both a great friend and mentor. A large portion of my research was performed at the Combustion Research Facility at Sandia National Laboratories in Livermore, CA. As a result, I was privileged to work closely with Dr. Richard R. Steeper, Dr. John E. Dec, Dr. Russell Fitzgerald and Dr. Nicolas Dronniou. This rare opportunity to work closely with such accomplished and talented researchers has been invaluable to my knowledge and development. I am sincerely grateful to my parents, Nick and Sharon, who have always supported me and instilled a work ethic and drive that has allowed me to be successful vii

8 both personally and academically. Finally, and most importantly I would like to thank my loving wife Christen whose support has been unwavering throughout my academic journey. Anything I have achieved is a direct result of her patience, sacrifice and frequent reminders of the bigger picture in life. viii

9 Contents Abstract Acknowledgements v vii 1 Introduction Background and Motivation PLIF Overview of Dissertation PLIF Technique and Optimization PLIF Theory Single-Line PLIF Two-Line PLIF Tracer Selection Wavelength Optimization Uncertainty Analysis Theory Uncertainty Analysis Parameters Single-Line Analysis Two-line Analysis Tracer Photophysics FQY Measurements in a Motored IC Engine Measurement Technique Pentanone Photophysics ix

10 3.3 Acetone Photophysics FQY Model Comparisons Pentanone Modeling Acetone Modeling Fluorescence Saturation Introduction Experimental Approach Optical Setup Data Acquisition Beam Width Measurements Experimental Results Saturation Wavelength Dependence Pressure and Composition Dependence Low-Load HCCI with NVO Experimental Setup Motored Engine Operation Fired NVO Engine Operation Data Acquisition Validation Experiments Measurement Interferences Fired NVO Results High-Load HCCI Experimental Setup Optical Engine PLIF System Engine Operation Data Acquisition and Processing Conventional Data Acquisition PLIF Data and Processing x

11 6.3.3 PLIF Data Processing Photophysics PLIF Measurement Uncertainty Accuracy Precision Motored Engine Thermal Stratification Motored Stratification Statistics Fired Engine Thermal Stratification Residual Mixing Comparison of Motored and Fired TS Correlation of Temperature and Reacting Zones Summary and Future Work PLIF Development Future Work: PLIF Development HCCI with NVO Future Work: HCCI with NVO HCCI Thermal Stratification Future Work: HCCI TS A Ketone Photophysical Parameter Fits 147 B Uncertainty Analysis Theory 151 C Camera Noise Characterization 154 Bibliography 157 xi

12 List of Tables 2.1 Excitation wavelengths, energy and energy stability inputs used for temperature and EGR precision calculations Input parameters used for temperature and EGR precision calculations, same as Excitation wavelength specifications for saturation experiments Experimental measurements of 10% non-linear threshold for various tracer and excitation wavelengths HCCI engine specification for NVO operation Fired engine operating conditions Engine Specifications Engine Operating Conditions A.1 Temperature dependent gaussian fit parameters for the absorption cross section of acetone and 3-pentanone for nm and K, taken from [1]. Temperature in kelvin A.2 Quadratic fit parameters of 3-pentanone absolute FQY in pure nitrogen measured in a motored engine. Fit parameters for a given image time (pressure) are used in conjunction with Equation A.2 to calculate the FQY for temperature within the stated range xii

13 A.3 Linear fit parameters of acetone absolute FQY in pure nitrogen measured in a motored engine. Fit parameters for a given image time (pressure) are used in conjunction with Equation A.3 to calculate the FQY for temperature within the stated range xiii

14 List of Figures Pentanone single-line PLIF measurement uncertainty estimates of temperature for potential excitation wavelength pairs. All precision is quoted based on ±1 standard deviation Acetone single-line PLIF measurement uncertainty estimates of temperature for potential excitation wavelength pairs. All precision is quoted based on ±1 standard deviation Toluene single-line PLIF measurement uncertainty estimates of temperature for potential wavelength pairs. Simulations are for an N 2 intake stream. All precision is quoted based on ± 1 standard deviation Pentanone two-line PLIF measurement uncertainty estimates of temperature (left) and EGR mole fraction (right) for potential excitation wavelength pairs. All precision is quoted based on ±1 standard deviation Acetone two-line PLIF measurement uncertainty estimates of temperature (left) and EGR mole fraction (right) for potential excitation wavelength pairs. All precision is quoted based on ±1 standard deviation Absolute FQY of 3-pentanone in nitrogen at 277 nm and 308 nm, measured under motored engine conditions. Each curve represents experiments at a given image timing (pressure) for a range of intake temperatures xiv

15 3.2 Comparison of absolute FQY results from motored engine and flowing cell Absolute FQY of acetone in nitrogen at 277 nm and 308 nm, measured under motored engine conditions. Each curve represents experiments at a given image timing and pressure for a range of intake temperatures Comparison of engine FQY data derived from quadratic fits and FQY model simulations. (a) Absolute FQY, (b) Normalized FQY for better comparison of slope Comparison of engine FQY data derived from linear fits and FQY model simulations. (a) Absolute FQY, (b) Normalized FQY for better comparison of slope Experimental setup for measurement of tracer saturation intensity (a) Representative fluorescence images without focusing lens (top) and with lens (bottom), laser propagation is from left to right. (b) Corrected axial signal saturation profile indicating fraction deviation from linearity (a) Series of beam profile images acquired along beam focus. (b) Measured beam half width along focus with fit results Percent deviation from linearity as a function of laser fluence, measured for 308 nm excitation of 3-pentanone at 1 bar in N 2 (derived from same data as Figure 4.2) Comparison of deviation from linearity for acetone, 3-pentanone and toluene at 248 nm excitation, and 1 bar of N Comparison of 10% non-linear threshold (top) and inverse absorption cross-section (bottom) for all tracers and wavelengths (a) Pressure dependence of 3-pentanone saturation with 308 nm excitation and a nitrogen bath gas. (b) Comparison of saturation characteristics for nitrogen and air bath gases with 308 nm excitation at 15 psi total pressure. Multiple curves included to demonstrate measurement repeatability xv

16 4.8 Pressure dependence of acetone saturation in pure nitrogen for excitation wavelengths of (a) 277 nm and (b) 248 nm Pressure dependence of toluene saturation with 248 nm excitation and a nitrogen bath gas Experimental schematic Sample PLIF image with valve, injector and piston window positions superimposed Measured pressure for NVO engine operation, (a) pressure vs. CAD showing valve events and measurement regimes; (b) pressure vs. volume (log-log scaling) (a) Measured average temperature and (b) air mole fraction measured for motored engine conditions with air intake temperature of 412 K and a manifold pressure of 1 bar (a) Measured temperature standard deviation and (b) air mole fraction standard deviation for motored conditions with air intake temperature of 412 K and manifold pressure of 1 bar Single-shot 277 nm and 308 nm LIF images of (a) residual gas recorded at +260 CD (no tracer), (b) carry-over 3P recorded at +260 CAD (prior to fuel injection, and (c) 3P recorded at +285 CAD (following DI fuel injection) Single-shot temperature and fuel mole fraction image pairs of carryover 3P signal during NVO recompression. Image capture timing is +260 CAD for all images. Three main combustion loads: (a) 7 mg/cycle, (b) 8 mg/cycle, (c) 9.5 mg/cycle Single-shot temperature and fuel mole fraction images during NVO recompression. Image capture time = +285 CAD. Three main combustion loads: (a) 7 mg/cycle, (b) 8 mg/cycle, (c) 9.5 mg/cycle Single-shot temperature and fuel mole fraction images during NVO expansion for early main injection at -320 CAD. Three NVO SOI timings shown: (a) +260, (b) +300, and (c) +330 CAD xvi

17 5.10 Single-shot temperature and EGR mole fraction images recorded at three image timings: (a) -215, (b) -65, and (c) -24 CAD. NVO SOI = +330 CAD; main SOI = -270 CAD. Images at -24 CAD (only) are binned 2x2. Note variable temperature color scale Demonstration of data correction for diagnostic uncertainty (a) Measured average temperature and (b) temperature standard deviation for several engine operating conditions. The error bars shown in (a) represent accuracy estimated from motored diagnostic characterization experiments (a) Measured average EGR mole fraction and (b) EGR standard deviation for several engine operating conditions. Error bars represent accuracy estimated from motored diagnostic characterization experiments Detailed schematic of optical HCCI engine showing location of laser sheets and imaging camera HCCI engine facility schematic PLIF experimental schematic for high load HCCI experiments PLIF image field of view with valves, injector, and piston window positions superimposed Measured absolute temperature for motored engine conditions with N 2 intake temperature of 170 C and 1 bar manifold pressure (a) In-cylinder pressure and temperature (calculated) for the baseline HCCI operating condition. (b) Temperature precision of single-line and two-line diagnostics of 3-pentanone ad acetone, calculated for baseline conditions Single-shot temperature images of TS development during main compression of motored engine. Diagnostic - 3-Pentanone, single-line 277 nm Comparison of single-line and two-line image quality xvii

18 6.9 Demonstration of measurement uncertainty correction, comparing corrected and uncorrect temperature standard deviation for singleline 3-pentanone at 308 nm excitation Temperature standard deviation (corrected) calculated from singleshot temperature images for both 3-pentanone (3P) and acetone (Ac) for different excitation wavelengths Demonstration of image binarization and pocket detection (b) for a single-shot temperature distribution (a) acquired in motored engine Evolution of cold pocket frequency and effective diameter during motored compression at 170 C intake temperature. Statistics derived from single-line 3-pentanone measurements at 277 nm Measured cylinder pressure tracers for 17-3 skipfired operation In-cylinder conditions for fired studies. The core temperature is estimated from measured pressure trace assuming adiabatic compression with variable specific heats. Data points correspond to PLIF image timings Single-shot temperature (left) and air mole fraction (right) image pair sequence showing evolution of residual gas mixing during the intake and early compression for fired HCCI operation. Diagnostic: two-line 277/308nm, 3-pentanone Single-shot PLIF temperature sequence of TS development for skipfired (cycle 20) HCCI engine operation. Diagnostic - single-line 277 nm, 3-pentanone. Engine conditions: skipfired, φ=0.4, T in =198 C, P in =100 kpa, 14% 3-pentanone in iso-octane Noise-corrected temperature standard deviation for cycle 18 (no hot residuals) and cycle 20 (hot residuals). Diagnostic: single-line 277nm, 3-pentanone. Engine conditions: skipfired, φ=0.4, T in =198 C, P in =100 kpa, 14% 3-pentanone in iso-octane xviii

19 6.18 Impact of direct fuel injection on measured temperature standard deviation for motored engine operation. Engine conditions are identical to skipfired experiments of Figure Diagnostic: single-line 277nm, 3-pentanone. Engine conditions: skipfired, φ=0.4, T in =198 C, P in =100 kpa, 14% 3-pentanone in iso-octane Impact of upper cylinder-wall temperature based on the noise corrected temperature standard deviation for motored and fired operation with varying coolant temperature. Diagnostic: single-line 277nm, 3-pentanone. Engine conditions: continuous fired, φ=0.32, T in =190 C, 17% 3-pentanone in iso-octane (a) Inverse binarization of fluorescence signal highlighting reaction zones. (b) Correlation between reaction area ratio (RAR) and CA10 combustion phasing Spatial correlation of temperature distribution before TDC (left) and early reaction zones (right) after TDC acquired for same cycle C.1 Characterization of PIMAX2:1003 image SNR versus signal based on the temporal SNR method. Polynomial fit results included for reference.156 xix

20 xx

21 Chapter 1 Introduction 1.1 Background and Motivation PLIF Planar laser-induced fluorescence (PLIF) is a non-intrusive, spatially resolved optical diagnostic that has evolved into a valuable tool for the investigation of flowfields and combustion systems [2]. PLIF is most frequently employed to measure species concentration (mole fraction) [3 7], however fluorescence sensitivities can be utilized to develop diagnostic strategies for temperature [8 11], pressure [12 14] and velocity [12, 15 17]. High characteristic signal levels make PLIF an appropriate choice for quantitative single-shot imaging and can provide temporal resolution of instantaneous flow phenomena. Alternative techniques such as Raman [18, 19] and Rayleigh [20,21] scattering can also be utilized for quantitative measurements. However, these techniques typically must be highly averaged (spatially or temporally) to overcome the inherently low signals and are less applicable when temporal resolution is of interest. PLIF is characterized by a two-stage process: laser-induced absorption which is followed closely by spontaneous emission (fluorescence). A typical PLIF implementation consists of a pump laser, a collection camera, and the associated laser sheet forming and collection optics. The laser output, tuned to an absorption feature of 1

22 2 CHAPTER 1. INTRODUCTION the probed species, is formed into a thin sheet and used to illuminate the flowfield of interest. A portion of the laser photons are absorbed, promoting a valence electron of the probed molecule to an excited state. Electronic excitation using ultraviolet wavelengths is most common, however excitation of vibrational transitions in the infrared (IR-PLIF) has also been demonstrated [22, 23]. A fraction of these excited molecules will subsequently return to the ground state while simultaneously emitting a photon, resulting in measurable fluorescence signal. This fluorescence signal is captured at 90 by a CCD camera, providing a two-dimensional image of the flow-field. The probed species can either occur naturally in the flow (i.e. OH, CH, NO, O 2 ) or be artificially seeded into the flow as a tracer. Because a majority of the electronically active species are combustion radicals, they can not be used to characterize the flow prior to reaction. As a result, seeded organic tracers such as acetone [24, 25], 3-pentanone [26, 27], and toluene [28, 29] are commonly used for measurements in non-reacting flowfields or in combustion systems prior to chemical reaction. To permit quantitative tracer-based techniques, substantial effort has gone into characterizing the spectroscopic behavior of these tracers [30 35]. Tracer-based PLIF has historically been used in internal combustion (IC) engine research with noted success and has helped to improve understanding of basic engine processes as well as advanced engine concepts. Engine applications typically focus on measuring in-cylinder fuel distribution [36 39]. Quantitative temperature measurements have also been demonstrated but with less frequency [40 44]. While these studies show favorable results, further technique refinement can improve measurement accuracy and precision. Diagnostic precision is especially important when studying small spatial fluctuations within the flowfield and limits the degree of nonuniformity that can be resolved. The current work explores the optimization of PLIF measurement precision through careful selection of tracer species and excitation wavelength. The optimization focuses on thermodynamic conditions relevant for homogeneous charge compression ignition engines (HCCI) as stratification has been found to influence HCCI combustion.

23 1.1. BACKGROUND AND MOTIVATION 3 HCCI Engines Homogeneous charge compression ignition (HCCI) is an emerging engine strategy that can provide both high thermal efficiency and low emissions [45], particularly in terms of soot and nitric oxide. HCCI is often characterized as combining benefits from spark-ignition (SI) gasoline and compression-ignition (CI) diesel engines. Fundamentally, HCCI utilizes a premixed charge similar to an SI engine but operates with a globally lean mixture that is ignited solely by compression similar to a diesel engine. However, unlike SI or CI engines HCCI combustion occurs volumetrically throughout the charge with no discernable flame-front propagation, thus reducing the total reaction time. Overall, HCCI has the potential to provide near diesel-like efficiencies with reduced emissions. The concept of HCCI has been around for nearly 30 years, although HCCI has only recently received substantial research attention as a viable and efficient engine strategy. The concept of HCCI was first proposed by Onishi et al. [46] for two-stroke engines and was later extended to the four-stroke cycle by Najt et al. [47]. Since these early studies, substantial research effort has been focused on characterizing overall HCCI performance and mapping out the load-speed operating range. Multiple reviews of HCCI concepts and technologies are available and provide a detailed discussion of current trends that will only be mentioned here briefly [48 50]. There are several factors that contribute to the improved efficiency of the HCCI engine over conventional SI engines [45]. First, the lack of spark-ignition and a propagating flame-front permits operation at fuel lean conditions and eliminates the need to operate near stoichiometric fueling. This means that intake throttling at partial load is no longer required thus reducing pumping losses and improving efficiency. Second, the engine can generally be operated at higher compression ratios as auto-ignition is actually a requirement. Third, the duration of combustion is shorter which reduces heat transfer and further improves efficiency. Attributes that improve engine efficiency also help to reduce the emission of several regulated pollutants. Specifically, the lean fuel/air mixture results in lower

24 4 CHAPTER 1. INTRODUCTION combustion temperatures and avoids the thermal formation of NO x. This is in contrast to diesel engines where peak in-cylinder temperatures can reach 2700 K [48], resulting in high levels of NO x formation. In addition, the lean premixed charge results in negligible soot formation due to the low combustion temperatures and the avoidance of localized fuel rich regions that generate high polycyclic aromatic hydrocarbons (PAHs) concentrations. Additional advantages of HCCI include the potential for lower equipment cost by avoiding the need for a high-pressure injector system and costly exhaust after-treatments, as well as the scalability of the engine stategy over a range of applications (light to heavy duty). HCCI engines can operate with a range of fuels, however this research considers only gasolinetype HCCI (single-stage fuel with no low temperature combustion) with a focus on light/medium applications at partial-load. Along with the benefits of HCCI come a number of technical challenges that must be addressed prior to widespread implementation. A primary challenge is the control of combustion phasing. Unlike SI or CI engines that utilize spark or direct fuel injection to control phasing, there is no direct means to control the onset of HCCI combustion. Instead the combustion phasing is kinetically-limited and is dependent only on the mixture properties and the temperature and pressure time history. A second challenge is the potential for elevated unburned hydrocarbons (UHC) and CO emissions. This results from the low combustion and exhaust temperatures associated with lean operation, and high fuel concentration in crevice volumes due to the premixed fueling. A third challenge is the narrow HCCI load range that is limited by combustion instabilities and misfires at the low-load limit, and high pressure rise-rates and engine knock at the high-load limit. HCCI is likely to be implemented in a hybrid system with spark ignition operation used for high-load, and HCCI used for low-load. However, expanding the HCCI load range is still important as it will maximize the benefit of the HCCI cycle and will reduce the frequency of mode transitions. The current work focuses on potential strategies to expand the HCCI operating range.

25 1.1. BACKGROUND AND MOTIVATION 5 Low-Load HCCI Many strategies for low-load extension incorporate extensive exhaust gas recirculation (EGR) to provide additional thermal energy during the compression stroke in an effort to enhance combustion at lighter loads. A common approach to achieve these high levels of internal EGR is through negative valve overlap (NVO) [51 53]. By advancing the exhaust valve closing (EVC) and delaying the intake valve opening (IVO) a substantial fraction of hot exhaust gas can be retained in-cylinder. Ignition may be further enhanced through pilot fuel injection during the NVO recompression [51, 52, 54, 55]. Exothermic reaction of pilot fuel during NVO increases charge temperature, and could produce reformed fuel species that may affect main combustion phasing [55, 56]. An advantage of this strategy is the potential for controlling main combustion phasing by varying NVO parameters including the amount of EGR, the NVO injection timing, and the NVO/main fuel-injection split. While performance of low-load NVO operation has been demonstrated [51,52,57 59], understanding of key processes can still be improved. Specifically, high residual gas fraction may result in temperature and compositional stratification near top dead center (TDC) that could further impact combustion. Currently, little is known of the degree of stratification that persists throughout compression for HCCI with NVO, motivating the development of in-situ visualization techniques to study this complicated process. High-Load HCCI To overcome the high pressure rise-rates (PRR) associated with high-load operation charge stratification has been investigated as it promotes sequential auto-ignition of the in-cylinder charge [60 63]. Here, both induced and naturally occurring variations in charge conditions (temperature and composition) can cause preferential ignition in localized regions followed sequentially by remaining portions of the charge. Past studies have shown that fuel stratification is only effective in altering the PRR when operating with fuels that exhibit two-stage ignition (low temperature chemistry) [64]. This indicates that thermal stratification alone will dominate gasoline high-load

26 6 CHAPTER 1. INTRODUCTION HCCI combustion. Despite the importance of thermal stratification (TS), little quantitative data is available regarding the distribution and evolution of TS. This provides further motivation for the development of high-fidelity, single-shot imaging diagnostics capable of resolving the small temperature fluctuations relevant for HCCI. Based on the challenging environment and the high measurement precision required for both low- and high-load HCCI measurements, tracer-based PLIF techniques have been selected. 1.2 Overview of Dissertation The focus of this dissertation is the development and optimization of tracer-based PLIF imaging diagnostics for specific application in HCCI engines. Chapter 2 provides the theoretical derivation of two PLIF diagnostic variations, a single excitation (single-line) and dual-excitation (two-line) strategy. Optimization efforts to maximize diagnostic performance for the HCCI engine domain are also highlighted. Chapter 3 focuses on engine-based measurements of tracer fluorescence quantum yield (FQY), a critical spectroscopic parameter for quantitative PLIF measurements. Chapter 4 investigates the fundamental fluorescence saturation behavior of common organic tracers and provides limitations on pump laser energies to ensure linear fluorescence behavior. Application of the two-line PLIF diagnostic for simultaneous measurement of temperature and composition in an HCCI engine with negative valve overlap (NVO) is discussed in Chapter 5. Topics include the validation of the technique in a motored engine, temperature and fuel concentration measurements during the NVO recompression and re-expansion, and characterization of the residual gas mixing during main compression. Chapter 6 concentrates on single-line PLIF measurements of thermal stratification (TS) in a low-residual, high-load HCCI engine. Studies include characterization of TS development under both motored and fired operation, and correlation between the temperature distribution and early chemical reaction.

27 1.2. OVERVIEW OF DISSERTATION 7 Lastly, Chapter 7 summarizes the overall development and application results and proposes future directions in terms of diagnostic improvements and additional engine studies.

28 Chapter 2 PLIF Technique and Optimization Development of trace-based PLIF diagnostics includes the selection of the excitation/collection strategy, tracer species, excitation wavelengths, and equipment. Potential diagnostic variations include single tracer with dual-wavelength excitation, dual tracer with single excitation and dual collection bands, and single tracer single excitation with dual collection bands. A variety of these strategies have been successfully demonstrated [26, 36, 44, 65 71]. Single-line excitation of acetone, 3-pentanone and toluene has been frequently used for fuel mole fraction or equivalence ratio measurements [36, 65, 66]. Similarly, Thurber et al. [67] utilized 248 nm excitation of acetone to measure temperature of mixing jets for isobaric, homogeneously seeded conditions. Kakuho et al. [68] performed temperature measurements, using 266 nm excitation of a combination of 3-pentanone and triethylamine (TEA) followed by dual-band collection of the spectrally separated fluorescence, to investigate the correlation between temperature and ignition time of HCCI combustion. Luong et al [69] utilized 266 nm excitation of toluene with dual-band collection to infer temperature based on the fluorescence emission spectral shift with temperature. Simultaneous measurements of temperature and fuel concentration were performed by Einecke et al. [26], through 248 nm and 308 nm excitation of 3-pentanone. Rothamer et al. [44, 70] later refined this two-line technique through optimization of excitation wavelengths at 277 nm and 308 nm. Additional examples of tracer-based PLIF techniques are available and are 8

29 2.1. PLIF THEORY 9 well-summarized by Schulz et al. [72]. Criteria for technique selection typically includes consideration of the application requirements, equipment, and availability of tracer photophysical data. The current development targets HCCI engines ranging from low-load HCCI with NVO, to highload HCCI with preheat. Based on these applications, a single-line temperature measurement has been selected for specific use in high-load HCCI engines where mixtures are expected to be compositionally homogeneous. To address the high residual fraction present in low-load HCCI operation with NVO, a two-line technique has been selected for simultaneous measurement of temperature and EGR mole fraction. The following sections discuss the development of these techniques, with emphasis on tracer and wavelength selection for optimized diagnostic performance. 2.1 PLIF Theory All PLIF diagnostics in this study are based on interpretation of fluorescence signal from various excitation wavelengths. The governing equation of fluorescence signal in the linear (weak excitation) regime is given by where S f S f = E [ ] hcν dv xtr P c σ (λ, T ) φ (λ, T, P, x) Ω kt 4π η coll (2.1) is the number of photons incident [photons/pixel], E is the local laser fluence [J/cm 2 ], h is Planck s constant [J s], c is the speed of light in a vacuum [cm/s], ν is the excitation laser frequency [cm 1 ], dv c is the probed volume [cm 3 ], x tr is the tracer mole fraction, P is the total pressure [MPa], k is the Boltzmann constant [J/K], T is the local temperature [K], σ is the absorption cross-section [cm 2 ], φ is the fluorescence quantum yield (FQY), Ω is the solid angle of collection, and η coll is the collection efficiency of the imaging optics (including transmission and collection efficiencies). As indicated in Equation 2.1, fluorescence is dependent on temperature, pressure, and composition through the absorption cross-section, fluorescence quantum yield and tracer number density (n tr = x tr P/kt). These dependencies permit the

30 10 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION development of various PLIF diagnostic strategies for measurement of temperature and composition. The accuracy of any technique depends on knowledge of the photophysical parameters (absorption cross-section, and FQY) over the range of desired experimental conditions, and will be discussed further in Chapter 3. Quantitative measurements also require that the laser energy (E) and collection terms (dv c, Ω/4π, η coll ) either be known or correctable. As these parameters are often difficult to measure accurately, suitable calibration schemes must be incorporated. The following sections highlight the development of a single-line, and a two-line PLIF technique, optimized for application in IC engines. While these diagnostics are presented in the context of engine research, they apply equally as well to other systems Single-Line PLIF As the name implies, the single-line excitation strategy derives temperature from the fluorescence signal of a single excitation wavelength. Application of this technique is restricted to conditions of uniform pressure and tracer mole fraction, as variations in fluorescence signal due to temperature and tracer number density are indistinguishable. The theoretical representation of the single-line scheme is shown in Equation 2.2. S f S cal f = E P T cal E cal P cal T σ(λ, T ) φ(λ, T, P, x i ) σ cal (λ, T ) φ cal (λ, T, P, x i ) (2.2) This relation was derived by normalizing the linear LIF relation, Equation 2.1, with a calibration signal (indicated with the cal superscript) acquired under homogeneous conditions of known temperature, pressure and tracer mole fraction. The calibration, or flat-field image, eliminates the need to directly measure the collection terms (dv c, Ω/4π, η coll ). In addition, the calibration measurements correct for spatial non-uniformities in laser sheet energy distribution, assuming minimal shot-to-shot variations in laser profile (sheet cross-section) and mean energy during the time between acquisition of calibration and data images. For engine experiments, the

31 2.1. PLIF THEORY 11 calibration images are typically acquired during motored operation at a time near bottom dead center (BDC) where the in-cylinder conditions are known and the mixture is homogeneous. The pressure term, P, in Equation 2.2 are assumed to be known based on single-point in-cylinder pressure measurements recorded during each cycle. Equation 2.2 can be further simplified by combining temperature-dependent terms and defining the single-line photophysical parameter, P P s, as shown in Equation 2.3. Substituting this relation into Equation 2.2 results in the principal equation for single-line temperature calibration, Equation 2.4. Performing the calibration correction in this fashion results in a measure of P P s, which itself is a function of temperature, pressure, and to a lesser extent species mole fraction. P P s is calculated as a function of these variables through knowledge of the absorption cross-section and FQY (as described in Chapter 3), and temperature is determined by iteratively solving Equation 2.4. P P s (T, P, x i ) = σ(λ, T )φ(λ, T, P, x i) T (2.3) P P s (T, P, x i ) = S f E cal P cal Sf cal E P P P s cal (2.4) Two-Line PLIF The two-line strategy derives temperature from the ratio of fluorescence signal resulting from two excitation wavelengths. This ratio eliminates the dependence on the tracer mole fraction, which increases the versatility of the technique and permits application in systems with spatially varying composition. This technique is particularly useful when investigating engine strategies utilizing significant residual gas fraction such as HCCI with NVO. The signal ratio is given by Equation 2.5, where subscripts 1 and 2 correspond to distinct excitation wavelengths λ 1 and λ 2. As with the single-line technique, the signal ratio is normalized by a calibration ratio acquired under known homogeneous conditions (T,P,x i ) to account for collection

32 12 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION terms and the laser energy distribution. S f2 /S f1 Sf2 cal/scal f1 = E 2 E1 cal E 1 E2 cal [ ] [ σ2 (λ 2, T )φ 2 (λ 2, T, P, x i ) σ cal 1 (λ 1, T )φ cal σ 1 (λ 1, T )φ 1 (λ 1, T, P, x i ) σ2 cal (λ 2, T )φ cal ] 1 (λ 1, T, P, x i ) 2 (λ 2, T, P, x i ) (2.5) Equation 2.5 is simplified by defining the ratio of photophysical parameters, R pp = σ 2φ 2 σ 1 φ 1, as shown in Equation 2.6. Performing the two-line calibration in this fashion results in a measurement of R P P which is itself a function of temperature, pressure and to a lesser extend composition. Temperature is determined by calculating R P P as a function of these variables through knowledge of the photophysical parameters (Chapter 3), and iteratively solving Equation 2.6. R pp (T, P, x i ) = σ 2φ 2 σ 1 φ 1 = S f2 S1 cal S f1 S2 cal R cal P P (2.6) The use of two excitation wavelengths allows temperature and composition to be measured simultaneously. Local composition is determined using the signal from either of the two excitation wavelengths and the measured temperature. The composition quantity being measured depends on where the tracer is seeded and how the data is processed. By seeding the fuel with tracer, measurement of the local fuel mole fraction (or concentration) is possible. Similarly, seeding of the intake air permits measurement of air mole fraction. If we further assume that in-cylinder contents consist only of intake air, EGR, and fuel, negative imaging of the EGR is possible by seeding both the fuel and air. Here, air-plus-fuel mole fraction is measured directly and the EGR mole fraction is determine using Equation 2.7. x EGR = 1 (x A + x f ) (2.7) This approach assumes that all tracer in the air and fuel is consumed during combustion and is not present in the EGR. A similar technique was first applied by Deshamps et al. [40] to estimate average EGR distributions, and was termed negative- PLIF due to the indirect method of measurement. Rothamer et al. [44] later refined this technique for quantitative simultaneous EGR and temperature measurements in HCCI engines. The same designation will be used throughout this work, but

33 2.1. PLIF THEORY 13 abbreviated to N-PLIF. Both direct fuel composition and N-PLIF measurements of EGR results are discussed in Chapter 5. To calibrate mole fraction measurements, the fluorescence signal for a given excitation wavelength is normalized by the corresponding calibration image. Solving this ratio for x tr results in Equation 2.8, where T is the measured temperature determined above. Assuming laser energy terms cancel as previously stated, all other variables in Equation 2.8 are known and the tracer mole fraction can be calculated directly. For fuel-seeded measurements, the tracer and fuel mole fraction are related by Equation 2.7, where x s is the mole fraction of tracer seeded into the liquid fuel. For N-PLIF measurements of EGR mole fraction, the air-plus-fuel mole fraction is first determined using Equation 2.10, where in this case x s is the tracer mole fraction seeded into the air and fuel, and x cal A = 1 for motored calibration images with no fuel present. Finally, EGR mole fraction is calculated using the characteristic N-PLIF relation, Equation 2.7. For the low-load NVO experiments considered in Chapter 5 only the air was seeded with tracer. Because the fuel constitutes an insignificant fraction of the total charge, seeding only the air does not introduce significant error in the EGR mole fraction measurements. x tr x cal tr = Ecal E P cal P T σ cal (λ, T ) T cal σ (λ, T ) φ cal (λ, T, P, x) φ (λ, T, P, x) S f Sf cal (2.8) x f = x tr x s (2.9) x tr x cal tr = x s(x A + x F ) x s (x cal A ) = x A + x F (2.10) The preceding theoretical development outlines the general framework for the two diagnostic variations. However, a number of additional selections must still be made including the tracer species, and the excitation wavelengths. The following sections detail these selections with particular emphasis on optimizing diagnostic performance for application in HCCI engines.

34 14 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION 2.2 Tracer Selection When selecting suitable tracer molecules, a number of factors must be considered including: (1) absorption spectrum optically accessible with high-power lasers, (2) overall fluorescence signal and sensitivity to temperature, (3) fluorescence sensitivity to oxygen quenching, (4) availability of photophysical data, and (5) evaporation characteristics (for fuel tracing). The relative importance of these criterion will vary somewhat depending on the desired measurement quantity (i.e. temperature, EGR mole fraction, fuel mole fraction). Ideally a tracer should be optically accessible by a high-power laser source such as an excimer or Nd:YAG laser which eliminates the need for more complex tunable laser sources. The tracer should also produce high fluorescence signal levels (due to a combination of high FQY and absorption) at reasonable seed concentrations, as well as exhibit good fluorescence sensitivity to temperature. These two factors are particularly important when single-shot measurements are desired, as the combination of these factors is what ultimately determines the diagnostic performance. Fluorescence sensitivity to oxygen should ideally be minimal as this tends to reduce the maximum signal levels and complicates the signal interpretation for highly stratified conditions. Availability of photophysical data is critical as this is required for signal interpretation, and limits the accuracy of the processed results. Finally, for cases when fuel tracing is of interest the tracer should exhibit evaporation characteristics similar to the base fuel to avoid distillation effects. This is often thought of in terms of matching the boiling point of the tracer and fuel, although more recent studies have indicated that this is not an exact criterion [73]. Potential fluorescent tracers range in size from small atomic and diatomic species (e.g. iodine, NO, SO 2, NO 2 ), up to larger hydrocarbons (e.g. aldehydes, aromatics, amines and ketones). The class of large hydrocarbon tracer molecules fulfill a majority of the selection criterion above, and will be considered here exclusively. Comprehensive discussions of potential tracers are included in [72, 94]. Common organic tracers used in engine research include acetone, 3-pentanone, biacetyl and toluene. Other tracer molecules such as acetaldehyde, formaldehyde, triethylamine

35 2.2. TRACER SELECTION 15 (TEA), fluorobenzene, and naphthalene have been used in previous studies but are not considered here due to either their toxicity or a lack of sufficient photophysical data. These larger molecules typically have broad-band absorption spectra due to the high density of energy states, permitting the use of high-power lasers. The potential excitation wavelengths include 248 nm (KrF Excimer), 266 nm (4th harmonic of Nd:YAG), 308 nm (XeCl Excimer), 351 nm (XeF Excimer), and 355 nm (3rd harmonic of Nd:YAG). Additional wavelengths can be derived through Raman shifting, including 277 nm (248 nm 1st Stokes in H 2 ), and 289 (266 nm 1st stokes in D 2 ). Toluene (C 6 H 5 CH 3 ) is a single-ring aromatic that is a component of commercial gasoline fuels at the percent level. The S 0 -S 1 (π, π ) absorption feature extends from 240 nm to 270 nm permitting excitation at 248 nm and 266 nm. The resulting fluorescence emission extends from roughly 260 nm to 400 nm at room temperature, and redshifts and widens as temperature is increased [74,75]. Toluene photophysics have been investigated at either high temperature or high pressures by Koban et al. [74] and Koch [75]. However, additional data at simultaneously high temperature and pressure is needed to better quantify the regime of engine conditions. In general, the toluene FQY is highly sensitive to temperature, decreasing by approximately 3 orders of magnitude or more from 300 K to 900 K [75]. As a result, toluene is often utilized for PLIF measurements of temperature [29, 69, 71], typically in pure N 2 environments. Like most aromatics, the fluorescence signal decreases dramatically in the presence of oxygen due to collisional quenching. The oxygen quenching not only reduces diagnostic performance through reduced signal-to-noise, but also complicates the signal interpretation. In general, toluene is a less suitable tracer for engine studies with high EGR, where the oxygen concentration can vary dramatically across the field of view. While such measurements are possible in theory, the correction for oxygen quenching effects at high temperatures and pressures is an unnecessary complexity when considering other candidate tracers. Toluene is better suited for PLIF measurements of temperature for a homogeneous composition field such as HCCI strategies with low residuals, and early fuel injection [71]. Toluene is also expected to be a good tracer for gasoline type fuels as the boiling point (B.P.)

36 16 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION of toluene (B.P.=383.8 K) is reasonably similar to that of iso-octane (B.P.=372.4 K) Biacetyl (CH 3 (CO) 2 CH 3 ) is a dione (diketone) that exhibits fluorescence behavior somewhat similar to acetone. Absorption for the first excited state S 0 -S 1 (n π ) and second excited state S 0 -S 2 (n π ) extend from 350 nm to 460 nm and 250 nm to 300 nm, respectively. Near-UV excitation to the second excited state is not applicable to fluorescence-based diagnostics as the second excited state is predissociative, resulting in negligible fluorescence. Both fluorescence emission ( nm) and longer-lived phosphorescence emission ( nm) are observed. Phosphorescence based diagnostics are not considered here based on the desired temporal resolution of the engine experiments. The fluorescence dependence on temperature, pressure and composition has been studied at some engine relevant conditions [76], however available data above 600 K is sparse. Biacetyl does exhibit oxygen quenching, but to a lesser extent than toluene [70, 76]. Rothamer et al. [77] concluded that biacetyl may be inferior to other available tracers based on comparisons of expected signal levels and general trends of temperature sensitivity. In addition, only two closely spaced high-energy excitation wavelengths (351 nm - XeF excimer, and 355 nm - 3rd harmonic Hd:YAG) are readily available, making development of a two-line diagnostic variation impractical. Based on these factors, biacetyl was not considered further for the optimization. Acetone (CH 3 COCH 3 ) and 3-pentanone (C 2 H 5 COC 2 H 5 ), are both aliphatic ketones with similar fluorescence behavior. Both exhibit S 0 -S 1 (n π ) absorption from approximately 225 nm to 320 nm, followed by fluorescence from 330 nm to 550 nm. The near-uv absorption spectrum permits excitation at a number of easily accessibly wavelengths including 248 nm, 266 nm, 277 nm, 289 nm, and 308 nm. The photophysics of these ketones have arguably received the most research interest over recent years, particularly at high temperatures and pressures [1,30,31,33 35,44,66,78 82]. Further discussion of these studies is provided in Chapter 3. Both tracers provide adequate fluorescence signal and temperature sensitivity, although typically less than toluene (except at high temperatures). One discernable quality between ketones is the evaporation characteristics. The boiling point of acetone (B.P.= 329 K) is much

37 2.3. WAVELENGTH OPTIMIZATION 17 lower than the typical gasoline reference fuel iso-octane (B.P.=373 K), indicating that acetone is not an appropriate selection to accurately track fuel evaporation. Conversely, 3-pentanone has a boiling point (B.P.=375 K) similar to iso-octane, making it better suited for accurate fuel mole fraction measurements (additional studies have shown that improvements over 3-pentanone can still be made ). This does not exclude acetone completely, as many intended measurement applications will utilize premixed seeding of pre-vaporized fuels. Given all these characteristics, acetone and 3-pentanone are the most common selection for quantitative tracerbased PLIF measurements. Based on the success of previous applications, and the availability of photophysical data, acetone and 3-pentanone will be further considered for both single-line and two-line diagnostic optimization below. Additionally, toluene will be included only in the single-line analysis for comparison. 2.3 Wavelength Optimization Having narrowed down the list of potential tracer molecules, wavelength selection can be performed to optimize overall diagnostic performance. The single-shot random uncertainty, also referred to as spatial precision, is of particular importance as this will dictate the minimum degree of stratification that can be resolved with a given diagnostic scheme. These random uncertainties can be thought of as measurement fluctuation over the imaged region, in the absence of any inhomogeneity. The measurement precision will be impacted by a number of factors including the signalto-noise ratio of the image, the shot-to-shot laser energy profile fluctuations, and the overall sensitivity of the fluorescence to temperature. In an effort to systematically account for all potential sources of random uncertainty a detailed uncertainty analysis was performed. This uncertainty analysis was first applied by Rothamer et al. [77] to optimize the wavelength selection for a two-line EGR/T diagnostic based on 3-pentanone. The same general uncertainty analysis is adopted in this study and is extended to the single-line PLIF scheme, and also to investigate different tracers for the two-line technique.

38 18 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION Uncertainty Analysis Theory The temperature and mole fraction measurement precision was determined by considering the propagation of uncertainties of individual measurement variables. Consider the general case in which an experimental result Y is a function of independent variable x n : Y = f(x 1, x 2,..., x n ) (2.11) Equation 2.11 essentially represents the calibration relation for the measured quantity such as Equation 2.4, 2.6 and 2.8. The uncertainty in measuring Y is determined by the root sum of squared uncertainties for the individual variables [83,84] as shown in Equation Y = ( ) 2 ( ) 2 ( ) 2 f f f x 1 + x x n (2.12) x 1 x 2 x n Here Y is the random uncertainty of the measured quantity Y, and x i is the uncertainty of each independent variable. For all subsequent calculations these uncertainties are assumed to be statistically well represented by the standard deviation. Temperature for the single-line technique is determined through measurement of the P P s, such that T = T (P P s ). Applying the error propagation relation, Equation 2.12, to this functionality results in Equation 2.13, where P P s is the random uncertainty of the P P s parameter. As indicated by Equation 2.4, P P s is itself a function of several measured quantities such that P P s = f(s f, Sf cal, E, P ). Applying the error propagation relation to P P s results in Equation No pressure terms are included in Equation 2.14 as the pressure is assumed to be constant across the cylinder and thus does not contribute to the spatial precision of a single-shot image. T = [ ( P ) 2 ( Ps P Ps P P s = S f + S f ( ) 2 T P P s (2.13) P P s S cal f ) 2 ( ) ] 2 1/2 P Sf cal Ps + E E (2.14)

39 2.3. WAVELENGTH OPTIMIZATION 19 The partial derivatives in Equation 2.14 are determined by differentiating Equation 2.4, and are given by: P P s S f = 1 S f P P s P P s S cal f = 1 Sf cal P P s P P s E = 1 E P P s (2.15) The collection of equations ( ) above constitutes the theoretical framework of the uncertainty analysis for the single-line PLIF technique. With these equations, the diagnostic performance can be assessed over a wide range of experimental conditions, as well as tracer species and excitation wavelengths. However, the uncertainties of individual variables ( E, S f ) still need to be determined. The random uncertainty in laser energy, E, is caused by shot-to-shot fluctuations in both total laser energy and profile distribution. They are included as a source of uncertainty because they are not being directly corrected for within the current measurement scheme. While it is possible to correct for these laser profile fluctuations on a shot-to-shot basis, the added complexity of a beam profiling system (beam sampler and camera) and the overall difficulty of an accurate correction make it prohibitive unless absolutely necessary. Many potential laser sources exhibit energy fluctuations on the order of ±1% and therefore have less impact on the overall uncertainty in comparison with the image shot-noise. Uncertainty estimates of the measured fluorescence signal, S f, are determined by modeling the noise characteristics of an intensified CCD (ICCD) camera. Previous studies [85] have shown that the signal-to-noise ratio of an ICCD camera is well represented by: SNR ICCD = fracη P C S f g tot sqrtη P C S f g 2 totf MCP + N 2 CCD (2.16) where SNR is the image signal-to-noise ratio, η P C photocathode quantum efficiency, S f is the number of photons incident at the photocathode [photons/pixel], g tot is the total photoelectron gain through the intensifier,f MCP is the characteristic noise factor of the micro-channel plate (MCP), and N CCD is the CCD read and dark noise [electrons/pixel].

40 20 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION The fluorescence signal per pixel used in Equation 2.16 is determined from the linear LIF Equation 2.1. The LIF relation requires estimation of a number of physical parameters such as the laser fluence, probed volume, and solid angle of collection. The laser fluence, E, is re-written in terms laser energy and beam dimensions as E = Ep where E n p l o w p is the total laser energy, n p is the number of pixels traversing the imaged laser sheet profile, l o is the pixel size in the object plane, and w is the sheet thickness. The probed volume is estimated based on square of the pixel size l o and the laser sheet thickness. Finally the solid angle of collection is estimated based on imaging parameters, as shown in Equation 2.17 where f/# is the imaging f-number and M is the magnification. Substituting all these relations into the linear LIF equation and simplifying results in Equation 2.18 [70]. In summary, the single-line measurement uncertainty (precision) can be estimated by solving Equations Ω = π 4 1 f/# 2 M 2 (1 + M) 2 (2.17) S f = E p hcν l o n p [ xtr P kt ] σ(λ, T )φ(λ, T, P, x i ) 1 ( ) 2 1 M (2.18) 16 f/# (1 + M) 2 Derivation of the uncertainty analysis for the two-line PLIF diagnostic variation follows a similar path to that shown for the single-line technique above. An indepth derivation has been previously presented [77] and will not be covered here further. The general equations for the two-line uncertainty analysis are presented in Appendix B for reference. Although the general equations differ, analysis of both diagnostic variations utilize the same estimates of fluorescence signal, laser stability and general equipment parameters Uncertainty Analysis Parameters Accurate prediction of fluorescence signal and resulting SNR requires the estimation of physical parameters of the diagnostic components. A list of excitation laser

41 2.3. WAVELENGTH OPTIMIZATION 21 wavelengths under consideration is shown in Table 2.1. The corresponding total laser energy and profile stability used for all simulations is also provided. These wavelengths were selected based on the absorption bands of the potential tracer molecules, with an emphasis on turn-key high-energy laser sources. This includes both fixed laser wavelengths (248 nm, 266 nm and 308 nm), and wavelengths derived through Raman shifting (277 nm and 289 nm). Additional wavelengths of 332 nm (XeCl 1st Stokes in N 2 ) and 351 nm (XeF) have been considered in past optimization studies [77], but were excluded for the current work. Attempts to generate sufficient 332 nm energy proved difficult with the current XeCl excimer (stable resonator). This is unfortunate as simulations by Rothamer et al. [70] indicate that the 332 nm / 266 nm combination should provide good diagnostic performance. The XeF excimer output at 351 nm was also excluded as this wavelength resides in the far wings of most absorption features, and the FQY model predictions at this wavelength are are not expected to be accurate given the limited experimental data. Laser energies listed in Table 2.1 represent achievable pump energies and were selected to be within the bounds of the linear excitation regime as discussed in Chapter 4. The 266 nm excitation is somewhat high, but linear excitation can be ensured by increasing the laser sheet thickness. This will reduce the overall resolution of the system but will not affect the overall diagnostic performance. The energy profile fluctuations in Table 2.1 are based on actual stability measurements where available. Profile stability of the 289 nm wavelength was extrapolated from the 266 nm data based on understanding of the impact of Raman shifting process. Uncertainty simulations also require photophysical data to both estimate the fluorescence signal and resulting camera SNR, and also to calculate the partial T derivatives (e.g. P P s ). Absorption cross-section data for all tracers was taken from Koch et al. [1], and is considered to be well characterized. FQY data is generated from available FQY models, provided by Koch et al. [75] for toluene, Thurber et al. [86] for acetone (except at 248 nm, where the Braeuer et al. [87] updated model was used), and Rothamer et al. [70] for 3-pentanone. These models exist with varying degrees of refinement and accuracy. The 3-pentanone model has received the most attention and refinement and is expected to be accurate over a wider range

42 22 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION of conditions. While perhaps less accurate, other models are still very useful in studying the measurement precision and are able to capture the general uncertainty trends. Remaining physical parameters of the optical and imaging systems required for simulation are listed in Table 2.2. The parameters are representative of realistic equipment specifications and have been selected to optimize total signal intensity. The parameters are identical to those used in previous work by Rothamer et al. [70], as the diagnostic components are the same. Some of these parameters such as the optical transmission and the solid angle of collection can be difficult to estimate, and result in some errors in the uncertainty analysis estimates. However, this will not influence comparison of the relative performance of diagnostic variations. Table 2.1: Excitation wavelengths, energy and energy stability inputs used for temperature and EGR precision calculations λ [nm] Laser Source Energy [mj] Energy Stability [±%] 248 KrF excimer Nd:YAG 4th harmonic KrF 1st Stokes (H 2 ) Nd:YAG 1st Stokes (D 2 ) XeCl excimer Single-Line Analysis Simulations of temperature uncertainty for the single-line technique were performed over a continuous range of temperature ( K) and pressure (1-50 bar) for constant tracer mole fraction. Simulations were performed with 3-pentanone, acetone and toluene for all potential excitation wavelengths listed in Table 2.1. Results are plotted as contoured color images to permit easy comparison of performance over the wide range of conditions. In general, these results are not limited to IC engine conditions alone and correspond to any experimental application with constant

43 2.3. WAVELENGTH OPTIMIZATION 23 Table 2.2: Input parameters used for temperature and EGR precision calculations, same as Camera Parameters Symbol Value Camera type Dual-frame ICCD CCD array size 1024 x 1024 Pixel binning (on chip) 8 x 8 Binned pixel in linear extent n p 128 Binned pixel size in object plane l o 0.5 mm Read Noise N ICCD 25 e per pixel Photocathode quantum efficiency eta P C 0.20 Total photoelectron gain g tot 13 MCP noise factor F MCP 2.56 Optics Parameters Symbol Value Lens f-number f/# 1.4 Image magnification M 0.21 Optical transmission η opt 0.4 Laser Parameters Symbol Value Sheet width 50 mm Laser energy E P See table 2.1 Laser profile stability E See table 2.1 Laser sheet thickness w 0.5 mm Mixture Parameters Symbol Value Tracer mole fraction in intake air x cal tr 1500 ppm EGR mole fraction x EGR 50% tracer mole fraction. Dashed curves in each plot represent bounding engine compression curves, with the top curve corresponding to naturally aspirated HCCI with high preheat, and the lower curve corresponding to boosted HCCI with minimal preheat. Temperature precision results for 3-pentanone are presented in Figure 2.1. All plots depict a large dynamic range of diagnostic performance, ranging anywhere

44 24 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION from 3 K to 40 K. The thermodynamic conditions of best performance vary for each wavelength and is related to the relative changes in total fluorescence signal and temperature sensitivity ( T/ P P S ) with temperature and pressure. The diagnostic performance improves with increasing pressure due to increased tracer number density (constant mole fraction) and hence higher fluorescence signals. The trends of diagnostic performance with increasing temperature varies for each excitation wavelength, as there is typically a tradeoff between low and high temperature performance. The shortest wavelength, 248 nm (Figure 2.1a), provides the best performance at low temperature due to high signal levels and temperature sensitivity. Conversely, longer wavelengths such as 289 and 308 nm (Figure 2.1d and 2.1e) perform poorly at low temperatures, but perform well at high temperatures. This variation in performance demonstrates the utility of the wavelength optimization, permitting thoughtful selection of excitation wavelength to maximize performance over a specific range of conditions. Close investigation of the 3-pentanone performance plots shown in Figure 2.1 indicates that excitation at 277 nm (Figure 2.1c) provides the best overall performance within the window of engine conditions. This wavelength is a good combination of low and high temperature performance and is preferred over 266 nm excitation given the lower laser profile energy fluctuations. In addition, the longer pulse duration of the excimer versus the Nd:YAG (25 ns and 6-8 ns respectively) reduces the potential for laser induced damage of windows and fluorescence saturation. Similar performance plots for acetone are presented in Figure 2.2. The overall acetone performance at all excitation wavelengths is lower than that of 3-pentanone. This can be attributed to the lower overall fluorescence signal for acetone resulting from lower absorption cross-section and FQY in comparison with 3-pentanone [1,75]. This was demonstrated by Koch et al. [75] who measured the absolute FQY for both acetone and 3-pentanone, and found that the FQY of 3-pentanone is approximately two times higher than for acetone (for 266 nm excitation at 300 K and 1 bar). At these conditions the absorption cross-section is approximately equivalent. Overall, this results in a fluorescence signal that is a factor of two higher for 3-pentanone, and thus a diagnostic performance that is improved by a factor of 2 assuming

45 2.3. WAVELENGTH OPTIMIZATION 25 Temperature Precision [K] Temperature [K] Engine Range Pressure [bar] (a) 248 nm (b) 266 nm Temperature [K] Engine Range Pressure [bar] Temperature [K] Engine Range Pressure [bar] (c) 277 nm Temperature [K] Engine Range Pressure [bar] (d) 289 nm Temperature [K] Engine Range Pressure [bar] (e) 308 nm Figure 2.1: 3-Pentanone single-line PLIF measurement uncertainty estimates of temperature for potential excitation wavelength pairs. All precision is quoted based on ±1 standard deviation.

46 26 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION shot-noise limited behavior. In addition, the temperature sensitivity of acetone is generally lower than 3-pentanone for moderate to high temperatures. Based on the performance data in Figure 2.2, 277 nm excitation again provides the best overall performance followed by 248 nm and 266 nm. The overlap in optimal excitation wavelength for acetone and 3-pentanone is not unexpected given that they both exhibit generally similar photophysical behavior. There is however, a large difference in absolute performance between tracers, and as a result, subsequent single-line engine measurements will primarily utilize 3-pentanone as the ketone tracer. For comparison, single-line toluene-based temperature uncertainty simulations were performed for 248 and 266 nm excitation. Additional wavelengths were omitted due to little or no absorption for wavelengths longer than approximately 275 nm (at low to moderate temperature). Several simulation parameters were adjusted for the toluene study in an effort to make results more realistic. First, the toluene mole fraction was reduced to 150 ppm to provide equivalent laser attenuation to the 3-pentanone studies (matched at 1000 K and 20 bar). Second, the total laser energy was reduced to 30 mj for both wavelengths to avoid fluorescence saturation (as described in Chapter 4). Third, the effective camera gain factor was reduced to avoid detector saturation at the lower temperature conditions. Finally, the bath gas was changed from air to N 2 due to the significant impact of oxygen quenching for toluene. The single-line toluene results are given in Figure 2.3. Excitation at 248 nm provides good performance at low temperatures, but quickly degrades with increasing temperature due to decreasing signal levels. This behavior highlights the performance tradeoff between temperature sensitivity and fluorescence signal. Toluene is often selected due to its high temperature sensitivity, however this can be detrimental under certain conditions due to the accompanying decrease in signal levels. Toluene excitation at 248 nm is a good example of this, where the temperature sensitivity is so high the fluorescence signal becomes the limiting factor as temperature is increased. Temperature precision at 266 nm is quite good over the entire test

47 WAVELENGTH OPTIMIZATION 27 Temperature Precision [K] Temperature [K] Temperature [K] Engine Range Engine Range Pressure [bar] (a) 248 nm Pressure [bar] (b) 266 nm Temperature [K] Engine Range Pressure [bar] (c) 277 nm Temperature [K] Engine Range Pressure [bar] (d) 289 nm Temperature [K] Engine Range Pressure [bar] (e) 308 nm Figure 2.2: Acetone single-line PLIF measurement uncertainty estimates of temperature for potential excitation wavelength pairs. All precision is quoted based on ±1 standard deviation.

48 28 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION matrix, and specifically within the engine bounds. While the temperature sensitivity for 266 nm is somewhat smaller than for 248 nm, the signal levels are generally higher, aided by the absorption cross-section increasing with temperature. Overall the estimated temperature uncertainty for toluene at 266 nm is superior to both acetone and 3-pentanone, however this is not the case when oxygen is added to the bath gas. Simulations with air (not shown) exhibit a dramatic drop off in performance, well below the performance of 3-pentanone. This shift in performance is a result of the precipitous drop in toluene FQY due to oxygen quenching. Koban et. al [88] measured a decrease in FQY of over two orders of magnitude (at 300 K and 1 bar) when switching the bath gas from nitrogen to air. Because the intended diagnostic application include both motored engine experiments with nitrogen and fired experiments in air, toluene was not considered for the single-line PLIF scheme. Based on the preceding uncertainty analysis, it is concluded that 3-pentanone excitation at 277 nm is the optimal choice for the single-line scheme in terms of both temperature precision and application versatility. Subsequently, this technique was used to investigate the thermal stratification development in a high-load HCCI engine under both motored and fired operation. Diagnostic validation and experimental results are presented in Chapter 6. Additional validation experiments were also performed with acetone to verify the validity of the uncertainty analysis, and to confirm the predictions of superior performance with 3-pentanone Two-line Analysis An uncertainty analysis of the two-line technique based on Equations B.5 and B.2 was performed to minimize measurement uncertainty for both temperature and EGR mole fraction measurements. This follows closely with the optimization performed by Rothamer et al. [70], but is extended to include both 3-pentanone and acetone. The assumed temperature and pressure matrix, excitation laser specifications (Table 2.1), and system parameters (Table 2.2) are equivalent to those used for the single-line analysis. All wavelength combinations considered utilize 308 nm as the longer wavelength. It has been found that combinations where both wavelengths are

49 2.3. WAVELENGTH OPTIMIZATION 29 Temperature Precision [K] Temperature [K] Engine Range Pressure [bar] (a) 248 nm Temperature [K] Engine Range Pressure [bar] (b) 266 nm 5 Figure 2.3: Toluene single-line PLIF measurement uncertainty estimates of temperature for potential wavelength pairs. Simulations are for an N 2 intake stream. All precision is quoted based on ± 1 standard deviation. below 308 nm results in a signal ratio that is not single-valued and complicates the temperature calibration. Interpretation of the two-line results is somewhat more complicated than for the single-line. Temperature uncertainty is dictated by the fluorescence signal from each wavelength, and the sensitivity of the signal ratio to temperature. Mole fraction uncertainty is dependent on the fluorescence signal from a single wavelength, the temperature sensitivity of fluorescence at that wavelength, and the accompanying temperature uncertainty. Representative uncertainty contour maps of both temperature and EGR mole fraction for 3-pentanone are presented in Figure 2.4. Again these results are very similar to those of Rothamer et al. [77], with slight differences arising due to small differences in simulation parameters. Considering temperature precision first, the 248/308 nm pair provides the best performance at low temperatures but quickly degrades at higher temperatures due to low signal levels at 248 nm and low ratio sensitivity. The 277/308 nm pair provides the best overall performance across the engine range, with the 266/308 nm performing only slightly below. The EGR mole

50 30 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION fraction uncertainties follow a similar trend. The 248/308 nm results in poor performance at higher temperatures, while the low temperature performance is comparable to the other combinations. The 277/308 nm pair again provides slightly better EGR mole fraction performance than 266/308 nm. Based on the combination of these results the 277/308 nm pair is found to be the optimal wavelength combination for 3-pentanone. This wavelength combination has been applied successfully in a number of studies [44, 77, 89] confirming the uncertainty analysis results. Analogous uncertainty simulations were performed with acetone as the fluorescence tracer to assess any performance benefits. Based on the temperature uncertainty results shown in Figure 2.5, the 248/308 nm combination provides the lowest temperature uncertainty, followed by 277/308 nm. However, the absolute temperature uncertainty for all acetone wavelength pairs are on average higher than for 3-pentanone. There may be some marginal benefits with acetone at high temperatures, but this is unclear given the inaccuracies of the FQY models. Trends in EGR mole fraction uncertainty, shown in Figure 2.5, are somewhat in contrast. Here the 277/308 nm combination provides the lowest EGR mole fraction uncertainty, while 248/308 nm is generally the highest. Overall, the EGR uncertainties achieved with acetone are lower than uncertainties for 3-pentanone and likely results from the lower temperature sensitivity of the acetone fluorescence. Because of the varying temperature and mole fraction trends for acetone, the optimal wavelength selection will depend on the relative importance of either measurements. The 277/308 nm combination affords a reasonable compromise. Based on the two-line uncertainty analysis results for all wavelengths and tracers, the 277/308 nm excitation of 3-pentanone provides the best overall performance and was used predominantly for the subsequent engine studies. Acetone was also investigated briefly for the same wavelength combination to validate the uncertainty results and to assess any high temperature benefits. It is interesting to note that the optimal 3-pentanone wavelength selections for the single-line and two-line overlap at 277 nm. This fortuitous conclusion means that single-line and two-line techniques can be performed within the same experimental setup, differing only in the data processing. This only improves the versatility of the overall diagnostic technique.

51 2.3. WAVELENGTH OPTIMIZATION 31 Temperature Precision [K] EGR Precision [mole fraction %] Temperature [K] Temperature [K] Engine Range Pressure [bar] (a) 248 nm / 308 nm Engine Range Pressure [bar] Temperature [K] Temperature [K] Engine Range Engine Range Pressure [bar] Pressure [bar] (b) 266 nm / 308 nm Temperature [K] Engine Range Pressure [bar] Temperature [K] (c) 277 nm / 308 nm Engine Range Pressure [bar] Figure 2.4: 3-Pentanone two-line PLIF measurement uncertainty estimates of temperature (left) and EGR mole fraction (right) for potential excitation wavelength pairs. All precision is quoted based on ±1 standard deviation.

52 CHAPTER 2. PLIF TECHNIQUE AND OPTIMIZATION Temperature Precision [K] EGR Precision [mole fraction %] Temperature [K] Temperature [K] Engine Range Pressure [bar] (a) 248 nm / 308 nm 3 3 Engine Range Pressure [bar] Temperature [K] Temperature [K] Engine Range Pressure [bar] 400 (b) 266 nm / 308 nm 5 3 Engine Range Pressure [bar] Temperature [K] Temperature [K] Engine Range Pressure [bar] (c) 277 nm / 308 nm 2 2 Engine Range Pressure [bar] 3 Figure 2.5: Acetone two-line PLIF measurement uncertainty estimates of temperature (left) and EGR mole fraction (right) for potential excitation wavelength pairs. All precision is quoted based on ±1 standard deviation.

53 Chapter 3 Tracer Photophysics Accurate knowledge of tracer photophysical parameters, namely the absorption cross-section and fluorescence quantum yield (FQY), is critical for diagnostic development and quantitative tracer-based measurements. Both parameters have been well studied for aliphatic ketones [1,30,31,33,80,86,87,90,91], but additional measurements are still required to broaden the range of experimental conditions particularly at simultaneous high temperatures and pressures. For the current work, a motored IC engine is employed as a test stand for measurements of FQY at engine relevant conditions. The near-uv absorption spectrum of the first excited singlet state of both acetone and 3-pentanone ranges from 225 to 320 nm, and is consistent with many cabonyl-containing compounds. Because of the high density of rovibronic transitions, the absorption feature is broadband (except at very low temperatures) and depends only on temperature and wavelength. Initial absorption measurements focused on room temperature conditions or below [92 96] and were later extended to elevated temperatures relevant for engine experiments [1, 30, 86]. Koch et al. [1] presents particularly useful absorption results in the form of gaussian fit coefficients for accurate representation of absorption of acetone and 3-pentanone as a function of wavelength and temperature. These fitting parameters were used for all data processing in this work, and are presented in Appendix A. For both ketones the absorption generally increases with temperature and the spectrum red-shifts to longer 33

54 34 CHAPTER 3. TRACER PHOTOPHYSICS wavelengths [1, 86]. The resulting fluorescence spectrum of acetone and 3-pentanone ranges from nm respectively. The excited state lifetime is dominated by fast intersystem crossing (S 1 T 1 ) to the first excited triplet, limiting the overall fluorescence quantum yield. The fluorescence has been shown to strongly depend on temperature, pressure and to a lesser extent composition. These functional dependencies of fluorescence are a result of the product of the absorption cross section and the fluorescence quantum yield (FQY). It is often easier to think of these quantities independently. In general, the FQY decreases with temperature (especially at shorter wavelengths), increases with pressure (up to a high-pressure limit), and decreases with decreasing excitation laser wavelength. These trends can be traced back to the non-radiative decay rate of the excited singlet which has been found to increase with excess vibrational energy in the excited state [75,78]. The FQY of ketones has been investigated by a number of researchers [30 32, 35, 86] and has mostly focused on FQY variation with independent changes in temperature or pressure (not simultaneous). More recent studies have expanded the measurement domain to include simultaneous high temperature and pressure conditions [33, 34, 87, 90]. Despite numerous FQY studies, the collection of data is still insufficient to accurately predict the ketone FQY behavior, especially at elevated temperatures and pressure relevant for engine experiments. This is especially problematic when investigating kinetically limited combustion systems such as HCCI, where measurement of temperature and composition distributions near TDC are critical. The following sections will include discussion about the measurement of FQY in a motored engine in an effort to characterize the FQY dependence with temperature for a range of relevant engine conditions. 3.1 FQY Measurements in a Motored IC Engine A single-cylinder optical IC engine can be utilized as a compression machine. It essentially represents a closed system reactor with time-varying temperature and

55 3.1. FQY MEASUREMENTS IN A MOTORED IC ENGINE 35 pressure. With sufficient optical access the IC engine can be used to study tracer photophysics at conditions relevant for the study of HCCI engine operation. Although the thermodynamic in-cylinder conditions are not as easily characterized as in a shock tube, rapid compression machine, or steady flow reactor, it is possible to use the engine facility as a test bench for a wide range of experiments including chemical kinetics [97 100] and photophysics [70, 80, 101]. Engine based measurements have the added benefit of fast data acquisition of up to 20 Hz (2400 RPM engine speed) allowing for repeated measurements and cycle averaging. In addition, the residence time at high temperature within the engine are typically much shorter than can be achieved in steady flow reactors and reduces the likelihood of pyrolysis and decomposition. This makes measurements at temperature extremes more feasible and allows direct characterization of engine conditions Measurement Technique In the current study, two different optical engines, both located at the Combustion Research Facility of Sandia National Laboratory, were used to acquire ketone FQY data. The specifics of each engine are described in detail in Chapters 5-6 and relevant engine parameters are shown in Table 5.1 and Table 6.1. Engine A, presented in Chapter 5, has a geometric compression ratio of 12:1 and imaging access limited to 24 btdc by the piston position. Engine B, presented in Chapter 6, has a higher geometric compression ratio of 14:1 and imaging access through TDC permitting data acquisition at higher temperatures and pressures. Intake pressure was held constant at 1 bar for all tests as subsequent engine studies are focused on naturally aspirated conditions. In each engine fluorescence images were acquired at various times during the compression stroke to access a range of pressures and temperatures. Similar measurements were repeated for a range of intake temperatures to determine the relative signal dependence on temperature for a given crank angle. In the absence of heat transfer variations, measurements at each crank angle correspond to a constant pressure. In reality, small pressure variations are seen near TDC as intake temperature is increased due to increasing heat transfer. However, these affected

56 36 CHAPTER 3. TRACER PHOTOPHYSICS image times reside in the high pressure limit [75] where little variation in FQY with pressure is observed. Engine measurements with constant intake temperature but varying intake pressure (not shown) confirm the minimal influence of pressure on FQY at crank angles near TDC. As a result, measurements at each crank angle are assumed to be at constant pressure, with all reported values corresponding to the average pressure. The relative data was converted to absolute FQY through acquisition of reference images at thermodynamic conditions where the FQY is already well characterized. These reference images were acquired near bottom dead center (BDC) where the pressure is effectively 1 bar, and the temperature can be well estimated [102]. The reference conditions were chosen to take advantage of previous absolute ketone FQY data acquired in a flowing cell for a range of temperatures and 1 bar pressure [31,35]. The calibration to absolute FQY is represented in Equation 3.1, where φ is the FQY, S is the measured fluorescence signal, P is the measured pressure, T is the estimated temperature, and σ is the absorption cross section at the corresponding wavelength and temperature. conditions and FQY. The ref subscript denotes the references images with known φ(t, P, x i ) = S P ref S ref P T σ ref (T ref ) φ ref (T ref, P ref, x i ) (3.1) T ref σ(t ) Equation 3.1 was derived by ratioing the fluorescence signal from the data and reference image given by the linear PLIF relation, Equation 2.1. The laser energy, probed volume, and collection efficiency were all constant between data and reference condition, and are eliminated by the ratio. In addition, because the mixture composition is constant throughout compression (in the absence of chemical reaction), a measurement of the exact tracer seeding level is not required and does not represent a source of error for the technique. This is contrary to flowing cell measurements where the tracer number density is a dominant source of error [75, 79]. Based on this, all terms on the right hand side of Equation 3.1 are known, measured or estimated, and thus the absolute FQY can be determined. The specific experimental setups are presented in more detail in Chapters 5-6,

57 3.1. FQY MEASUREMENTS IN A MOTORED IC ENGINE 37 and was nearly identical for both engines. The in-cylinder pressure was measured in each engine using a high-speed pressure sensor and recorded at 1/4 CA resolution and averaged for consecutive engine cycles. A premixed fueling system located upstream of the intake surge tank was utilized to ensure a truly homogeneous in-cylinder tracer distribution. All experiments were performed with both N 2 and air bath gases to study the impact of oxygen quenching. Intake temperatures ranging from C were used to study the FQY dependence on temperature as described above. The in-cylinder core gas temperatures during compression were determined by first calculating the in-cylinder temperatures at BDC. Temperatures at BDC for Engine A were estimated based on GT-Power engine simulations tuned to match the measured pressure trace. Temperature at BDC for Engine B were calculated using methods previously developed for that facility [102]. Finally, the in-cylinder core gas temperatures during compression were calculated assuming adiabatic compression of the measured pressure trace with variable specific heats recalculated at each time step. This calculation method indirectly accounts for heat transfer effects through the measured pressure trace, but assumes an adiabatic core region consistent with the interrogation region of the laser. The thermodynamic data used to calculate mixture gamma was taken from the thermodynamic database employed in the Lawrence Livermore National Laboratory (LLNL) iso-octane chemical kinetic mechanism [103]. Engine-based FQY measurements were performed for excitation wavelengths of 277 and 308 nm based on the detailed uncertainty analysis of Chapter 2. Laser excitation at 277 nm was generated by Raman shifting the output of a 248 nm KrF excimer in H 2 (1st Stokes), while the 308 nm excitation was provided by a XeCl excimer laser. The laser energy fluctuation was verified to be less than 1.5 % consistent with the constant laser energy assumption implicit in Equation 3.1. Pump laser energies were selected to be in the linear regime and are within the limits set in Chapter 4. In addition, all experiments have been repeated with laser energies decreased by as much as a factor of 4 with minimal variation in processed FQY results. The fluorescence signal was imaged using an intensified CCD camera (PIMAX2-1003) equipped with a 85 mm f/1.4 glass lens. No additional collection

58 38 CHAPTER 3. TRACER PHOTOPHYSICS filter was used as the glass lens sufficiently rejected all scattered laser light. The measured fluorescence signal is thus averaged over the entire fluorescence bandwidth, although no correction for the wavelength dependent camera quantum efficiency has been performed. Three separate images sets were required for data processing at each crank angle. First, background images were acquired in the motored engine with laser excitation on but without tracer seeding to correct for dark current accumulation and any fixed-pattern background signal. The tracer seeding system was then turned on, and allowed to reach steady-state before additional data acquisition. Next 50 reference (flat-field) images were acquired followed by 50 data images at the selected timing. The reference and data images were acquired with minimal delay between sets to further ensure minimal laser energy drift. All images and pressure measurements were averaged over the 50 cycles prior to subsequent data processing. After background correction, the data and reference images were corrected for laser attenuation along the propagation direction assuming a Beer s law dependence. For this correction the absorption cross-section was calculated at the corresponding core image temperature using the gaussian fit data in Appendix A. Finally a spatially averaged signal was calculated over a subregion of the data and reference images, and used as inputs in Equation 3.1. The FQY results were found to be insensitive to the location of the averaging subregion (tested regions near entrance, exit, center, right, left and total region), indicating that the influence of attenuation is small and that there was no preferential variation in results based on spatial location in-cylinder. Absolute uncertainty of these engine-based FQY measurements is estimated to be ±6% based on error propagation analysis. This uncertainty is dominated mostly by uncertainty in the in-cylinder temperature calculations required for data processing. The following sections outline the results obtained for 3-pentanone and acetone over the range of conditions tested.

59 PENTANONE PHOTOPHYSICS Pentanone Photophysics Representative 3-pentanone absolute FQY results from both engine A and B are shown in Figure 3.1a for a constant pressure of 12 bar in a nitrogen bath gas. Good agreement in the FQY measurements is observed for the two engines in the narrow region of temperature overlap. Measurement agreement between engines was typically within 5-7%. The consistency in the results from both engines is encouraging given the differences in facilities and demonstrates the robustness of the measurement technique. The multiple data points for Engine B represent experiments completed with varying laser intensities, slightly varying engine conditions, and spanning several months in time. The low scatter of these repeated measurements further demonstrates the consistency of the technique and results. The FQY measurements from Engine A are nearly identical to those previously used by Rothamer et al. [70] as targets for re-optimization of the 3-pentanone FQY model originally formulated by Koch et al. [75, 104]. The remaining discussion will focus on the higher temperature and pressure results from Engine B as these represent a larger deviation from the parameter space of other studies. Instead of using the FQY measurements as model optimization targets, simple quadratic fits have been generated at each image timing (pressure). This method is thought to be more accurate than model optimization, particularly when recent studies have demonstrated that current FQY model functionality cannot accurately predict trends over an extensive range of conditions and wavelengths [33, 59, 87]. While the FQY model can be optimized to match these wavelengths, pressures, and temperatures the accuracy at other conditions is sure to be sacrificed. However, given that the quadratic fits have no physical significance, extrapolation to conditions outside the fitting range was avoided. Quadratic fitting results are shown in Figure 3.1a for 12 bar total pressure, and in the compiled set of image timings in Figure 3.1b where each curve represents a different image timing and thus pressure. The expected trends of FQY with temperature can be seen in Figure 3.1b where the FQY decreases with increasing temperature. The increase in FQY with increasing excitation wavelength is also captured. Finally

60 40 CHAPTER 3. TRACER PHOTOPHYSICS the influence of total pressure can be seen for the lower temperature images and is indicated by the slight translation of curves to higher FQY as pressure is increased. This is mostly seen for the 5-12 bar image sets, as higher pressures reside in the high-pressure limit as indicated by [31]. Beyond this point there is good overlap between curves of increasing pressure, indicating that there is little influence of bath gas total pressure when above bar. Experiments with air bath gas exhibited nearly identical trends with temperature. In general, there was a slight decrease in the absolute FQY values consistent with the effect of oxygen quenching to reduce FQY. All fitting results for the 3-pentanone FQY data are presented in Appendix A and are accompanied by the range of applicable temperatures and pressures. Absolute FQY x nm P=12 bar 3 Pentanone N 2 Bath 308 nm Engine A Engine B Data Fit Temperature [K] (a) Absolute FQY x nm 308 nm 305 5bar 320 8bar bar bar bar bar bar bar Pentanone N Bath Temperature [K] (b) Figure 3.1: Absolute FQY of 3-pentanone in nitrogen at 277 nm and 308 nm, measured under motored engine conditions. Each curve represents experiments at a given image timing (pressure) for a range of intake temperatures. The absolute accuracy of the engine FQY measurements is dominated primarily by the accuracy of the in-cylinder temperature calculations that relies on the assumption of an adiabatic core region. Previous in-cylinder temperature measurements using TDL absorption sensors [ ] do provide some absolute temperature measurements but have not been refined to the extent that they can answer the question of adiabaticity in the engine core. Past imaging studies do indicate that thermal stratification does convect into the core due to convection and heat transfer

61 3.3. ACETONE PHOTOPHYSICS 41 in the near-wall region [71,108], thus indicating some degree of non-adiabatic behavior. However this non-adiabatic behavior will likely shift the absolute FQY curves with temperature, while maintaining the overall slope [71]. While this does result in some inaccuracy for absolute temperature measurements, the degree of stratification (related to the slope of the FQY curves) is the primary focus of the HCCI studies here and should be well preserved. The accuracy of these measurements can further be tested by comparing the motored engine FQY results with applicable FQY measurements made in a flowing test cell. Cheung et al. [90] have recently completed FQY measurements of 3-pentanone for excitation wavelengths of 248, 266, 277 and 308 nm from K and 0-20 bar. While these experimental conditions are on the low temperature extreme of the engine measurements, they do provide some basis for comparison. Figure 3.2 presents a comparison of the engine FQY data with measurements by Cheung et al. [90] for a total pressure of 4.5 bar in nitrogen. The agreement between the measurement techniques is quite good, and well within the quoted measurement uncertainty of ±7% for the Cheung et al. [90] data. Unfortunately higher temperature data (>800 K) is not available. Nevertheless, this limited comparison does confirm the accuracy of the lower temperature engine FQY measurements. 3.3 Acetone Photophysics Similar engine FQY measurements were performed for acetone in both Engine A and B, utilizing an identical technique and experimental setup with only a change in the seeded tracer. Acetone FQY measurements are shown in Figure 3.3a at a total pressure of 8 bar in air. The range of test conditions for Engine A is more limited for the acetone experiments, however excellent agreement is again evident where temperatures coincide. The acetone data was found to be well represented by linear fits and the resulting fit parameters are presented in Appendix A. The linear acetone FQY curves at each crank angle are presented in Figure 3.3b. As with 3- pentanone, the acetone curves show the expected trend in FQY with temperature,

62 42 CHAPTER 3. TRACER PHOTOPHYSICS Absolute FQY 11 x P=4.5 bar 3 Pentanone N 2 Bath 308 nm 277 nm Engine Cell Cheung et al. Data fit Temperature [K] Figure 3.2: Comparison of absolute FQY results from motored engine and flowing cell wavelength and pressure. Acetone exhibits stronger pressure dependence at 277 nm in comparison with 3-pentanone as indicated by the larger spacing between consecutive lower pressure FQY curves. The absolute FQY for 3-pentanone is found to be higher than acetone at both wavelengths over the entire temperature range. In addition the slope of the FQY curves is generally larger for 3-pentanone than for acetone. Experiments performed with acetone in air again showed a decreased magnitude of FQY, and a modest deviation in slope. Only one applicable fundamental acetone FQY study at coincident elevated temperature and pressure is available for comparison. Löffler et al. [33] acquired relative acetone signal data for 248 and 308 nm excitation at K and 1-20 bar, normalized by the fluorescence signal at room temperature and pressure for constant tracer number density. The relative 308 nm data was converted to absolute FQY using known absorption cross section data and absolute FQY data at ambient conditions [32]. The converted Löffler et al. [33] data are shown in Figure 3.3a in comparison with engine FQY measurements at the same gas conditions. Similar to 3-pentanone, there is good agreement between the engine and cell FQY data for acetone at 308 nm, well within the quoted 6% uncertainty for the Löffler et al. [33]

63 3.4. FQY MODEL COMPARISONS 43 data. Although no higher temperature cell data is available, this does confirm the accuracy of the moderate temperature engine FQY measurements of acetone. Absolute FQY 7 x P=8 bar Acetone Air Bath 308 nm 277 nm Engine A Engine B Löffler et al. Data Fit Temperature [K] (a) Absolute FQY x nm 308 nm 305 4bar 320 8bar bar bar bar bar bar bar 2 Acetone Air Bath Temperature [K] (b) Figure 3.3: Absolute FQY of acetone in nitrogen at 277 nm and 308 nm, measured under motored engine conditions. Each curve represents experiments at a given image timing and pressure for a range of intake temperatures 3.4 FQY Model Comparisons In conjunction with experimental investigations of tracer photophysics, significant research efforts have focused on the development of simplified models for accurate prediction of absolute FQY dependencies. Photophysical models are particularly helpful when developing new tracer-based diagnostic strategies that are optimized for specific experimental conditions as was shown in Chapter 2. In addition, these models provide useful insight into the key processes dominating energy transfer for large complex molecules like ketones and aromatics. The first comprehensive photophysical model for acetone was proposed by Thurber et al. [86]. The semi-physical model is based on excitation and multi-step decay of an average molecule, simplifying the need to model the complicated molecular energy distributions of the ground

64 44 CHAPTER 3. TRACER PHOTOPHYSICS and excited states. The model considers essential energy transfer mechanisms including intramolecular non-radiative decay (combining intersystem crossing, internal conversion and photodissociation), vibrational relaxation, and electronic quenching of oxygen. The model was originally tuned to match experimental FQY data at either elevated temperature at 1 bar, or elevated pressure at room temperature and has been shown to accurately predict the behavior at these conditions [86]. Braeuer et al. [87] expanded the measurement parameter space by measuring relative FQY at 248 nm for concurrent high temperatures and pressures ( K and 1-20 bar). Here comparison with the initial Thurber model indicated that the pressure-related FQY increase was over-predicted by the model, and new rate parameters were suggested. However the modeling work of Braeuer et al. [87] only included data at 248 nm excitation, and would likely not be as reliable at longer wavelengths. The acetone photophysical model has since been adapted for 3-pentanone by Koch et al. [75,104], using a similar formulation but with re-optimized rate parameters to better match 3-pentanone behavior. As with acetone, the initial 3-pentanone model tuning was based on FQY data at either high temperatures or high pressures. Subsequent model comparisons with FQY data at simultaneous high temperature and pressure [34,70,80] exhibited similar trends, but also highlighted errors in model predictions for conditions far outside the original tuning range. Rothamer et al. [44] later refined this model to better match 3-pentanone fluorescence behavior at 277 and 308 nm characterized in an IC engine. This refined 3-pentanone FQY model has been successfully used for quantitative two-line PLIF data calibration to measure temperature and composition in HCCI engines [44, 89]. Most recently, Cheung et al. [90] have further refined the 3-pentanone FQY model using the most comprehensive set of experimental data, and includes both flowing cell and engine-derived FQY data. Collectively, modeling efforts for both ketones have improved the fundamental understanding of FQY dependencies, however additional measurements and modeling refinements are required to expand the measurement domain and improve model accuracy. Comparisons between the preceding engine FQY data and available models have been completed here to further elucidate current model performance at high

65 3.4. FQY MODEL COMPARISONS 45 temperatures and pressures Pentanone Modeling Comparison of engine-based 3-pentanone FQY measurements (quadratic data fits) and available model simulations are provided in Figure 3.4a. Models derived by Koch [75], Rothamer [44] and Cheung [90] have been included. In addition to an absolute comparison, normalized FQY curves are presented in Figure 3.4b to better access the relative slope which is important when considering measurements of stratification. The absolute and normalized FQY data in Figure 3.4 is best predicted by the Cheung model, followed by the Rothamer and Koch models. The superior performance of the Cheung model is likely a result of the extensive FQY database used for model optimization. This agreement is particularly impressive given the wide range of excitation wavelengths ( nm) included in the Cheung model development. Despite small differences in absolute magnitude, the Chueng model accurately predicts the normalized slope of FQY with temperature. As a result, the Chueng model can be used for direct image calibration when performing stratification measurements where absolute accuracy is not critical. The Rothamer model captures general 3-pentanone FQY trends, but because the initial model tuning at 277 and 308 nm only considered temperatures up to 800 K the accuracy is expected to degrade outside this range. The Rothamer model has been successfully employed for data calibration [70, 89] at the moderate temperatures and pressures considered during model tuning ( K, 1-12 bar). Similarly, inaccuracies in the Koch model arise from the limited set of FQY data available (asynchronous high temperature and pressure) during initial model development. Although the Koch model captures the general trends and could be used for theoretical diagnostic development and optimization, it is not well suited for direct image calibration. The FQY comparisons in Figure 3.4 also provide insights into future model refinements. In general the absolute FQY is under-predicted by both the Cheung and Rothamer model, likely indicating that the impact of vibrational relaxation is

66 46 CHAPTER 3. TRACER PHOTOPHYSICS under-predicted or the non-radiative decay rate is over-predicted for this range of conditions. Conversely the over-prediction of absolute FQY by the Koch model is evidence of either an over-predicted vibrational relaxation or under-predicted nonradiative decay rate. 277 Eng. 277 Koch 277 Rothamer 277 Cheung 308 Eng. 308 Koch 308 Rothamer 308 Cheung 1 Absolute FQY [10 3 ] P=18 bar 3 Pentanone N Temperature [K] (a) Normalized FQY P=18 bar 3 Pentanone N Temperature [K] (b) Figure 3.4: Comparison of engine FQY data derived from quadratic fits and FQY model simulations. (a) Absolute FQY, (b) Normalized FQY for better comparison of slope Acetone Modeling A similar comparison between engine measurement and model is presented in Figure 3.5 for acetone. Simulation results for the original Thurber model [79], and the refined Braeuer model [87] have been included. Here the absolute FQY is overpredicted by all models, while the normalized slope is under-predicted. This is consistent with the observations by Braeuer [87], who noted that the original Thurber model over-predicted the relative importance of vibrational relaxation, resulting in higher calculated FQY at high temperatures as pressure is increased. The impact of model adjustments by Braeuer are not as apparent in Figure 3.5a for longer wavelengths. This is because the non-radiative decay rate in the Braeuer model was

67 3.4. FQY MODEL COMPARISONS 47 altered for higher excess energies only (due to 248 nm excitation) and does not have much impact on longer wavelengths. The overall deviation between acetone model predictions and measurement is not unexpected given the limited degree of acetone model refinement in comparison to 3-pentanone. Although no model improvements are suggested here, the engine-based FQY measurements in conjunction with other studies [33, 87] provide an expanded database for future model refinement. 277 Eng. 277 Thurber 277 Braeuer 308 Eng. 308 Thurber 308 Braeuer Absolute FQY [10 3 ] P=18 bar Acetone N Temperature [K] (a) Normalized FQY P=18 bar Acetone N Temperature [K] (b) Figure 3.5: Comparison of engine FQY data derived from linear fits and FQY model simulations. (a) Absolute FQY, (b) Normalized FQY for better comparison of slope. The preceding discussion demonstrates the utility of engine-based FQY measurements as an alternative to fundamental flowing cell measurements. The close agreement between engine and cell FQY data confirms the accuracy of the measurement despite some uncertainty in the calculated in-cylinder temperature. Additional comparisons of engine data with FQY model predictions indicate that most models predict general trends, however only a few provide sufficient accuracy to be considered for direct image processing. Based on the success of past studies, the Rothamer 3-pentanone model has been employed for data calibration of HCCI engine measurements at moderate temperatures and pressures encountered in engine A (Chapter 5). For the higher temperatures and pressures achieved in engine

68 48 CHAPTER 3. TRACER PHOTOPHYSICS B HCCI experiments (Chapter 6), direct FQY data fits for 3-pentanone and acetone are used for data processing. The accuracy of these methods will be discussed further for each application.

69 Chapter 4 Fluorescence Saturation 4.1 Introduction A majority of PLIF strategies, including those presented in Chapter 2, assume linear fluorescence behavior. This implies that the excitation laser fluence (energy per unit area) is sufficiently low such that the fluorescence signal is proportional to laser energy. Operation in this regime is desirable as it reduces the complexity of signal interpretation, particularly in comparison with intermediate excitation (between linearity and saturation). At high laser fluences fluorescence saturation can occur due to substantial ground state depopulation. At these conditions the laser absorption rate dominates the state to state energy transfer, resulting in a fluorescence signal that is independent of laser power and electronic quenching rates. As a result, saturated LIF has been used to measure the concentration of minor species such as OH, NO, CH, CN, C 2 and NH [ ] without substantial knowledge of quenching rates. Saturated LIF measurements are less common for seeded organic tracers such as ketones and aromatics. Complete saturation is difficult to achieve for these species due to the low absorption cross-sections associated with the symmetry-forbidden transition of the carbonyl functional group. However, knowledge of the energy threshold of non-linear behavior is particularly important for diagnostic development 49

70 50 CHAPTER 4. FLUORESCENCE SATURATION as the threshold will limit either the diagnostic performance or minimum spatial resolution that can be achieved. This chapter investigates the threshold of nonlinear fluorescence behavior for common tracers including acetone, 3-pentanone and toluene at typical excimer excitation wavelengths of 248, 277 and 308nm. Although toluene is not used for any PLIF diagnostics in the current work, it is included here for completeness. Experiments were performed over a range of total pressures, for both air and nitrogen bath gases to investigate the impact of vibrational energy transfer and electronic quenching on saturation. The study highlights the 10% nonlinear threshold and sets bounds for peak laser intensities that result in linear LIF excitation. 4.2 Experimental Approach Saturated LIF can be studied by either varying the laser energy for fixed beam dimensions, or by varying the beam dimensions (focusing) at constant energy. While both techniques have merit, the focused beam technique will be used exclusively. By focusing the laser beam to a waist inside a test cell, each acquired fluorescence image contains data over a continuous range of laser fluences along the propagation direction. This reduces the time for data acquisition and makes parametric studies of tracer partial pressure, and total pressure more manageable. This technique was first used by Petersen et al. [114] to study the saturation of 3-pentanone at 266nm excitation. Petersen observed that the onset of fluorescence saturation occurs at lower laser fluences than previously thought, motivating additional studies Optical Setup The experimental setup used to study fluorescence saturation is shown in Figure 4.1. Laser excitation at 248nm and 308nm were provided by a KrF and XeCl excimer laser respectively (both were Lambda Physik CompPexPro 102). The 277 nm excitation corresponds to the 1st Stokes Raman shift of 248 nm in H 2 and was generated by passing the KrF output through a high-pressure H 2 -filled Raman

71 4.2. EXPERIMENTAL APPROACH 51 cell (not shown). The pump laser beams were focused with a 150 mm cylindrical fused silica lens and then passed through the test cell. Relevant specifications for all excitation wavelengths are shown in Table 4.1. The relative laser energy fluctuations were monitored by sampling the beam with a fused silica wedged beam sampler located upstream of the test cell. The energy of the partial reflection was measured by temporally integrating (via Tektronics TDS 7104 scope) the signal from a high speed photo-detector (Thorlabs DET210). The absolute laser energy was calibrated before each test run using a pyroelectric energy meter (Ophir PE50-BB). Each wavelength was tested individually at all desired conditions before switching delivery optics for a subsequent wavelength. The large aperture test cell (152 mm long, 114 mm x 114 mm square crosssection) is a custom design, fabricated from aluminum (6061-T6) with a maximum pressure rating of 8 bar. Optical access in the test cell is provided by three fused silica windows (58mm clear aperture, 12.5mm thick), two opposed for line of sight laser access, and one orthogonal for imaging access. The static test cell is connected to a vacuum and filling assembly that evacuates and fills the cell with tracer and bath gas before each experiment. To generate a specific mixture, the evacuated cell is first filled monometrically with tracer vapor to the desired partial pressure, before being diluted with bath gas to the final total pressure. Sub-atomospheric pressures were monitored with an isolated MKS Baratron ( torr), while pressures above atmospheric were monitored with a Setra 280E pressure transducer (0-100 psia). The induced fluorescence was imaged using an intensified CCD camera (Princeton Instruments PIMAX2:1003) equipped with either a 50 mm f/1.2 lens (ketones) or a 105 mm f/4.5 UV Nikkor lens (toluene) depending on the fluorescence emission band of the tracer being tested. The images were binned 4x4 prior to readout, resulting in a pixel size in the image plane of mm and mm for the ketone and toluene experiments respectively. Black paint applied to the rear internal cell wall minimized the impact of reflected fluorescence signal and resulted in an insignificant background signal. No camera filters were used for the ketone experiments as the glass camera lens sufficiently attenuated any scattered laser light. A band-pass filter was used for the toluene experiments to reject elastically scattered excitation

72 52 CHAPTER 4. FLUORESCENCE SATURATION at 248 nm. The excimer lasers and camera controller were syncronized using a delay generator (SRS DG535) to ensure that the laser pulse was centered in the exposure time. Beam Profiling Camera Test Cell ICCD Camera Tracer Vapor Translation Stage Focusing Optics P N 2 / Air Integrating Sphere Vacuum Excitation Laser 248 / 277 / 308nm Photodiode Figure 4.1: Experimental setup for measurement of tracer saturation intensity Data Acquisition For each experimental condition, the test cell is first evacuated and 50 background images are acquired with laser emission but without tracer present. These images are averaged and used to correct for dark signal accumulation, any analog to digital offset, and any fixed pattern background noise. The cell is then monometrically filled with tracer vapor to the desired partial pressure, followed by dilution gas to the desired total pressure. A sufficient time delay between filling and data image acquisition is used to ensure proper mixing. Comparison of saturation results using the filling procedure described above, and those using a homogeneous premixed mixture of tracer and diluent show no difference. This confirms that no significant error is induced when preparing mixtures directly in the test cell. After mixture preparation and mixing delay, 50 data images are acquired with the focused beam traversing the cell. Finally, the focusing lens is removed, and 50 flat-field images are acquired. These flat-fields provide an unsaturated (linear) reference image that

73 4.2. EXPERIMENTAL APPROACH 53 corrects for laser absorption and any inhomogeneities in the image. All image sets are averaged before data processing. Representative background corrected flat (top) and data (bottom) images are shown in Figure 4.2a. The effects of saturation are represented in the data image by the darkened region towards the right side of the bottom image and results from the ground state depopulation. The location of the beam focus was adjusted horizontally in the test cell to ensure that the input region (left side of image) was in the linear fluorescence regime. This ensures that the resulting saturation profile originates in the linear regime. The linearity of the input region was confirmed by matching the fluorescence signal of the data and flatfield at the beam input (after correcting for the 8% reflection loss of the focusing lens). Fluorescence saturation profiles were generated by first ratioing the backgroundcorrected data and flat images. An axial 1-D profile was then generated by summing 20 horizontal line profiles centered around the vertical beam center. This profile was then normalized to unity on the beam input side (left) where the data image is known to be in the linear regime. A typical saturation profile is shown in Figure 4.2b, and represents the fractional deviation from linearity along the beam propagation direction. To convert this profile into more meaningful saturation data, the beam width as a function of axial distance must be determined in order to calculate the variation in laser fluence Beam Width Measurements Beam width measurements were performed using a CCD-based beam profiling system depicted in Figure 4.1. The system consists of a beam sampler (4 fused silica wedge), a profiling CCD camera (Spiricon UBS-L230), and a linear translation stage. The beam sampling wedge is positioned downstream of the focusing lens, and provides a partial surface reflection of the incoming beam that is directed towards the camera. The reflected beam was then passed through a series of reflective neutral density filters and was imaged directly on the CCD chip. Careful selection of the ND filters ensured that the readout signal was sufficiently high but not saturated.

74 54 CHAPTER 4. FLUORESCENCE SATURATION [mm] Relative Signal [a.u.] [mm] (a) Fluorescence Images Propagation Distance [mm] (b) Corrected Profile Figure 4.2: (a) Representative fluorescence images without focusing lens (top) and with lens (bottom), laser propagation is from left to right. (b) Corrected axial signal saturation profile indicating fraction deviation from linearity. Camera dark noise was accounted for by performing Spiricon s proprietary Ultracal baseline correction algorithm before and periodically during experiments to reduce the impact of the background noise. The entire camera and filter assembly was attached to a precision linear translation stage, allowing measurements to be made in numerous planes along the beam focus. For beam profile images far from the focus where energy density was low, multiple laser pulse accumulations were captured to maximize signal. Jitter in the pointing stability of the laser did not impact the profile measurements at these locations. Profile images near the focus, where signal was sufficiently high, were acquired on a single-shot basis. Prior to acquiring beam profile data, the camera position was adjusted to locate the beam focus by minimizing the measured width in real-time. Subsequent beam profile measurements were then performed at equal distances before and after the focus location and extended well outside the Rayleigh range. Finally the beam width was calculated for each image location, as described in Section below. An alternative and common method of beam width measurement utilizes a scanning knife edge. For this technique a knife edge, oriented perpendicular to the dimension of interest, is scanned across the beam to determine the distance between

75 4.2. EXPERIMENTAL APPROACH 55 the 10% and 90% clip locations. The clip width is then converted into a beam width using an appropriate scaling factor [115], and is typically assumed to be This measurement can be performed at various locations along the beam propagation direction to determine the beam width as a function of distance for a focused beam. In general, the technique is straightforward and only requires a translation stage and some means of measuring beam energy. This technique was applied by Petersen et al. [114] for the study of 3-pentanone fluorescence saturation at 266 nm. The scanning knife edge was used for initial test in the current study, but was found to be too time intensive. In addition, the scaling factor used to convert from clip width is only exact for a perfect gaussian beam (TEM 00 ), and can vary for higher order beams (e.g. excimers) depending on the mode content. As a result, the CCD-based profiling system above was adopted. Generally, results acquired by each method provided consistent results. Beam Profile Image Processing A series of representative beam profile images taken along the laser propagation direction are shown in Figure 4.3a, where the bottom image corresponds to beam focus. Each images was converted into an accumulated 1-D profile along the beam width by summing line profiles centered around the peak energy location in the vertical direction. The beam width was then calculated from these profiles using the second moment method, the accepted ISO standard for beam width (ISO :2005). The second moment (σx) 2 in the x-direction is calculated using Equation 4.1: σ 2 x = x2 x 1 x2 (x x) 2 I (x) dx x 1 I (x) dx (4.1) where I(x) is the transverse intensity profile determined above, x is the distance along the profile, and x is the beam centroid in the x-direction determined from Equation 4.2. The resulting beam half width, w, is then twice the standard deviation

76 56 CHAPTER 4. FLUORESCENCE SATURATION (a) Beam profile images Beam Half width [mm] λ=308 nm w 0 =0.076 mm Model Fit Meas. Width M 2 = Propagation Distance [mm] (b) Measured beam widths Figure 4.3: (a) Series of beam profile images acquired along beam focus. (b) Measured beam half width along focus with fit results. as shown in Equation 4.3 x = x2 x 1 x2 x 1 I(x)xdx I(x)dx (4.2) w x = 2σ x (4.3) Camera-based beam width measurements using the second moment technique are influenced by the CCD background noise, particularly in the wings of the beam profile. The quadratic term in Equation 4.1 heavily weights the profile wings which have low signal, and are often non-zero due to imperfect baseline subtraction. This noise can result in significant errors in the beam width calculation if not eliminated or accounted for. To address this issues the Spiricon UltrCal procedure, which has been shown to reduce these errors [116], was performed frequently to improve baseline correction. In addition, a truncated calculation of the second moment has been performed using an adaptive aperture to set the bounds for integration. This is represented in Equation 4.1 by the finite integral bounds, which are infinite for the ideal second moment calculation. After processing all images acquired along the propagation direction, the beam half widths were plotted as a function of distance about the focus as shown in

77 4.2. EXPERIMENTAL APPROACH 57 Figure 4.3b. The data was then fit with Equation 4.4 ( ) λm w x (z) = w (z z 0 ) 2 (4.4) πw 0 where w 0 is the half width of the beam waist, M 2 is the beam propagation parameter, and z 0 is the location of the beam waist. Equation 4.4 is a form of the gaussian beam spot size equation that has been modified for real laser beam propagation through inclusion of the propagation parameter M 2 [117]. M 2 can be interpreted as a measure of beam quality relating the real beam divergence to an ideal gaussian, thus providing the number of times diffraction limited (M =1 for ideal gaussian). With the fit parameters determined, Equation 4.4 can be used to calculate the beam width at any location, x, and is used to determine the variation in laser cross-section area along the focus. Representative curve fit results are shown in Figure 4.3b. This procedure was completed for each wavelength to determine the propagation characteristics, and the results are shown in Table 4.1. The beam width measurements were carried out multiple times during the measurement campaign to assess the beam stability. Two sets of beam width data taken on different days are shown in Figure 4.2b. The good agreement confirms the laser output stability, which was assumed to be constant throughout all experiments. Table 4.1: Excitation wavelength specifications for saturation experiments Excitation Total Energy M 2 Beam Waist Wavelength [nm] [mj] [mm] Variation in laser fluence along the beam focus was determined from the total laser energy measurement and the beam width data. As mentioned above the saturation profiles generated in Figure 4.2b correspond to a subregion of the total vertical beam dimension. The subregion height was determined based on the number of pixels used for summation and the measured spatial resolution of the imaging system.

78 58 CHAPTER 4. FLUORESCENCE SATURATION This coupled with the beam width data determined the beam cross-sectional area along the subregion focus. The corresponding fraction of total laser energy within the beam subregion was determined by normalizing the integrated signal within the subregion by the total integrated signal along the vertical direction. Finally, the laser fluence, I L, was determined from I L = E p /2wh p where E p is the laser energy in the profile subregion, w is the beam half width, and h p is the height of the profile subregion. The metric used to describe saturation is the percent deviation from linearity, δ Lin, and is determined from Equation 4.5. Here S f is the axial fluorescence profile, and Sf Lin is the corresponding linear fluorescence profile. The ratio of fluorescence signals S f /Sf Lin is exactly what is shown in Figure 4.2b. δ Lin = 100 (1 S f Sf Lin ) (4.5) The percent deviation is then plotted versus the variation in laser fluence and is presented in Figure 4.4 for 308 nm excitation of 3-pentanone. At low laser fluence (<70 mj/cm 2 ) the deviation is zero and thus the fluorescence is in the linear regime. As the fluence is increased, the percent deviation is increased due to increasing ground state depopulation. Complete saturation is not achieved at the maximum fluence for the current studies, as indicated by the continuously increasing percent deviation with increasing fluence. At full saturation the percent deviation will asymptote to a constant value other than 100%. The laser fluence was not increased to the point of saturation for the current work, as the main focus was to determine the 10% non-linear threshold value for each tracer molecule. This threshold has been selected as an indicator of strong non-linear fluorescence behavior, assuming that deviations below 10% due not significantly impact PLIF measurements. The 10% threshold is highlighted in Figure 4.4, by the dashed red line and was found to be 320 mj/cm 2 for 3-pentanone at the specified conditions. The aforementioned data acquisition and processing procedure has been used to study the saturation characteristics of three common tracers as discussed below.

79 4.3. EXPERIMENTAL RESULTS Deviation from Linearity [%] Laser Fluence [mj/cm 2 ] Figure 4.4: Percent deviation from linearity as a function of laser fluence, measured for 308 nm excitation of 3-pentanone at 1 bar in N 2 (derived from same data as Figure 4.2) 4.3 Experimental Results The saturation characteristics of acetone, 3-pentanone and toluene are studied using the experimental techniques described above. The investigation is considered in two parts. First, the dependence of saturation intensity on excitation wavelength is considered. Second, the dependence on total pressure and composition (air or nitrogen) is presented. These parametric variables have been selected in an effort to better understand the critical energy transfer processes and to assess the relative importance of absorption, vibrational relaxation, and electronic quenching of oxygen. Effects of gas temperature, which is somewhat more complicated to vary experimentally, is not considered in the current work. 4.4 Saturation Wavelength Dependence Fluorescence saturation data similar to Figure 4.4 was acquired for acetone, 3- pentanone and toluene over a range of wavelengths (248,277,308 nm) to assess the functional dependence. All experiments were performed at a total pressure of 1

80 60 CHAPTER 4. FLUORESCENCE SATURATION bar with nitrogen as the bath gas. Tracer partial pressures, which were selected to provide high fluorescence signals, varied between species. These differences are not expected to impact the results as initial data showed minimal sensitivity of saturation to tracer partial pressure. Saturation profiles for each tracer at an excitation wavelength of 308 nm are presented in Figure 4.5 for comparison. Acetone and 3-pentanone exhibit very similar saturation profiles, with difference in magnitude of only 10% for the highest laser fluence achieved. The similarities between these ketones is not unexpected given the analogous physical structure and photophysical behavior. Differences in magnitude likely arise from the collective differences in absorption cross-section and specific energy transfer rates. In comparison, the saturation intensities for toluene are generally much lower. Specifically, toluene exhibits approximately 27% deviation from linearity at a laser fluence of 100 mj/cm 2, while the ketone fluorescence is still effectively linear at these conditions. This large difference in saturation intensity is due in part to the high absorption cross-section of toluene which is factor of times higher than for either ketone. Deviation from Linearity [%] Pent. Acetone Toluene 248 nm Laser Energy Density [mj/cm 2 ] Figure 4.5: Comparison of deviation from linearity for acetone, 3-pentanone and toluene at 248 nm excitation, and 1 bar of N 2.

81 4.4. SATURATION WAVELENGTH DEPENDENCE 61 A comprehensive set of saturation data was acquired for each tracer at several excitation wavelengths. The resulting 10% non-linear thresholds for these measurements are summarized in Table 4.2. Both ketones exhibit a minimum in saturation intensity at 277 nm (based only on the wavelengths tested), with intensity increasing for longer and shorter wavelengths. These trends seem to follow an inverse relationship with absorption cross-section, which attains a maximum between nm at room temperature. Overall, the saturation intensities of acetone are slightly higher than for 3-pentanone. In contrast, the toluene saturation intensity was found to be a at least factor of 10 less than the ketones for the same excitation wavelength of 248 nm. Table 4.2: Experimental measurements of 10% non-linear threshold for various tracer and excitation wavelengths Tracer Excitation 10% Non-linear Wavelength [nm] Threshold [mj/cm 2 ] 3-Pentanone Acetone Toluene For all tracers tested, the excited state lifetime is dominated by fast intersystem crossing, transfering excited state molecules from the excited singlet to the excited triplet state. Due to the inherently long lifetime of the excited triplet state, minimal ground state repopulation from the excited states occurs during the timescale of the laser pulse. As a result, the saturation processes is likely more dominated by the absorption rate than any other mechanism of ground state repopulation. To demonstrate this, the saturation intensities for all three tracers are compared with the inverse of absorption cross-section in Figure 4.6. As expected, the trends of both saturation intensity and absorption cross-section are well correlated, confirming that the absorption rate dominates the saturation intensity at these room temperature

82 62 CHAPTER 4. FLUORESCENCE SATURATION % Dev. [mj/cm 2 ] / σ [10 19 cm 2 ] Pentanone Acetone Toluene Wavelength [nm] Figure 4.6: Comparison of 10% non-linear threshold (top) and inverse absorption cross-section (bottom) for all tracers and wavelengths and pressure conditions. However, the changes in saturation intensity and absorption cross-section are not exactly one to one (i.e. a 50% decrease in σ does not correlate exactly with a 50% increase in saturation intensity), and indicates that other secondary energy transfer processes are also influential. Potential secondary mechanisms such as vibrational relaxation and electronic quenching are considered below. 4.5 Pressure and Composition Dependence The total pressure of the mixture increases the collision frequency of tracer molecules with colliding partners and thus increases the potential for vibrational relaxation. This in turn can increase the saturation intensity through competition with the nonradiative decay rate that dominates the excited state for many tracer molecules. This is analogous to the increase in FQY with increasing pressure [75]. Elevated pressure can also increase the rate of intermolecular vibrational re-distribution (IVR) of the

83 4.5. PRESSURE AND COMPOSITION DEPENDENCE 63 ground-state and could help to avoid hole-burning effects due to high laser pumping. Changes in bath gas composition, specifically the addition of oxygen, may further increase the saturation intensity by adding an additional pathway for ground-state re-population through electronic quenching. While these processes collectively can impact saturation, the relative importance of each is unknown. To investigate these effects, saturation data of each tracer species was acquired for a range of total pressures (0-90 psi) in both nitrogen and air. All experiments were performed with a constant partial pressure of seeder tracer for each species. The pressure dependence of 3-pentanone saturation is explored in Figure 4.7a for 308 nm excitation and nitrogen dilution. The increase in saturation intensity with pressure follows intuition as the increased collision frequency is expected to improve population re-distribution and enhance vibrational relaxation. For these conditions, the saturation intensity increased from 320 to 680 mj/cm 2 when varying the total pressure from psi. The relative importance of oxygen quenching is shown in Figure 4.7b through comparison of saturation profiles for nitrogen and air at 15 psi total pressure. The close agreement between the air and nitrogen data suggests that oxygen quenching does not impact saturation for 3-pentanone with 308 nm excitation. Figure 4.7b includes several repeated data sets for both bath gases and demonstrates the repeatability of the measurements. Interestingly, the pressure dependence observed in Figure 4.7a for nitrogen is not seen in similar data for air where instead the saturation characteristics are constant with increasing pressure. Additional experiments at 277 and 248 nm showed no signs of pressure or composition dependence. This implies that the pressure-related phenomenon shown in Figure 4.7a are exclusive to excited states with low excess vibrational energy (energy above thermal equilibrium in excited state) that are associated with long excitation wavelengths such as 308 nm. Similar to 3-pentanone, the pressure dependence of acetone is considered in Figure 4.8a for 277 nm excitation and nitrogen dilution. Acetone data at 308 nm exhibited minimal dependence on pressure or oxygen compositions and is not shown. The results in Figure 4.8a are in stark contrast with the 3-pentanone results at 308 nm. For acetone the pressure dependence is reversed, with increasing pressures

84 64 CHAPTER 4. FLUORESCENCE SATURATION Deviation from Linearity [%] P tot =15 P tot =30 P tot =45 P tot =60 P tot =75 P tot = Laser Energy Density [mj/cm 2 ] (a) Nitrogen 3 Pentanone 308 nm N 2 Bath Deviation from Linearity [%] N 2 Air Laser Energy Density [mj/cm 2 ] (b) Nitrogen vs. Air 3 Pentanone 308 nm P tot =15 psi Figure 4.7: (a) Pressure dependence of 3-pentanone saturation with 308 nm excitation and a nitrogen bath gas. (b) Comparison of saturation characteristics for nitrogen and air bath gases with 308 nm excitation at 15 psi total pressure. Multiple curves included to demonstrate measurement repeatability. resulting in decreasing saturation intensity. Such behavior was unexpected and conflicts with the notion of enhanced vibrational relaxation improving the saturation characteristics. A similar inverse pressure dependence is also seen for acetone at 248 nm excitation, as shown in Figure 4.8b. In fact, the magnitude of the pressure influence appears to increase at shorter wavelengths. In general, the inverse pressure scaling indicates that other energy transfer mechanisms are at work. One possible explanation is the existence of a collisionally assisted non-radiative decay pathway. In a recent study of 3-pentanone photophysics by Cheung et al [90], a pressure dependent k nr term was added to the FQY model. This was done to better match experimental data that exhibited a decrease in FQY with increasing pressure. In addition, the effect of the collisionally assisted k nr was found to increase with excitation energy (excess vibrational energy in excited state). Assuming this energy transfer pathway also holds true for acetone and with a larger magnitude, it could result in the reversed pressure scaling shown here. Although not depicted graphically, acetone saturation was found to have minimal dependence on oxygen concentration as evidenced by nearly identical saturation behavior for both nitrogen and air dilution.

85 4.5. PRESSURE AND COMPOSITION DEPENDENCE 65 Deviation from Linearity [%] P tot =0 psi P tot =15 psi P tot =30 psi P tot =60 psi P tot =90 psi Laser Energy Density [mj/cm 2 ] (a) 277 nm Acetone 277 nm N 2 Bath Deviation from Linearity [%] P tot =0 psi P tot =15 psi P tot =30 psi P tot =60 psi P tot =90 psi Laser Energy Density [mj/cm 2 ] (b) 248 nm Acetone 248 nm N 2 Bath Figure 4.8: Pressure dependence of acetone saturation in pure nitrogen for excitation wavelengths of (a) 277 nm and (b) 248 nm. Saturation pressure dependence data was also acquired for toluene at 248 nm excitation in nitrogen and is shown in Figure 4.9. A small decrease in saturation intensity is observed at low pressures when transitioning from pure vapor pressure to 15 psi total pressure. Increases in pressure beyond this have little influence. Composition dependence data for toluene in air is not considered here, as the acquisition of high-quality data was difficult. Due to the dramatic impact of oxygen quenching on fluorescence signal for toluene, the dynamic range between nitrogen and air experiments was cumbersome. Toluene concentrations that provided reasonable performance in air saturated the detection system, and conversely, concentrations optimized for nitrogen experiments were not sufficient for the air bath gas. As a result, no composition dependence data for toluene is considered here. In general, toluene is expected to have larger compositional dependence in comparison with ketones given the higher electronic quenching rates. The preceding discussion provides additional insight into the fluorescence saturation process for typical PLIF tracers. Specifically, the rate of absorption was found to dominate the saturation behavior for the room temperature conditions considered. Other processes such as vibration relaxation, ground-state population re-distribution, and electronic quenching were found to be secondary. These secondary processes did impact the results for some conditions, but overall did not

86 66 CHAPTER 4. FLUORESCENCE SATURATION Deviation from Linearity [%] P tot =0 psi P tot =15 psi P tot =30 psi P tot =60 psi P tot =90 psi Laser Energy Density [mj/cm 2 ] Toluene 248 nm N 2 Bath Figure 4.9: Pressure dependence of toluene saturation with 248 nm excitation and a nitrogen bath gas. dramatically impact saturation characteristics for a majority of tracer, wavelength, and pressure combinations. The 10% non-linear thresholds presented in Table 4.2 are particularly useful when designing PLIF experiments and ultimately sets the limit of excitation energy for measurements in the linear regime. While it is generally preferred to assess fluorescence linearity in-situ for a given experimental setup, these baseline numbers provide a general guideline and can be used if in-situ characterization is difficult. Based on these results, the pump laser energies for the HCCI engine applications presented in Chapter 5-6 have been chosen to be below the measured linearity thresholds.

87 Chapter 5 Low-Load HCCI with NVO Development of the gasoline homogeneous charge compression ignition (HCC) engine is currently focused on extending the operating range at both low and high load extremes. At the low limit, many strategies incorporate extensive internal exhaust gas recirculation (EGR, throughout this work refers to internal exhaust gas recirculation), providing additional thermal energy that can help stabilize and control combustion phasing. A common approach is the negative valve overlap (NVO) strategy based on injecting and reacting/reforming a small quantity of fuel during NVO recompression [51, 52, 118]. An advantage of this strategy is the potential for controlling main combustion phasing by varying NVO parameters such as the amount of EGR, the NVO injection timing, and the NVO/main fuel-injection split. While performance of low-load NVO operation has been demonstrated, our understanding of the strategy can still be improved. Fuel reformation during NVO provides both heat and reaction products that affect main combustion phasing. While pressure records can provide an estimate of exothermicity during NVO, the extent of reaction, the composition of reformed gases, and the relative importance of thermal and chemical influences on main combustion are mostly unknown. Previous optical diagnostic studies have provided some limited details of NVO chemistry such as laser-induced fluorescence (LIF) detection of formaldehyde during low-temperature reactions of two-stage fuels, and evidence of high-temperature reactions via OH chemiluminescence [119]. The current work addresses the gap in understanding of 67

88 68 CHAPTER 5. LOW-LOAD HCCI WITH NVO the NVO strategy through the development and application of an optical diagnostic providing simultaneous temperature and composition distributions during HCCI operation. Planar laser-induced fluorescence (PLIF) is capable of measuring two-dimensional temperature and concentration distributions with a high degree of accuracy and precision. PLIF of tracer molecules such as 3-pentanone, acetone, and toluene have seen widespread use in IC engine research, typically for imaging of in-cylinder fuel distribution. Tracer-based PLIF measurements of in-cylinder temperature distributions have previously been demonstrated by Einecke et al. [26, 41], Fujikawa et al. [42] and Kakuho et al. [43] among others. Although similar in theory, each diagnostic method varies in terms of tracer, excitation wavelengths, excitation and collection scheme, and calibration method. In previous work by Rothamer et al. [44,70,120], a two-line (i.e., dual-wavelength) PLIF technique using 3-pentanone (3P) as the fluorescent tracer was developed for simultaneous measurements of temperature and composition. The technique was optimized for in-cylinder engine conditions using available data for 3-pentanone absorption cross-section and fluorescence quantum yield (FQY) at elevated temperature and pressure [31, 34]. Wavelength selection was based on a comprehensive uncertainty analysis that lead to the selection of 277nm and 308nm. Validation of this technique was performed in a motored engine under known conditions to assess measurement accuracy and precision. Additional demonstrations were also performed under fired-engine operation for both conventional HCCI and HCCI with NVO. The present work extends the application of the two-line PLIF technique to portions of the NVO engine cycle that were previously unexplored, particularly NVO recompression and re-expansion. Both fuel-seeded and air-seeded variations of the technique are applied. Fuel-seeded measurements of temperature and fuel mole fraction in the NVO recompression provide better understanding of in-cylinder conditions at the end of exhaust and early in the NVO recompression period. Similarly, measurements during NVO re-expansion help characterize gas temperatures following recompression reactions that ultimately affect main combustion phasing. Air-seeded measurements of temperature and EGR mole fraction during the main

89 5.1. EXPERIMENTAL SETUP 69 compression explore a range of NVO operating conditions, studying the effects of main and NVO injection timing on in-cylinder charge evolution. This suite of measurements is intended to address general aspects of NVO engine operation while assessing the feasibility of the diagnostic technique. The focus of this work is more on the application of the two-line PLIF technique and less about specifics of NVO engine operation. 5.1 Experimental Setup Measurements were performed in a single-cylinder optical engine configured for NVO experiments. The pent-roof head is a GM prototype, housing two intake valves, a single exhaust valve, and a vertical, near-centrally located 8-hole injector. The head is also equipped with two spark plug ports that have been plugged for the current HCCI experiments. The Bowditch piston is equipped with a 65-mm fused silica window, providing imaging access from below. A sample PLIF image is shown in Figure 5.2 with the valves, injector and piston window superimposed in the field of view. For all PLIF images shown throughout this work, laser propagation is from image bottom to top, with exhaust and intake valves on the left and right respectively. Optical access for the laser sheets is provided by a 25-mm-tall fused silica ring incorporated into the cylinder liner. Short dwell cams (145-CAD duration) enable NVO operation, providing the high internal EGR levels required to burn high octane fuels in this engine facility. Engine specifications are provided in Table 5.1, with further details available in [121]. All engine measurements were performed at 1200 RPM, with iso-octane as the base fuel. Tracer was seeded in either the liquid fuel (20% 3-Pentanone by volume), or the intake air as described below. The crank angle convention for the NVO work is -360<CAD<360, with 0 crank-angle degrees (CAD) corresponding to top dead center (TDC) of main compression The basic PLIF diagnostic system consists of two lasers, a Raman cell, one dualframe ICCD camera, and associated optical components. A schematic of the PLIF diagnostic setup is presented in Figure 5.1. The 308-nm laser pulse is generated

90 70 CHAPTER 5. LOW-LOAD HCCI WITH NVO Table 5.1: HCCI engine specification for NVO operation Bore 92 mm Stroke mm Geometric Compression Ratio 11.5 IVO / IVC* -285 / -140 CAD EVO / EVC* 140 / 285 CAD *intake/exhaust valve open/close by a Lambda Physik ComPexPro 102 XeCl excimer laser. The 277-nm pulse is generated by Raman shifting the output of a 248-nm KrF excimer laser (Lambda Physik ComPexPro 102, stable resonator) to the first Stokes wavelength (277nm) in H 2. The Raman cell (RC) used in the current experiments (custom LightAge RC) is 72 cm in length, and is configured with a 60-cm focal length (f.l.) spherical lens at the entrance and a 40-cm f.l. spherical lens at the exit. The RC is filled with high-purity H2 to a total pressure of 57 bar, providing approximately 10% energy conversion efficiency to the first Stokes wavelength. The output of the RC is passed through an equilateral dispersing prism to spatially separate the residual pump and shifted wavelengths, allowing the unwanted beams to be collected downstream with a beam dump. The current setup utilizes an extended beam path length of 5-6 m for each excitation wavelength, allowing the inherent laser beam divergence to expand the beam to the desired imaging sheet dimension. Each excitation beam is passed through a 1-m f.l. cylindrical lens to form the beam waist, before being spatially overlapped on a dichroic element coated to reflect 277 nm and transmit 308 nm. Finally the beams pass through a vertical elevator assembly that rotates the beam 90 and elevates the beam path to the desired location in the optical engine. The resulting laser sheets are approximately 45-mm wide and 0.5-mm thick and are aligned to be parallel to the fire deck, and displaced 3 mm below. The sheets diverge slightly as they pass through the engine cylinder wall due to its effective negative focal length. Typical laser energies per pulse during experiments were mj for 277 nm and mj for 308 nm. Laser energies were measured several

91 5.1. EXPERIMENTAL SETUP 71 ICCD Camera 248 nm 277 nm 308 nm Raman Shifter 248 nm Excimer Laser 308 nm Excimer Laser Optical Engine Laser Sheets Figure 5.1: Experimental schematic Figure 5.2: Sample PLIF image with valve, injector and piston window positions superimposed.

92 72 CHAPTER 5. LOW-LOAD HCCI WITH NVO optical components upstream of the cylinder and an additional 4-6% energy loss is included in the above laser energies. These laser fluences were selected to be below the threshold values presented in Chapter 4, ensuring linear fluorescence behavior. The resulting fluorescence signal is transmitted through the fused silica piston window, and imaged onto the camera using a 85-mm f/1.4 Nikon lens. The ICCD camera (Princeton Instruments PIMAX2:1003, P46 phosphor, SB slow-gate photocathode) is equipped with an interline transfer CCD array allowing the collection of two images with inter-frame timings as low as 2 µs. A sharp cut-on 325-nm long-pass filter is used to reject scattered 308-nm laser light. The 1024x1024-pixel CCD array is binned 8x8 on chip resulting in read-out images of 128x128 pixels (hereafter referred to as full-frame). This provides an image spatial resolution of approximately 0.5 mm, matching the out-of-plane resolution provided by the sheet thickness. In some instances the images are binned an additional 2x2 during post-processing to improve signal-to-noise ratio while sacrificing spatial resolution. For all experiments, the ICCD camera inter-frame time is set to 5 µs, and the gate widths for the two images are set to 800 ns. The camera and laser system were sychronized to the engine by a quarter-crank-angle-degree-resolution crankshaft encoder Motored Engine Operation Motored engine operation is used to generate a homogeneous tracer-air mixture of known temperature, pressure and composition. The premixed seeding system consists of a positive displacement pump delivering tracer to a heated cell mounted on the intake air line upstream of the surge tank. The premix cell is maintained at a temperature approximately 40 K above the boiling point of the tracer to ensure rapid evaporation. The tracer flow rate is determine by measuring the mass of tracer delivered with a scale, and measuring the corresponding delivery time with a stopwatch. The air flow rate to the engine was metered with a sonic orifice. Typical seeding levels of 3-pentanone in the intake stream corresponded to a 3-pentanone mole fraction of between % depending on engine conditions. The motored

93 5.1. EXPERIMENTAL SETUP 73 operation mode is used for two purposes: diagnostic validation and image calibration. For purposes of diagnostic validation, motored experiments were performed for a range of intake conditions with intake temperatures ranging from K, and manifold pressures from bar. These experiments were carried out using more conventional cam phasing resulting in approximately zero valve overlap (NVO 0). Results are described in Section 5.2. For the purpose of image calibration, motored operation was employed between fired data sets to acquire homogeneous calibration images required for data processing. In these experiments, cam phasing was adjusted for valve overlap of -150 CAD (denoted NVO 150) Fired NVO Engine Operation The NVO operation scheme applied in this work utilizes a dual-injection strategy to achieve HCCI combustion. A representative pressure trace for fired NVO operation is shown in Figure 5.3, illustrating the valve events and range of injection timings. The largest pressure peak, centered at 0 CAD, is associated with main compression and combustion, and the lesser peak, centered at -360 CAD, is due to NVO recompression resulting from early exhaust valve closing. Fuel injected during the NVO mixes with hot exhaust gases from the previous main combustion and undergoes compression during the NVO. If sufficient oxygen is present and ignition temperatures are reached, exothermic reaction can occur. The exothermic reaction can elevate residual temperatures enough to affect the subsequent main combustion phasing. The NVO reactions also may produce partially reformed fuel species that could further affect main combustion phasing. For all conditions in the current experiments, appreciable apparent heat release is observed during the NVO recompression stroke (seen in Figure 5.3b as an upturn at the left end of the bottom pumping loop) due to relatively early NVO injection and globally lean operating conditions. A list of fired engine operating conditions is presented in Table 5.2. Main and NVO injection timings sweeps as well as main combustion load sweeps were investigated. During the injection timing sweep experiments, total fuel per cycle is held constant at 9.5 mg/cycle, with 1.5 mg/cycle

94 74 CHAPTER 5. LOW-LOAD HCCI WITH NVO (a) (b) Figure 5.3: Measured pressure for NVO engine operation, (a) pressure vs. CAD showing valve events and measurement regimes; (b) pressure vs. volume (log-log scaling) injected during NVO and the remainder injected during main intake / compression. For the main load sweep experiments, NVO injections are held constant at 1.5 mg/cycle while varying the main injections from mg/cycle. Average fuel mass delivered per 50 cycles was measured using a positive displacement flow meter (Max Machinery Model 213 Piston Meter) providing an estimated ±3% accuracy for these low-load conditions. Because the TB-PLIF diagnostic requires a fluorescence tracer, its application is limited to times in the engine cycle when sufficient tracer is present. Tracer can be seeded in the fuel or in the air. Considering fuel seeding first, NVO fuel injection occurs as early as +260 CAD (see Figure 5.3a), and LIF images can be obtained from +280 until the tracer decomposes during NVO reaction. The earliest main fuel injection was at -320 CAD, allowing imaging as early as -300 CAD, prior to intake valve opening (IVO). Again, imaging can continue throughout compression until tracer decomposition and oxidation occurs. For all fuel seeded experiments, tracer was mixed in the fuel at 20 vol%. For the air-seeded experiments, images can be captured soon after IVO, once tracer-seeded air enters the cylinder, continuing through the compression stroke. In addition, some measurements were made during

95 5.1. EXPERIMENTAL SETUP 75 Table 5.2: Fired engine operating conditions. Air Mass Flow Rate: g/s Intake Air Temperature: 360 K Manifold Pressure: 1 bar Main Start of Injection (SOI): -320 to -100 CAD Main Fuel Amount: mg/cycle NVO Start of Injection (SOI): +260 to +330 CAD NVO Fuel Amount 1.5 mg/cycle Engine Speed 1200 RPM Coolant Temperature: 90 C Residual Gas Fraction (mass): 48-54% Tracer Mass: 1-2 mg/cycle NVO recompression based on residual tracer that escapes main combustion through sequestration in cold crevice volumes. The addition of 3-pentanone does impact the HCCI combustion characteristics of iso-octane as has been previously studied [122]. In general, the addition of 3- pentanone to the liquid DI fuel advances the combustion phasing in comparison to pure iso-octane operating. The phasing is further advanced by pre-mixing the tracer into the intake air due to the effective addition of the heat of vaporization prior to delivery in-cylinder. For the current work, all quantitative comparisons are made between experiments with like seeding methods, and as such the influence of combustion does not alter any conclusions Data Acquisition Quantitative PLIF measurements require three separate image sets for data processing: a data set, calibration set and background set. Each data set consists of 100 images (50 images for each excitation wavelength) taken at the desired engie condition and image timing. Calibration images (also referred to as flat-field images) are taken under homogeneous motored operation at known conditions. The calibration images are taken shortly after acquisition of data images to minimize effects of laser profile fluctuations, ensuring accurate profile correction. Image timing

96 76 CHAPTER 5. LOW-LOAD HCCI WITH NVO for the calibration images was carefully selected to ensure a uniform distribution and minimal absolute temperature uncertainty. Background images are acquired with laser emission, but without tracer seeding, to correct for any laser scatter or fixed pattern signal. Cylinder pressure was recorded simultaneously with data and calibration images to permit cycle-by-cycle corrections of any pressure fluctuations. 5.2 Validation Experiments Homogeneous motored experiments provide a controlled means of validating the PLIF diagnostic, as well as assessing measurement accuracy and precision. For this purpose, motored data was recorded over a range of intake temperature and pressure conditions. Figure 5.4 presents measured average temperature and air mole fractions acquired under motored conditions with an air inlet temperature of 412 K and a manifold pressure of 1 bar. Average temperatures were calculated over a 30x30 pixel region in the center of each image, and averaged among all 50 data images. The measured temperatures in Figure 5.4a are compared to an isentropic compression calculation based on the measured pressure trace and variable specific heats. There is good agreement between measurement and prediction, as expected since both represent the core region of the cylinder. For the full range of motored operating conditions examined, the average temperature error is estimated to be less than ±4%. Average air-mole-fraction measurements are shown in Figure 5.4b, calculated in a similar fashion as the average temperatures. Because the motored experiments contain no measurable EGR mole fraction, the expected value for these conditions is 100% air. The measurements are systematically 4-5% higher than the expected value and indicate an average mole fraction error of ±5% for the conditions tested. This offset is likely due to a combination of inaccuracy in the 3-pentanone FQY model and propagation of uncertainties in measured temperature that are used to correct the mole fraction measurement. Given the systematic offset, it may be reasonable in practice to apply a constant correction factor; however this has not been applied here.

97 5.2. VALIDATION EXPERIMENTS 77 Temperature [K] Measured Average (± 4% error bars) Isentropic Compression Crank Angle [ CA] (a) Air Mole Fraction [%] Measured Average (± 5% error bars) Crank Angle [ CA] Figure 5.4: (a) Measured average temperature and (b) air mole fraction measured for motored engine conditions with air intake temperature of 412 K and a manifold pressure of 1 bar. (b) Measurement precision is determined by calculating the standard deviation of both temperature and mole fraction over the same 30x30 pixel region used above. Measurement precision calculations corresponding to the same experimental conditions as Figure 5.4 are presented in Figure 5.5. For these conditions, minimum single-shot temperature and mole fraction precisions of ±7 K and ±3.7% have been achieved, respectively. These standard deviation results are generally higher than the theoretical estimates presented in Chapter 2 but do follow the predicted trends. Deviations between simulation and experiment are attributed to inaccuracies in the theoretical fluorescence signal estimates. Accuracy of the predictions could be further improved through a more accurate characterization of system components (e.g. collection efficiency, quantum efficiencies etc.). In general, the favorable agreement in performance trends does demonstrate the utility of the uncertainty analysis and the excitation wavelength selection.

98 78 CHAPTER 5. LOW-LOAD HCCI WITH NVO Temperature Std. Dev. [K] Mole Fraction Std. Dev. [%] Crank Angle [ CA] (a) Crank Angle [ CA] (b) Figure 5.5: (a) Measured temperature standard deviation and (b) air mole fraction standard deviation for motored conditions with air intake temperature of 412 K and manifold pressure of 1 bar. 5.3 Measurement Interferences When performing fired NVO engine experiments there are a number of possible signal interferences that can make quantitative measurements difficult. Of particular importance for this work are fuel droplets and residual gas fluorescence. Fuel droplets persisting after injection can cause laser extinction as the beam traverses the cylinder, as well as excessively high signal levels if tracer is present in the fuel. Unwanted fluorescence from exhaust gas species can cause problems by altering the signal ratio and leading to erroneous measurements. This is particularly important for NVO engine operation where high levels of exhaust gas residuals are retained in-cylinder. accuracy. Each of these interferences was investigated to ensure measurement In-cylinder droplet lifetimes were assessed by measuring Mie-scattering over a range of delay times following injection. The measurements were performed using the 308 nm laser sheet alone in the configuration specified in Figure 5.1. A 45 mm f/1.8 UV lens (Cerco 2073) was used for these experiments to transmit the elastically scattered laser light. Initial measurements under cool, motored conditions indicate

99 5.3. MEASUREMENT INTERFERENCES 79 that a large fraction of droplets can persist for up to 80 CAD after injection for early NVO and main injections at SOI +260 and -320 respectively. However, under fired NVO engine conditions, droplet lifetimes fall dramatically due to substantially higher temperature, allowing droplet-free LIF measurements much sooner after injection. For fired operation, Mie images were acquired following NVO injection of 1.5 mg of fuel at +260 CAD, and main injection of 8 mg at -320 CAD. The latter timing was selected to determine if fuel (and tracer) might evaporate fast enough to allow LIF imaging during the re-expansion prior to IVO. Results of the fired tests indicate that, at typical operating conditions, LIF images can be recorded as early as 15 CAD after start of injection. The Mie images were recorded in the same image plane location as the two-line PLIF measurements to ensure consistency. At locations outside of this plane, droplet lifetimes could be different Measurements of residual fluorescence were performed for a range of fired NVO operating conditions to assess the significance of this interference in the three measurement regimes (NVO recompression, NVO re-expansion, and main compression). The experiments were performed at conditions identical to later parametric studies of injection timing and combustion load to identify any operating conditions where interferences may be substantial. Sources of residual interference include partial products of combustion and unburned tracer. Since some combustion intermediates such as aldehydes contain the same chromophore as 3-pentanone they can contribute fluorescence to the measurements. The relative importance of interference from unburned tracer and residual fluorescence can be assessed by comparing results with and without tracer seeding. Figure 5.6 presents this comparison. The single-cycle image pairs were recorded during NVO recompression using the same setup shown in Figure 5.1. Each pair comprises 277 nm and 308 nm fluorescence images that are background corrected, but not corrected for laser profile variation. Figure 5.6a shows the recorded signal with no tracer added, thereby representing partial combustion product fluorescence alone. A weak signal is seen in the 308 nm image, but minimal corresponding signal is visible at the same locations in the 277 nm image. Since the wavelength pair is optimized specifically for the tracer

100 80 CHAPTER 5. LOW-LOAD HCCI WITH NVO Signal 277 nm [counts] 300 Signal 308 nm [counts] (a) (b) (c) Figure 5.6: Single-shot 277 nm and 308 nm LIF images of (a) residual gas recorded at +260 CD (no tracer), (b) carry-over 3P recorded at +260 CAD (prior to fuel injection, and (c) 3P recorded at +285 CAD (following DI fuel injection)

101 5.3. MEASUREMENT INTERFERENCES 81 3-pentanone, the uncorrelated fluorescence seen in the Figure 5.6a image pair is evidence of fluorescence from non-tracer species. A likely candidate is formaldehyde that photo-dissociates disproportionately at the shorter, 277 nm, wavelength. Figure 5.6b presents the signal for the case of tracer seeding of the fuel, but is acquired before NVO injection, thus displaying any residual tracer fluorescence in addition to combustion product fluorescence. Here the signals are approximately an order of magnitude higher than Figure 5.6a with noticeable correlation between the two wavelengths. These results indicate that, in this case, the signal is dominated by unburned 3-pentanone, referred to henceforth as carry-over tracer. It should be noted that this signal from carry-over tracer does not represent a source of error for our fuel concentration measurements (NVO recompression and re-expansion) since it can be interpreted in the same fashion as tracer from the NVO injection. Carryover signal is a potential source of error for EGR measurements (main compression) as the technique accuracy relies on complete tracer consumption. However, for current experimental conditions, the carry-over signal is weak and any resulting error is insignificant. The image pair in Figure 5.6c represents the same experiment as the previous pair, but with image acquisition 25 CAD after fuel injection. Examination of gray scales reveals that these signals are two orders of magnitude higher than those of Figure 5.6a. Based on measurements for a range of operating conditions, we conclude that residual-gas fluorescence is not a significant source of error for quantitative PLIF measurements during NVO recompression. Similar measurements were also performed during NVO re-expansion and main compression, indicating that residual fluorescence does not present a problem in these regimes. Note however, that other operating conditions do result in high interference signal levels from residual gas, specifically conditions with later NVO injections (SOI nvo >+345) where substantial fuel reforming occurs during NVO reaction. The test for interfering fluorescence is straightforward (i.e. LIF imaging without tracer addition) and should be applied for any new operating conditions.

102 82 CHAPTER 5. LOW-LOAD HCCI WITH NVO 5.4 Fired NVO Results Carry-over Tracer Measurements The results of the above interference LIF measurements during NVO recompression indicate that the main source of residual gas fluorescence is unburned 3-pentanone. This tracer is carried over from the previous main combustion and likely originates as unreacted crevice gases ejected into the measurement field of view during the main expansion and exhaust strokes. This carry-over signal offers an opportunity to make temperature and fuel mole-fraction measurements of unburned fuel during the exhaust stroke. Such measurements require that the tracer faithfully follows any unburned fuel through compression, expansion, and exhaust strokes. Simple CHEMKIN simulations as well as prior studies [72] suggest that chemical reaction of 3-pentanone tracer should parallel the reaction of the iso-octane fuel for engine time scales. Thus, it is reasonable to assume that any unburned fuel in crevices, for example, will be accompanied by a proportional amount of tracer. Any departure from this constant ratio of tracer to unburned fuel affects only the composition measurement and not temperature. The following fired experiments illustrate the carry-over tracer measurements. Three main combustion loads of 7, 8 and 9.5 mg/cycle were selected with a constant NVO injection mass of 1.5 mg/cycle. Main injection was set at -270 CAD and NVO injection at +260 CAD. Increases in main combustion load lead to an increase in apparent main heat release and increased residual temperature. Higher residual temperatures can in turn affect apparent heat release and phasing of the subsequent NVO recompression, demonstrating the cyclic feedback of NVO engine operation. Representative temperature and fuel mole fraction measurements of unburned fuel are presented in Figure 5.7 for a constant image timing of +260 CAD (near EVC). Due to low tracer concentrations at these conditions, signal-to-noise ratio (SNR) for these images is low. A 3x3 median filter has been applied to the processed temperature and mole fraction images to improve clarity (applied to these low-snr images only). The images depict characteristic regions of unburned fuel

103 5.4. FIRED NVO RESULTS 83 that are formed as the crevice volume fluid is drawn across the field of view during the exhaust. The unburned fuel is preferentially located on the right (intake) side of the cylinder, likely due to the bulk flow toward the exhaust valves. The temperature images in Figure 5.7 indicate a wide range of temperatures, with cooler temperatures frequently located near the core of the fuel regions where fuel concentration is higher, and hotter temperature located near the perimeter where crevice fuel is mixed with hotter residuals. Average temperatures (calculated over regions with sufficient tracer mole fraction) rise with increasing main combustion load as expected. Fuel mole fraction measurements in Figure 5.7 indicate peak unburned levels of ppm. These values are difficult to verify but are at least consistent with unburned hydrocarbon emissions measurements made during similar experiments. Because the unburned fuel/tracer does not occupy a large portion of the cylinder volume, the measured temperature may not represent the overall average in-cylinder temperature. However the technique does provide insight into temperature and concentration of unburned fuel prior to the NVO recompression. NVO Recompression Measurements The NVO recompression is an important regime that has received only modest optical diagnostic attention [44, 53, 56, 119, 123, 124]. Because exothermic reaction and fuel reforming during the NVO can significantly affect the main combustion event that follows, it is important to understand NVO reactions. The two-wavelength PLIF diagnostic can contribute to understanding by characterizing temperature and fuel distribution prior to recompression reactions. To demonstrate, we measured temperature and fuel mole fraction during NVO recompression for main-combustion loads of 7, 8 and 9.5 mg/cycle. Injection times for these experiments were +260 and -270 CAD for the NVO and main injections respectively. The NVO fuel/tracer is injected near EVC so that LIF images could be recorded during recompression. Representative single-shot results are shown in Figure 5.8 for a constant imagecapture timing of +285 CAD. The spatial variation in local fuel mole fraction is visibly high, consistent with the relatively short time between start of injection and

104 84 CHAPTER 5. LOW-LOAD HCCI WITH NVO Temperature [K] 700 Fuel Mole Fraction [%] (a) (b) (c) Figure 5.7: Single-shot temperature and fuel mole fraction image pairs of carry-over 3P signal during NVO recompression. Image capture timing is +260 CAD for all images. Three main combustion loads: (a) 7 mg/cycle, (b) 8 mg/cycle, (c) 9.5 mg/cycle.

105 5.4. FIRED NVO RESULTS 85 image acquisition. High levels of temperature stratification are also seen, resulting from both fuel evaporation in areas with high fuel fraction, and stratification originating from the exhaust event. (This extensive stratification masks somewhat the typical strong correlation between temperature and mole fraction images as seen for example in Figure 5.10 a, where high EGR regions correlate with high temperatures.) Measured temperatures, averaged over the entire field of view (where tracer is present) of 50 images, follow the expected increasing trend with main-combustion load: the average values are 700 K (Figure 5.8 a), 730 K (Figure 5.8 b) and 780 K (Figure 5.8 c). Concurrent thermocouple measurements of exhaust port temperature were 549 K, 575 K and 604 K for the three loads. Any agreement in temperature measurements may be fortuitous given the small fraction of total cylinder volume being probed, but this does confirm the expected trend of increasing temperature with load. Not surprisingly, the thermocouple values are significantly lower than in-cylinder values since they are time averages of pulsating exhaust flows in a cooled port. However, temperature differences (i.e. T-cylinder - T-port) match well for the lower two loads, while for the highest load the difference increases. Additional measurements were performed at later image capture times during NVO recompression, but the extent of visible tracer signal decreases as fuel passes through the measurement plane under these conditions, limiting the area over which temperatures can be measured. The above results demonstrates the viability of PLIF measurements during NVO; further work is warranted to push these measurements later in the recompression. NVO Re-expansion Measurements NVO heat release and fuel reformation influence the residual temperature and composition, thereby affecting main combustion phasing. Experiments show that tracer from the NVO fuel injection is consumed during NVO reactions. Thus, in order to make PLIF measurements, we devised an alternative strategy based on advancing the main fuel injection to -320 CAD. This allows PLIF imaging as early as -300 CAD (15 CAD before IVO). Temperatures obtained at this time could be useful

106 86 CHAPTER 5. LOW-LOAD HCCI WITH NVO Temperature [K] 900 Fuel Mole Fraction [%] (a) (b) (c) Figure 5.8: Single-shot temperature and fuel mole fraction images during NVO recompression. Image capture time = +285 CAD. Three main combustion loads: (a) 7 mg/cycle, (b) 8 mg/cycle, (c) 9.5 mg/cycle

107 5.4. FIRED NVO RESULTS 87 in understanding the relative importance of thermal and chemical effects of NVO reactions. We verified that fuel injected at this time does not react until the end of main compression, and does not change main combustion phasing compared to a more conventional main-injection timing of -270 CAD. We tested this technique over a range of operating conditions by varying NVO-SOI timing from +260, to +330 CAD. NVO and main fueling were held constant at 1.5 and 8 mg/cycle, respectively. Sample temperature and fuel mole fraction results are shown in Figure 5.9. As expected, there is a high spatial variation in fuel mole fraction due to the short time scale for mixing between injection and image acquisition. A large degree of stratification is also seen in the temperature images, in locations where tracer is present and measurements are possible. This temperature stratification results both from fuel evaporation in areas with high fuel fraction, and from stratification produced during NVO reactions. Average temperatures for the +260, +300 and +330 NVO SOI timings were calculated to be 790 K, 800 K and 800 K respectively. These measurements indicate that there is not a significant change in residual temperatures for the injection timings tested. However, because of the relatively large image area without signal, these measurements may not be representative of the entire field of view or cylinder volume. Additional measurements were performed at later times in the NVO re-expansion, but the effective measurement area decreases as fuel passes through the imaging plane. These measurements demonstrate the feasibility of fuel-seeded measurements of temperature during NVO re-expansion, providing a means to potentially correlate NVO SOI timing with NVO residual temperatures and subsequent main-combustion phasing. Note also that the composition images obtained simultaneously could be used to compensate for evaporative cooling as a way of estimating NVO average temperatures absent the effects of fuel injection. Main Combustion Measurements Measurements of temperature and EGR mole fraction (utilizing the N-PLIF formulation), were performed in the main-compression stroke using intake air seeded

108 88 CHAPTER 5. LOW-LOAD HCCI WITH NVO Temperature [K] 900 Fuel Mole Fraction [%] (a) (b) (c) Figure 5.9: Single-shot temperature and fuel mole fraction images during NVO expansion for early main injection at -320 CAD. Three NVO SOI timings shown: (a) +260, (b) +300, and (c) +330 CAD.

109 5.4. FIRED NVO RESULTS 89 with tracer. Being able to make such measurements is important when studying the mixing of hot residual gases with fresh intake charge prior to combustion. We tested our diagnostic over the following range of injection timings: with main injection held constant at -270 CAD, NVO SOI timing was varied between +260 and +330 CAD; and with NVO injection constant at +260 CAD, main SOI timing was varied between -270 and -100 CAD. Image-capture timing ranged from intake through compression. Representative results for a sweep of image timings are presented in Figure Injection timing for these experiments was NVO SOI = +330 CAD and main SOI = -270 CAD. The amount of fuel injected during the NVO was 1.5 mg/cycle, with a total injection of 9.5 mg/cycle. Typical single-shot images acquired during the intake stroke, Figure 5.10 a, show large non-uniformities in both temperature and EGR mole fraction as the hot residuals mix with cooler intake air. A characteristic pattern is often observed, with cooler intake flow across the cylinder center from right to left (intake to exhaust), and hotter residuals toward the rear of the cylinder (top of image). Early in the compression stroke, Figure 5.10 b, much of the nonuniformity is mixed out. Reduced levels of stratification in both temperature and mole fraction persist through main compression, as evident in the late compression images, Figure 5.10 c. These qualitative comparisons demonstrate the utility of simultaneous EGR and temperature measurements, but a more quantitative means of comparison is required. In an effort to quantitatively compare PLIF measurements for varying operating conditions, we calculated average and standard deviation statistics of temperature and mole fraction for all pixels in the field of view, and averaged over all 50 data images. Standard deviation was selected as an indicator of charge stratificationalthough this eliminates any spatial information regarding stratification, it does track overall fluctuations. Interpretation of standard deviation is further complicated by the convolution of physical standard deviation (i.e., physical, in-cylinder stratification) with diagnostic uncertainty. Because uncertainty varies with imagecapture timing and engine conditions, i.e. with changing signal levels and photophysical parameters, comparisons between engine conditions are challenging. Four calculations of temperature standard deviation (for the experiment of Figure 5.10)

110 90 CHAPTER 5. LOW-LOAD HCCI WITH NVO Temperature [K] 600 EGR Mole Fraction [%] (a) (b) (c) Figure 5.10: Single-shot temperature and EGR mole fraction images recorded at three image timings: (a) -215, (b) -65, and (c) -24 CAD. NVO SOI = +330 CAD; main SOI = -270 CAD. Images at -24 CAD (only) are binned 2x2. Note variable temperature color scale.

111 5.4. FIRED NVO RESULTS 91 are presented in the four data sets of Figure Since the 2x2 binning process suppresses diagnostic noise (which is characterized by a relatively short length scale), we can estimate the magnitude of diagnostic uncertainty by looking at the difference between the Full-Frame and Binned 2x2 data sets in Figure This difference is small at -215 CAD where the diagnostic uncertainty is low and statistics are dominated by physical fluctuations in in-cylinder temperature. Conversely, the difference is higher at -24 CAD due to increase diagnostic uncertainty at higher cylinder temperatures. These results suggest that a method of correction is needed as follows. Two techniques to decouple physical standard deviation from diagnostic uncertainty have been investigated. The first technique (labeled Corrected in Figure 5.11) utilizes the difference of standard deviations for full-frame and 2x2 binned images described above to estimate diagnostic uncertainty. For this technique we assume that the physical standard deviation ( Tphys 2 ) and diagnostic uncertainty ( T diag 2 ) sum in quadrature resulting in the measured standard deviation ( Tmeas) 2 as shown in 5.1 written for the full-frame images. A similar relation also holds for the 2x2 binned images (subscript bin). Subtracting these relations, we arrive at Equation 5.2 which relates the diagnostic uncertainty and measured standard deviation for the full-frame and 2x2 binned images. Equation 5.2 can be further simplified by introducing f, the ratio of diagnostic uncertainty for the full-frame and 2x2 binned images shown in Equation 5.3. Finally Equation 5.4 represents the diagnostic uncertainty as a function of measured standard deviation of the full-frame and binned images and f. Assuming the measurements are in the shot-noise-limited regime, and that diagnostic uncertainty is dominated by camera shot noise, the factor f will be exactly 2. When including additional sources of uncertainty such as laser energy and profile fluctuations, the factor is slightly smaller. T 2 meas = T s phys + T 2 diag (5.1) T 2 meas T 2 meas bin = T 2 diag T 2 diag bin (5.2)

112 92 CHAPTER 5. LOW-LOAD HCCI WITH NVO Temperature Std. Dev. [K] Full Frame Binned 2x2 Corrected Binned Filt Crank Angle [ CA] Figure 5.11: Demonstration of data correction for diagnostic uncertainty. f = T 2 diag T 2 diag bin (5.3) The Corrected data set in Figure 5.11 represents physical standard deviation calculated by subtracting the square of diagnostic uncertainty (Equation 5.4 from the square of the measured standard deviation (full-frame data). The second technique (labeled Binned-Filt. in Figure 5.11) represents a simpler decoupling method that uses a 3x3 median filter (in addition to the 2x2 binning) to more completely suppress the diagnostic noise. The large difference in Full-Frame and Corrected data near TDC of Figure 5.11 indicates, as expected, that diagnostic uncertainty is high in this regime, and that quantitative statistical comparison would be difficult without some form of correction. The similar results of the Corrected and the Binned-and-Filtered techniques indicate that both achieve the intended goal. Although not shown here, the two techniques perform equally well for correction of mole fraction standard deviation. With this correction, we are now in a position to compare average and standard deviation statistics of the NVO test conditions. Figures 5.12 and 5.13 compare data for three operating conditions selected primarily to demonstrate a range of NVO-experiment data. The triangular data points

113 5.4. FIRED NVO RESULTS 93 correspond to the same data from Figures 5.12 and 5.13; the others differ in NVO and main-injection times. Looking first at Figure 5.12a, the closely spaced data points indicate little variation of average temperatures for the three cases presented at the latest image timing (-24 CAD); the temperature spread between the three conditions amounts to 13 K. Examining the trend of the data through the cycle, one sees temperatures dropping during intake (data at -215 through -140 CAD) as hotter EGR mixes with cooler intake air. (Recall that for these N-PLIF experiments, the air is seeded, so that image capture begins only after sufficient air is present in the field of view.) This trend of decreasing average temperature during intake is influenced by the measurement plane location and the relative proportion of EGR and fresh intake. Average temperatures are expected to decrease through intake for a measurement plane initially containing high EGR levels, and increase for a measurement plane initially containing low EGR levels. Later during compression, average temperatures climb rapidly due to compressive heating. Temperature stratification, as represented by standard deviation in Figure 5.12b, is highest early in the intake event before EGR and fresh charge are well mixed. The stratification decreases through the end of intake and into compression, reaching an apparent minimum value at -65 CAD. The stratification then increases slowly through compression. The latter increase is due to both the charge compression, which increases both average temperature and spatial temperature differences, as well as heat transfer with the cylinder walls. Although there are visible differences in stratification for the various engine conditions in Figure 5.12b, it is difficult to discern any significant pattern given the data scatter. However, the standard deviation trends for each of the operating conditions is comparable. The large spread in temperature stratification for the earliest image timing in Figure 5.12b (-215 CAD) merits additional comment. The spread is likely due to differences in the operating conditions in terms of both injection timing and NVO heat release. For the data points corresponding to a main SOI of -270, the fuel has been injected prior to image acquisition, potentially adding temperature stratification through evaporative cooling. Conversely, for the main SOI of -100 CAD, the fuel is injected well after the image capture. NVO exothermicity also varies considerable between these operating

114 94 CHAPTER 5. LOW-LOAD HCCI WITH NVO Temperature [K] SOI NVO = 460 SOI main = 270 SOI NVO = 390 SOI main = 270 SOI NVO = 460 SOI main = 100 Temperature Std. Dev. [K] SOI NVO = 460 SOI main = 270 SOI NVO = 390 SOI main = 270 SOI NVO = 460 SOI main = Crank Angle [ CA] (a) Crank Angle [ CA] (b) Figure 5.12: (a) Measured average temperature and (b) temperature standard deviation for several engine operating conditions. The error bars shown in (a) represent accuracy estimated from motored diagnostic characterization experiments. conditions, understandably producing different stratification results. T diag = T meas 2 Tmeas 2 bin ( 1 1 f 2 ) 1 2 (5.4) EGR mole-fraction measurements for the selected engine conditions are shown in Figure Average EGR mole-fraction values in Figure 5.13a follow similar trends as temperature during intake, where values decrease as EGR mixes with intake air. This indicates that at the current measurement plane, the distribution at -215 CAD contains a high portion of EGR, resulting in decreasing temperature and EGR mole fraction during intake as additional fresh charge is drawn into the cylinder. The observed rise in EGR during late compression requires more explanation. If our diagnostic field of view contains a representative sample of cylinder contents, then average EGR fraction should remain constant. The contrary rise seen in Figure 5.13a could be evidence of vertical stratification in the cylinder, with layers of different EGR fractions being swept through the imaging plane by the rising

115 5.4. FIRED NVO RESULTS 95 piston. Some portion of the EGR fraction increase could also be associated with errors in measured temperature as well as inaccuracies in the FQY model used during calibration, since these propagate through to the mole-fraction calculations. In previous studies [44], this error propagation has been shown to induce errors in average mole fraction of up to 10%, particularly for extreme conditions near TDC. The mole fraction accuracy of 5% reported above holds over a majority of the tested engine range but may be larger at the extreme temperatures near TDC. Finally, decomposition of the tracer would also lead to an apparent increase in EGR mole fraction throughout compression. However, chemical simulations indicate that tracer and fuel decomposition are well correlated, implying that no tracer decomposition is expected prior to the main heat release. In addition, because iso-octane does not exhibit any low-temperature chemistry and our latest image capture timing during compression (-24 CAD) is much earlier than the onset of exothermic reaction, it is expected that tracer decomposition at this time will be negligible. Similar to average temperature data in Figure 5.12a, Figure 5.13a shows no significant differences in average EGR mole fraction for the three conditions presented, indicating that this is no substantial difference in engine breathing. Each EGR standard deviation data set in Figure 5.13b follows a trend generally similar to temperature standard deviation. Stratification is highest for the earliest points, before substantial mixing has occurred. (The increase in stratification at -180 CAD is unexpected and will required a more refined study to determine the cause.) Stratification then decreases, reaching a steady value after -42 CAD. This behavior suggests that mixing has slowed by this point, and this conclusion is consistent with the rising temperature stratification in Figure 5.12b (strong mixing would decrease temperature stratification). As a final observation, comparing the three data sets in Figure 5.13b indicates that there is little difference in EGR stratification for the operating conditions tested. Again, this is not surprising - although injection timing can significantly affect combustion phasing, it has less effect on engine breathing and in-cylinder mixing.

116 96 CHAPTER 5. LOW-LOAD HCCI WITH NVO EGR Mole Fraction [%] SOI NVO = 460 SOI main = 270 SOI NVO = 390 SOI main = 270 SOI NVO = 460 SOI main = Crank Angle [ CA] (a) EGR Mole Fraction Std. Dev. [%] SOI NVO = 460 SOI main = 270 SOI NVO = 390 SOI main = 270 SOI NVO = 460 SOI main = Crank Angle [ CA] Figure 5.13: (a) Measured average EGR mole fraction and (b) EGR standard deviation for several engine operating conditions. Error bars represent accuracy estimated from motored diagnostic characterization experiments. (b)

117 Chapter 6 High-Load HCCI The homogeneous charge compression ignition (HCCI) engine strategy can provide both high efficiency and low emissions. However, the limited load range is a technical challenge that must be addressed prior to widespread implementation. The high-load limit specifically is constrained by high cylinder pressure rise rates (PRR) during combustion that can lead to engine knock as fueling is increased. Previous research has shown that naturally occurring thermal stratification (TS) of the incylinder charge is central to high-load HCCI operation [108, 125]. The dominant mechanism for TS development is thought to result from wall heat transfer and convection of the near-wall cold gas [108]. However, TS may also result from the retained combustion residual gas, and fuel injection. The resulting localized temperature variations in the bulk-gas results in sequential auto-ignition, a process in which the hottest regions ignite first followed by progressively cooler regions [108,126]. This effectively slows the combustion heat release rate (HRR) and could potentially be exploited to expand the high-load limit. While recent studies have begun to investigate the impact of thermal stratification, additional characterization is still needed. Chemiluminescence was previously employed to indirectly study the effects of TS on HCCI combustion [108]. Here, the auto-ignition was found to occur in localized zones that are randomly dispersed within the bulk-gas despite operation with a fully premixed fuel/air charge. 97

118 98 CHAPTER 6. HIGH-LOAD HCCI High-speed chemiluminescence image sequences further showed that reactions occurred sequentially throughout the charge. Although this has been attributed to TS [108, 126], this has not been directly verified. It was also found that TS in the boundary layer has only a secondary effect on the reduction in PRR as the timing of reactions in these regions is well after the maximum PRR [108]. Additionally, multi-zone Chemkin engine simulations demonstrated the necessity of including thermal stratification in HCCI simulations to better match experimental cylinder pressure measurements and apparent heat release (AHR) profiles [125]. In addition, the benefit of a given amount of TS was found to be amplified by retarding the combustion phasing [125]. While these studies do provide evidence of TS, they do not specifically address the distribution, spatial-scale, and evolution of TS during HCCI operation. This fact has motivated the development of several fluorescencebased diagnostics to directly investigate TS. In an initial work, Dec et al. [71] developed a single-line PLIF temperature diagnostic using toluene as the fluorescence tracer with 266 nm excitation. This diagnostic was used to study the distribution and evolution of TS in a motored HCCI engine and represents the first direct quantification of TS. The current work has developed similar single-line [127] and additional two-line [77,89,127] techniques using either 3-pentanone or acetone (both ketones) with excitation at 277 nm and 308 nm. All single-line techniques provide sufficient temperature precision to resolve small temperature fluctuations, and were used to study TS development in the same motored HCCI engine as [71]. These experiments provided validation of the diagnostic techniques, and aided further quantification of the TS development. The current work further extends the application of the ketone-based diagnostics to fired HCCI operation and investigates the differences and similarities between motored and fired conditions. First, a two-line diagnostic for temperature and composition is applied to study the residual mixing process and to determine the potential impact of residuals on TS for low-residual HCCI operation. Low-residual conditions are achieved with conventional valve timing, resulting in a small quantity of hot combustion residuals. Strategies such as negative valve overlap (NVO) [89] that increase residual gas retention are not considered. The impact of combustion residuals

119 6.1. EXPERIMENTAL SETUP 99 is further studied by comparing single-line temperature results acquired for skipfired cycles with and without hot residuals. Next, single-line temperature measurements during the compression stroke are used to characterize the TS development for motored and fired operation. The influence of early direct fuel injection is considered by comparing motored measurement results with fully premixed fuel or early direct injection. In addition, the impact of boundary wall temperature is investigated through comparison of premixed fired results with varying coolant temperature. Finally, spatial correlation between the temperature distribution and subsequent reaction zones are used to study the progression of reaction within the cylinder core, and to demonstrate the importance of TS in HCCI combustion. 6.1 Experimental Setup Optical Engine The high-load HCCI experiments were performed in an optical engine located at Sandia National Laboratories Combustion Research Facility in Livermore, California. The optically accessible engine is derived from a Cummins B-series mediumduty diesel engine that has been converted for single-cylinder operation. Detailed description of the engine and facility is available in [108, 128], and will only be summarized here. An engine schematic highlighting key components is shown in Figure 6.1, and the relevant specifications are listed in Table 6.1. The pancake style combustion chamber (flat head, flat piston-crown) has a 0.98 liter displacement, and a geometric compression ratio of 14:1. Laser sheet access is provided by three windows integrated into a spacer ring that forms the top portion of the cylinder wall. The spacer ring assembly is recessed into the head to provide laser and imaging access up to the firedeck. Laser access through TDC was possible due to the simplified engine geometry (unlike previous experiments in pentroof engine where image timing was limited to 24 btdc due to piston position). A Bowditch piston outfitted with a large piston-crown window (70 mm diameter) and

120 100 CHAPTER 6. HIGH-LOAD HCCI 277 nm 308 nm Figure 6.1: Detailed schematic of optical HCCI engine showing location of laser sheets and imaging camera. right-angle mirror permits imaging access from below. A drop-down cylinder liner with hydraulic-piston activation allowed rapid cleaning of internal window surfaces between experiments. A schematic of the engine facility and related subsystems is shown in Figure 6.2. Intake flow of either air or nitrogen was metered with a sonic-nozzle orifice and was continuously adjusted to maintain a constant intake pressure of 100 kpa for all experiments. The intake stream was electrically heated by a main and auxiliary heater to the desired intake temperature. The auxiliary heater, located near the engine, provided precise temperature control and was used for fired experiments to adjust combustion phasing in real-time. Intake temperatures ranging from 100 C- 205 C were used for calibration experiments, while a constant intake temperature of 170 was used for all motored studies. For fired experiments, the intake temperature was adjusted to maintain CA50 combustion phasing at 367 CA, typically ranging from 200 C-210 C. All motored and fired experiments were taken at an engine speed of 1200 RPM. A summary of engine operating conditions is presented in Table 6.2 Both premixed and direct injection (DI) fuel delivery methods were employed

121 6.1. EXPERIMENTAL SETUP 101 Table 6.1: Engine Specifications Displacement liters Bore 102 mm Stroke 120 mm Connecting Rod Length 192 mm Geometric Compression Ratio 14:1 Clearance Volume Height 8.05 mm No. of Valves 4 IVO 0 CA IVC 202 CA EVO 482 CA EVC 8 CA Swirl Ratio 1.3 Engine Speed 1200 rpm for this study. All motored experiments and a fraction of fired experiments utilized the premixed operation. The premixed fueling system consisted of a positivedisplacement pump that delivered the fuel / tracer mixture to an electrically heated fuel vaporizer shown in Figure 6.2. The vaporizing mixture chamber was located upstream of the intake plenum to ensure a truly homogeneous mixture in-cylinder. Fuel delivery rate was determined by monitoring the mass of fuel delivered (with a lab scale) over 60-second intervals. Direct injection of fuel was also used for a series of skipfired experiments. For this, fuel was supplied by a gasoline-type direct injector (GDI 8-hole, 70 included angle) mounted in the center of the cylinder head. A positive displacement fuel-flow meter was used to monitor the amount of fuel supplied.

122 102 CHAPTER 6. HIGH-LOAD HCCI Figure 6.2: HCCI engine facility schematic PLIF System The PLIF diagnostic system used for the high-load experiments is similar to the system described in Chapter 5, differing only in overall optical path length and optical components. The PLIF diagnostic system is shown schematically in Figure 6.3. Excitation at 308 nm was provided by a XeCl excimer laser (Lambda Physik ComPexPro 102), providing 200mJ max energy output. The 277-nm pulse was generated by Raman shifting the 248 nm output of a KrF excimer laser (Lambda Physik Com- PexPro 102, 400 mj max output) to the 1st Stokes wavelength in H 2. The Raman cell (custom fabrication) was 112 cm in length with a 37 mm clear aperture, and was operated at a total pressure of 57 bar. A 600 mm focal length spherical lens focused

123 6.1. EXPERIMENTAL SETUP 103 Table 6.2: Engine Operating Conditions Intake Temperature C Intake Pressure 100 kpa Coolant Temperature 100 C Equivalence Ratio Base Fuel Iso-Octane Fuel Delivery Fully Premixed Direct Injection Tracer Seeding [% Liq. Vol.]: 3-Pentanone 17% Acetone 14% the 248 nm beam through the Raman cell, while an appropriately positioned 500 mm lens collimated the output. This configuration resulted in approximately 15% energy conversion efficiency to the 1st Stokes, an improvement from past experiments [44,89]. The resulting 277 nm beam was spatially separated from the residual pump beam through a equilateral dispersing prism. A long propagation distance for each wavelength was used to form the sheet width dimension based on the inherent divergence of the laser outputs. After propagating across the optical tables near the laser output, the beams were transported to the engine optical table by two vertical periscopes. Each wavelength was then passed through a 1 m focal length cylindrical lens to form the beam waist, before being spatially overlapped with a dichroic mirror coated to reflect 277 nm and transmit 308 nm. Lastly the beams were sent through the vertical elevator assembly that rotates the beam 90 and elevates the laser sheet to the desired vertical location in the engine. The current engine windows were designed such that the inner and outer radius of curvature is equal. Because of the thickness, the windows acts as a positive cylindrical lens. To counteract the induced beam convergence, a 200 mm cyl. lens was added to the optical path 650 mm upstream of the engine input window. The

124 104 CHAPTER 6. HIGH-LOAD HCCI Figure 6.3: PLIF experimental schematic for high load HCCI experiments Figure 6.4: PLIF image field of view with valves, injector, and piston window positions superimposed. resulting laser sheet diverged slightly across the field of view, increasing in width by approximately 4-5 mm. Final laser sheet dimensions were approximately 43 mm in width and 0.5 mm in thickness. Typical laser energies per pulse during experiments were mj for 277 nm and mj for 308 nm. Laser energies were measured several optical components upstream of the engine, so the actually laser energies in-cylinder are likely 8-10% lower due to surface reflection losses.

125 6.2. ENGINE OPERATION 105 The overlapped sheets were aligned nearly parallel to the firedeck, with a small downward angle to avoid collinear back reflections from the rear engine window. As shown in the inset in Figure 6.3, the last turning mirror in the vertical setup before the engine was affixed to a vertical translation stage that allowed the laser sheet to be adjusted to any desired position below the firedeck. Unless otherwise noted, the laser sheet was positioned at the mid-plane of the combustion chamber for all image timings, ranging from 4 to 16 mm below the firedeck. The resulting fluorescence transmitted through the piston-crown window was imaged onto the camera using an 85 mm f/1.4 Nikon lens. The intensified CCD camera (Princeton Instruments, PIMAX2:1003) is equipped with an interline transfer CCD array allowing the collection of both images with an interframe timing of 5µs (2µs minimum). The 1024x1024 pixel CCD array was binned 8x8 on chip to increase signal level while sacrificing some spatial resolution. The resulting image spatial resolution was approximately 0.55 mm per pixel, roughly matching the outof-plane resolution provided by the sheet thickness. The intensifier gate width for all experiments was set to 900 ns, and was synchronized to the laser output using a pulse delay generator (SRS DG535). The delay generator was itself synchronized with the engine controller permitting image acquisition at any desired image timing. The camera and laser systems were operated at 10 Hz for all experiments, allowing one image acquisition per cycle at the 1200 RPM engine speed. 6.2 Engine Operation Motored experiments with fully premixed seeding for a range of intake temperatures were used to calibrate the fluorescence dependence on temperature for both excitation wavelengths. More fundamentally, these measurements were used to characterize the FQY at elevated temperature and pressure as was presented in Chapter 3. In addition, premixed motored operation was used to study the fundamental development of thermal stratification for a constant intake temperature of 170 C. For all motored experiments the engine was statically preheated to 100 C by electrical

126 106 CHAPTER 6. HIGH-LOAD HCCI heaters on the cooling water and lubricating oil circulation systems. Prior to data acquisition, the engine was motored until the measured cylinder wall temperature reached steady-state, typically 3-5 minutes. During this time the intake mass flow was adjusted to maintain a 1 bar intake pressure, and the flow of the premixed fuel/tracer was adjusted to provide an equivalence ratio of 0.4 (based on intake mass flow). Thermal stratification was also studied under fired operation to assess any fundamental similarities and differences between motored and fired operation. Initial experiments utilized a 17/3 skipfired (17 motored cycles followed by 3 fired cycles) to minimize thermal loading on the optical engine. Before firing the engine a similar warm up procedure was used as above to elevate the cylinder wall temperature prior firing. Once fuel injection is initiated, the air flow and amount of injected fuel are adjusted to maintain 1 bar intake pressure and an equivalence ratio of 0.4. Intake temperature was also adjusted to set combustion phasing based on real-time calculations of CA50. Data acquisition was initiated once engine operation was stable. Skipfired operation in this fashion resulting in reasonably steady fired operation, and sufficiently long run times for data acquisition. Additional premixed fired experiments were completed to eliminate any impact of the direct injection above. For these experiments the equivalence ratio was lowered to 0.32 to reduce the thermal loading on the engine for the continuous firing. 6.3 Data Acquisition and Processing Conventional Data Acquisition Several conventional (non-optical) diagnostics were used to monitor and characterize engine performance. In-cylinder pressure measurements were performed with a transducer (AVL QC33C) mounted in the cylinder head. The pressure transducer output was recorded for 50 consecutive cycles at 1/4 CA resolution and was adjusted (pegged) to match the intake pressure near BDC where the cylinder pressure is effectively constant. A camera reference pulse was recorded with pressure to

127 6.3. DATA ACQUISITION AND PROCESSING 107 provide a cycle-specific indication of PLIF image acquisition. This allowed investigation of relative trends in engine performance and PLIF data on a cycle-to-cycle basis. For fired conditions, the calibrated pressure was used to calculate the apparent heat-release rate (AHRR), and the 50% burn point (CA50) of the cumulative AHRR. Calculations were performed for each cycle assuming constant specific heats and negligible heat transfer [129], and then averaged for the 50 cycles. Real-time calculations of AHRR and CA50 were used to adjust intake temperature to maintain CA50 at approximately 367 CA (7 atdc). Intake temperatures were measured using thermocouples mounted in the two intake runners near the cylinder head. Exhaust gas temperature was also monitored with a similar thermocouple in the exhaust port. Based on these measurements, incylinder BDC temperature was estimated following the procedure provided in [102]. Also, upper cylinder-wall temperatures were inferred from a thermocouple embedded in the spacer ring approximately 2 mm beneath the surface and 5 mm below the firedeck. The crank angle (CA) convention for all data presented is 0 <CA<720, with 360 CA corresponding to top dead center (TDC) of combustion PLIF Data and Processing The current study of thermal stratification in HCCI engines includes application of both single-line and two-line PLIF diagnostics. These measurements are derived from the same experimental setup as shown in Figure 6.3. For all experiments the signals from both excitation wavelengths are recorded, and implementation of either diagnostic strategy is only a matter of post-processing. This provides an easy means of comparing the diagnostic performance for identical conditions. The theoretical derivation and post-processing for both PLIF diagnostic variations were described in Chapter 2. Three separate image sets were acquired for the quantitative PLIF image processing: a data, flat-field and background set. Each data set consisted of 50 images pairs taken over consecutive cycles at the desired imaging timing. Flat-field (calibration) images are taken for motored conditions of known temperature and pressure,

128 108 CHAPTER 6. HIGH-LOAD HCCI and are used to correct for fixed pattern (consistent shot-to-shot) non-uniformities of the camera and laser sheet. The flat-field images are acquired shortly after the data images to minimize the impact of laser profile fluctuations (typically within seconds). Finally background images were acquired in the motored engine, with laser emission, but without tracer present to correct for any laser scatter or fixed pattern background. Both the flat-field and background images are averaged over the 50 cycles prior to data processing. The acquisition timing of the flat-field images requires additional discussion. For the current work, two different flat-field images have been tested. The first was acquired at BDC under motored conditions, and was verified to have a flat and homogenous distribution at known in-cylinder conditions. This flat-field image timing is held constant, and used to process subsequent data for all image timings. Use of this flat-field provides an absolute temperature measurement at the desired data image timing. Image processing in this fashion does require a laser attenuation correction along the beam propagation, as the absorption cross-section and tracer number density increase during compression. A similar calibration image timing was applied in Chapter 5. A second flat-field was derived from the average of the 50 data images. Use of this flat-field results in measure of the relative fluctuations about the average temperature for a given image timing. Here the average temperature is calculated based on adiabatic compression of the measured pressure tracer assuming variable specific heats. The flat-field normalization accurately corrects for laser attenuation (absorption identical for data and averaged data images), and no additional correction is needed. However the uniformity of this average derived flat-field must be considered. If the temperature-induced fluctuations of fluorescence signal are relatively small and completely random, sufficient averaging would eliminate any variation producing a homogeneous flat-field. If the fluctuations are not random, a fixed pattern in the averages would result, and normalization by this image would eliminate the fixed pattern temperature distribution and alter the thermal stratification statistics slightly. Comparison of the BDC flat-field and data-averaged flat-field (attenuation corrected) show minimal differences. In addition, temperature statistics

129 6.3. DATA ACQUISITION AND PROCESSING 109 derived from the processed temperature data are very similar using either flat-field, with only a slight difference associated with small errors in the attenuation correction. A similar flat-field normalization was applied by Dec et al. [71], who found that the averaged data images were very uniform (no detectable repeating pattern) and would not impact the measured thermal stratification. This average derived flat-field is only applicable to the thermal stratification study where fluctuations are expected to be random. The main focus of the current study is to quantify the amount and evolution of thermal stratification during compression. This is related more to the relative fluctuations in temperature, making absolute measurement of temperature less critical. Therefore, the data-averaged flat-field will be used for a majority of the image post-processing, and will be indicated by the relative temperature display (± 50 K). Processing with the alternative flat-field provided nearly identical results. For the fired measurements of residual gas mixing during intake and early compression, the temperature distribution is not random, and the motored BDC temperature must be used. These measurements can be differentiated by the absolute temperature scale used for image presentation PLIF Data Processing Photophysics The accuracy and precision of quantitative PLIF measurements is highly dependent on the photophysical data or model used to interpret variation in fluorescence signal. The absorption cross-section data is based on fit parameters summarized in Appendix A. The FQY data was determined based on motored calibration experiments performed in this engine over a range of intake temperatures. The results of these calibration experiments were presented in Chapter 3. The current processing routine utilizes FQY data fits at each image timing, instead of a tuned FQY model as was previously employed in Chapter 5. This technique is assumed to be

130 110 CHAPTER 6. HIGH-LOAD HCCI more accurate. The in-cylinder temperature range of the FQY calibration measurement conditions were carefully selected to bound the expected temperature range for subsequent experiments, in an effort to avoid errors due to data fit extrapolation. 6.4 PLIF Measurement Uncertainty Accuracy Absolute temperature accuracy was initially assessed by comparing the measured PLIF temperature with estimated temperatures assuming adiabatic compression of the measured in-cylinder pressure trace. This temperature calculation method indirectly accounts for global heat transfer through the measured pressure trace, but assumes an adiabatic core region. Comparison results are shown in Figure 6.5, for motored engine operation with a nitrogen intake temperature of 170 C and 1 bar manifold pressure. The average PLIF temperatures were calculated over a central 30x30 pixel subregion, and averaged over the 50 image sequence. The PLIF derived temperatures are in excellent agreement with the adiabatic temperature calculations for all diagnostic variations, and are within 1% throughout compression. This result is not surprising given that the FQY fit data used for image processing was generated from engine calibration experiments assuming the same adiabatic temperatures within the cylinder core. Therefore, the temperature comparison in Figure 6.5 does not completely characterize the absolute accuracy, but does confirm the consistency of the FQY fits and image processing technique. The absolute accuracy is mainly dominated by the adiabatic core assumption used for engine FQY calibration, and is difficult to assess. Subsequent temperature images (e.g. Figure 6.7) indicate that the core interrogation region is not completely adiabatic near TDC, as evidenced by the cold gas pockets within the field of view. Because these pockets are randomly dispersed across the cylinder volume, the average temperatures are likely somewhat lower than the adiabatic temperatures. Assuming the hottest regions within each temperature image are effectively adiabatic, the measured temperature spread provides some indication of error induced

131 6.4. PLIF MEASUREMENT UNCERTAINTY 111 Temperature [K] Premixed N 2 Intake P in =1 bar T in =170 C φ= nm(±2% error bars) nm 277 / 308 nm Adiabatic Comp Crank Angle [ CA] Figure 6.5: Measured absolute temperature for motored engine conditions with N 2 intake temperature of 170 C and 1 bar manifold pressure by the adiabatic assumption. At 360 CA, the highest temperature regions are typically K above the average, meaning the measured average temperatures are likely K low at 1000 K. This error decreases for earlier image timings as the adiabatic assumption is assumed to be more accurate at lower temperatures. Based on this, the absolute accuracy is conservatively estimated to be ±2.5% throughout compression, and is represented in the error bars in Figure 6.5. It should be noted that the errors associated with the adiabatic assumption will only affect the absolute value of temperature and will have little affect on the measured fluctuations about the average. The accuracy could potentially be improved by altering the FQY calibration measurements to only consider the higher temperature regions where the adiabatic assumption is more accurate. However, the main focus of the current work is to assess the evolution and distribution of thermal stratification, making the highly accurate absolute temperatures measurements less critical.

132 112 CHAPTER 6. HIGH-LOAD HCCI Precision Measurement precision (spatial) is critical when studying thermal stratification as this indicates the minimum detectable temperature fluctuation for a single-shot image. If the precision limits are much greater than the actual in-cylinder temperature fluctuations, resolution of the coherent structures will be poor and any statistical analysis will be difficult. The thermal stratification for fully premixed HCCI is expected to be relatively small, motivating the development of highly sensitive diagnostics. The theoretical temperature precision was investigated in Chapter 3 for excitation wavelength optimization. This analysis indicated that the main factors controlling the measurement precision were the image signal-to-noise ratio (SNR), the inherent tracer temperature sensitivity, and the random fluctuations in laser sheet profile. Similar precision calculations are provided here, but have been updated to include the engine FQY measurements of Chapter 3, and actual fluorescence signal levels achieved in HCCI engine experiments. The fluorescence signal was measured in the motored engine over a range of image timings ( ) during the compression stoke, for an intake temperature of 170 C. The measured cylinder pressure trace and calculated temperature during compression are shown in Figure 6.6a. The resulting image SNR was determined by a camera calibration, relating digitized read out signal and SNR over the dynamic range of the camera. Results and further discussion of the camera calibration are provided in Appendix C. The single-line precision was calculated using Equations , while the two-line precision was calculated applying Equations B.5-B.6. Precision calculations were performed for both acetone and 3-pentanone for comparison. The semi-empirical temperature precision results are presented in Figure 6.6b, and provide a substantial amount of insight and information. The precision for most diagnostic schemes varies significantly throughout the engine cycle, changing by a factor of two or more in some cases. This variation is not unexpected given the large span of in-cylinder conditions, but further emphasizes the importance of diagnostic optimization for the intended range of experimental conditions. Considering the twoline results first, the 3-pentanone based diagnostic provides the best performance at

133 6.4. PLIF MEASUREMENT UNCERTAINTY 113 Pressure [bar] 30 Premixed N 25 2 Intake P in =1 bar T 20 in =435 K φ= Crank Angle [ CA] (a) In-cylinder conditions Temperature [K] Temperature Precision (1σ) [K] P 277 3P 308 3P 277/308 AC 277 AC 308 AC 277/ Crank Angle [ CA] (b) Temperature Precision Figure 6.6: (a) In-cylinder pressure and temperature (calculated) for the baseline HCCI operating condition. (b) Temperature precision of single-line and two-line diagnostics of 3-pentanone ad acetone, calculated for baseline conditions lower temperatures and pressures (early image timings) and progressively degrades towards TDC. The increase in temperature uncertainty is due to a combination of decreasing 277 nm signal, and reduced temperature sensitivity of the ratio of photophysical parameters (R pp ). Conversely, highest acetone uncertainty occurs early in the cycle, reaches a minimum (340 CA) and then increases slightly towards TDC. This is a result of initially low fluorescence signal and sensitivity that increases towards TDC. The performance cross-over point between acetone and 3-pentanone is approximately 340 CA (890 K, 16 bar). As shown in Figure 6.6b, single-line temperature precision is much improved over the two-line technique (308 nm excitation of acetone excepted). The performance improvement at high temperatures (late compression) is most significant, where the temperature precision is improved by a factor of 4-5. Explanation for superior singleline diagnostic performance are threefold. First, single-line temperature is derived from one single-shot image, while the two-line technique involves the ratio of two single-shot images (both are normalized by an averaged flat-field image with high SNR, and does not contribute to the noise content). The resulting two-line ratioed image will have a higher SNR, based on the square root of the sum of squares of SNR

134 114 CHAPTER 6. HIGH-LOAD HCCI for each image, leading to higher temperature precision. Secondly, the temperature sensitivity of the two-line technique is inherently less than the single-line sensitivity. Ketones generally exhibit decreasing fluorescence with temperature (per unit mole fraction) at almost all wavelengths due to the increasing non-radiative decay rate and thus decreasing FQY [75,79]. Ratioing these signals only reduces the sensitivity, thus increasing the measurement uncertainty. Lastly, the single-line technique is sensitive to density gradients through the inverse temperature dependence of P P s. This density term is eliminated for the two-line technique through the signal ratio. It should be stressed however that the two-line technique is more versatile, and can be applied to systems with varying temperature and tracer concentration, while the single-line technique is limited to compositionally homogenous mixtures. Among the potential single-line variations, 277 nm excitation of 3-pentanone provides the best overall performance, with temperature uncertainties of 5 K or below throughout compression. Excitation of 3-pentanone at 308 nm also provides excellent performance at higher temperatures. The high simultaneous performance of both 3-pentanone excitation wavelengths at high temperature provides a unique ability to acquire two temporally or spatially shifted temperature measurements within a single engine cycle. This capability will be investigated further in Section Acetone excitation at 277 nm also provides good performance, while the 308 nm performance at low temperatures is inadequate. Based on these results, all subsequent engine studies will focus on the application of 3-pentanone based diagnostics at 277 and 308 nm. The theoretical uncertainty analysis in Chapter 2 indicated similar overall trends, but the absolute magnitude in temperature precision was not exact due to over-predicted fluorescence signals and SNR, and inaccuracies in the FQY models used for calculation. Despite this, the resulting optimized line selection was the same for both calculations.

135 6.5. MOTORED ENGINE THERMAL STRATIFICATION Motored Engine Thermal Stratification Several mechanisms can contribute to the development of thermal stratification including wall heat transfer and turbulent convection, direct fuel injection, and retained residuals. Wall heat transfer is expected to be the most dominant. As the in-cylinder temperatures begin to rise during compression, the temperature differential between the cylinder wall and bulk-gas provide a driving force for heat transfer. This cooler near-wall fluid can subsequently convect into the cylinder core, resulting in significant thermal stratification throughout the charge. Direct fuel injection can further increase thermal stratification through evaporative cooling, and variable compression due to localized thermodynamic gamma variation with fuel concentration. In addition, any retained residuals from a previous fired cycle can also add to thermal stratification, depending on the mixing efficiency. For the premixed motored experiments considered in this section, only the wall heat transfer mechanism is active (no DI fuel or combustion). Other mechanisms will be discussed further in the fired engine results section. In-cylinder temperature distributions were measured using the single-line 3- pentanone diagnostic for motored engine operation with a 170 C and 1 bar intake. A pure nitrogen intake stream was used to suppress combustion, however premixed fuel was added at an equivalence ratio of 0.4 to better represent the ratio of specific heats (gamma) of the mixture and the resulting compression temperatures. These conditions are identical to those used for the measurement precision assessments above, and the pressure and temperatures profiles are represented in Figure 6.6a. Because these conditions are run fully premixed with no hot retained residuals (motored), the temperature distribution is expected to be homogenous during early compression. Thermal stratification induced by wall heat transfer will not be significant until the gas temperature is well above the wall boundary temperature, and even then additional time is required for fluid motion to convect the cooler near-wall fluid into the core region of the charge. As a result, the TS development is thought to predominately occur in the latter portion of compression. To assess the development of TS, a sequence of single-shot temperature images

136 116 CHAPTER 6. HIGH-LOAD HCCI from CA are presented in Figure 6.7. Each image was selected to represent the average extent of temperature fluctuations for that image timing, and are plotted in terms of temperature difference about the average ( T). A constant image color scale of ±50 K was used for easy comparison of temperature fluctuations at each timing. As expected, the measured temperature distribution early in compression ( CA) is essentially homogeneous. Initial emergence of localized hot and cold pockets is seen at 330 and 340 CA, followed by a progressive increase in TS towards 360 CA. Increase in TS during this time results from an increase in both the frequency of the hot and cold regions, and the amount overall temperature difference between these regions. These observations are consistent with previous temperature measurements in this engine using a toluene-based, single-line PLIF diagnostic [108], and supports related KIVA simulations of a matching all-metal engine [63]. It is important to note that the small variations in temperatures depicted in Figure 6.7 would be difficult to discern without a diagnostic scheme with sufficient performance. A visual demonstration of the superior resolving power of the singleline technique is shown in Figure 6.8 for the 360 CA image timing. Here single-shot 277 and 308 nm single-line temperature are compared with the corresponding twoline 277/308 nm result processed from the same data images). The temperature distribution between both single-line measurements is in good agreement, confirming the consistency of the FQY data used for calibration. Both variations provide low temperature uncertainty on the order of 4 K, and the thermal stratification structures are well resolved. This is in contrast to the two-line results, Figure 6.8c, which has a substantially higher uncertainty of 25 K. Although the general hot and cold structures can be observed, the increased measurement noise masks much of the fine structure. In addition, the high noise level makes statistical analysis of the two-line results difficult and inaccurate. As a result, single-line measurements have been applied whenever possible (i.e. homogeneous tracer mole fraction distribution).

137 6.5. MOTORED ENGINE THERMAL STRATIFICATION 117 T=T T AVE (a) 305 CA (b) 320 CA (c) 330 CA (d) 340 CA (e) 345 CA (f) 350 CA (g) 355 CA (h) 360 CA Figure 6.7: Single-shot temperature images of TS development during main compression of motored engine. Diagnostic - 3-Pentanone, single-line 277 nm

138 118 CHAPTER 6. HIGH-LOAD HCCI T=T T AVE (a) 277 nm (b) 308 nm (c) 277/308 nm Figure 6.8: Comparison of single-line and two-line image quality Motored Stratification Statistics To facilitate a quantitative characterization of thermal stratification, a number of statistical parameters have been employed to analyze the PLIF data. Standard deviation has been selected as a metric to characterize the magnitude of temperature fluctuations for each image timing. The standard deviation was calculated over all active pixels (masked image area shown in Figure 6.8) within each image, and averaged for the 50 image sequence. A statistical correction was also applied to correct for the impact of measurement uncertainty on calculated image statistics. Despite the good performance of the single-line diagnostics, evidence of measurement noise can still be observed through the speckle or graininess that is particularly evident in the 305 CA images. Although these shot-noise induced fluctuations are small, they can have a large impact on statistics if not properly accounted for. Fortunately the temperature uncertainty has already been well characterized, as discussed in Section and seen in Figure 6.6. Assuming the measurement uncertainty is random and independent of the measured temperature field, it can be subtracted in quadrature from the averaged image standard deviation, providing a better indication of the physical temperature fluctuations in-cylinder. Without this correction, the varying noise content would

139 6.5. MOTORED ENGINE THERMAL STRATIFICATION 119 Temperature Std. Dev. (1σ) [K] Uncorrected Std. Dev. Meas. Uncertainty Corrected Std. Dev Crank Angle [ CA] Figure 6.9: Demonstration of measurement uncertainty correction, comparing corrected and uncorrect temperature standard deviation for single-line 3-pentanone at 308 nm excitation. fictitiously alter the trends of thermal stratification, and would make comparison of results from differing diagnostics and parametric studies difficult. A demonstration of the impact of this uncertainty de-convolution is shown in Figure 6.9, for the single-line 3-pentanone measurement at 308 nm. Here the uncorrected standard deviation at early image timings ( CA) is dominated by measurement noise, as indicated by the nearly equivalent uncorrected standard deviation and measurement uncertainty. Upon deconvolution, the corrected standard deviation more accurately follows the qualitative TS observations from images above. The impact of the noise correction decreases towards TDC as both the measurement uncertainty improves and the physical TS increases. This measurement uncertainty correction has been applied to all subsequent standard deviation data to best quantify the TS development. The uncertainty-corrected temperature standard deviations are shown in Figure 6.10 for the baseline motored compression conditions. Data for 3-pentanone at 277 and 308 nm as well as acetone data at 277 nm have been included for comparison.

140 120 CHAPTER 6. HIGH-LOAD HCCI Temperature Std. Dev. (1σ) [K] Premixed N 2 Intake P in =1 bar T in =170 C φ=0.4 3P 277 nm 2 3P 308 nm AC 277 nm Crank Angle [ CA] Figure 6.10: Temperature standard deviation (corrected) calculated from singleshot temperature images for both 3-pentanone (3P) and acetone (Ac) for different excitation wavelengths. The standard deviations for all measurement techniques are in good agreement, particularly for later image timings, and confirm the consistency of the measurements and noise correction. The slightly larger scatter in measurements at 305 and 320 CA is due to the increased error in noise correction when uncertainties are on the order or larger than the physical temperature fluctuations. Overall, the qualitative trends seen in the temperature images of Figure 6.7 are well represented. The standard deviation for all techniques are near 0 K at 305 and 320 CA, consistent with the minimal stratification seen in images for these timings. The standard deviation then progressively increases toward TDC, achieving a maximum of approximately 10 K at 360 CA. These general TS trends are consistent with single-line toluenebased PLIF temperature measurements of Dec et al. [71] previously performed in the same optical engine. In an effort to characterize the spatial scale of the thermal stratification, a simplified calculation of the cold region effective diameter has been performed. The cold regions or pockets have been selected for study as they can reduce the pressure

141 6.5. MOTORED ENGINE THERMAL STRATIFICATION 121 (a) Single-shot temperature (b) Binary cold pocket Figure 6.11: Demonstration of image binarization and pocket detection (b) for a single-shot temperature distribution (a) acquired in motored engine. rise rate through sequential auto-ignition, and are important for operation near the high-load HCCI limit. To calculate the effective diameter, each single-shot temperature image was first binarized to highlight the cold pockets as seen in Figure A threshold of 10 K below the average temperature at each image timing was selected for the binarization criterion. Next a pixel connectivity image processing routine was applied to identify each object and determine the corresponding cold pocket area. Sample results of the object detection scheme are presented as the green outlined elements in Figure 6.11b. Lastly, the cold pocket diameter was calculated assuming a circular cross-section of the calculated area. In actuality, the cold structures have an irregular shape, and a more rigorous auto-correlation analysis would be required to completely define the spatial scale, but has not been completed here. In addition, the effective diameter only considers the projected area within the probe laser sheet, but is still thought to represent the general evolution of cold regions during compression. The frequency of cold pockets at each image timing was also calculated, based on the cumulative sum of cold regions identified in the 50 image sequence. To minimize the influence of small-scale measurement noise, only cold pockets with an area greater than or equal to 3 mm 2 were considered. The cold pocket effective diameter and frequency during motored compression are shown in Figure The effective diameter shown here, is derived from the average of the top 5%.

142 122 CHAPTER 6. HIGH-LOAD HCCI Cold Pocket Frequency [a.u.] Premixed N 2 Intake P in =1 bar T in =170 C φ=0.4 T thres =T ave 10 K Crank Angle [ CA] Figure 6.12: Evolution of cold pocket frequency and effective diameter during motored compression at 170 C intake temperature. Statistics derived from single-line 3-pentanone measurements at 277 nm. 6.6 Fired Engine Thermal Stratification The TS studies described above were restricted to motored operation as it provided a well-controlled measurement environment for initial diagnostic studies. The absence of combustion in these studies was considered acceptable given that TS is formed prior to combustion and is not directly influenced by subsequent chemical reactions. However, these studies raise the question of exactly how the TS will vary when transitioning to fired operation. Specifically, the presence of residual gases is a key difference between motored and fired operation that could impact the relative TS behavior. The current engine configuration is characterized as low-residual HCCI due to the conventional valve timing and low residual gas fraction (4-6%). Additional hot residuals are not intentially retained to enhance auto-ignition as was done in Chapter 5. However, the small quantity of retained residuals could increase the TS during compression if the charge mixing is incomplete. This is important because any additional thermal stratification that persists near TDC will further reduce the PRR and could alter the combustion phasing Effective Diameter [mm]

143 6.6. FIRED ENGINE THERMAL STRATIFICATION Residual Mixing This study of the residual mixing process was completed in two phases. First, the two-line PLIF diagnostic was utilized to provide simultaneous temperature and mole fraction distributions to directly visualize the residual mixing process. Second, single-line PLIF measurements of temperature during the remainder of compression were employed to characterize the overall TS development. Using these single-line results, the impact of residual gas mixing was indirectly studied by comparing TS results for cycles during the skipfired sequence with and without hot residuals. The residual mixing study was performed using a 17-3 skipfired sequence (17 motored cycles followed by 3 fired cycles), as seen in the pressure traces in Figure The skipfired sequence was selected to reduce the thermal loading on the optical engine, and permitted prolonged, steady operation at the high-load equivalence ratio of 0.4. Direct fuel injection occurred during cycles only and ensured skipfired operation. The three sequential fired cycles were chosen to ensure near steady-state retained-residual temperature and quantity. The importance of using three fired cycles is reflected in the peak pressure of the fired cycles in Figure 6.13, which increases significantly between cycles Here the residual temperature increases from cold air to hot combustion products which advances the combustion phasing and thus increases the peak pressure. The small peak pressure difference between cycle 19 and 20 confirms that near steady-state operation is achieved. The in-cylinder temperatures and pressures for cycle 20 of the skipfired sequence are shown in Figure The core temperature was estimated based on the measured pressure trace, assuming adiabatic compression with variable specific heats, and is identical to the technique used for the FQY calibration experiments. The open circles in Figure 6.14 correspond to the image timings used for the current study. These temperature and pressure profiles are nearly identical to those for the premixed motored and continuously fired experiments, with only a small variation in magnitude.

144 124 CHAPTER 6. HIGH-LOAD HCCI Motored Fired Cylinder Pressure [bar] Cycle 1 Cycles 2 16 Cycle 17 Cycle 18 Cycle 19 Cycle 20 Cycle 1 Figure 6.13: Measured cylinder pressure tracers for 17-3 skipfired operation. Direct Imaging of Residual Gas Mixing Simultaneous two-line temperature and air mole fraction measurements were performed during intake and early compression to track the mixture evolution. Application of the two-line technique was required during this portion of the engine cycle as substantial compositional stratification is expected, particularly during intake stroke. The tracer was fully premixed with the intake air to provide a direct measurement of air mole fraction, while pure iso-octane was directly injected with a start of injection (SOI) of 80 CA. All images contained in the residual mixing study were taken with a laser sheet positioned 16 mm below the firedeck. Sample single-shot temperature and air mole fraction image pairs acquired during the intake stroke from CA are shown in Figure Note that the temperature and mole fraction color scale change for the latter images to better emphasize the stratification. As expected, the earliest images show a high degree of stratification in both temperature and mole fraction, as the colder inducted air mixes with the hot retained residuals from the previous combustion event. The image pairs exhibit good spatial correlation and exhibit an inverse relationship between temperature and air mole fraction, where high-temperature regions correspond to

145 6.6. FIRED ENGINE THERMAL STRATIFICATION 125 Pressure [bar] Skipfired P in =1 bar T in =198 C φ= Temperature [K] Crank Angle [ CA] Figure 6.14: In-cylinder conditions for fired studies. The core temperature is estimated from measured pressure trace assuming adiabatic compression with variable specific heats. Data points correspond to PLIF image timings. lower air mole fraction (higher residuals) and vice versa. Additionally, the residual gas mole fraction is systematically higher in the upper left corner of the images, corresponding to the exhaust valve orientation within the field of view as shown in Figure 6.4. As seen in Figure 6.15, the amount of stratification rapidly decreases as image acquisition progresses through the intake stroke, and by 100 CA (middle of intake event) only a small fraction of the initial stratification persists. Mixing continues through intake valve closing (IVC) and into early compression but is less dramatic. By 220 CA the residual mixing process is essentially complete, and the temperature and air mole fraction distributions are effectively homogeneous. Later in compression, thermal stratification begins to develop, as seen by the hot and cold regions in the 305 CA and 320 CA images. These variations are not due to hot residual gases, as the air mole fraction distribution is homogeneous (see images) and remains homogeneous for the rest of the compression stroke. Instead, these

146 126 CHAPTER 6. HIGH-LOAD HCCI temperature fluctuations presumably result from near-wall heat transfer and convection [71]. The observed onset of TS development, is similar to the motored engine data shown in Figure 6.7. The results of the two-line residual-mixing study have two main implications. First, the residual mixing process for low-residual operation is fast, so it is completed by early compression, and it is not likely a significant source of TS near TDC. Second, because of the homogeneous composition after 220 CA, single-line PLIF temperature measurements can be applied without significant error due to mole fraction inhomogeneities. This is particularly useful given the dramatic improvement in measurement precision for the single versus two-line techniques (5 K versus 20 K respectively) as previously demonstrated in Figure 6.6b. As a result, additional single-line PLIF studies of TS development in the fired HCCI engine were performed, as described below. TS Development in Fired Engine Single-line temperature results during compression for fired operation are presented in Figure These images correspond to the same skipfired conditions shown above for the residual mixing study, with pure tracer seeded into the intake air stream and iso-octane fuel directly injected early in the intake stroke (SOI=80 CA). The DI fueling does not impact the temperature measurements given the small fuel mole fraction for the conditions tested. The fired image sequence has been truncated at 350 CA (instead of TDC) as the onset of tracer decomposition is expected to occur during the early reactions after this crank angle. The onset of decomposition has been indirectly validated through PLIF images of early formaldehyde formation. Previous studies have demonstrated that formaldehyde production can be a good indicator of tracer removal [89, 130]. A series of experiments were performed without tracer seeding to image formaldehyde fluorescence using 308 nm excitation (not shown). Images acquired for progressively advancing image times near TDC indicated the onset of formaldehyde production to be after 350 CA, bounding the window for quantitative PLIF measurements. As shown in Figure 6.16, small temperature non-uniformities are initially seen at 305 CA and

147 6.6. FIRED ENGINE THERMAL STRATIFICATION 127 Temperature [K] Air Mole Fraction [%] Temperature [K] Air Mole Fraction [%] (a) 20 CA (e) 160 CA (b) 40 CA (f) 220 CA (c) 60 CA (g) 305 CA (d) 100 CA (h) 320 CA Figure 6.15: Single-shot temperature (left) and air mole fraction (right) image pair sequence showing evolution of residual gas mixing during the intake and early compression for fired HCCI operation. Diagnostic: two-line 277/308nm, 3-pentanone.

148 128 CHAPTER 6. HIGH-LOAD HCCI 320 CA, and progressively increase in both magnitude and frequency during the remainder of compression. By 350 CA a significant amount of thermal stratification is observed, with peak-to-peak temperature fluctuations on the order of 25 K. These large temperature variations presumably lead to sequential auto-ignition, and are critical for high-load operation. A comparison with the motored engine images in Figure 6.7 and those in Ref. [71] shows that the development of TS is similar for fired and motored operation. The 17-3 skipfiring sequence provides a unique opportunity to directly study the impact of residuals on TS development. As previously mentioned, the 3rd fired cycle (cycle 20) has a near steady-state amount of residuals due to the previous two fired cycles. The first fired cycle (cycle 18) however, is preceded by a motored cycle and does not contain any hot combustion residuals. Therefore, comparison of images and statistics from cycle 18 and 20 can elucidate any fundamental impact of hot residuals on TS. This comparison is particularly useful as the boundary wall temperature is effectively constant for both cycles, and will not impact the relative trends in TS. Single-line temperature measurements have been performed in both cycle 18 and cycle 20 within the same skipfired sequence. For better quantitative comparison, the temperature standard deviation was calculated for each individual image and averaged over a 50 image series. A measurement uncertainty correction was applied to the calculated standard deviation to de-convolve the physical temperature fluctuations and the measurement precision. Without this correction, the measurement noise would fictitiously increase the apparent stratification and the calculated statistics would not reflect the actual in-cylinder temperature variation. To apply this correction, the measurement precision was first estimated assuming the dominant sources of measurement uncertainty arise for the image signal-to-noise ratio (SNR) and the random laser energy profile fluctuations. Assuming these variable are random and independent, their effect will combine in quadrature (square root of the sum of squares), and the induced temperature precision can be determined using Equation 6.1, where T is the temperature precision [K], SNR is the image signal-to-noise ratio, E is the laser profile fluctuation [%], P P s is the single-line photophysical parameter (defined in Equation 2.3) evaluated at the corresponding

149 6.6. FIRED ENGINE THERMAL STRATIFICATION 129 T=T T AVE (a) 305 CA (b) 320 CA (c) 330 CA (d) 340 CA (e) 345 CA (f) 350 CA Figure 6.16: Single-shot PLIF temperature sequence of TS development for skipfired (cycle 20) HCCI engine operation. Diagnostic - single-line 277 nm, 3-pentanone. Engine conditions: skipfired, φ=0.4, T in =198 C, P in =100 kpa, 14% 3-pentanone in iso-octane.

150 130 CHAPTER 6. HIGH-LOAD HCCI temperature and pressure, and T P P s is the relative change in temperature with the single-line photophysical parameter. This equation was derived from the detailed uncertainty analysis of the single-line PLIF technique presented in Chapter 2 (see Equations ). T meas prec = T P P s ( ( ) ) E 2 P P s (T, P ) (6.1) SNR T phy = ( ) Timg 2 std.dev. Tmeas prec (6.2) To determine the image SNR, a separate camera noise characterization has been completed to relate the SNR to measured fluorescence signal. This camera characterization is presented in Appendix C. Here SNR was measured as a function of light intensity to allow easy calculation of SNR based on the average fluorescence signal measured in each image [127]. The laser profile fluctuation has previously been characterized for these laser systems based on shot-to-shot profile variations measured for a homogeneous distribution, and was found to be approximately 1% for the excimer pump laser used for the current study. The photophysical parameter and partial derivative in Equation 6.1 were calculated using the engine FQY quadratic fits and absorption cross-section data described in the tracer photophysics section. Finally, the temperature precision calculated with Equation 6.1 is subtracted in quadrature from the image standard deviation ( T img std.dev. ) using Equation 6.2, providing a metric for the physical temperature fluctuations in-cylinder ( T phys ). All temperature standard deviations presented here have been corrected in this manner. Comparison of the noise-corrected temperature standard deviation for cycle 18 and cycle 20 is shown in Figure The close agreement in temperature standard deviation for these cycles indicates that combustion residuals have little to no impact on TS development, for low-residual HCCI operation with conventional valve timing. This fact is consistent with the two-line visualization of residual mixing discussed above, and it implies that any differences in TS development between motored and fired operation is not related to the presence of combustion residuals.

151 6.6. FIRED ENGINE THERMAL STRATIFICATION 131 Temperature Std. Dev. [K] Cycle 20 w/ Residuals Cycle 18 w/o Residuals Crank Angle [ CA] Figure 6.17: Noise-corrected temperature standard deviation for cycle 18 (no hot residuals) and cycle 20 (hot residuals). Diagnostic: single-line 277nm, 3-pentanone. Engine conditions: skipfired, φ=0.4, T in =198 C, P in =100 kpa, 14% 3-pentanone in iso-octane Comparison of Motored and Fired TS Comparison of results for motored and fired operation is helpful in assessing any fundamental differences or similarities in TS development. It could also be helpful for adjusting operating conditions of future motored data to ensure that they are representative of fired conditions. Because the mixing study above showed minimal impact of residuals for low-residual HCCI operation, any added differences in TS can be attributed to other mechanisms. The motored versus fired discussion will be given in two parts. First, differences in TS between early DI and premixed fueling are considered. Second, direct comparisons between premixed motored and premixed fired results are presented to investigate the influence of cylinder-wall temperature. Impact of Early Direct Injection on TS Previous studies have shown that early direct injection and fully premixed fueling result in nearly identical combustion performance [64], indicating that both techniques provide an effectively homogeneous fuel distribution. As a result, the early DI used for skipfired experiments was expected to have minimal impact on TS development. However, comparison of temperature standard deviation results for skipfired (early

152 132 CHAPTER 6. HIGH-LOAD HCCI DI) and premixed motored conditions with identical intake temperatures exhibit differences in the magnitude of TS, as seen in Figure Here the skipfired TS is systematically 2-3 K higher than premixed motored operation, indicating that the early DI may impact TS development more than previously thought. In order to test TS sensitivity to direct injection, a series of motored engine experiments with either DI or premixed fueling were performed. Statistical results from this study are shown in Figure 6.18, and they indicate that direct fuel injection can result in higher TS during compression. This difference in TS between DI and premixed is similar to that seen in the skipfired / premixed comparison, and likely means that the difference in motored and skipfired statistics is mainly a result of the direct injection. The systematic increase in standard deviation with direction injection indicates that fuel/charge-gas mixing is not complete despite similar engine performance noted in previous studies [64]. Apparently, a small amount of fuel stratification persists which increases TS through evaporative cooling and variable compression resulting from localized variation in the thermodynamic gamma (ratio of specific heats) of the mixture. This slight fuel stratification has been confirmed with supplemental fuel mole fraction measurements (not shown) made under identical in-cylinder conditions. These results showed some amount of fuel stratification that persisted towards TDC despite the very early start of injection at 80 CA. Impact of Wall Temperature on TS To eliminate the problems associated with direct injection described above, singleline temperature measurements were performed for fired operation with fully premixed fueling. The engine was continuously fired, since premixed operation does not provide a means for skipfired operation. These continuously fired experiments were performed with a reduced equivalence ratio of 0.32 (compared to 0.4 for previous data) to avoid high thermal loading on the optical engine which can lead to continuously advancing of combustion phasing. Complementary single-line temperature measurements for motored operation with identical intake temperature and fueling were performed for comparison. Noise-corrected temperature standard deviations

153 6.6. FIRED ENGINE THERMAL STRATIFICATION 133 Temperature Std. Dev. [K] Premixed Direction Injection Crank Angle [ CA] Figure 6.18: Impact of direct fuel injection on measured temperature standard deviation for motored engine operation. Engine conditions are identical to skipfired experiments of Figure Diagnostic: single-line 277nm, 3-pentanone. Engine conditions: skipfired, φ=0.4, T in =198 C, P in =100 kpa, 14% 3-pentanone in isooctane. for the motored and fired operation at these conditions are compared in Figure Examination of Figure 6.19 reveals that for premixed operation the magnitude of TS for motored operation is now systematically higher than fired operation for these premixed conditions. One potential source of this difference is the internal surface temperatures. The high gas temperatures achieved during combustion naturally increase the wall temperature, especially for continuous fired operation. This is particularly important for optical engines, which have less effective cooling systems due to the cylinder windows and bowditch piston assembly. The increased surface temperature for fired operation reduces the temperature differential driving nearwall heat transfer, which could ultimately reduce the thermal stratification. For the current study, the measured wall temperature increases from 127C to 141C between motored and continuously fired operation respectively. To assess the impact of wall temperature, additional single-line measurements were performed for fired operation with the coolant temperature reduced to 60C, compared to 100C for all other measurements. This reduced the wall temperature for the fired operation to 121C (down from 141C), which is close to the 127C wall temperature for motored operation (with 100C coolant). Figure 6.19 compares the

154 134 CHAPTER 6. HIGH-LOAD HCCI temperature standard deviation results for these data. As expected, the reduced coolant and wall temperatures result in higher TS that is closer to, but does not exactly match, the premixed motored data. These measurements confirm the sensitivity of TS to boundary wall temperature, and provide an explanation for the standard deviation differences between motored and fired operation. A potential reason for the remaining difference in TS between motored and fired (low coolant) data is that the piston-crown surface temperature, which is not strongly affected by the coolant, is presumably still at a much higher temperature for fired operation. It was initially thought that the difference in wall temperature would play a minor role in TS development, given the already large temperature differential between bulk-gas temperature and wall temperature late in compression. However, based on the temperature statistics shown in Figure 6.19, changes in wall temperature of only 20C can significantly impact TS development. Temperature Std. Dev. [K] Motored T cool =100 C, T cyl =127 C Fired T cool =100 C, T cyl =141 C Fired T cool =60 C, T cyl =121 C Crank Angle [ CA] Figure 6.19: Impact of upper cylinder-wall temperature based on the noise corrected temperature standard deviation for motored and fired operation with varying coolant temperature. Diagnostic: single-line 277nm, 3-pentanone. Engine conditions: continuous fired, φ=0.32, T in =190 C, 17% 3-pentanone in iso-octane Correlation of Temperature and Reacting Zones As discussed above, it is generally considered that the temporal variation in the auto-ignition timing of the various parts of the charge is the result of sequential

155 6.6. FIRED ENGINE THERMAL STRATIFICATION 135 auto-ignition of progressively cooler regions. Considering the importance of thermal stratification for HCCI combustion at all but the lowest loads, it is valuable to investigate the validity of this hypothesis. This was done by comparing the temperature distribution prior to TDC, with reaction zones after TDC. Such a comparison requires the ability to image areas of combustion reaction simultaneously with temperature in the same engine cycle. Chemiluminescence imaging is a common technique used to visualize reaction zones [108,126], which is based on emission from excited-state intermediate combustion species. However, such measurements are line-of-sight averaged across the cylinder volume and may not exactly correlate with planar temperature results within the thin laser sheet. An alternative fluorescence based technique utilizes tracer consumption during combustion to indicate reaction regions. This technique is often applied in spark-ignition engine studies to visualize the remaining reactant areas during flame-front propagation [ ]. For the current study, images acquired after TDC contain regions of low fluorescence signal that are presumably indicative of early combustion reactions. These regions are deemed early reaction zones as consumption of the large tracer molecules occurs early in the combustion process and is coincident with iso-octane fuel decomposition. A sample image of tracer consumption is shown in Figure 6.20a, where the darker regions with low fluorescence intensity signify early reaction zones. These low-signal regions were verified to correlate with exothermic reaction by comparing variations in measured reaction zone area from the fluorescence images with variations in the 10% burn point (CA10) based on the cumulative AHRR. To facilitate the comparison, fluorescence images were first binarized based on a chosen threshold, and inverted to highlight the reaction zones as shown in Figure 6.20a. The reaction area was then calculated and normalized by the total PLIF measurement area, to determine the reaction area ratio (RAR). Finally, the RAR was compared with CA10, calculated from the individual cycle cylinder-pressure trace, on a cycle-to-cycle basis as seen in Figure 6.20b. The excellent correlation between RAR and CA10 shown in Figure 6.20b confirms that the tracer consumption in the central bulk-gas regions is representative of the early reactions of the total charge. This means that tracer consumption after TDC can be used for reaction zone visualization as expected. In

156 136 CHAPTER 6. HIGH-LOAD HCCI (a) Fluorescence signal Binary reaction zone image CA10 [ CA] Correlation Coeff. = Cycle Number (b) Combustion phasing correlation Reaction Area Ratio Figure 6.20: (a) Inverse binarization of fluorescence signal highlighting reaction zones. (b) Correlation between reaction area ratio (RAR) and CA10 combustion phasing. addition, because the TS distribution measurements prior to TDC result from a single-line technique, the remaining excitation wavelength in the current setup can be used for reaction imaging in the same cycle. To investigate the spatial correlation between temperature and combustion zones, single-line temperature distributions have been compared with the images of reaction zones described above, and are presented in Figure This comparison was achieved by temporally shifting the 277 and 308 nm excitation laser pulses to provide a single-line 277 nm temperature image before TDC, and a 308 nm reaction

157 6.6. FIRED ENGINE THERMAL STRATIFICATION 137 image after TDC, all within the same engine cycle. Temperature images 6.21a-c were acquired at a constant imagine timing of 350 CA and provide individual-cycle temperature distributions prior to the onset of reaction. The corresponding reaction images were taken at times ranging from CA to CA and show the progression of reaction throughout the measurement area. As expected, the reaction area increases as the image timing is delayed from TDC due to the progress of chemical reaction. It should be noted that charge motion between temperature and reaction image acquisition does slightly affect the spatial correlations shown in Figure 6.21 but does not significantly impact the general observations. The first image pair 6.21a indicates that the earliest reactions are localized to the highest temperature regions. As time progresses, the reaction zones spread to include cooler regions as evidenced in pair 6.21b. As the reaction progresses further still (6.21c), only the coolest temperature areas remain un-reacted. Earlier temperature images acquired at 345 CA also show favorable spatial correlations with reaction zones as shown in Figure 6.21d. Collectively these imagines directly confirm the progression of sequential auto-ignition, in which reaction initiates in the highest temperature regions followed by progressively cooler zones. Furthermore, it demonstrates the overall importance of thermal stratification for high-load HCCI operation.

158 138 CHAPTER 6. HIGH-LOAD HCCI Temperature [K] Early Reaction Zones (a) 350 CA / CA (b) 350 CA / 365 CA (c) 350 CA / CA (d) 345 CA / 365 CA Figure 6.21: Spatial correlation of temperature distribution before TDC (left) and early reaction zones (right) after TDC acquired for same cycle