Residual Gas Mixing in Engines

Transcription

1 Residual Gas Mixing in Engines by Andrew G. Bright A thesis submitted in partial fulfillment of the requirements for a degree of Master of Science (Mechanical Engineering) at the University of Wisconsin Madison 2004

2 ii

3 i Abstract The mixing of fresh charge with residual gases was studied in a spark-ignition engine using planar laser-induced fluorescence (PLIF) of a homogenous air/fuel/tracer mixture. An adjustable, dual-overhead cam cylinder head and throttled operation provided a range of elevated residual gas fractions. The bulk residual fraction was measured with a sampling valve and exhaust emissions were recorded for 15 experimental conditions covering two engine speeds and five valve overlap strategies. Residual gas fractions ranged from 24% to 40% at 600 RPM and 21% to 45% at 1200 RPM. Indicated mean effective pressure ranged from 146 kpa to 271 kpa across all conditions, with variability levels consistently below 6%. Calculated heat release confirmed the high dilution levels with universally slow burning rates. A non-intensified CCD camera was used to capture the PLIF signal and operated with a peak signal-to-noise ratio of 21:1. The negative-plif imaging technique was verified with a quantitative measure of intake charge homogeneity, and a fuel-cutoff experiment that isolated unwanted fluorescence signal from residuals. Data images were analyzed with first and second statistical moments of pixel intensity, as well as an ensemble PDF curve. All fired conditions showed a clear increase in spatial variation from the homogeneous condition, a trend that was qualitatively verified visually in the corrected data images. Inhomogeneity in the compressed charge increased rapidly above 35% residual gas fraction, independent of engine speed or overlap strategy. The intake cam advance valve overlap strategy was found to provide reduced spatial variation over equivalent symmetric valve overlaps and exhaust cam retard overlaps.

4 ii Acknowledgements First to thank for the completion of this project is my advisor, Professor Jaal B. Ghandhi for giving me the opportunity to pursue my graduate work at the Engine Research Center. Prof. Ghandhi has been an exceptional point of reference for the myriad challenges that have presented themselves over the past two years. The support staff at the ERC also have to be thanked, particularly Sally Radecke and Susan Strzelec in the office, for tolerating my approach to procedure and paperwork. Also, Ralph Braun has provided the supplies and access to shop facilities essential to completing this project. Very little would have been accomplished without the help of fellow students here, past and present. Matt Wiles got me started in the engine lab and familiarized me with all aspects of the laser imaging procedure. Randy Herold has been an invaluable aid throughout the project with the optical system and emissions analyzers. Lonny Peet provided his time in completing the accumulator fuel system, which has been a major improvement in the lab. Brian Albert, Dennis Ward, Bob Iverson, Tongwoo Kim, Soochan Park, Jared Cromas, Nate Haugle, Karen Bevan, Daniel Rodriguez and Anton Kozlovsky have all given substantial help along the way. Cheers to all. The Wisconsin Small Engine Consortium generously assumed funding support midway through this project. The representatives of Briggs & Stratton, Fleetguard/Nelson, Harley-Davidson, Kohler, Mercury Marine, MotoTron and the Wisconsin Department of Commerce are to be thanked. Preliminary funding came through a grant from the National Science Foundation, to which I am equally grateful.

5 iii Table of Contents ABSTRACT...I ACKNOWLEDGEMENTS... II TABLE OF CONTENTS...III LIST OF FIGURES...VI LIST OF TABLES... X 1. INTRODUCTION MOTIVATIONS FOR RESIDUAL GAS STUDY Small Engines Issues High-Dilution Automotive Engines Homogeneous-Charge Compression-Ignition PROJECT OBJECTIVES OUTLINE BACKGROUND RESIDUAL GAS EFFECTS ON COMBUSTION Combustion Thermodynamics Flame Speed Effects Oxides of Nitrogen Formation Cycle-to-Cycle Variations BULK RESIDUAL GAS FRACTION MEASUREMENT Measurement Principle Sampling Valves Sampling Valve Operation ONE-DIMENSIONAL STUDIES OF RESIDUAL GAS Early Work Recent Work PLANAR LASER-INDUCED FLUORESCENCE Laser Source Tracer Chemical Selection Camera PLIF MEASUREMENTS IN ENGINES d Quantification of SI Engine Flow Inhomogeneity Direct Visualization of Residual Gas Negative Visualization of Residual Gas EXPERIMENTAL SETUP SINGLE-CYLINDER RESEARCH ENGINE... 36

6 iv Base Engine Optical Access Cylinder Head and Combustion Chamber Valvetrain Timing System Dynamometer Engine Fluid Systems Engine Aspiration Systems Fuel Delivery System Engine Control System COMBUSTION DATA ACQUISITION Cylinder Pressure Measurement Sampling Valve Emissions Bench OPTICAL MEASUREMENT SYSTEM Laser Source Laser Optics Camera Optical Triggering ENGINE OPERATING CONDITIONS SELECTION CRITERIA Optical Engine Considerations Establishing Engine Conditions COMBUSTION ANALYSIS Cylinder Pressure Data Heat Release Analysis EXHAUST GAS EMISSIONS MEASUREMENT Emissions Measurement Procedure Emissions Analysis Emissions Measurements BULK RESIDUAL GAS FRACTION MEASUREMENT Sampling Valve Measurement Technique Residual Gas Fraction Calculations Residual Gas Fraction Measurements IMAGING SYSTEM DEVELOPMENT AND ANALYSIS PLIF IMAGE PROCESSING Image Acquisition Procedure Image Correction Procedure Median Filtering Image Statistics Probability Distribution Function Image Presentation IMAGING SYSTEM PERFORMANCE Camera Selection... 96

7 v Region of Interest and Spatial Resolution Signal-to-Noise Ratio MicroMax Comparison with Intensified CCD ASSESSMENT OF INTAKE CHARGE HOMOGENEITY First and Second Moments of Homogeneous Data Homogeneous Image PDF DIRECT-INJECTION TEST OF IMAGING TECHNIQUE Skip-Direct Injection Experiment Skip-DI Imaging and Results RESIDUAL GAS MIXING SAMPLE IMAGING DATA CORRELATION OF SPATIAL-MEAN PIXEL INTENSITY WITH MEASURED RESIDUAL GAS FRACTION CORRELATION OF RESIDUAL GAS FRACTION TO IMAGE INTENSITY VARIATION Cycle-Averaged Image Intensity COV Correlation Lower Residual Fraction Case-to-Case Comparison Higher Residual Fraction Case-to-Case Comparison PRIOR-CYCLE EFFECT ON IMAGE INTENSITY VARIATION ENGINE OPERATING CONDITIONS EFFECT ON DATA IMAGE INTENSITY VARIATION Symmetric Overlap Increase Intake Cam Advance Exhaust Cam Retard SUMMARY AND CONCLUSIONS PROJECT SUMMARY RESULTS SUMMARY CONCLUSIONS RECOMMENDATIONS FOR FUTURE WORK REFERENCES APPENDIX A ENGINE OPERATING CONDITIONS APPENDIX B IMAGE STATISTICS

8 vi List of Figures FIGURE 1.1. STRATEGIES PURSUED FOR HCCI CONTROL IN CURRENT RESEARCH. REPRINTED FROM [9] FIGURE 2.1. EXPERIMENTAL MEASUREMENTS OF GASOLINE LAMINAR FLAME SPEED IN EXHAUST GAS-DILUTED MIXTURES RELATIVE TO UNDILUTED MIXTURES, SU(0), FOR A RANGE OF DILUENT FRACTIONS, EQUIVALENCE RATIOS AND INITIAL BOMB PRESSURES. REPRINTED FROM [3] FIGURE 2.2. SAMPLE CYLINDER PRESSURE DATA FOR IN-CYLINDER SAMPLING IN A SMALL 2- STROKE ENGINE, WITH VALVE LIFT DURATION MEASURED BY AN INDUCTIVE PROXIMITY SENSOR SHOWN. REPRINTED FROM [12] FIGURE 2.3. CORRELATION OF MEASURED [CO2] TO LOCAL N2 TEMPERATURE USING CARS. THE PLOT ON THE LEFT IS FOR DATA ACQUIRED AT 30 BTDC WITH A CORRELATION COEFFICIENT OF THE PLOT ON THE RIGHT IS AT 5 BTDC WITH A CORRELATION OF REPRINTED FROM [20] FIGURE 2.4. EXPERIMENTAL SETUP FOR RAMAN SCATTERING MEASUREMENTS IN A MODERN 4- VALVE PENT-ROOF COMBUSTION CHAMBER. REPRINTED FROM [8] FIGURE 2.5. RESIDUAL GAS MOLE FRACTION VS. CRANK ANGLE, BASED ON ENSEMBLE- AVERAGED CONCENTRATION MEASUREMENTS OF VARIOUS SPECIES. REPRINTED FROM [8] FIGURE 2.6. LEVELS OF VARIANCE IN DATA FOR ENSEMBLE-AVERAGED MEAN RESIDUAL GAS MOLE FRACTION GIVEN IN FIGURE 2.5. REPRINTED FROM [8] FIGURE 2.7. ABSORPTION AND EMISSION PROPERTIES OF 3-PENTANONE IN LIF APPLICATIONS [17] FIGURE 2.8. MEASURED TEMPERATURE DEPENDENCY OF LIF SIGNAL OF ACETONE AT ATMOSPHERIC PRESSURE, NORMALIZED TO ROOM TEMPERATURE CONDITION. REPRINTED FROM [18] FIGURE 2.9. MEAN H2O PLIF SIGNAL TREND WITH INTAKE MAP. REPRINTED FROM [22] FIGURE CYCLIC VARIATION IN H2O PLIF SIGNAL FOR INCREASING LOAD. REPRINTED FROM [22] FIGURE CORRELATION OF LOAD-NORMALIZED RESIDUAL GAS FLUCTUATION TO CCV OF 0-0.5% HEAT RELEASE DURATION USING H2O PLIF. REPRINTED FROM [22] FIGURE COMPARISON OF FLOWFIELD EFFECT ON RESIDUAL GAS DISTRIBUTION AS MEASURED BY NEGATIVE-PLIF. BOTH CONDITIONS ARE 1200 RPM, ΗVOL = 0.6. REPRINTED FROM [23] FIGURE MEAN RESIDUAL GAS DISTRIBUTION ACROSS COMBUSTION CHAMBER (DIRECTION ALONG PENT-ROOF AXIS) FOR TWO BULK FLOWFIELD CONDITIONS. IMAGE DATA TAKEN WITH NEGATIVE-PLIF AT SPARK TIMING (27 BTDC) RPM, ΗVOL = 0.6. REPRINTED FROM [23] FIGURE 3.1. VALVETRAIN TIMING LAYOUT FOR DOHC CYLINDER HEAD... 41

9 FIGURE 3.2. COMPARISON OF MEASURED CYLINDER PRESSURE TRACES AT WALL-MOUNT LOCATION TO CONVENTIONAL ROOF-MOUNT. MOTORING ENGINE CONDITION WITH OHV HEAD, 1200 RPM FIGURE 3.3. IN-CYLINDER SOLENOID-ACTUATED SAMPLING VALVE MOUNTED TO BLOCK-HEAD SPACER RING. TEFLON SAMPLED GAS LINE TRAVELS TO AN ADJACENT ICE BATH AND THEN TO THE ANALYZER FIGURE NM LASER PULSE SEPARATION AND DELIVERY OPTICS (PLAN VIEW) FIGURE 3.5. LASER SHEET-FORMING OPTICS SETUP FOR 266 NM PLIF IMAGING FIGURE 3.6 MICROMAX CAMERA MANUAL SUMMARY OF DIF-MODE TIMING. IMAGE EXPOSURE TIMES ARE SHOWN IN THE SECOND LINE. READY AND SCAN ARE OUTPUT SIGNALS FROM THE CAMERA CONTROLLER, EXT. SYNC IS THE INPUT TRIGGER TTL, LASER OUTPUT SHOWN IS FOR A DOUBLE-PULSE LASER, THIS EXPERIMENT ONLY USES THE FIRST PULSE. REPRINTED FROM [24] FIGURE 3.7 SCHEMATIC FOR TTL TIMING OF LASER PULSE AND CAMERA, SYNCHRONIZED WITH MOTOTRON SKIP-FIRING IGNITION BY A ONE-AND-ONLY-ONE CIRCUIT FIGURE 4.1 SUMMARY OF FOUR VALVE OVERLAP STRATEGIES. BASELINE CAM TIMING IS INDICATED BY THE DASHED LINE IN ALL PLOTS. ARROWS INDICATE CAM SHIFT FROM BASELINE. THE BASELINE OVERLAP DURATION IS 20, THE 600 RPM EXTENDED OVERLAPS ARE 30 DURATION, AND THE 1200 RPM CONDITIONS ARE 60 OVERLAP DURATION FIGURE 4.2 HEAT RELEASE RATE AND CUMULATIVE HEAT RELEASE FOR ALL CAM STRATEGIES AT 600 RPM LOW LOAD FIGURE 4.3 HEAT RELEASE RATE AND CUMULATIVE HEAT RELEASE FOR ALL CAM STRATEGIES AT 600 RPM MID LOAD FIGURE 4.4 HEAT RELEASE RATE AND CUMULATIVE HEAT RELEASE FOR ALL CAM STRATEGIES AT 1200 RPM MID LOAD FIGURE 4.5 SKIP-FIRING SEQUENCE EXAMPLE (1200 RPM BASELINE OVERLAP SHOWN). SAMPLING VALVE IS ACTUATED ON COMPRESSION STROKE OF SKIP-FIRED CYCLE (SEE TABLE 4.5) FIGURE 4.6 SAMPLE PRESSURE DATA FOR SKIP-FIRED CYCLE WITH SAMPLING VALVE ACTUATION. THE AVERAGE FIRED CYCLE PRESSURE TRACE AND THE SAMPLING VALVE LIFT TRANSDUCER SIGNAL FOR THAT SKIP-FIRED CYCLE (NO PHYSICAL UNITS) ARE OVERLAYED RPM EXHAUST CAM RETARD CONDITION SHOWN FIGURE 4.7 FREQUENCY HISTOGRAM OF PRIOR-CYCLE IMEP FOR SKIP-FIRING OPERATION AT 600 RPM LOW LOAD SYMMETRIC OVERLAP INCREASE CONDITION. DATA COMPILED FROM 100 CONSECUTIVE SAMPLED CYCLES FIGURE 4.8 FREQUENCY HISTOGRAM OF PRIOR-CYCLE IMEP FOR SKIP-FIRING OPERATION AT 1200 RPM EXHAUST RETARD CONDITION. DATA COMPILED FROM 100 CONSECUTIVE SAMPLED CYCLES FIGURE 5.1 SAMPLE 100-IMAGE MEAN BACKGROUND IMAGE. PIXEL INTENSITY SCALE IS ON RIGHT FIGURE IMAGE MEAN FLATFIELD IMAGE, 30 BTDC 600 RPM MID LOAD EXHAUST RETARD CONDITION. FLATFIELD IMAGES HAVE BEEN BACKGROUND-SUBTRACTED FIGURE 5.3 SAMPLE RAW DATA IMAGE (NO CORRECTIONS), 30 BTDC 1200 RPM EXHAUST RETARD CONDITION vii

10 viii FIGURE 5.4 SAMPLE HOMOGENEOUS IMAGES ACQUIRED AT 30 BTDC FOR THE 1200 RPM, ZERO OVERLAP CONDITION DEMONSTRATING VERTICAL BANDING IN THE CORRECTED IMAGES. SEE SECTION FOR IMAGE PRESENTATION CONVENTION FIGURE 5.5 LOCATION OF ROI WITHIN COMBUSTION CHAMBER, DOHC CYLINDER HEAD. DISTANCE H IS BETWEEN LASER SHEET PLANE AND PISTON FACE, AND IS TABULATED FOR IMAGE TIMINGS IN TABLE FIGURE 5.6 CAMERA NOISE CHARACTERIZATION, AS A FUNCTION OF SIGNAL INTENSITY - MICROMAX FRAME-STRADDLING CCD. REPRINTED FROM [14] FIGURE 5.7 COMPARISON OF THEORETICAL SHOT NOISE INTENSITY VARIATIONTO MEASURED ( σ y µ y) HOMOGENOUS PIXEL INTENSITY VARIATION FIGURE 5.8 PROBABILITY DISTRIBUTION FUNCTION FOR PIXEL INTENSITY IN HOMOGENEOUS IMAGE SETS AT FOUR IMAGE TIMINGS FOR ALL THREE ENGINE SPEED/LOAD POINTS. BASELINE VALVE OVERLAP. EACH PDF CURVE CONTAINS INFORMATION ABOUT 100 CORRECTED HOMOGENOUS IMAGES FIGURE 5.9 DIRECT-INJECTION EXPERIMENT CYLINDER PRESSURE TRACE COMPARISON WITH DOHC BASELINE VALVE OVERLAP. 600 RPM FIGURE 5.10 DIRECT-INJECTION EXPERIMENT CYLINDER PRESSURE TRACE COMPARISON WITH DOHC BASELINE VALVE OVERLAP RPM FIGURE 6.1 SAMPLE HOMOGENEOUS IMAGE SEQUENCE, 60 BTDC FIGURE 6.2 SAMPLE DATA IMAGE SEQUENCE, HIGH RESIDUAL FRACTION CONDITION, 60 BTDC FIGURE 6.3 SAMPLE DATA IMAGE SEQUENCE, MID-RANGE RESIDUAL FRACTION, 60 BTDC. 112 FIGURE 6.4 SAMPLE DATA IMAGE SEQUENCE, LOW RESIDUAL FRACTION CONDITION, 60 BTDC FIGURE 6.5 CORRELATION OF MEAN IMAGE INTENSITY RATIO TO MEASURED RESIDUAL FRACTION FOR ALL 15 EXPERIMENT CONDITIONS FIGURE 6.6 PIXEL INTENSITY COV VS. RESIDUAL GAS FRACTION FOR ALL ENGINE CONDITIONS AT 30 BTDC. SHOT NOISE-LIMITED MAXIMUM SNR WAS ~22:1 FOR THIS IMAGE TIMING FIGURE 6.7 PIXEL INTENSITY COV VS. RESIDUAL GAS FRACTION FOR ALL ENGINE CONDITIONS AT 45 BTDC. SHOT NOISE-LIMITED MAXIMUM SNR WAS ~20:1 FOR THIS IMAGE TIMING FIGURE 6.8 PIXEL INTENSITY COV VS. RESIDUAL GAS FRACTION FOR ALL ENGINE CONDITIONS AT 60 BTDC. SHOT NOISE-LIMITED MAXIMUM SNR WAS ~18:1 FOR THIS IMAGE TIMING FIGURE 6.9 PIXEL INTENSITY COV VS. RESIDUAL GAS FRACTION FOR ALL ENGINE CONDITIONS AT 99 BTDC. SHOT NOISE-LIMITED MAXIMUM SNR WAS ~15:1 FOR THIS IMAGE TIMING FIGURE 6.10 SAMPLE DATA IMAGES FOR 600 RPM, LOW-RESIDUAL CONDITION FIGURE 6.11 SAMPLE DATA IMAGES FOR 1200 RPM, LOW-RESIDUAL CONDITION FIGURE IMAGE PIXEL INTENSITY PDF FOR 600 RPM LOW-RESIDUAL CONDITION FIGURE IMAGE PIXEL INTENSITY PDF FOR 1200 RPM LOW-RESIDUAL CONDITION.121 FIGURE 6.14 SAMPLE DATA IMAGES FOR 600 RPM, HIGH-RESIDUAL CONDITION. 45 BTDC

11 ix FIGURE 6.15 SAMPLE DATA IMAGES FOR 1200 RPM, LOW-RESIDUAL CONDITION FIGURE 6.16 PRIOR-CYCLE IMEP VS. IMAGE INTENSITY COV. 600 RPM LOW LOAD, SYM. ( σ y µ y) INCREASE 60 BTDC. YR = 40.4%, IMEP=152 KPA, COVIMEP = 6.0%, =5.2% FIGURE 6.17 PRIOR-CYCLE IMEP VS. IMAGE INTENSITY COV ( RPM, SYM. INCREASE 60 σ ) y µ y BTDC. YR = 43.7%, IMEP=253 KPA, COVIMEP = 1.2%, n =7.3% FIGURE 6.18 MEAN IMAGE INTENSITY VARIATION VS. CA AT 600 RPM LOW LOAD, ALL OVERLAPS FIGURE 6.19 MEAN IMAGE INTENSITY VARIATION VS. CA AT 600 RPM MID LOAD, ALL OVERLAPS FIGURE 6.20 MEAN IMAGE INTENSITY VARIATION VS. CA AT 1200 RPM, ALL OVERLAPS FIGURE 6.21 INTAKE ADVANCE DATA IMAGES AT 600 RPM MID LOAD. 45 BTDC FIGURE 6.22 EXHAUST RETARD DATA IMAGES AT 600 RPM MID LOAD. 45 BTDC FIGURE 6.23 INTAKE ADVANCE DATA IMAGES AT 1200 RPM. 45 BTDC FIGURE 6.24 EXHAUST RETARD DATA IMAGES AT 1200 RPM. 45 BTDC FIGURE 6.25 INTAKE ADVANCE 100-IMAGE PIXEL INTENSITY PDF AT 600 RPM MID LOAD, 45 BTDC FIGURE 6.26 EXHAUST RETARD 100-IMAGE PIXEL INTENSITY PDF AT 600 RPM MID LOAD, 45 BTDC FIGURE 6.27 INTAKE ADVANCE 100-IMAGE PIXEL INTENSITY PDF AT 1200 RPM 45 BTDC FIGURE 6.28 EXHAUST RETARD 100-IMAGE PIXEL INTENSITY PDF AT 1200 RPM 45 BTDC

12 x List of Tables TABLE 1.1. SAMPLE RESULTS FROM A HIGH-DILUTION STOICHIOMETRIC DISI ENGINE. CASE 1 REPRESENTS THE BASELINE ENGINE RUNNING THROTTLED WITH PORT FUEL INJECTION. CASE 2 IS A 70-CAD WIDENED VALVE OVERLAP WITH DIRECT INJECTION, SUPPLEMENTED WITH A SECONDARY AIR INJECTION AND A HIGH-ENERGY VARIABLE-GAP IGNITION SYSTEM. BOTH CONDITIONS ARE AT 1500 RPM AND 400 KPA BMEP. [5]... 4 TABLE 3.1. FIXED INTERNAL DIMENSIONS OF GM-TRIPTANE ENGINE. VALVE TIMINGS ARE FOR INTERNAL SINGLE CAMSHAFT USED FOR OHV ENGINE OPERATION TABLE 3.2. MAJOR COMBUSTION CHAMBER DIMENSIONS FOR GM-TRIPTANE ENGINE WITH DOHC ADJUSTABLE-CAM CYLINDER HEAD TABLE 3.3 FUEL PROPERTIES FOR PURE ISO-OCTANE AND THE 20% 3-PENTANONE TRACER BLEND USED FOR THIS EXPERIMENT TABLE 3.4. HORIBA EXHAUST EMISSIONS ANALYZER BENCH SUMMARY TABLE 3.5 TRIGGER TIMING DELAYS FOR OPTICAL MEASUREMENT SYSTEM. DELAYS ARE RELATIVE TO THE LEADING EDGE OF THE TRIGGER SIGNAL FROM THE CRANKSHAFT ENCODER TABLE 4.1. AIR/FUEL ENGINE OPERATION PARAMETERS FOR THE THREE EXPERIMENTAL SPEED/LOAD POINTS. THESE VALUES WERE HELD CONSTANT FOR EACH CAM STRATEGY. 67 TABLE 4.2 MEAN EFFECTIVE PRESSURE DATA FOR 100-CYCLE AVERAGE PRESSURE DATA AT ALL EXPERIMENTAL CONDITIONS. PERCENTAGES SHOWN ARE CHANGES RELATIVE TO THE BASELINE OVERLAP CONDITION FOR THE INDIVIDUAL SPEED/LOAD POINTS AT EACH CAM STRATEGY TABLE 4.3 FLAME DEVELOPMENT ANGLES AND OVERALL BURNING ANGLES FOR DIFFERENT OVERLAP STRATEGIES, DETERMINED BY A SINGLE-ZONE HEAT RELEASE CODE. PERCENTAGES INDICATED ARE CHANGES RELATIVE TO THE BASELINE OVERLAP CONDITION AT EACH SPEED/LOAD POINT TABLE 4.4 SUMMARY OF EXHAUST EMISSIONS SPECIES MEASUREMENTS, CONCENTRATIONS SHOWN ARE CORRECTED TO A WET BASIS FROM THE RAW READINGS. AIR/FUEL RATIO AND COMBUSTION EFFICIENCY COEFFICIENT HAVE BEEN CALCULATED FROM THE CONCENTRATION DATA TABLE 4.5 SAMPLING VALVE OPERATION FOR ALL EXPERIMENTAL CONDITIONS. SAMPLING FREQUENCY IS LISTED AS THE NUMBER OF FIRED CYCLES BETWEEN SAMPLED CYCLES (SEE FIGURE 4.5) TABLE 4.6 SUMMARY OF BULK RESIDUAL GAS FRACTION MEASUREMENTS AT ALL EXPERIMENTAL CONDITIONS. PERCENTAGES SHOWN ARE CHANGES RELATIVE TO THE BASELINE OVERLAP CONDITION AT EACH INDIVIDUAL SPEED/LOAD POINT TABLE 5.1 DISTANCE FROM PISTON FACE TO LASER SHEET ROI FOR EXPERIMENT IMAGE TIMINGS

13 TABLE 5.2 VALUES OF SPATIAL-MEAN DATA IMAGE INTENSITY AND RESULTING SHOT NOISE- LIMITED MAXIMUM SNR FOR THREE SPEED/LOAD POINTS. EACH SET IS THE MEAN VALUE FOR THE FIVE VALVE OVERLAP STRATEGIES TABLE 5.3 DIRECT INJECTION EXPERIMENT ENGINE CONDITIONS AND UNBURNED HYDROCARBON EMISSIONS MEASUREMENTS. * INDICATES THE APPROXIMATE IGNITION TIMING TABLE 5.4 DIRECT INJECTION EXPERIMENT IMAGING RESULTS. 100-IMAGE MEAN SIGNAL LEVEL FOR FLATFIELD, SKIP-FIRED, AND MOTORED SKIP-DI PLIF DATA TABLE 6.1 COMPARISON OF LOWER-RESIDUAL CONDITIONS AT 600 AND 1200 RPM. ( σ y µ y) DEVELOPMENT OF IMAGE [%] WITH CRANK ANGLE TABLE 6.2 COMPARISON OF HIGHER-RESIDUAL CONDITIONS AT 600 AND 1200 RPM. ( σ y µ y) DEVELOPMENT OF [%] WITH CRANK ANGLE xi

14 1 1. Introduction 1.1. Motivations for Residual Gas Study Residual gas plays an important role in the combustion development process in fourstroke cycle spark-ignition (SI) engines. This type of internal combustion has to this day been the dominant prime-mover in automobiles and utility engine applications. Residual gas is present in all engines and has important implications to the designer in terms of engine stability and pollutant emissions. Residual gas is especially significant in its role as a diluent species during combustion. This property provides the major benefit to increased residual gas fractions reduction in NO x generation during combustion. NO x is a major pollutant species in internal combustion engine exhaust. The advent of variable valvetrain actuation (VVA) systems in recent years has provided much more freedom to the spark ignition engine designer to utilize the exhaust residual for pollutant reduction and load control, in addition to improvements in volumetric efficiency across the engine speed and load range. VVA, commonly performed by mechanical or electro-hydraulic phase-shifting of the camshaft, is becoming increasingly common on new automotive engine designs. More information about the participation of residual gas in engine flows preceding combustion reactions will be critical to achieving the maximum potential (in terms of SI engine emissions and efficiency) of this and other dilution-controlling technologies.

15 Small Engines Issues 2 Small engines can be defined as the category of internal combustion engines below 500 bhp used for non-automotive applications, principally in power equipment, motorcycles and marine transportation. Despite sharing similar if not identical operation fundamentals, small engines have unique engineering considerations to automotive SI engines. When faced with new challenges related to emissions regulations, small engine manufacturers do not have the luxury of simply adopting mature technologies from the automotive industry. Of particular concern is NO x emissions, which have only been reduced to environmentally acceptable levels in cars by universal use of three-way exhaust catalysts (TWC). For many small engines, the unit cost of the automotive TWC exceeds that of the entire engine, and as such this technology is not deemed practical in the category. Instead of aftertreatment, focus is being placed on charge dilution strategies for NO x reduction, and the simplest delivery mechanism is through internal recirculation via residual gas. Since VVA systems also fall outside the cost-acceptable realm of most small engine designs, elevated residual gas fractions will likely be provided by fixed camshaft profiles. This presents a strong challenge to the combustion chamber designer, with the need to accommodate high-dilution mixtures throughout the engine speed and load range without negatively impacting performance felt by the user. More must be learned about charge composition development at high dilution levels in small engines for this worthy goal to be achieved.

16 High-Dilution Automotive Engines 3 New applications of high residual gas dilution occur in novel engine designs. Olafsson et al. in [5] describe a high-dilution spark ignition engine designed at Saab to reduce fuel consumption and NO x emissions. The engine has a similar objective as seen with direct injection spark ignition (DISI) engines which typically operate without intake throttling and thus enjoy large improvements in part-load fuel efficiency. The critical drawback to DISI engines is that by using excess fresh air, the highly effective and durable three-way catalyst cannot be used to control NO x, CO and HC emission. By utilizing the exhaust gas residual instead of excess air, Olafsson et al. were able to operate at overall stoichiometric conditions with a 10% reduction in part-load fuel consumption from the conventional SI engine. This engine design requires complicated engine systems such as continuously variable camshaft phasers to control residual dilution, air-assisted in-cylinder fuel injection, and most notably, a variable spark plug gap to consistently ignite dilute mixtures. Sample results from this project are presented in Table 1.1.

17 Case 1 Case 2 Change MAP [kpa] BMEP [kpa] PMEP [kpa] COV of IMEP [%] BSFC [g/kwh] % BSNOx [g/kwh] % BSHC [g/kwh] % BSCO [g/kwh] % Exhaust Temp [C] % HR [CAD] % HR [CAD] IGN timing [btdc] Table 1.1. Sample results from a high-dilution stoichiometric DISI engine. Case 1 represents the baseline engine running throttled with port fuel injection. Case 2 is a 70-CAD widened valve overlap with direct injection, supplemented with a secondary air injection and a high-energy variable-gap ignition system. Both conditions are at 1500 RPM and 400 kpa BMEP. [5] Homogeneous-Charge Compression-Ignition Homogeneous Charge Compression Ignition (HCCI) is a rapidly developing new engine combustion strategy that could combine some of the best operating characteristics of SI and diesel engines. In particular, HCCI can achieve the part-load fuel efficiency of diesel engines with substantially reduced in-cylinder soot and NO x emissions on the level of SI engines. Like knock in homogeneous charge SI engines, HCCI involves a controlled autoignition that can be obtained with a variety of petroleum-based fuels. Controlling the autoignition of a mixture is separated into 2 strategies: altering the fuel mixture reactivity kinetics and altering the time-temperature history of the mixture. Cooled external EGR is often explored for the former, given the usual need to delay the onset of compression

18 ignition. The latter strategy commonly involves significant heating of the fuel/air charge 5 which can encourage the onset of autoignition in engines with lower compression ratios. Figure 1.1. Strategies pursued for HCCI control in current research. Reprinted from [9]. This lower-compression ratio configuration would enable dual-mode operation with part-load HCCI combustion transitioning to full-load spark ignition combustion. Intake air heating, while convenient in a laboratory, is not deemed practical for mobile applications. Instead, the focus is being placed on the use of VVA to deliver high residual fractions for heating of the charge. High-dilution operation may be a likely application of HCCI for improving the efficiency of gasoline automotive engines [9, 10]. For this and a variety of other reasons, the mixing and chemical kinetics of the exhaust gas residual is a growing topic of research.

19 1.2. Project Objectives 6 Four broad objectives have been identified for this research: 1. To provide high-quality, spatially and temporally resolved, two-dimensional quantification of residual gas mixing with fresh homogenous air/fuel charge through a range of positions in the SI engine cycle. 2. To supplement and correlate the mixing data with engine-out operating information such as cylinder pressure data and exhaust emissions analysis for a range of residual gas dilution levels. 3. To extract conclusions from the residual gas mixing measurements and engine performance data that will be helpful to the field in designing high-dilution engines. 4. To aid in the development of Planar Laser-Induced Fluorescence as an invaluable combustion diagnostic in SI engines Outline This thesis will be divided into six subsequent chapters. Chapter 2 presents the project background in the form of a literature review of residual-effected SI combustion, sampling valve measurements, prior optical studies of residual gas and the use of PLIF in engines. Chapter 3 contains a detailed, design-oriented discussion of the experimental facility including the research engine, combustion diagnostic instrumentation, and the optical system. Chapter 4 will present the engine operating conditions covered in the project, including the basis for their selection and the measurements of bulk residual gas fraction at

20 each condition. Chapter 5 will discuss the development of the imaging technique, 7 particularly the selection criteria for the hardware and processing steps and subsequent performance of the data images. Chapter 6 will contain the residual gas mixing data derived from the PLIF images, with discussion. Finally, chapter 7 contains project summary, conclusions and recommendations.

21 8 2. Background 2.1. Residual Gas Effects on Combustion Recycled exhaust gas has a substantial effect on combustion processes by acting as a diluent, meaning that it does not participate in the oxidation of the fuel but is present and absorbing the released energy in a quantity significant enough to reduce flame speed and gas temperature [2]. Decreasing flame front speed inherently lengthens the time to reach 10, 50, and 90% mass-fraction burned levels, extending combustion reactions further into the expansion stroke. If the engine control system is not able to adjust other parameters properly, residual gas dilution can slow the burning rate to a point where partial-burn and misfire cycles emerge with severe penalties on emissions and performance. The temperature-mitigating effect of residual gas is well-known as a strategy for reducing oxides of nitrogen (NO x ) production in internal combustion engines Combustion Thermodynamics Residual gas in a spark-ignition engine running at a stoichiometric air/fuel ratio is composed predominantly of N 2, CO 2, H 2 O and O 2. Engines that operate fuel-rich of stoichiometry, such as small air-cooled utility engines, will see significant CO and H 2 and very little remaining O 2 in the residual gas. In most SI engines, pollutant species such as NO x and unburned hydrocarbon compounds (HC) normally sum to 1% or less by volume [1].

22 9 Based on this composition, it can be seen that when added to a mixture of vaporized fuel and air, residual gas will lower the mass-specific heating value of the mixture. For constantvolume combustion, the first law of thermodynamics can be expressed as U ( T, p ) = U ( T, p ) (2.1) reactants i i products ad f where T ad is called the adiabatic flame temperature and is easily calculated from a balanced reaction equation by assuming adiabatic conditions, ideal gas behavior, and no dissociation of reactants or products into minor species [4]. These assumptions make exact calculations difficult but the trend of in-cylinder flame temperature vs. initial reactant composition becomes clear. Residual gas species reduce the total enthalpy (formation plus sensible) of the reactants, which is related to the initial internal energy by the universal gas constant, and thus reduce the flame temperature from that of undiluted air/fuel mixtures Flame Speed Effects The effect of reducing adiabatic flame temperature is observed in reduced burning velocity. Combustion in an SI engine occurs via a turbulent, thin-sheet wrinkled flame structure, which, despite being inherently complex is locally modeled closely by laminar flame propagation rates. The laminar flame speed, S L has been measured [24], and for conventional hydrocarbon fuels has been found to obey the power law equation:

23 S L S α T p u = L,0 T0 p0 β (2.2) 10 where the reference values are standard temperature and pressure and S L,0, α and β are tabulated constants for particular combinations of fuel and equivalence ratio. The term T u represents the unburned gas temperature just ahead of the reaction zone in the flame front. Rhodes and Keck [3] studied gasoline combustion with controlled residual concentration in a constant-volume bomb experiment and quantified a laminar flame speed correction factor for Equation (2.2) given the inclusion of a residual gas fraction in the reaction, based on the data of figure 2.1: 0.77 L r L r r S ( x ) = S ( x = 0)( x ) (2.3) Decreasing the flame temperature and velocity represents a significant challenge to maintaining appropriate engine performance. If, for whatever reason, reactant preheating temperatures fall below 1900 K, flame velocity will be at or near the partial-burn and misfire lower limit [5]. This situation might typically arise if the exhaust valve opens prior to completion of flame propagation, or if the flame is prematurely extinguished [1]. Partial burn and misfire are extreme symptoms of cycle-to-cycle variation (CCV) in engine power output. Besides contributing to unwanted engine roughness characteristics, the incomplete combustion of the fuel charge represents a very significant emission of HC pollutants.

24 11 Figure 2.1. Experimental measurements of gasoline laminar flame speed in exhaust gasdiluted mixtures relative to undiluted mixtures, S u (0), for a range of diluent fractions, equivalence ratios and initial bomb pressures. Reprinted from [3] Oxides of Nitrogen Formation Another major consequence of the dilution effect of residual gas is reduced NO x formation. NO x is a primary ingredient in photochemical smog found in the lower atmosphere mainly above major cities. It also is known to contribute to acid rain. NO x is also regrettably known for being somewhat inextricably linked with engine performance and efficiency. Rate equations for the formation of NO x are non-linear functions of time, elevated temperature and availability of nitrogen and oxygen molecules. Peak NO x formation at optimal combustion phasing occurs close to stoichiometric air/fuel ratio, which also represents the operating point for peak engine stability, power output and efficiency [4].

25 Cycle-to-Cycle Variations 12 Increased residual fractions are expected to locally affect small-scale mixture homogeneity, which describes imperfect distribution of fuel vapor within the air and residual charge. It is assumed that low to moderate spatial inhomogeneity will affect combustion only during the earliest stages near the discharge of the spark plug and the formation of a flame kernel. The scales of non-uniformity are larger or of the same order of the enflamed volume during these critical early instants. As the flame front area grows much larger, the effect of inhomogeneity is averaged out in a global sense [7, 8]. The variation of air/fuel ratio and residual dilution in the vicinity of the spark gap has an important effect on cycle-to-cycle variations (CCV) in SI engines. Local mixtures outside the ignition limit or too dilute to rapidly transition into a fully developed turbulent flame are common causes of misfire and high CCV [1]. In their literature review of cyclic variation, Ozdor et al. [6] summarized several studies of mixture inhomogeneity on flame development. They point to a general uncertainty in applicable length scales of non-uniformities, but to a demonstrated effect of controlled in-cylinder turbulence (particularly swirling motion) at time of spark on reducing CCV. At the time of writing (1994), they point out that none of the dozens of papers reviewed were able to quantify the impact of spatial inhomogeneity of residual gas on CCV.

26 2.2. Bulk Residual Gas Fraction Measurement 13 In this project, residual gas mixing quantifications will be performed for varying levels of residual gas fraction. This quantity, denoted y r, is defined as the mass of burned exhaust gases carried over from the previous cycle s combustion process relative to the total cylinder mass. Like most other in-cylinder quantities, y r is subject to cycle-by-cycle variation in magnitude. However, cycle-averaged values can be measured using in-cylinder gas sampling as will be discussed in this section Measurement Principle The exhaust gas emissions analyzer bench has become a standard engine test cell instrument and typically provides concentration measurements of CO 2, CO, O 2, NO and HC present in a stream of exhaust gas. Given this measurement capability, the most direct way of quantifying total cylinder residual gas fraction is by the relation: x r ( x% CO 2 ) comp = (2.5) x% ( CO 2 ) exh which defines a ratio of mole fractions of CO 2 in the cylinder during the compression stroke (after IVC) and the exhaust system downstream of the engine, typically after passing through a mixing volume. It is important that this calculation be made on a wet basis, where the absence of water vapor in NDIR CO 2 analyzers is accounted for. Water is always condensed

27 out of the exhaust sample lines since it is damaging to instruments. There are a few 14 techniques for correction and they typically involve knowledge of fuel chemistry, CO 2 and CO dry basis readings and intake air relative humidity [1] Sampling Valves Extracting an emissions analyzer sample during the compression stroke from the closed cylinder is most directly performed with a category of hardware known as the fastacting sampling valve. Sampling valves have been employed as early as 1927 to aid the study of chemical and physical processes in engine combustion. Zhao and Ladommatos [14] document a more comprehensive summary of valve designs employed in the engine literature. Most sampling valves covered were either of the outward-opening poppet type or inward-opening needle type. Needle valves hold advantages of smaller tip diameters, which can be advantageous in space-confined combustion chamber surfaces, and also a lack of physical intrusion into the combustion chamber volume. Poppet valves benefit from better sealing performance, aided by combustion pressures and potential for smaller crevice volumes via flush-mount machining. It is proposed by the authors that needle valve sampling volumes will be slightly larger in reach across the combustion chamber. Although mechanical and electro-hydraulic sampling valves have been used for engine studies in the past, the most popular actuation mechanism is electromagnetic force. Typically driven by a linear solenoid, this design must feature a high traction force to counteract a strong return spring used for valve sealing and high armature acceleration for

28 minimum lift duration [11]. Utilization of programmable research/calibration-type digital 15 engine control systems has greatly improved control of valve response. Additionally, monitoring the valve stem lift with an inductive proximity sensor in the back side of the valve body can provide necessary feedback for exact location of the valve window [12] Sampling Valve Operation For sampling of residual gas mixtures, the ignition system should be synchronized to shut off during the cycle of valve actuation to prevent alteration of the residual concentration. Monitoring the effect of skip-firing the engine is important in controlling the quality of the analyzed residual gas mixture. It is expected that after the misfire of the sampled cycle, the following cycle will be strong due to the residual gas being composed of additional unburned fuel/air. It is necessary to ensure that the next sampled cycle follows a cycle that is representative of the steady-state engine performance. One example from the literature is that Hinze & Miles, in [7], found that the third cycle following the skip-fired cycle had an average IMEP equal to the steady 100-cycle average for a 32 kpa MAP, 800 RPM condition. For residual fraction measurement, sampling valve opening frequency must be optimized for maximum sample gas flow rate and minimum deviation of sampled cycle characteristics from steady-state conditions.

29 16 Figure 2.2. Sample cylinder pressure data for in-cylinder sampling in a small 2-stroke engine, with valve lift duration measured by an inductive proximity sensor shown. Reprinted from [12]. One other concern with global residual fraction measurements with fast-acting sampling valves is that the volume of sampled gas must be representative of the total cylinder charge. In designing the UW/ERC poppet-type sampling valve in [15], Foudray referenced sources that indicated that a minimum of 10% to 25% of cylinder volume is adequate to characterize cylinder composition, depending on degree of stratification. Although that research was focused on 2-stroke cycle engine exhaust scavenging, the same criteria are believed to hold for the 4-stroke cycle engine. Using a bellows flow meter, Foudray estimated a sampling mass flow to be within a range of 33% to 66% of per-cycle cylinder mass. Leakage was measured to be approximately 3% of the sample flow rate and neglected in calculations.

30 2.3. One-Dimensional Studies of Residual Gas 17 Raman scattering has been used for many years to provide in-cylinder temporallyresolved measurements in IC engines. Three papers are reviewed here where this onedimensional optical technique has been used to characterize residual gas participation in SI engine flows. Line spectroscopy studies hold advantages over two-dimensional imaging in the reduced impact of optical access and the ability in many cases to track individual chemical species without the use of tracers. They are inherently limited by their one-dimensional nature and within that, a limited spatial resolution Early Work Lebel and Cottereau in [20] performed an early study of residual gas effects on SI combustion. They measured simultaneous CO 2 concentration and N 2 temperature using a Coherent Anti-Stokes Raman Scattering (CARS) setup, with a fixed measurement region 1 cm long and 100 µm in diameter. CO 2 was chosen to track residual gas, while charge temperature was monitored to ensure that same-cycle burned gases in the firing engine were not present in the measurement region. Laser beam intensity referencing was used to allow comparison of single-shot measurements. Correlations were reported, at a single operating condition, between [CO 2 ] and temperature, cycle peak cylinder pressure (PP) and location of peak pressure (LPP) at instants before and after ignition and two locations near and far from the spark plug.

31 18 Very poor correlation was found between [CO 2 ] and PP/LPP in measurements taken 1 mm from the spark plug and 5 btdc (considered end of ignition delay). Since this is counter-intuitive, the authors conclude that, given their limited measurement region, it indicates that the residual gas is not perfectly mixed at the end of the compression stroke. The only meaningful correlation reported in this paper is between increasing [CO 2 ] and increasing T (figure 2.3), which is somewhat obvious given the charge heating property of residual gas. As local temperature readings did not correlate with pressure data, this would reinforce the statement that residual gases (and thus local charge temperatures) are stratified late in the compression stroke. Direct correlations of [CO 2 ] with PP/LPP yielded coefficients from -0.2 to 0.2, limiting the authors to very basic conclusions for effects of local residual gas concentrations on engine performance with this technique. Figure 2.3. Correlation of measured [CO2] to local N2 temperature using CARS. The plot on the left is for data acquired at 30 btdc with a correlation coefficient of The plot on the right is at 5 btdc with a correlation of Reprinted from [20].

32 Recent Work 19 Hinze and Miles at Sandia National Laboratories performed two subsequent lineimaging studies of residual gas mixing [7, 8], developing a detailed statistical quantification for mean and fluctuating inhomogeneity components. Both studies utilized a laser measurement volume in an axially centered position, in which CO 2, H 2 O, N 2, O 2 and C 3 H 8 concentrations were recorded. Binning on the CCD array divided the volume into individual adjacent measurement points which established the spatial resolution. Data was presented in 15 CAD increments from start of intake to TDC compression. Homogenous propane/air mixtures were supplied at stoichiometric conditions. Neither paper presents engine performance data. Figure 2.4. Experimental setup for Raman scattering measurements in a modern 4-valve pent-roof combustion chamber. Reprinted from [8].

33 20 Ensemble-averaged measurements were taken to describe mean stratification of fresh charge and residual gas, while 500-cycle single-shot images were analyzed to establish a cycle-to-cycle fluctuating component. These data were used to generate spatial covariance functions of species mole fractions (based on the adjacent measurement points), which were broken down into fluctuation components coming from system noise, turbulence, and bulk composition. These covariance functions, once developed, could be used to extract integral length scales of local residual gas fraction fluctuation (the scale over which turbulent fluctuations remain correlated.) In their first paper [7], Miles and Hinze utilized a side-valve, side-spark optical engine to test this technique at the same engine operating conditions in two bulk flowfields a semi-quiescent condition and a high-swirl condition. The measurement volume was 11 mm long and 0.49 mm in diameter, divided into 12 measurement points. The quiescent flow was shown to homogenize rapidly, with fluctuations in residual gas concentration nearly eliminated by 150 btdc. For the swirling flow, the measurement volume was radially traversed away from the centerline to two additional measurement regions. Gradients were observed throughout the cycle in the mean concentration data between these volumes which suggested a bulk charge stratification which persisted throughout the compression stroke. Rms fluctuations in the mixture composition at spark time were 5 times higher in the swirling condition (5% vs. 1% for quiescent at -15 CAD.) Mixing length scales for both conditions were found to vary from 2 to 5 mm. In the second paper [8], Hinze and Miles moved to a more conventional pent-roof, 4- valve cylinder head for their measurements and chose to focus on a single engine condition representative of idle. Figure 2.5 shows the reported development of the ensemble-averaged

34 21 residual gas fraction during the engine cycle. In this experiment, the measurement volume was 14.5 mm long and 0.27 mm in diameter divided into 16 sub-regions, improving the spatial resolution by nearly a factor of two. During the intake stroke, the authors were able to track residual gas backflow into the intake and a later period where all the residual gas has been re-inducted away from the measurement volume. The largest gradients in the measurement volume occurred at BDC, as shown in Figure 2.6, with significant gradient breakdown during compression similar to the first project. Length scales encountered at -180 CAD were on the order of 1 cm. Rms fluctuation (1%) and mixing length scale range (2-4 mm) at spark time were comparable to the previous experimental computations. Figure 2.5. Residual gas mole fraction vs. crank angle, based on ensemble-averaged concentration measurements of various species. Reprinted from [8].

35 22 Figure 2.6. Levels of variance in data for ensemble-averaged mean residual gas mole fraction given in figure 2.5. Reprinted from [8] Planar Laser-Induced Fluorescence Planar laser-induced fluorescence (PLIF) is an increasingly popular advanced combustion diagnostic. PLIF has the ability to provide quantitative two-dimensional measurements in single-phase or multi-phase flows with exceptional spatial and temporal resolution. A general summary of a PLIF measurement system is a high-energy, pulsed laser sheet propagating through a flowfield containing a suitable fluorescent tracer species resulting in absorption and subsequent emission of photons at a characteristic wavelength of the tracer molecules. With a process time response on the order of nanoseconds, individual laser shots can be captured by a CCD camera for correction and analysis.

36 23 Detailed discussion of PLIF theory has been presented in the literature [13, 15] and will not be repeated here. Instead, a summary of the important characteristics of the system components used in this project are covered, including laser source, camera, and tracer chemical Laser Source The traditional laser source for PLIF work in engines is the Nd:YAG laser, which offers high-power laser pulses at four harmonic wavelengths, 1064 nm, 532 nm, 354 nm and 266 nm. Laser pulses are delivered at an optimal repetition rate, most commonly 10 Hz. Individual pulses are on the order of 8 ns duration with maximum energies exceeding 100 mj. Nd:YAG lasers can operate with external triggering and can thus be synchronized with engine events, although the low repetition rate typically precludes sequential measurements in the engine cycle. Pulsed laser operation requires attention to shot-to-shot variation in laser beam intensity and profile when making quantitative measurements Tracer Chemical Selection Since neither air nor iso-octane fluoresce under the range of wavelengths supplied by the Nd:YAG laser, a tracer chemical is doped into the intake charge at a controlled concentration. Tracer addition can occur by either on-the-fly seeding of the intake air or by pre-mixing in solution with the fuel, depending on the targeted measurement. Maximum tracer concentration must yield maximum fluorescence signal without significant laser power

37 24 attenuation or influence on combustion performance. The most popular class of tracers for combustion PLIF is the di-ketone group, and the preferred match for iso-octane research is 3- pentanone, based on its closely-related distillation curve. Tracer-matching is far more important in multi-phase PLIF where evaporation rates must be matched than in prevaporized homogenous charge studies. Relative Absorption, Fluorescence Optical Properties of 3-Pentanone Absorption Fluorescence λ (nm) Figure 2.7. Absorption and emission properties of 3-pentanone in LIF applications [17]. The excitation wavelengths for di-ketones fall in the ultraviolet, with an absorption range of nm [17]. Thurber et al. performed important studies on the temperature [18] and pressure [19] dependence of acetone fluorescence at various excitation wavelengths. It was shown that temperature dependence is practically eliminated on the range of K using 289 nm. Likewise, an optimal wavelength for neglecting pressure effects is shown

38 25 to be 308 nm. Making the extension of the acetone behavior to 3-pentanone, tuning the laser wavelength to a value near 289 nm is highly beneficial in quantifying engine flows which are at all temperature-stratified. Figure 2.8. Measured temperature dependency of LIF signal of acetone at atmospheric pressure, normalized to room temperature condition. Reprinted from [18] Camera The di-ketone tracer group emits photons in a broadband range of nm [17]. This visible light is best collected by a high-resolution scientific-grade CCD camera. Charge-coupled devices contain a photo-sensitive pixel array, which when impacted by photons, convert the photon energy to electron charge potentials with a quantum efficiency

39 26 that is a property of the device. The individual pixel charges are read out sequentially into a registry where they are amplified and digitized for computer processing [14]. There are four sources of noise important in making quantitative measurements with CCD images: dark, read, pattern and shot noise. Dark noise arises from thermal generation of electrons in the array and is limited with cooled (thermo-electric or cryogenic) CCD chips. Read noise is a property of the array readout circuit and the programmed readout rate. Fixed pattern noise can be traced from sources on either the CCD chip or the imaging subject, and is unique in this discussion in that it can be eliminated with standard background and flatfield image correction. Shot noise is typically the limiting noise element in high-fidelity CCD imaging such as found in PLIF studies. Shot noise is completely independent of the CCD type and arises from the probabilistic nature of photon impingement on the pixels. The shotnoise limited signal-to-noise ratio is equal to the square root of the number of photons incident per CCD pixel, based on Poisson statistics [13] PLIF Measurements in Engines As mentioned in the previous section, planar laser-induced fluorescence is a powerful IC engine diagnostic tool due to its two-dimensional nature and superior spatial and temporal resolution. Previous studies at the UW/ERC have achieved sufficient spatial resolution to calculate scalar dissipation and used it to quantify the degree of mixedness in stratified DISI flows [15, 16]. Additionally, using two high-shuttering speed intensified CCD cameras, Rothamer [13] was able to simultaneously image unburned and burned mixtures to quantify

40 flame-front equivalence ratio in a stratified-charge DISI engine. For the current study of 27 residual gas mixing in engines, it is important to first present basic techniques for quantifying spatial charge inhomogeneity from PLIF intensity data and then introduce the limited literature on residual gas studies using this technique d Quantification of SI Engine Flow Inhomogeneity Baritaud and Heinze conducted an early application of PLIF in an SI engine at the Institut Français du Pétrole (IFP) in 1992 [21]. The subject of their experiment was quantification of the development of fuel/air stratification in a PFI engine. A major portion of this paper discusses the statistical means for describing charge inhomogeneity in PLIF images. The authors define a total standard deviation for a set of N single-shot images, based on the idea that a single image s inhomogeneity can be quantified by its standard deviation about the spatial mean (σ n ). By ensemble-averaging this value after normalizing each by the mean image intensity ( I n ), the influence of the pulse-to-pulse variation in laser intensity is removed: 1 N σ n σ tot = (2.6) N I n n= 1

41 28 The total standard deviation σ tot is presented as a relative value, since absolute measures of charge inhomogeneity cannot be correlated with individual engine cycles without bias error from the pulse energy variations. To extract the maximum potential information from the data images, the simple standard deviation was broken down into fine-scale and large-scale contributions by employing a basic spatial Fourier transform. First, a 3x3 smoothing procedure was twice performed on the I x J pixel data image, with the resulting smooth field termed Φ(I n (i,j)). The large scale contribution to the inhomogeneity, arising from gradients in large-scale structures in each data image n is: i, j ( ( I (, )) ) 2 n i j In 1 σ n,lf = Φ (2.7) IJ After ensemble averaging, the relative large scale variation is: σ LF σ 1 N n,lf = (2.8) N i= 1 I n Likewise, small-scale fluctuations in each image can be tracked by examining the fluctuation in the raw image intensities relative to the smoothed image: 1 σ ( ( ( )) ( )) 2 n,hf = Φ In i, j In i, j (2.9) IJ i, j

42 29 This value is again ensemble averaged on a normalized basis: σ HF σ 1 N n,hf = (2.10) N i= 1 I n If the ensemble-averaged pixel intensity field (, ) I i j is used in place of the singleimage data in equation (2.9), a hybrid fluctuation arises which can describe the variation of the large-scale inhomogeneities from cycle-to-cycle: n 1 σ ( ( ( )) ( ) ) 2 n,cyc = Φ In i, j In i, j (2.11) IJ i, j Importantly, (, ) I i j is biased by laser pulse variations, which limited its usefulness in this n initial study. Finally, this value can also be ensemble-averaged to a relative basis. σ cyc σ 1 N n,ccv = (2.12) N i= 1 I n The authors indicate that it is difficult using metrics such as σ tot, σ LF, σ HF, and σ cyc to separate single-cycle inhomogeneity effects from cycle-to-cycle variations captured in the data images.

43 Direct Visualization of Residual Gas 30 Direct visualization of combustion residual species such as H 2 O and NO 2 is possible, although challenging, with PLIF. In [22], Johansson et al. used water as a residual tracer, which required use of strategy known as 2-photon LIF, which is unique in its requirement for an interaction of two photons at 248 nm to detect the water molecule. This approach yields inherently lower signal levels than a single-photon LIF study like those done on fuel tracers. Additionally, the authors were unable to provide a homogeneous distribution of water molecules at a known concentration, which prevented signal calibration and therefore quantification of the H 2 O intensity data. The objective of this study was to observe the influence of residual gases on cycle-bycycle variations in engine power output. The optical access system required a vertical laser sheet only 6 mm in height. The laser sheet centerline was passed 4.5 mm below the spark plug and water concentration images were obtained for a range of engine loads (based on intake MAP.) Cylinder pressure-derived heat release data were compiled to correlate residual gas levels with initiation and propagation of SI combustion. The engine was operated on homogeneous natural gas at 700 rpm, and the images were acquired 1 before spark time. Imaging was performed with an intensified CCD gated to 100 ns exposure. Resulting noise levels due to low signal strength and maximum intensifier gain were roughly 20%. The conclusions made on ensemble-averaged water intensity data were fairly basic, essentially confirming predicted trends in increasing residual gas concentration near the spark plug with decreasing load. When normalized by the equivalence ratio of the data set,

44 31 the duration of 0-0.5% heat release was shown to correlate well with the CCV of the water concentration normalized by load point. This is thought to strengthen the argument that fluctuation in residual gas near the spark plug is a major contributor to CCV in SI engines. Unfortunately, quantitative values of the observed fluctuations were not available. Figure 2.9. Mean H2O PLIF signal trend with intake MAP. Reprinted from [22]. Figure Cyclic variation in H2O PLIF signal for increasing load. Reprinted from [22].

45 32 Johansson et al. also attempted correlations with pressure and heat release data for the single-cycle measurements. Although laser power intensity fluctuations were corrected in this experiment by shot-resolved power meter readings, the poor SNR and small imaging region created a large amount of scatter in these correlations. The correlation between duration of 0-0.5% HR and [H 2 O] was optimized for radius of ROI within the image. At a low-load condition, a peak 60% correlation was shown at a radius of 2.9 mm. This correlation degraded with decreasing residual fraction, which was satisfactory since the magnitude of the fluctuations relative to the image noise was expected to also decrease. Figure Correlation of load-normalized residual gas fluctuation to CCV of 0-0.5% heat release duration using H2O PLIF. Reprinted from [22].

46 Negative Visualization of Residual Gas 33 Residual gas can also be tracked with PLIF images by examining the negative of the intensity field provided by a homogeneous air/fuel/tracer charge. Following up on the early work described in Section 2.5.1, Deschamps and Baritaud at IFP [23] performed a negative- PLIF visualization of burned gas distribution in an SI engine. Because this project sought to observe separately the distributions provided by external EGR as well as internal residual gas, the upstream intake air was chosen to be seeded with biacetyl. Air seeding via a carburetor imparted more uncertainties and challenges than premixed fuel solutions. A 25- mm wide horizontal laser sheet was passed 4 mm below the spark plug parallel to the ridge of the cylinder head s pent roof. For the internal residual gas study, five engine effects were examined: fuel type, fuel distribution, tumble level, spark plug location and volumetric efficiency. Mean image intensity profiles in the direction of the sheet across the pent roof were examined, but only in a qualitative manner. The enhanced tumble experiment was conducted with propane to remove fuel stratification effects. With enhanced tumble, mixing along the roof ridge direction was observed to be more difficult during the intake stroke than during compression, where it is assumed that the tumble motion normal to the laser sheet is broken down by turbulence. However, by the end of compression, the enhanced tumble condition shows both a higher concentration and flatter linear distribution than the standard case. The increased concentration suggested that lower tumble levels leave a portion of the residual gas trapped in the bottom of the combustion chamber. Increased charge motion then not only helps

47 34 distribute the residual gas vertically in the combustion chamber, but laterally to create a more homogenous mixture. Another property of enhanced tumble operation proposed by the authors is improved SI combustion efficiency which often correlates with increased intake MAP, reducing bulk residual fraction. Figure Comparison of flowfield effect on residual gas distribution as measured by negative-plif. Both conditions are 1200 RPM, η vol = 0.6. Reprinted from [23].

48 35 Figure Mean residual gas distribution across combustion chamber (direction along pent-roof axis) for two bulk flowfield conditions. Image data taken with negative-plif at spark timing (27 btdc) RPM, η vol = 0.6. Reprinted from [23]. With varying volumetric efficiencies, changes in the distribution of residual gas in the data images taken at -30 CAD are explained primarily through assumed changes and asymmetries in the intake port flows, imparting different bulk flowfields. The residual gas concentration in the image ROI decreases with increasing volumetric efficiency as expected. Deschamps and Baritaud conclude in this section of the paper that the interacting parameters they studied were too complex for control of residual gas distribution in an engine, and suggest choosing external EGR as a delivery mechanism instead. The remainder of the paper discusses EGR effects in a similar manner, only with the addition of emissions work.

49 36 3. Experimental Setup 3.1. Single-Cylinder Research Engine This project was performed on a single-cylinder, optically-accessible research engine mated to a regenerative AC dynamometer. For improved control of residual gas dilution, a dual overhead cam cylinder head was integrated. Calibrated air flow was delivered from a critical flow orifice rack and control of air-assisted fuel injection and spark timing was provided by a commercial engine control and calibration system Base Engine The base engine block for this project is the GM Research Triptane Base 4, originally designed for alternative fuels research in the late 1950 s. It is of two-part construction, with cast iron crankcase and cylinder barrel. The crankcase contains a balancing shaft and a single fixed two-lobe camshaft for pushrod actuation of an overheadvalve system. The cylinder barrel has been re-lined recently and contains a liquid coolant jacket. The firedeck surface includes a groove for an o-ring seal with the cylinder head spacer ring. The major fixed dimensions of the Triptane engine are provided in table 3.1.

50 Bore [mm] 92.4 Stroke [mm] 76.2 Displacement [cc] 511 Connecting Rod Length [mm] Exhaust Valve Open [CAD] 115 Exhaust Valve Close [CAD] 365 Intake Valve Open [CAD] 349 Intake Valve Close [CAD] Table 3.1. Fixed internal dimensions of GM-Triptane engine. Valve timings are for internal single camshaft used for OHV engine operation Optical Access The major feature of the Triptane engine is the Bowditch-type optical-access piston/cylinder geometry. The extended-height cylinder barrel accommodates the aluminum Bowditch piston and allows for mounting of the 45 mirror, which passes through the cylinder barrel and allows for a periscope view of the combustion chamber via a transparent piston cap. The piston cap is fabricated of aluminum and is fastened to the Bowditch piston with an internally threaded steel retaining ring. The cap contains an axially-centered 47 mmdiameter 10 mm-thick sapphire window. The fit of the cap into the retaining ring is indexed and the assembly locks with a small screw-fastened key. Cylinder sealing for the windowcap and cap-retainer surfaces is performed with Viton O-rings.

51 38 The piston rings used for this experiment are common to optical engine studies and unique in that they operate without a lubricating oil film on the cylinder wall. Custom manufactured by the C. Lee Cook Company based on dry gas compression technology, they are composed of a spring-loaded oil control ring and a bronze-impregnated Nylon rider ring below the mirror and an additional rider ring above the mirror. The single compression ring is of a butt-cut design and is made of Vespel. The compression ring groove is located in the steel retaining ring at a maximum height that does not cross the firedeck surface gap. The final component of the optical access system is the steel spacer ring fastened between the block and head, with an inside diameter matching the engine bore. The 25 mmtall ring contains four equally-sized window ports. Two ports contain 16.5 mm-thick quartz windows for laser sheet propagation through the combustion chamber. The other two ports are utilized for combustion diagnostics described in Section 3.2. Although the piston cap crosses the plane of the windows near TDC, the compression ring stays below the spacer ring throughout the cycle Cylinder Head and Combustion Chamber In the interest of generating a range of residual gas fractions for this experiment, a means of independent cam phasing was required. Since the base engine s OHV camshaft is of fixed geometry and difficult to access within the crankcase, a dual overhead camshaft (DOHC) single-cylinder research cylinder head of near-identical bore was obtained from GM Research Labs. Originally designed and used for gasoline direct- injection (GDI) studies, the

52 39 cylinder head contains intake and exhaust cams that are independently phase adjustable via taper-split drive pulleys. As an additional lab improvement, the DOHC cylinder head provided a combustion chamber geometry that is consistent with modern multi-valve SI engines. The pent-roof cylinder head contains two intake valves and two exhaust valves with an axially-centered M14 spark plug. The GDI injector bore (tangential, wall-guided orientation) was plugged in this project. One intake port is cast in a helical approach for swirl generation, which can be varied with a butterfly throttle on this port alone. This throttle was left full-open for this project to generate maximum flowfield turbulence. The major consideration in the integration of this cylinder head was its effect on compression ratio. The DOHC head s pent-roof occupies a 49.5 cc volume, whereas the Triptane engine s traditional OHV head is designed with a flat pancake roof. The requirements of the optical access system prevented modification to either the spacer ring or the Bowditch piston, so the highest-compression flat-top piston crown was used in this project. With an OHV setup, this piston yields a CR of over 12:1 for compression ignition studies, but with the DOHC head, we are able to obtain only 5.95:1. The upside of this arrangement is that the engine free-spins without any valve-piston interference, allowing infinite valve timing flexibility and reduced risk of catastrophic engine damage. During initial testing, the low compression engine was demonstrated to operate stably at elevated dilution conditions. The major combustion chamber dimensions for this project are summarized in Table 3.2.

53 Compression Ratio 5.95:1 Top-Ring Crevice Volume [cc] 4.13 Exhaust Valve Inner Seat Diameter [mm] 29.5 Intake Valve Inner Seat Diameter [mm] Table 3.2. Major combustion chamber dimensions for GM-Triptane engine with DOHC adjustable-cam cylinder head. Mating the 116 mm square bolt pattern of the DOHC head to the 117x86 mm rectangular pattern of the Triptane firedeck required fabrication of four hardened steel mating blocks for an offset 2-screw fastening method. The deck surface of the DOHC head was lowered by to accommodate a replaceable graphite head gasket to seal against the spacer ring. As mentioned, the spacer ring sealed to the firedeck with a Viton O-ring. Additional lab modifications for the DOHC setup are described in subsequent sections Valvetrain Timing System The major feature of the DOHC cylinder head is its independent cam phasing adjustment. This is accomplished with indexed taper-split pulleys on each camshaft. The inner flange half is permanently fastened and keyed to the camshaft along with a backing graduated degree wheel. The outer pulley half is fastened through radial slots against the internal taper. The slots were machined so as to allow access to any practical cam phasing arrangement.

54 Prior to this project, an external, belt-driven half-speed shaft was added to the 41 laboratory to provide a timing signal for engine control software. For the DOHC arrangement, this half-speed shaft was linked to the camshaft pulleys at a 1:1 ratio via a Gates 1 -width 52 -length 3/8 -pitch trapezoidal tooth timing belt. An automotive OEM 1 torsional belt tensioner was added to the half-speed shaft assembly for final belt tensioning. Despite its apparent complexity, this dual-belt timing system eliminated the need for a very long single timing belt and camshaft extensions, as well as providing required modularity with the OHV engine setup. Figure 3.1. Valvetrain timing layout for DOHC cylinder head.

55 42 The degree wheels for each camshaft were first calibrated to valve open/close events. The graduations on the degree wheels are to be read against markers bolted to the valve cover with the engine rolled to TDC compression. This baseline point was used because it is assumed that both cams will always be on the base circle at this time. Using a valve lift threshold of open/close timing and proper oil pressurization of the hydraulic valve lifters, the engine is manually rolled over with a crankshaft degree wheel, watching a dial indicator on the appropriate valve stem. This procedure does suffer from significant degree value uncertainty of valve timing, due to the effect of engine rotational speed on hydraulic lifter response. However, engine operation data have shown the timing system to be highly repeatable in terms of engine-out performance. A related procedure was developed for selecting valve timings during the experiment. The engine was rolled to TDC compression, and the cam taper was broken with the timing belt still taught. The pulley half is backed off slightly (not completely off) to allow relative rotation between the two halves. The flange half contains a hex nut that can be used to turn the camshaft to a different degree wheel position relative to the timing belt, which remains locked to the crankshaft. The pulley was then re-fastened on the taper. It is important in this procedure that the engine always is manually rotated in its proper counter-clockwise direction to avoid the multi-degree backlash in the belt tensioner and that the timing belt tension is preserved throughout the process.

56 Dynamometer 43 The crankshaft of the Triptane engine is connected, via a flywheel, to a three-phase 440 VAC General Electric dynamometer. The control system is a Reliance Electric Max Pak Plus VS Drive box. The dyno system can provide motoring or generating operation up to 1500 RPM and 30 kw load. Manual selection of engine speed is performed with a rheostat and feedback control loop, which is periodically optimized Engine Fluid Systems The Triptane engine and DOHC cylinder head are liquid cooled in a conventional block-thru-head loop, linked by external hoses. A 50/50 water-antifreeze mixture was used for corrosion resistance. The external, motor-driven circulating pump is pressure-fed by a standing column reservoir. An electric water heater is operated continuously during experiments to bring the engine to operating temperature and a feedback-controlled solenoid valve is used to meter cold building water through a copper counter-flow heat exchanger to maintain a system set point during operation. Previous optical engine projects have determined the optimum coolant temperature of 68 C. Oil pressure and flow rate was also provided by an external pump, with a commercial in-line filter. For this project, a Triptane internal post-main bearing oil gallery was selected to provide an external high-pressure feed to the DOHC head cam journals and valve lifters. A low-pressure drain line was also installed from the head to the reservoir in the crankcase. Engine oil selected for optical studies is SAE 40, for its higher viscosity and resistance to

57 infiltrating the combustion chamber. With low speeds, loads and temperatures, engine oil 44 grade selection is not considered critical to Triptane performance. To provide further defense against oil fouling of the optical access system, vacuum pumps are applied to both the crankcase and the cylinder head valve cover. Crankcase vacuum cuts down on blow-by past the oil control ring, which can quickly foul the turning mirror, camera lens and back surface of the piston window. Valve cover vacuum was necessary in this project to reduce oil migration past the valve stem seals, which fouls all combustion chamber windows Engine Aspiration Systems Intake air is metered through a critical flow orifice rack, where the upstream pressure was varied to obtain set mass flow rates. Three orifice diameters are used to provide an adequate range of air flow at the common engine operating speeds of 600 and 1200 RPM: 0.125, and All three orifices are calibrated with a bellows flow meter for the range of upstream pressures providing choked flow. Mass air flow rate is then determined from a density correction. During the experiment, the supply air (separated, filtered and humidity-controlled central compressed air) is monitored for consistent upstream properties with the calibration condition. For this project, only the smallest orifice is used, since all experiment conditions were throttled or sub-atmospheric intake manifold absolute pressure (MAP). Also for this consideration, a new pressure-tested rigid copper intake runner was fabricated to link the DOHC head with the existing 14 gal intake air surge tank. Intake MAP is monitored

58 45 with a Wallace & Tiernan 0.1-psi resolution absolute pressure gauge at the surge tank. An atmospheric intake vent is opened for all transient dyno speed selection periods. For the exhaust, a new steel runner was fabricated. Approximately 10 cm downstream of the cylinder, the runner contains an axially-located K-type thermocouple used to confirm thermal steady-state firing operation. Engine exhaust emissions are sampled from the near-exit centerline of a 12 gal mixing tank located 2.25 m downstream of the ports. This surge tank is positioned to allow for modularity with the OHV head setup and contains a perforated tube diffuser entrance for gas mixing. Exhaust back pressure is manually controlled with a gate valve and monitored on an absolute pressure gauge. All conditions in this project were set to 1.0 bar absolute back pressure Fuel Delivery System The fuel used in all measurements of this experiment is 80% iso-octane, 20% 3- pentanone, by volume. The tracer concentration was set as being the maximum level that did not attenuate the laser sheet intensity across the combustion chamber. Since this experiment involved a homogeneous, pre-vaporized mixture, the tracer was assumed to faithfully track the fuel. More thorough discussions of the co-evaporation properties of iso-octane and 3- penatnone are found in [14] and [15], where direct-injection spray studies mandated the extra consideration. Relevant properties of the fuel mixture are given in Table 3.3.

59 46 Iso-octane 20% 3-pentanone / 80% iso-octane Molecular Weight H:C Ratio Stoich. Air/Fuel Ratio Table 3.3 Fuel properties for pure iso-octane and the 20% 3-pentanone tracer blend used for this experiment. This optical project required that the intake air and vaporized fuel/tracer solution be thoroughly pre-mixed before entering the combustion chamber. To accomplish this, the fuel injector is mounted far (1.3 m) upstream of the intake ports. Based on previous optical engine studies, an Orbital air-assisted fuel injection system is used to provide the highest level of initial fuel/tracer atomization. Degree of homogeneity is analyzed in Section 4.3. The Orbital air-assist injector is primarily used in North America in the Mercury Marine Optimax 2-stroke outboard engine line, where it operates in a GDI arrangement. In this system, a fuel injector draws fuel from a pressurized fuel rail and fires into a mixing volume on the entrance to an adjacent air injector, which is within a separate 80 psi compressed air rail. A fixed delay of 4 ms occurs for initial air/fuel mixing before the air injector fires the mixture into the intake runner for a fixed duration of 3 ms. The critical operation parameter with this system is that the fuel supply rail be held at 10 psid above the compressed air rail (= 90 psi). Fuel delivery rate is varied with the initial fuel injection pulse width. Due to the use of pure iso-octane fuel and 3-pentanone tracer, the OEM high pressure pump and recirculating pressure regulator system are not used for the Orbital fuel rail. Pressurized fuel/tracer and air are instead supplied by a clean accumulator system. Mixed

60 47 fuel/tracer solution is drawn into the Tobul piston-type accumulator (Model 4.5A ) by a vacuum pump. The accumulator, which contains Teflon-encapsulated seals for chemical resistance, is then pressurized with nitrogen. After passing through a safety shutoff valve and a 0.5 micron filter, the fuel is regulated with a Go single-stage regulator (Model 00-HO2073) to the necessary 10 psi differential pressure. An Orange Research differential pressure gauge (Model 1516D6) is used to directly monitor this value during the experiment. Regulated compressed air for the injector is supplied by a medical-grade cylinder. Fuel delivery rate was calibrated before the experiment using a gravimetric technique for varying durations of fuel injector pulsewidth. The engine control system was operated using its internal timing generator mode set at 600 RPM. Permanent injector characterization settings along with fuel-air delay and air injector duration were fixed in software, with only the fuel injector duration varied. The air-assist injector rail was mounted on an atmosphericpressure mixing volume and drained through a ¼ tube into a capped glass beaker on an Ohaus Scout II digital scale. Overall mass flow rate (over a six minute duration after initial flow equilibrium) was then converted to a mass per injection value. As expected, the mass flow rate was linear across the delivery rate range of this project (10-18 mg/inj). Gain and offset were then entered into the engine control software. There is substantial uncertainty in the overall air/fuel ratio with this experimental setup. Particular sources are from the just-described injector calibration, which is not performed in a negative-pressure environment as found in the intake system, the well-known chamber sealing inefficiencies of the optical engine piston ring pack, uncertainties in the intake orifice calibration and small air leaks in the intake system.

61 Engine Control System 48 Control of fuel injection, ignition, sampling valve and camera triggering for this project was provided by the MotoTron commercial engine control and calibration package. Using triggers from an interpolated crankshaft encoder and a camshaft Hall-effect encoder, the ECU software is able to generate output signals with 1/16-CAD resolution. A major advantage of this particular package is its built-in support for the Orbital air-assist fuel injection system, including characterization and calibration software inputs and integrated electronic driver circuits. Furthermore, it provides up to six spark ignition TTL signals, which can be given independent timings and pulse widths for triggering of external systems. For this project, Mototron engineers generously provided the laboratory with a new programmed feature for skip-fired operation. When enabled, this mode provides a userdefined number of firing cycles to occur before cutting the ignition on a single cycle. During that skip-fired cycle, a TTL signal of user-defined timing and duration is activated for triggering of a sampling valve or camera. The ignition coil used for this experiment is a Mercury Marine DFI model, with maximum spark energy of 150 mj. Due to the high residual dilution levels and low compression ratio of this project, the maximum coil dwell was used at all times. Likewise, the AC Delco spark plug (Model ) was gapped to 2.1mm, a very large amount, but one that was proven to consistently sustain spark propagation.

62 3.2. Combustion Data Acquisition 49 To add relevance to the optical studies of this project, multiple combustion diagnostics were needed. Most important is cylinder pressure data, which in addition to allowing for optimization of performance at each running condition, allows for analysis of cylinder heat release rate and cycle-by-cycle power variations. To quantify bulk residual gas fraction at each running conditions, a solenoid-actuated in-cylinder sampling valve is used in conjunction with an exhaust gas emissions analyzer. This emissions bench is also used to describe general running trends in pollutant formation Cylinder Pressure Measurement Cylinder pressure was measured with an AVL model QC42D-E C109 water-cooled piezoelectric transducer. The charge output was sent to a Kistler model 5010 amplifier, operated with a medium time constant. The amplified signal was logged on a Hi-Techniques A/D conversion PC running REVelation software. Pressure readings were recorded in sets of 100 cycles at 0.25 CAD resolution, based on a simultaneous signal from a high-resolution BEI optical crankshaft encoder. The pressure transducer/amplifier were calibrated using a hydraulic dead-weight tester with an excellent resulting linearity (R 2 > 99.9%). The relative pressure signal was pegged to the intake MAP at -180 CAD in software. The REVelation program provided instant display of an averaged trace, IMEP and COV of the recorded data in the laboratory, but additional statistics and cycle-resolved data were only accessible using a binary data extraction program. Extracted pressure traces for each final

63 running condition were entered into a single-zone heat release code which operates in the 50 Engineering Equation Solver (EES) environment, for analysis of cumulative and instantaneous heat release rate. An important consideration in this project was the location of the AVL pressure transducer (M14 thread) within the combustion chamber. Space constraints in the 4-valve DOHC head prevented traditional roof access. An unused window in the optical spacer ring was selected and a special tapped window was machined. There were initial concerns about dynamic effects on the pressure trace at this location from the piston, which passes the window location near TDC, placing the transducer in the top ring crevice volume. A test was performed with the OHV head, using two pressure transducers (Figure 3.2). One was mounted in that head s roof location and the other in the window port. Both were logged simultaneously and compared. The window transducer followed the shape of the roof transducer exactly, except for a small (< 5%) deviation of peak pressure near TDC. The behavior of the cylinder wall location was deemed acceptable for this project.

64 51 Pressure Trace Comparison Roof Mount Wall Mount 1500 p [kpa] CAD Figure 3.2. Comparison of measured cylinder pressure traces at wall-mount location to conventional roof-mount. Motoring engine condition with OHV head, 1200 RPM Sampling Valve Residual gas fraction levels are measured by comparing the concentrations of CO2 trapped in-cylinder before ignition to that in the combustion products in the exhaust. To obtain the in-cylinder CO2 concentration, a solenoid-actuated sampling valve is installed in the head/block spacer ring in a modified steel window opposite the pressure transducer. The design details of this sampling valve are covered in [12]. The sampling valve outlets into a 1 m long ¼ Teflon line connected to an ice bath. The condensing ¼ stainless steel tube coil removes water from the stream and two external

65 52 coalescing filters remove impurities before passing the sampled gas stream through a 10 m long ¼ Teflon line to the emissions bench. The sampling valve 48 VDC driver circuit is triggered by the skip-fire MotoTron TTL signal. Skip-firing mode is used for cylinder sampling to ensure faithful measurement of pre-ignition trapped charge composition. Sampling valve timing and duration are optimized for the engine running condition to provide maximum flow rate to the emissions bench. The target flow rate is 2.5 lpm, measured by a rotameter at the bench entrance. Figure 3.3. In-cylinder solenoid-actuated sampling valve mounted to block-head spacer ring. Teflon sampled gas line travels to an adjacent ice bath and then to the analyzer. Before the experiment, the leakage rate past the sampling valve seat was measured during three continuous-firing baseline engine operating conditions without opening the

66 53 valve. A Hewlett-Packard ml soap bubble flow meter was used for this experiment. At the 600 RPM 64 kpa MAP load condition, the leakage was 1.11 ml/sec. At 600 RPM 46 kpa MAP, the leakage was reduced to 0.57 ml/sec. Finally, at 1200 RPM 50 kpa MAP, the leakage rate was 1.03 ml/sec. Given these values and a worst-case sample flow rate of 1.5 lpm, the highest possible leakage gas concentration was 4.4%. Since most conditions were assumed to be below this value, the sealing performance of the sampling poppet valve was deemed acceptable Emissions Bench A five-gas Horiba emissions analyzer was used to measure steady-state exhaust species concentration sampled from the engine exhaust mixing tank. After exiting the mixing tank, the sample was transported to the emissions bench by an electrically heated line. The line was temperature-controlled to 190 C to avoid hydrocarbon and water condensation. Adequate flow is provided by a vacuum pump in the bench and a regulated manifold tree to the individual analyzers. Before entering the infrared analyzers, water was condensed from the stream in a 0 C refrigerant bath. The five analyzers were CO, CO2, O2, HC and NOx, each paired to a signal amplifier. The CO and CO2 analyzers were calibrated through a Stec gas divider to a second-order polynomial fit of voltage vs. concentration. The remaining three analyzers were linear in operation and required only two-point calibrations. Table 3.4 summarizes the Horiba emissions bench hardware.

67 54 Gas Span Level Horiba Analyzer # Analyzer Type Horiba Amplifier # CO % AIA-23 ND-Infrared OPE-135 CO 2.56 % AIA-23 ND-Infrared OPE-115 O % MPA-21A Paramagnetic OPE-335 NO x 101 ppm CLA-22A Chemiluminescent CLA-22A HC 6286 ppm (C 3 H 8 ) FIA-23A Flame Ionization Detector (FID) Table 3.4. Horiba exhaust emissions analyzer bench summary. FIA-23A All amplifier output signals were passed through an A/D converter card and logged on a PC using LabView 6.0. LabView was used to automatically perform the voltage calibration and average multiple samples for final data reporting. NO x measurement during this experiment was precluded by the inability of the analyzer to achieve a steady-state during the short firing duration of the optical engine. The engine could not be fired continuously for more than five minutes, where NO x readings were still increasing for all conditions. Therefore, NO x will not be reported in the results section. To rapidly switch from exhaust emissions measurement to the sampling valve stream during the experiment, the front side calibration port of the CO 2 analyzer was used to receive the line from the ice bath. In this arrangement, the low-flow sampling valve stream was supplied exclusively to the CO 2 analyzer, which was the only measurement needed. Before and after sampling valve measurement runs, the valve plumbing system was purged with nitrogen.

68 3.3. Optical Measurement System 55 The mixing of the residual gas with the fresh homogeneous fuel/air charge in the combustion chamber of the engine was performed using planar laser-induced fluorescence (PLIF). The measurement system consisted of a laser source, beam-transport and sheetforming optics and a camera. The optical system was synchronized with the engine crankshaft to capture PLIF images at several crank angles during the compression stroke Laser Source The laser used in this project was the Spectra-Physics GCR-170 Nd:YAG. The fundamental 1064 nm output was frequency quadrupled to 266 nm, providing a pulse duration of 4-5 ns and a peak energy of 90 mj/pulse. Pulse-to-pulse energy stability is listed as <10%. The laser was externally triggered at its design repetition rate of 10 Hz, with an optimized delay between flash lamp and q-switch trigger signals of 186 µs. Laser power, which could only be measured safely before the last two optical elements ~35 cm upstream of the engine entrance, was monitored with a ScienTech Mentor MD-10 power meter with a UV-sensitive power head. Laser energy at this point was adjusted to 30 mj/pulse +/- 3 mj, although this value is an integrated time average and can not be resolved to a pulse-by-pulse basis. The Nd:YAG laser was able to supply more power, but damage to the engine quartz windows near the focused laser sheet prevented higher powers.

69 Laser Optics 56 The 266 nm output of the Nd:YAG laser was separated from the higher harmonics by a Pellin-Broca prism located at the exit of the laser. The visible and infrared beams were captured by beam dumps. The 266nm beam was again turned 90 by a dichroic mirror to traverse the length of the laser table to a second Pellin-Broca Prism for final wavelength separation. A dichroic mirror directed the 266 nm beam into a 1 m focal length spherical lens, designed to focus the laser at the center of the engine bore. After the spherical lens, the beam was vertically traversed by two right-angle prisms from laser output height to engine window height. The traverse distance is ~10 cm and was adjusted by a micrometer translation stage. A 100 mm positive focal length cylindrical lens was located 60 cm downstream of the spherical lens to develop the laser sheet. Finally, a 2 in diameter dichroic mirror was used to direct the laser sheet into the engine s through the quartz windows in the space ring. The optical system is presented in Figures 3.4 and 3.5 for clarity.

70 57 Figure nm laser pulse separation and delivery optics (plan view). Figure 3.5. Laser sheet-forming optics setup for 266 nm PLIF imaging.

71 Camera 58 The primary camera used for PLIF imaging in the Triptane engine was the Roper Scientific MicroMax. This camera lies in a category of scientific CCD cameras known as the frame-straddling type and is specified with a nominal quantum efficiency of 45% at the peak fluorescence wavelength range from Figure 2.7. The MicroMax CCD is front-side illuminated and cooled thermoelectrically to -20 C. A twin-blade fast mechanical shutter is used to protect the device from combustion luminosity. The CCD array measures 1300x1030 pixels with a pixel size of 6.7 µm. In this experiment, the camera was binned on-chip 6-by-6 to increase PLIF signal and shorten read-out time. The binning selection is discussed in Section The readout rate for the device is 5 MHz, with 12-bit digitization. The MicroMax camera is not intensified. The principal design feature of the MicroMax is a Double-Image Feature (DIF) mode designed for particle-image velocimetry (PIV), which allows two separate exposures to be captured on the CCD array in rapid succession without mechanical shuttering. This is accomplished by interline transfer on-chip, which is the reason why this camera was used in the project. In DIF mode, the mechanical shutter is pre-opened once the camera is finished reading the previous image and the chip is actively drained of charge before the exposure trigger arrives. The CCD is divided into alternating columns of masked and unmasked pixels in this mode and the unmasked pixels are charged during the first exposure for a time programmed in software as short as 1 µs. The camera then performs the interline transfer of the charged pixels over to the masked pixels, which requires 200 ns, or some longer programmed duration. Then, a second exposure is taken on the unmasked pixel columns.

72 59 The second exposure has no place to be shifted to, so it must be exposed until the mechanical shutter is closed for read-out. A diagram of DIF-mode operation is shown in Figure 3.6. The basis for selecting the MicroMax over other high-quality CCD cameras is discussed in Section 5.2. In this project, the exposure time was set to 10 µs, with the second image from each engine cycle deleted from the analysis. When pixels were binned 6-by-6, the resulting readout time was 0.3 sec, easily fast enough to keep up with our engine skip-firing frequency. MicroMax images were saved on a Pentium III Windows PC operating Roper Scientific WinView/32 v as multiple-frame 12-bit grayscale TIFF files. The lens selected for the primary camera was an 85 mm Nikkor f/1.4 model mounted on a 20 mm Kenko extension tube in addition to a C-mount to F-mount adapter. The selection criteria for this lens are detailed in [15]. Figure 3.6 MicroMax camera manual summary of DIF-mode timing. Image exposure times are shown in the second line. Ready and Scan are output signals from the camera controller, Ext. Sync is the input trigger TTL, Laser Output shown is for a double-pulse laser, this experiment only uses the first pulse. Reprinted from [24].

73 Optical Triggering 60 The requirements of the optical triggering system were threefold: 1) To supply the Nd:YAG laser with an uninterrupted 2-pulse trigger sequence at 10 Hz; 2) to ensure that the camera triggers were delivered on skip-fired cycles only; and 3) to gate both the primary and reference camera with sub-microsecond resolution to capture the 5 ns laser pulse. To perform this, the BEI crankshaft encoder and the MotoTron engine control were each utilized with an interface at a special TTL logic circuit. The clock signal for the laser and both cameras was provided by the shaft encoder, via a TTL counter box supplied by Mercury Marine. The A-pulse (divided by 4 to one-per- CAD frequency) and the Z-pulse (TDC) were input into the counter box. An advance or delay value of 0-99 CAD before or after TDC could be selected with surface-mounted control switches. One TTL pulse per engine revolution was output from the counter box at the input advance/delay timing. A Berkeley Nucleonics model 555 pulse/delay generator was used to supply the highprecision trigger signals to the optical system. This device was operated in external gate mode with an input clock signal from the counter box. At 600 RPM, the counter box frequency (one per revolution) was already at 10 Hz, while at 1200 RPM, the pulse/delay generator had to be operated in divide-by-2 mode, where synchronization with the compression stroke had to be verified. The outputs of the pulse/delay generator were used to provide the signals summarized in Table 3.5, with delays relative to the leading edge of the counter box pulse:

74 Channel Device Width Delay A Flash Lamp ms µs B Q-Switch ms µs C MicroMax (primary camera) ms µs D PI-Max (alternate camera) ms µs 61 Table 3.5 Trigger timing delays for optical measurement system. Delays are relative to the leading edge of the trigger signal from the crankshaft encoder. The laser trigger pulses were delivered to the Nd:YAG directly to provide the uninterrupted 10 Hz operation. The camera triggers were sent to two one and only one TTL logic circuits (chip #4013), where they provided the clock input to the circuit diagram shown in Figure 3.7. The MotoTron skip-fire TTL signal, previously used for activating the sampling valve, was supplied as the enable input, at a timing of -180 CAD. On receipt of the enabling signal, the circuit outputs the next clock signal and only that one pulse. This allows the camera to capture the laser sheet that is fired on the compression stroke of the skip-fired engine cycle and prevents the MicroMax camera from being triggered during a combustion cycle. With the camera exposure set to 10 µs, the circuit s insertion loss (predicted to be ~10-20 ns) does not affect the capturing of the laser pulse.

75 Figure 3.7 Schematic for TTL timing of laser pulse and camera, synchronized with MotoTron skip-firing ignition by a one-and-only-one circuit. 62

76 63 4. Engine Operating Conditions 4.1. Selection Criteria Utilizing the adjustable-camshaft feature of the cylinder head, conditions for varying levels of residual gas dilution were established at our two operating speeds. For this project, five categories of cam-phasing strategies were covered: a baseline valve overlap, an enlarged valve overlap symmetric about TDC exhaust, an enlarged valve overlap with the intake cam advanced from the baseline, an enlarged valve overlap with the exhaust cam retarded from the baseline, and finally a zero-overlap (IVC=EVO) setting. At each of these five categories, three engine conditions were established: a low load at 600 RPM, a mid load at 600 RPM, and a low load at 1200 RPM. The result was a test matrix of 15 distinct engine operating conditions Optical Engine Considerations The objective of this project was to study conditions of high residual gas fraction. Severe limitations in operating a fired optically accessible engine made this objective challenging. The low compression ratio (5.95:1), despite being conducive to increased trapped residual gas levels, impacts the ignition and combustion stability of the engine. To address this, fuel was delivered at consistently rich conditions to aid the spark ignition (with the added benefit of increased tracer density for optical measurement). Additionally, the

77 spark plug was gapped to 2.1mm and the ignition coil was permanently set on maximum 64 dwell. The second major limitation was from thermal loading of the oil-less ring pack for the Bowditch piston. Despite the relatively cold block temperature ( 3.1.6), the Triptane engine could not be fired continuously for more than four minutes at the 1200 RPM and 600 RPM mid load conditions or six minutes for the 600 RPM low load conditions. Mass loss through the non-metallic ring pack and the temporally evolving nature of the mass loss under firing operation introduce uncertainty into cylinder pressure analyses ( 4.2). Secondly, time-limited firing operation effects steady-state exhaust gas emissions measurement uncertainty ( 4.3). A final mention before proceeding must be given to the consideration of engine speed, bulk flowfield and manifold wave dynamics when extending optical engine combustion data to conventional SI engines Establishing Engine Conditions A baseline valve overlap duration of 20 was selected, based on a presumed-typical value for a 4-valve engine of 510 cc displacement operating at RPM. For simplicity, this overlap duration was positioned symmetrically about TDC exhaust. Intake air mass flow rates were established at the baseline valve overlap for the three engine speed/load combinations, and were held consistent for the other overlap cases. In establishing the baseline air flow rates, fuel delivery and spark timing were both freely adjusted to optimize IMEP and COV of IMEP. The 600 RPM low load condition was set by varying the air mass flow rate to find a comfortable minimum for stable combustion operation. Based on the measurement

78 requirements of both the optical and sampling valve techniques, a combustion stability 65 criteria of <10% COV of IMEP was established. Since the objective was to increase dilution with increased valve overlap from this baseline condition, an absolute minimum air delivery rate was not chosen. The air mass flow rate chosen for the five 600 RPM low load conditions, 144 mg/cycle, resulted in an intake MAP of approximately 50 kpa at the baseline overlap. The 1200 RPM low load condition was established by adjusting the intake mass flow rate to provide the same MAP as 600 RPM low load at the baseline overlap. The result was 181 mg/cycle, although IMEP and exhaust temperatures were significantly higher at the increased speed. The 600 RPM mid load point was established by increasing the intake mass flow rate to a comfortable upper limit for safe engine operation. At 208 mg/cycle, peak cylinder pressure was at 15 bar and steady exhaust temperature was at 400 C, both acceptably close to the upper limit. Intake MAP increased from the low load condition to 61 kpa. A mid load point at 1200 RPM could not be established due the safety limits in combustion pressures and temperatures. Likewise, full-load atmospheric-map firing operation could not be performed at either speed, so no quantitative pressure-based reference for load points could be established. Therefore, only intake air mass flow rate is used as a basis for engine load. With air mass flow rates fixed, the advanced valve overlap conditions could be established. Both fuel mass and spark timing were varied and 10% COV was used an upper limit for combustion stability. Since the engine was less tolerant of increased overlap at 600 RPM compared to 1200 RPM, different overlap levels were tested for the two speeds. At all 600 RPM increased-overlap conditions (both loads), a total valve overlap duration of 30 was set. This 10 increase from baseline was established from the maximum amount tolerable at

79 the low-load condition at either intake-advance, exhaust-retard or symmetric-increase. At RPM, a 40 increase in overlap from baseline (60 total) was achieved for all three cam strategies. 12 Symmetric Overlap Increase 12 Intake Cam Advance 10 Exh. Int. 10 Exh. valve lift [mm] Int Baseline RPM Exhaust Cam Retard 1200 RPM 12 Zero Valve Overlap valve lift [mm] Exh. Int Exh. Int CAD atdc CAD atdc Figure 4.1 Summary of four valve overlap strategies. Baseline cam timing is indicated by the dashed line in all plots. Arrows indicate cam shift from baseline. The baseline overlap duration is 20, the 600 RPM extended overlaps are 30 duration, and the 1200 RPM conditions are 60 overlap duration. Fuel delivery rate was set by the highest injection mass required in the increased overlap conditions for each of the three speed/load combinations. Fuel mass flow was then fixed for all overlap conditions at the particular speed/load, to provide consistency in the

80 optical measurements. A summary of the targeted air/fuel delivery rates for each of the 67 speed/load combinations is shown in Table 4.1. Spark timing was not fixed in this experiment, but was optimized at each of the 15 test conditions to provide consistent combustion phasing. Mass Air Flow Fuel Injection Targeted AFR 600 RPM, Low Load 144 mg/cycle 14.5 mg/cycle 9.93:1 600 RPM, Mid Load 208 mg/cycle 18.0 mg/cycle 11.56: RPM, Low Load 181 mg/cycle 18.0 mg/cycle 10.06:1 Table 4.1. Air/fuel engine operation parameters for the three experimental speed/load points. These values were held constant for each cam strategy. A complete summary of engine operating conditions recorded in the laboratory can be found in the master summary in Appendix A Combustion Analysis Cycle-resolved cylinder pressure data were recorded throughout this experiment using the system discussed previously. Ensemble-averaged indicated mean effective pressure (IMEP) and pumping mean effective pressure (PMEP) were directly calculated by the post-processing program along with COV of IMEP, which was used as a basis for combustion stability. Averaged pressure traces were also used to compute heat release information for each of the 15 test conditions.

81 Cylinder Pressure Data 68 Table 4.2 contains a summary of the IMEP and PMEP data for the test conditions. The data are organized by valve overlap strategy and include a percent change relative to the baseline overlap condition for the particular speed/load point. As previously discussed, combustion phasing was kept consistent (location of peak pressure ~ 15 atdc) for all conditions. Cam Strategy Baseline Overlap Symmetric Increase Intake Advance Exhaust Retard Zero Overlap Speed [RPM] IMEP [kpa] COV of IMEP /Load (gross) [%] PMEP [kpa] 600 low () 4.1 () 46.2 () 600 mid () 2.05 () 35.3 () 1200 low () 1.32 () 51.1 () 600 low (+ 3%) 6.03 (+ 47%) 45.0 (- 3%) 600 mid (+ 4%) 2.07 (+ 1%) 30.3 (- 14%) 1200 low (+ 11%) 1.2 (- 10%) 35.3 (- 31%) 600 low (+ 6%) 3.89 (- 5%) 45.4 (- 2%) 600 mid (+ 3%) 2.46 (+ 17%) 31.1 (- 12%) 1200 low (+ 14%) 1.84 (+ 40%) 32.2 (- 37%) 600 low (+ 7%) 3.77 (- 8%) 43.5 (- 6%) 600 mid (+ 4%) 2.64 (+ 20%) 29.1 (- 18%) 1200 low (+ 6%) 3.64 (+ 176%) 35.1 (- 31%) 600 low (- 4%) 2.99 (- 27%) 51.5 (+ 12%) 600 mid (- 2%) 1.2 (- 41%) 39.2 (+ 11%) 1200 low (+ 0%) 0.87 (- 50%) 53.3 (+ 4%) Table 4.2 Mean effective pressure data for 100-cycle average pressure data at all experimental conditions. Percentages shown are changes relative to the baseline overlap condition for the individual speed/load points at each cam strategy. From the trends shown in Table 4.2, it can be seen that the baseline overlap selected was not an ideal setting for any of the three speed/load points. The 600 RPM low load condition was least sensitive to the cam-phasing changes, but nevertheless showed a

82 69 consistent increase in IMEP with extended overlaps. The 600 RPM mid load points enjoyed larger volumetric efficiency improvements (12 to 18 %) with the increased overlap durations with minor increases in IMEP, indicating that the pumping improvements were nearly overshadowed by losses due to charge dilution and reduced compression/expansion times. The 1200 RPM points show the most significant reductions (> 30%) in pumping work, and also the largest gains in IMEP, with the notable exception of the exhaust cam retard case. The zero-overlap condition predictably reversed the trend due to penalties in pumping work Heat Release Analysis A single-zone heat release code was used to process ensemble-averaged cylinder pressure data. The code, which runs in the EES equation solver, uses an iterative optimization scheme within the numerical integration to obtain wall heat transfer characteristics. Combustion efficiency was calculated at each condition, using exhaust gas measurements ( 4.3.2), with a resulting range from 72% to 90%. The limits of integration were optimized for each of the 15 engine conditions to obtain appropriate cumulative heat release traces. The graphical results of the analysis are presented in Figures , organized by speed/load point.

83 RPM Low Load Baseline Symmetric Incr. Zero Overlap Intake Advance Exhaust Retard Heat Release Rate [kj/deg] CAD atdc Cumulative Heat Release Baseline Symmetric Incr. Zero Overlap Intake Advance Exhaust Retard CAD atdc Figure 4.2 Heat release rate and cumulative heat release for all cam strategies at 600 RPM Low Load.

84 RPM Mid Load Baseline Symmetric Incr. Zero Overlap Intake Advance Exhaust Retard Heat Release Rate [kj/deg] CAD atdc Cumulative Heat Release Baseline Symmetric Incr. Zero Overlap Intake Advance Exhaust Retard CAD atdc Figure 4.3 Heat release rate and cumulative heat release for all cam strategies at 600 RPM Mid Load.

85 Baseline Symmetric Incr. Zero Overlap Intake Advance Exhaust Retard Heat Release Rate [kj/deg] CAD atdc Cumulative Heat Release Baseline Symmetric Incr. Zero Overlap Intake Advance Exhaust Retard CAD atdc Figure 4.4 Heat release rate and cumulative heat release for all cam strategies at 1200 RPM Mid Load.

86 73 The most immediate trend from the plot sequence is the long tail of the heat release curves for this engine, which extend far into the expansion stroke and approaches the EVO timing. Even the baseline and zero overlap conditions demonstrate protracted burn durations, demonstrating the influence of the low compression ratio and inherently high trapped residual mass of the engine. Table 4.3 summarizes the flame development duration and overall burning duration, defined as crank angle 0-10% cumulative heat release and 10-90% heat release [1], respectively. Once again, trends relative to the baseline 20 overlap are included at each condition. Cam Strategy Baseline Overlap Symmetric Increase Intake Advance Exhaust Retard Zero Overlap Speed [RPM] /Load Flame Development Angle (0-10% HR) Overall Burning Angle (10-90% HR) 600 low 53 () 59 () 600 mid 36 () 60 () 1200 low 36 () 68 () 600 low 61 (+ 15%) 79 (+ 34%) 600 mid 41 (+ 14%) 67 (+ 12%) 1200 low 57 (+ 59%) 87 (+ 28%) 600 low 61 (+ 14%) 75 (+ 27%) 600 mid 43 (+ 20%) 67 (+ 12%) 1200 low 57 (+ 59%) 84 (+ 24%) 600 low 61 (+ 14%) 78 (+ 32%) 600 mid 42 (+ 17%) 70 (+ 17%) 1200 low 60 (+ 68%) 93 (+ 37%) 600 low 40 (- 25%) 60 (+ 2%) 600 mid 27 (- 27%) 59 (- 2%) 1200 low 33 (- 7%) 72 (+ 6%) Table 4.3 Flame development angles and overall burning angles for different overlap strategies, determined by a single-zone heat release code. Percentages indicated are changes relative to the baseline overlap condition at each speed/load point.

87 74 The extended valve overlaps have a more pronounced effect on the heat release data than the IMEP data of Table 4.2. At 600 RPM low load, the 10 increases in overlap duration show up clearly in the 10-90% burn angle, with a consistent 25% increase, and a smaller influence on the flame development angle. The overall increase in the burn duration was lowest for the elevated load (600 mid) conditions. At 1200 RPM, where the largest valve overlap extensions occurred and the highest residual fractions were anticipated, the effect on early flame development was most severe. Since the baseline overlap at 1200 RPM was predicted to be one of the lowest-residual conditions, this result was not surprising. Higher turbulence levels at 1200 RPM help explain the lessened impact on 10-90% burn duration at the high-overlap conditions. Once again, the 1200 RPM exhaust cam retard condition is an outlier, showing a larger impact than either symmetric increase or intake advance Exhaust Gas Emissions Measurement Downstream exhaust gas emissions were recorded for the 15 established test conditions as part of the residual fraction measurement described in Section 4.4. CO2 readings were critical for those measurements, but the additional CO, HC and O2 readings are used here to better-quantify air/fuel ratio and combustion efficiency.

88 Emissions Measurement Procedure 75 Downstream emissions were recorded during continuously-fired operation with the sample line feeding all five gas analyzers. Readings were acquired at 30 second intervals during the time allowed for engine firing four minutes at 600 RPM mid load and 1200 RPM, six minutes at 600 RPM low load. At that time, the engine had to be stopped (1200 RPM) or motored (600 RPM) for several minutes to avoid damage to the piston rings. The readings were then graphically reviewed to determine the region of steady-state behavior, and the corresponding measurements in that region were averaged to yield the reported measurement value. The NOx analyzer was not able to achieve steady-state in any of the measured conditions. Rather than report suspect values, NOx measurements will not be included in the project. With the highly fuel-rich operating conditions and low exhaust temperatures, readings were predicted to be low (<< 1000 ppm) and should be negligible in combustion calculations Emissions Analysis Since all analyzers were recorded as dry measurements from the exhaust sample line, a dry-to-wet correction factor had to be applied to raw the emissions bench data. This value, K exh, is defined as:

89 K exh = n exh ( nexh + nh O exh ) 2, (4.1) 76 where n exh and n H2 Oexh, are derived from a carbon balance and the fuel s combustion stoichiometry (see 3.1.8). The concentration of H 2 gas in the rich combustion products was determined by the equation: [ H ] 2 dry ( H : C) [ CO] fuel = (4.2) dry 4 Knowledge of the incomplete combustion species, the water concentration and the dry-to-wet correction factor allows for computation of the air/fuel ratio by way of a chemical balance. Combustion efficiency was determined by the wet-basis mole-fraction equation [25]: N 100 [ HC] + [ CO] + [ CO ] {[ ] CO, T = 298 K + [ 2 ] H, 298 [ ] 2 T = K + HC, T = 298 K } N CO h H h HC h ηc (%) = 100 (4.3) h fuel, T = 298K 2 (4.4) The molar enthalpy of combustion for the unburned HC is assumed to be equal to that of the fuel. All concentration measurements are to be used on a wet basis.

90 Emissions Measurements 77 Steady-state downstream exhaust emissions measurements are summarized in Table 4.4. Analyzer readings shown are on a wet basis, after correction. Cam Strategy Baseline Overlap Symmetric Increase Intake Advance Exhaust Retard Zero Overlap Speed [RPM] /Load [CO 2 ] [CO] [O 2 ] HC [ppm C 1 ] AFR η c 600 low , % 600 mid % 1200 low % 600 low , % 600 mid % 1200 low % 600 low , % 600 mid % 1200 low % 600 low , % 600 mid % 1200 low , % 600 low , % 600 mid % 1200 low % Table 4.4 Summary of exhaust emissions species measurements, concentrations shown are corrected to a wet basis from the raw readings. Air/fuel ratio and combustion efficiency coefficient have been calculated from the concentration data. The combination of fuel-rich stoichiometry and the abnormally large crevice volume of the Bowditch piston contributed to the very high hydrocarbon readings shown in Table 4.4. Pressure data summarized in Section were taken simultaneous with these emission measurements and did not indicate a single misfire (less than 1 bar IMEP) for any of the 15

91 78 conditions. When the heat release data of section are considered, it seems likely that incomplete combustion at EVO was a major culprit in HC concentration, particularly at the two low-load conditions. It is notable that the intake cam advance strategy demonstrated the lowest CO readings in the experiment and the lowest HC readings for extended overlap at all three speed/load points. The 600 RPM mid load condition was least sensitive to overlap strategy in terms of pollutant emissions. Combustion at the 600 RPM low load condition fared the poorest at the symmetric overlap increase condition, with comparable results found at the two asymmetric overlap extensions. At 1200 RPM, the worst-performing strategy was once again the exhaust retard case, consistent with the pressure and heat release data. The O2 readings were higher than would be anticipated at the equivalence ratio indicated by our recorded CO concentrations. Since the highest O2 measurements were consistently found at 600 RPM low load (~ 1%), where IMEP was lowest, it is proposed that the high readings are a consequence of thermally-dependent sealing inefficiencies in the ring pack during cylinder scavenging. The exhaust system was kept under positive back pressure (above ambient) throughout the experiment and was newly constructed and leak-tested. Furthermore, the emissions vacuum sampling line was tested with pure N2 gas with a successful zero reading on the O2 analyzer. Air/fuel ratio calculations were consistent across both of the 600 RPM load points. A significant variation was encountered at the 1200 RPM condition, which is believed to be a consequence of difficulty reading the time-limited measurement RPM were the hardest-running conditions in the experiment, with the high piston speeds and gas temperatures. All analyzer-derived AFR readings were higher than the targeted delivery

92 rates, indicating some error in the fuel injector and/or intake air orifice calibrations 79 (discussed in ) Bulk Residual Gas Fraction Measurement For this project, the primary basis for comparing optically measured residual gas mixing phenomena is the bulk in-cylinder residual gas mass fraction (y r ). To measure this value, the fast-acting sampling valve was installed in the Triptane engine to measure incylinder CO2 as described in Section Sampling Valve Measurement Technique The sampling valve experiment was performed simultaneously with the exhaust emissions measurements described in the previous section. In-cylinder CO2 readings were taken under skip-fired operation at each condition immediately prior to switching to continuous-firing for the downstream sampling and pressure data logging. The skip-fired sequence is graphically presented in the pressure trace of Figure 4.5. The settings for the sampling valve and the sampled gas flow rates are summarized in Table 4.5.

93 80 Figure 4.5 Skip-firing sequence example (1200 RPM baseline overlap shown). Sampling valve is actuated on compression stroke of skip-fired cycle (see Table 4.5). 600 RPM 1200 RPM Sampling Valve Open 56 btdc 75 btdc Valve Open Duration 17 ms 14 ms Sampling Valve Close 5 atdc 25 atdc Sampling Frequency 4 cycles 6 cycles Table 4.5 Sampling valve operation for all experimental conditions. Sampling frequency is listed as the number of fired cycles between sampled cycles (see Figure 4.5). Like the downstream emissions measurements, the skip-fired in-cylinder readings were limited by the continuous-firing time limits of the engine. In a similar procedure, the

94 81 steady-state reading of the CO2 analyzer was determined graphically. Measurements taken at the two low load condition sets at the smaller baseline and zero overlaps demonstrated a weaker flow to the bench, which required fully opening the flow control valve on the analyzer entrance, resulting in flow pulsations to the instrument. All other conditions were able to operate with a non-pulsating flow delivered to the bench at the analyzer s optimal flow rate of 2.5 lpm Fired Cycle Sampled Cycle Valve Lift Transducer Signal P [kpa] CAD atdc Figure 4.6 Sample pressure data for skip-fired cycle with sampling valve actuation. The average fired cycle pressure trace and the sampling valve lift transducer signal for that skipfired cycle (no physical units) are overlayed RPM exhaust cam retard condition shown.

95 A concern with measuring in-cylinder composition with this technique is that the 82 mixture trapped in-cylinder on the skip-fired cycle be representative of the bulk cylinder charge composition. Particularly concerning was the possibility of the influence of partialburn or misfire on the cycle prior to the sampling valve open event. Even at high dilution conditions, this was not found to be a major problem, as can be seen for the histograms of prior-cycle IMEP shown in Figures 4.7 and 4.8, which are the two conditions having the slowest burning rate for the two engine speeds.. Figure 4.7 Frequency histogram of prior-cycle IMEP for skip-firing operation at 600 RPM low load symmetric overlap increase condition. Data compiled from 100 consecutive sampled cycles.

96 83 Figure 4.8 Frequency histogram of prior-cycle IMEP for skip-firing operation at 1200 RPM exhaust retard condition. Data compiled from 100 consecutive sampled cycles Residual Gas Fraction Calculations The residual gas fraction was calculated on a molar basis by comparing the mole fractions of CO2 in the compressed charge, xco2,cc, with the downstream exhaust measurement, xco2,exh. xco 2, cc xr = (4.5) x CO2, exh

97 The dry-to-wet factor for the exhaust, K exh, was determined in Equation 4.1 and used to 84 convert the denominator term from its raw dry-basis analyzer reading. The dry-to-wet factor is different in the compressed charge, and is defined in [26] to be: K cc 1+ Kexhx = 1+ x r r (4.6) Equations 4.1, 4.5, 4.6 and the raw CO 2 analyzer readings can be used to iteratively solve for x r. It is assumed that the molecular weights of the compressed charge mixture and the exhaust gas mixture are equal, and that the mole fraction x r is then equal to the mass fraction y r. Additionally, the small relative humidity in the dried intake air is neglected Residual Gas Fraction Measurements The results of the bulk residual gas fraction study are presented in Table 4.6 organized by cam phasing strategy. As in previous sections, the percent change from the baseline is provided for y r. Values for x and K CO 2, cc cc are in the master conditions summary in Appendix A.1. The intake cam advance strategy is shown to yield the smallest increases in residual fraction. For the two 600 RPM loads, the symmetric overlap extension provided the largest residual gas fraction increase. Since the baseline overlap condition at 600 RPM low load was already at 37 % residual fraction, it is not surprising that the engine was not especially tolerant of increased valve overlap durations there. The largest residual gas fraction was

98 85 measured at 1200 RPM with the exhaust cam retarded, which is consistent with the analysis of the pressure, heat release, and emissions measurements. In general, the range of residual fractions covered (21.9% %) is believed to be representative of those encountered in a high-dilution engine design. It is not a major concern for this project that the lower end of the measured y r range does not correlate well with typical values in conventional SI engines (5% - 25%). Cam Strategy Baseline Overlap Symmetric Increase Intake Advance Exhaust Retard Zero Overlap Speed [RPM] /Load y r 600 low () 600 mid () 1200 low () 600 low (+ 7%) 600 mid (+ 21%) 1200 low (+ 60%) 600 low (+ 3%) 600 mid (+ 10%) 1200 low (+ 50%) 600 low (+ 6%) 600 mid (+ 10%) 1200 low (+ 64%) 600 low (- 24%) 600 mid (- 17%) 1200 low (- 20%) Table 4.6 Summary of bulk residual gas fraction measurements at all experimental conditions. Percentages shown are changes relative to the baseline overlap condition at each individual speed/load point.

99 5. Imaging System Development and Analysis PLIF Image Processing Before discussing the selection of the camera for the experiment, an overview of the procedure for correcting planar laser images in engines will be covered. First the image acquisition sequence at each measurement condition will be described, followed by the numerical corrections applied to the images to extract a faithful representation of the tracer molecules in the laser sheet. Finally, the statistical processes used to evaluate and quantify the inhomogeneity of the fresh charge in the engine are discussed Image Acquisition Procedure Each PLIF measurement consists of three series of TIFF-format grayscale intensity images a background image sequence, a flatfield image sequence, and a data image sequence. These images are all acquired at the same crank angle timing (engine motoring/firing), with the camera focused on the laser sheet plane and the room lights turned off. Fifty background images are acquired before the data measurement and 50 more are acquired afterward. The background images are taken with the laser sheet firing through the combustion chamber, but without tracer addition (i.e. no fuel injection). Background images contain signal from blemishes in the turning mirror and piston window, as well as from

100 scattered laser sheet light impinging on cylinder head roof surfaces. These images are 87 subtracted from all subsequent images to better isolate the laser sheet. A sample 100-image mean background is shown in Figure 5.1, where all four valves, the spark plug and the periphery of the piston window are all visible in the full-ccd image. Figure 5.1 Sample 100-image mean background image. Pixel intensity scale is on right. Following the first set of 50 background images, 100 flatfield images are acquired with the fuel injector activated and the ignition disabled. With the far-upstream fuel injection ( 3.1.8), this mode is considered to provide a very nearly homogeneous in-cylinder mixture. Laser-induced fluorescence images of the homogeneous tracer distribution are used to perform a flatfield correction of the data images, whereby the laser sheet intensity profile is normalized. In this project, the flatfield image sets are additionally useful, since they can be used to define the homogeneous image condition, when the mean flatfield correction is applied to the 100 individual flatfield images. Since the flatfield images are acquired while motoring, the residual gas contains fuel/tracer and these images can be used for comparison

101 with the fired data images which contain residual gases. Provided the intake charge is 88 thoroughly pre-mixed ( 5.3), the corrected homogeneous images provide the statistical definition for a completely mixed cylinder charge, limited by the detection system signalto-noise characteristics. Figure 5.2 contains a mean flatfield image acquired 30 btdc, and demonstrates the spatial variation in laser sheet intensity. Figure image mean flatfield image, 30 btdc 600 RPM Mid Load Exhaust Retard condition. Flatfield images have been background-subtracted. Data images were acquired with the engine in skip-fired operation. Data images contain fresh air/fuel/tracer charge mixing with combustion residuals. Like the homogeneous images, data images are background subtracted and flatfield corrected. A sample raw data image is shown in Figure 5.3. The correction technique will be developed in the following section.

102 89 Figure 5.3 Sample raw data image (no corrections), 30 btdc 1200 RPM Exhaust Retard condition Image Correction Procedure Image processing in this experiment was performed using the Matlab programming environment. First, 100 background images are read in and converted from a 12-bit integer value to double precision (to avoid truncation of low signal in subsequent computations). An ensemble mean background image, B i, j, is computed by the formula: B i, j i, j, n 1 = ( Bi, j, n) n= 1 (5.1) th ( i j) ( i j) B = intensity of pixel, of n background image for all, in CCD array The 100 flatfield images are then read in and converted to double precision. An ensemble mean flatfield image is computed by Equation 5.2:

103 i, j, n ( i, j, n ) i j F = F B 100 i, j, n= 1 (5.2) th ( i j) ( i j) F = intensity of pixel, of n flatfield image for all, in CCD array 90 The mean maximum pixel value from the 100 flatfield images, F max, is stored and used as a scaling factor for a normalized mean flatfield image used for data image correction, F0 1. i, j This image will be divided into the individual data images. In order to preserve the full range of the original pixel count scale after division, F0 1 is created as a normalized zero-to-one i, j double precision array. F 0 1 = i, j ( i j) F F max i, j (5.3) for all, in mean flatfield image To avoid amplifying noise-level pixels in the flatfield correction, pixels in F0 1 below i, j 25% of the maximum value are forced to zero. Zero-value pixels are not included in the flatfield correction. The homogeneous images and data images are corrected by the normalized flatfield after background subtraction (Equations 5.4 and 5.5, respectively).

104 H i, j, n F = i, j, n F B 0 1 i, j i, j for each of n=100 flatfield images (5.4) ( Fi, j) 91 D i, j, n Q = i, j, n F B 0 1 i, j i, j (5.5) for each of n=100 engine raw data images ( Qi, j) In this project, the data and homogenous image corrections are performed on a region of interest completely located within the laser sheet. The coordinates of the ROI within the CCD array vary based on the exact alignment of the laser sheet at each condition. The ROI properties are discussed in Section Median Filtering Homogeneous and data image sets ( Hi, j, nand D i, j, n) are transformed with the 3-by-3 median spatial filter built in to Matlab s Image Processing Toolbox. This operation was performed after all corrections from the previous section and before any statistical calculations or output presentation. Median filtering has shown to be a valuable technique for noise removal in images while preserving gradients.

105 Image Statistics 92 The two-dimensional spatial mean pixel intensity and standard deviation were computed on the homogenous and data image ROI using built-in array operators in the Image Processing Toolbox. Since the fluorescence intensity of the 3-pentanone tracer was not calibrated to an absolute scale (due to shot-to-shot laser power variation and thermal gradients in-cylinder), a coefficient of variation had to be defined to quantify inhomogeneity in each image n ( COV ), relative to the mean signal. n COV where and n = ( I ) ( σ ) I ( I ) ( σ ) n I n n n (5.6) = standard deviation of pixel intensities on ROI of image n = spatial mean pixel intensity on ROI of image n During the experiment, a significant error in the flatfield correction occurred, imposing vertical bands on the corrected images (Figure 5.4). This phenomenon has been a common occurrence in prior PLIF studies [13, 15, 16] and can been attributed to slight temporal variation in the laser sheet profile during the measurement. The non-physical structures introduce error into the two-dimensional statistics, so a modified column COV was used to calculate the relative standard deviation only in the direction of the vertical banding. This quantity, although technically a coefficient of variation, will be referred to symbolically as ( y y) σ µ indicating statistics performed in the vertical direction across an entire image ROI and then ensemble-averaged.

106 ( σ y µ y) n N = 1 σ j N j= 1 ( I j ) (5.7) where I and σ are the mean pixel intensity and standard deviation j j th for each pixel column j in the n image ROI ( columns wide) J 93 If the laser sheet is aligned with the camera s pixel array, the column COV term better indicates variation in pixel intensity due to engine flows only. As a convention in the nomenclature, over-bars will used to indicate spatial-mean values within an image and angle brackets < > will used to denote ensemble-mean values. Y r = 21.87%, <(σ y / µ y )> = 1.74% 1200 RPM Homogeneous, 30 btdc 0.9 I max I max 220 Figure 5.4 Sample homogeneous images acquired at 30 btdc for the 1200 RPM, zero overlap condition demonstrating vertical banding in the corrected images. See Section for image presentation convention.

107 Probability Distribution Function 94 The term ( ) σ µ was developed in the previous section to quantify intensity y y variation in individual data images, and also the mean stratification for a particular set of images. Since this ensemble mean encompasses a large amount of pixel values, the probability density function (PDF) is a relevant tool for presenting graphically the measure of central tendency for a set of 100 images. Pixel intensities were gathered from the regions of interest in both corrected data and homogeneous images. A two-dimensional count of these data would be biased by the vertical intensity banding shown in Figure 5.4. Therefore, as in the ( ) σ µ calculations, a vertical column-based analysis was developed. For each column j in each image ROI, the y y mean intensity, I j, was computed and used to generate a pseudo-image, i, j D, of the ROI where each pixel was normalized by it s column mean. (, ) D i j D = i, j i, j j I i= 1.. M j= 1.. N i= 1 for corrected data image D (5.8) M 1 I j = D( i, j) for each column j in the ROI (5.9) M The relative intensities of D i, jon the (M x N) region of interest are decoupled from the flatfield correction errors and can be processed by two-dimensional statistics. Each image n in the 100-image data set is analyzed by the image histogram function in the Matlab Image Processing Toolbox. Since this function only operates on unsigned integers, the

108 double-precision values of D i, jare first multiplied by 10,000 counts to avoid truncation 95 before conversion to16-bit integer class. The histogram bin widths are set to 4 counts (total = 16,384 bins) and the bin counts from each image are summed for the data set to form the full histogram. The full histogram is integrated and then normalized by the integral value to form the PDF curve. The bin scale is normalized to the mean intensity from the D i, jpseudo-images. At each condition where the PDF is examined, the homogenous image set is overlaid on the skip-fired data for comparison Image Presentation Figure 5.4 provides an example of the PLIF data image presentation format for this project. Regions of interest from 25 successive corrected images are compiled into a tiled array. The images are in successive left-to-right, top-to-bottom order from the 100-frame raw data file, taken at the same image timing at the engine cycle imaging frequency (typically one every 6 th cycle). Figure 5.4 also includes an intensity scale at right. As labeled, the upper and lower limits of this scale are the 90% and 50% levels, respectively, of the maximum intensity found in the ROI at that imaging condition. These cutoff levels were used for visual representation help to better illustrate regions of high and low fresh charge concentration in the fired data images. In all future images this same grayscale criteria has been employed.

109 5.2. Imaging System Performance 96 An objective of this project was to apply the highest-fidelity image-based measurement system to residual gas mixing processes. The evaluation criteria for the camera are signal-to-noise ratio and spatial resolution. The MicroMax CCD camera, described in Section 3.3.3, was selected after comparison with other alternatives. The lens selection resulted in a small, centrally-positioned laser sheet region of interest in the combustion chamber. A test target was used to determine the spatial resolution of the system. Finally, image data were analyzed to determine a maximum range of SNR Camera Selection The MicroMax camera is categorized as a frame-straddling CCD, and was analyzed in a comparative study on PLIF detection systems by Rothamer and Ghandhi in [14]. The highest-performing device in that study (assuming a strong PLIF signal) was from the slowscan category of cameras. The Apogee AP7 CCD, which had produced SNR greater than 80:1 in prior non-firing direct-injection mixing studies [15, 16], was not capable of shuttering residual background luminosity during the skip-fired compression stroke. The 16-bit CCD saturated at the lowest shutter duration time (20 ms), and was eliminated from consideration in the project. The MicroMax, with the electronic interline transfer on-chip shuttering feature described in Chapter 3, was able to easily reject this luminosity at an exposure of 10 µs.

110 Region of Interest (ROI) and Spatial Resolution 97 As demonstrated in the background image in Figure 5.1, the full CCD array captures a region of the combustion chamber located below the spark plug and slightly off-center toward the exhaust valve-side of the pent-roof axis. Figure 5.5 more exactly locates the ROI relative to the cylinder head and bore diameter, while Table 5.1 contains the vertical distance separating the ROI from the piston face at the four experimental image timings. Figure 5.5 Location of ROI within combustion chamber, DOHC cylinder head. Distance h is between laser sheet plane and piston face, and is tabulated for image timings in Table 5.1.

111 Image Timing [CAD atdc] h [mm] Table 5.1 Distance from piston face to laser sheet ROI for experiment image timings. 98 The small pixel size (6.7 µm) of the 1300-by-1030 pixel MicroMax CCD required binning the pixels 6-by-6 to obtain an equivalent pixel size in the imaging plane on the order of 185 µm. This value was measured by Wiles [15] in a bench experiment to optimize the photonic flux on the CCD array given our use of a large-aperture lens and also the fixed dimensions of the lab s Bowditch piston (~45cm camera-roi distance). Bin size in this experiment was determined prior to test target measurement using a less-precise scale image. The equivalent pixel size, and thus the physical size of the ROI, was computed using an image taken of the USAF 1951 optical test target. The target was back-illuminated and placed in the laser sheet plane above the piston with the cylinder head removed. Using the standard 4% contrast criterion, the image s spatial resolution was measured to be 4.0 line pairs per mm (lpmm). The equivalent pixel size was measured on the target to be 174 µm in the imaging plane. The binned pixel size is comparable to that of the focused laser sheet thickness [15] and therefore does not reduce system spatial resolution. All regions of interest in this project were 163 pixels long in the direction parallel to the laser sheet propagation. Condition-to-condition variation in laser alignment resulted in a range of ROI widths (across sheet profile) of pixels. Therefore, the ROI sizes

112 recorded in Chapter 6 range from 18-by-28 mm to 23-by-28 mm in real size within the 99 cylinder Signal-to-Noise Ratio Rothamer and Ghandhi [14] demonstrated that, in the presence of a sufficient threshold signal level, the frame-straddling camera will operate in the shot noise-limited regime. This is important, as it reduces the influence of other CCD noise sources ( 2.4.3) from the calculation of the maximum possible SNR. This condition is verified graphically in Figure 5.6, where a characteristic curve for the MicroMax camera is shown with a reference line indicating the shot noise-limited slope of ½. Figure 5.6 Camera noise characterization, as a function of signal intensity - MicroMax frame-straddling CCD. Reprinted from [14].

113 100 Signal-to-noise evaluations of the data images were based on the two-dimensional mean intensity level in the background-subtracted data images (not flatfield-corrected). This value was calculated on the laser sheet ROI described in the previous section. Day-to-day and also set-to-set variation in laser power caused a range of mean signal levels to occur. Table 5.2 contains mean signal level and shot noise-limited SNR (Equation 5.10) for each of the four experiment image acquisition times. Due to the increased number density from compression, the later timings were expected to yield higher signal, and thus SNR. 600 RPM Low Load 600 RPM Mid Load 1200 RPM Low Load Image Timing [atdc] Mean Signal Intensity [counts] : : : 1 Shot Noise-Limited Maximum SNR : : : : : : : : : 1 Table 5.2 Values of spatial-mean data image intensity and resulting shot noise-limited maximum SNR for three speed/load points. Each set is the mean value for the five valve overlap strategies. The signal levels of Table 5.2, when located on the abscissa of Figure 5.6 provide a similar range of SNR as published by Rothamer and Ghandhi. The equation used for shot noise-limited maximum signal-to-noise ratio is:

114 101 SNRmax = I * G out (5.10) The output gain term G out is related to digitization and equal to 4.4 e - /ADU for the MicroMax camera. On examination of the characteristic curve in Figure 5.6, it can be seen that the camera is operating in the shot-noise limited region, although the lower-signal images at 99 btdc are very near the transition away from shot noise-limited behavior. Flatfield images, and thus homogenous images, have higher signal levels since they do not contain residual gases. Due to the marginal signal level at some data image conditions, there was interest in exploring potential benefits to using an intensified slow-scan camera in place of the MicroMax MicroMax Comparison with Intensified CCD Intensified cameras are commonly used in combustion PLIF measurements since their extremely short gating time capabilities allow maximum light rejection. The MicroMax camera was compared with the Roper Scientific PI-Max intensified slow-scan camera. In a moderate signal level environment, the two devices were found to provide very comparable signal-to-noise ratios, and the MicroMax yielded a slightly lower measure of spatial variation on the homogeneous image condition. Given the moderate-signal conditions and the intensified camera s fundamental tendency to blur gradients, the MicroMax camera was selected to maximize spatial performance of the detection system.

115 5.3. Assessment of Intake Charge Homogeneity 102 A major experimental assumption of this project is that the fresh charge contains only air, fuel and tracer and that it is homogenously mixed before IVO. Far-upstream air-assisted fuel injection of pure hydrocarbon fuels has previously been verified for similar laboratory setups in [13, 15, 16]. Nevertheless, the homogenous PLIF images will be analyzed here to quantify degree of homogeneity First and Second Moments of Homogeneous Data ( y y) The homogenous images were tested by comparing their intensity variation σ µ with the theoretical level of shot noise for the mean flatfield signal level at the four timings at all 15 engine conditions. The column statistics were performed on unfiltered corrected flatfield images, since the non-linear filtering is not predicted by shot noise theory. The mean signal level ξ m was that of the 100 flatfield images after background subtraction. From this value, the theoretical shot noise level can be defined in terms of a normalized standard deviation as: m ( σ µ ) = (5.11) shot where AD = 1 G AD ξ out m ξ

116 103 A plot of the homogeneous image spatial variation compared to the theoretical shot noise flatfield is shown in Figure 5.7. A 1-to-1 line is shown indicating the shot noise floor for the ( y y) σ µ metric. All homogeneous images lie correctly above the shot noise floor, although the small offset, close grouping and similar slope of the data points indicate that the variations measured in all homogeneous conditions are primarily influenced by shot noise on the CCD. The small offset can be partially attributed to slight misalignment of the laser sheet vertical banding to the pixel columns where ( y y) contribution of read and dark noise. σ µ is calculated as well as the <(σ y / µ y )> hmg [%] <(σ / µ)> shot [%] Figure 5.7 Comparison of theoretical shot noise intensity variation homogenous pixel intensity variation ( σ y µ y). ( σ µ ) shot to measured

117 Homogeneous Image PDF As a second check of the homogeneous fresh charge and also a demonstration of the pixel intensity PDF function described in Section 5.1.5, the 12 homogeneous image sets taken at the baseline valve overlap are shown in Figure 5.8. This plot shows the strong grouping about the mean pixel intensity in the corrected images. The 600 RPM Low Load condition is the curve showing the lowest peak, and also the condition of highest ( σ y µ y) and lowest SNR in the experiment. Additional PDF figures will be shown in the next chapter, accompanying data images, which will indicate that most homogeneous image sets fall near the taller curves in Figure 5.8 (peaks near 30) and also that the skip-fired residual gas data PDF s are substantially lower in profile.

118 PDF I / I mean Figure 5.8 Probability distribution function for pixel intensity in homogeneous image sets at four image timings for all three engine speed/load points. Baseline valve overlap. Each PDF curve contains information about 100 corrected homogenous images Direct-Injection Test of Imaging Technique Residual gas mixing was measured in this experiment using what amounts to a negative PLIF approach. The presence of residual gases in the image ROI is denoted by regions of low fluorescence intensity. The fresh charge, in which the tracer molecules are homogenously distributed, is assumed to provide the entire PLIF signal. If molecules trapped or re-inducted with the residual gas fluoresce in the laser sheet, that signal would then skew the identification of the residual in the image. A test was then performed to isolate this signal from the predominant PLIF signal from the fresh air/fuel/tracer charge. Gasoline direct-injection (GDI) hardware and the MotoTron skip-fire sequence were used to do this.

119 Skip-Direct Injection Experiment 106 In order to isolate the fluorescence signal of the residual gases, fuel injection had to be shut off on the image cycle. Direct fuel injection is capable of achieving this, so the Triptane engine s pushrod OHV cylinder head, which is outfitted with access for a highpressure Chrysler Pressure-Swirl injector, was used. The details of this combustion chamber and fuel injection strategy are well documented in [13, 15, 16]. A summary of the engine conditions for the GDI experiment are covered in Table RPM 1200 RPM CR 9.8 : 1 m [mg/cycle] air m [mg/cycle] fuel AFR IMEP [kpa] COV of IMEP [%] imap [kpa] EOI [ atdc] IGN [ atdc] -20* -20* T exh [ C] y r [%] HC [ppm C 1 ] 13, Table 5.3 Direct injection experiment engine conditions and unburned hydrocarbon emissions measurements. * indicates the approximate ignition timing.

120 107 To isolate the residual gas fluorescence signal on the image cycle, the direct injection was disabled on the intake stroke preceding the image timing. The only means to accomplish this was by using the one available skip-fire TTL output from MotoTron to perform both fuel injection and ignition timing. The MotoTron sequence was set to trigger the SOI timing, via a Berkeley Nucleonics Model 555 pulse/delay generator. The pulse generator provided the fuel injection driver signal and, on a time delay, the ignition coil signal. This operation disabled both fuel injection and ignition on the image cycle, but introduced an uncertainty into the ignition timing, due to the time-based delay from the ECU signal. This resulted in the high COV of IMEP shown for early-injection GDI operation. The remainder of Table 5.3 shows that the OHV cylinder head provides a significantly higher geometric compression ratio than the DOHC head (CR = 5.95:1). This is reflected in the much lower residual gas mass fraction measured here (compare with Table 4.6). Based on the significant change in combustion chamber geometry, the experimental conditions covered in Chapter 5 could not be matched. Instead, worst-case conditions for HC emissions were set at 600 and 1200 RPM (within engine load limits). Hydrocarbon emissions were considered indicative of the presence of unburned residual 3-pentanone (or other fluorescent compounds) in the residual gas mixing images. By advancing the ignition timing to phase location of peak pressure (LPP) near TDC, HC levels from crevice volume outgassing were maximized. Measured HC was found to be comparable to the imaging conditions in Table 4.4. Figures 5.9 and 5.10 present pressure data at both 600 RPM and 1200 RPM relative to the baseline overlap condition at each speed.

121 Direct injection, OHV head Baseline overlap, DOHC head p cyl [kpa] CAD (atdc) Figure 5.9 Direct-injection experiment cylinder pressure trace comparison with DOHC baseline valve overlap. 600 RPM Direct injection, OHV head Baseline overlap, DOHC head p cyl [kpa] CAD (atdc) Figure 5.10 Direct-injection experiment cylinder pressure trace comparison with DOHC baseline valve overlap RPM.

122 Skip-DI Imaging and Results 109 Two engine speeds were examined (Table 5.3); images were acquired at 10 btdc compression. First, the flatfield condition was imaged using upstream homogenous fuel injection. The mean signal from the 100 flatfield images was compared to two directinjection conditions. First, the engine was operated with both skip-direct injection and skipfiring enabled. This operation provided the best case for low concentration of unburned fluorescent compounds in the combustion chamber, since the prior cycle involved complete combustion. Second, the skip-fire ignition signal was disabled and the engine was motored with the skip-direct injection. This condition provided the worst case for hydrocarbon concentration in the residual gas, simulating a prior cycle with partial burn or misfire quality. The results of the imaging experiment are shown in Table 5.4, with all mean intensity values representing that of an ROI within the laser sheet after background subtraction. 600 RPM 1200 RPM Motored Flatfield Skip-Fire / Skip-DI Skip DI, motoring Table 5.4 Direct injection experiment imaging results. 100-image mean signal level for flatfield, skip-fired, and motored skip-di PLIF data. The data of Table 5.4 indicate a very low signal from the fired residual gas less than 3% of the flatfield intensity for both engine speeds. Under non-fired operation, where

123 misfire residual gas is simulated, the signal level rose to nearly 30% of the flatfield. The 110 engine operating conditions outlined in the next chapter were established to avoid high cyclic-variability operation. Therefore, provided that the combustion is nearly complete, the low fluorescence intensity levels shown by the skip-di/skip-fire test indicate that the technique is faithful in depicting residual gas mixing with homogenous fresh air/fuel charge.

124 Residual Gas Mixing 6.1. Sample Imaging Data Before entering into statistical analysis of residual gas mixing, sample data images representative of engine conditions with high, medium, and low residual gas fractions are presented in Figures A sample homogeneous image set from the same engine timing is shown in Figure 6.1. All engine images are presented in sets of 25 consecutive frames, and the grayscale assignment follows the procedure outlined in Section Y r = 0%, <(σ y / µ y )> = 1.94% 600 RPM Homogeneous Images, 60 btdc 0.9 I max 280 Y r = 44.9%, <(σ y / µ y )> = 8.9% 1200 RPM Exh. Retard, 60 btdc 0.9 I max I max I max Figure 6.1 Sample homogeneous image sequence, 60 btdc. Figure 6.2 Sample data image sequence, high residual fraction condition, 60 btdc.

125 112 Y r = 35.7%, <(σ y / µ y )> = 4.7% 600 RPM Sym. Increase, 60 btdc 0.9 I max 120 Y r = 21.9%, <(σ y / µ y )> = 4.0% 1200 RPM Zero Overlap, 60 btdc 0.9 I max I max 0.5 I max 100 Figure 6.3 Sample data image sequence, mid-range residual fraction, 60 btdc. Figure 6.4 Sample data image sequence, low residual fraction condition, 60 btdc. All data image shown have been background-subtracted, flatfield-normalized and 3 x 3 median-filtered. All pixel variance data, ( ) σ µ, are ensemble-mean values, computed in the vertical direction ( 5.1.4). This figure sequence serves to introduce the data produced by the measurement system outlined in Chapters 3 and 5, and also visually demonstrates the effect of elevated residual gas fractions on the homogeneous air/fuel/tracer mixture shown in Figure 6.1. A more thorough discussion of the correlation between residual fraction and ( ) σ µ will be covered in Section 6.3. y y y y

126 6.2. Correlation of Spatial-Mean Pixel Intensity with Measured Residual Gas Fraction 113 The ratio of the spatial-mean fluorescence signal of the skip-fired data image to that of the motored flatfield image can be assumed to correlate with the bulk residual gas fraction if important assumptions are made. The region of interest captured by the imaging system ( 4.3.1) is small relative to the combustion chamber and inherently two-dimensional, and therefore extensions of the ROI image properties to entire cylinder volume are uncertain. Ensemble-averaging of the maximum number of fired data images available can improve the characterization, but nevertheless assumptions about the ROI have to be made. Secondly, knowledge of the different in-cylinder conditions between skip-fired and motored operation indicates an influence of the temperature-dependent fluorescence intensity on any calculations. The flatfield condition, without any combustion products, contains a cooler mixture, while the skip-fired data images contain tracer molecules that have been heated by the residual gases shown to be mixing with them. Figure 2.8 demonstrates the di-ketone group s decreasing intensity yield with increasing temperature at 266-nm laser excitation, and therefore a ratio of fired data to motored data will slightly over-predict our residual gas fraction (Equation 6.1). For these calculations, only background subtraction was applied to the raw CCD image data, to preserve the detection system s absolute scaling at each image timing and engine condition. For each of the 15 engine speed/load/overlap conditions, the intensity ratio was calculated as the mean of the four image acquisition timings by Equation 6.1:

127 I ratio 1 I data Idata I data I data = I I I I ff ff ff ff (6.1) 114 The correlation between I ratio and y r is shown in Figure 6.5. The correlation can be considered surprisingly good, given our assumptions about the ROI. The temperature dependence of fluorescence appears in this plot as the offset between the slope of the data point grouping, which is not far from parallel to the 1:1 line. The influence of residual gas charge heating on PLIF uncertainty indicated by this correlation certainly has important implications for the results of this project. A likely source of scatter in Figure 6.5 is slight inconsistencies in mean laser sheet power between corresponding flatfield and skip-fire imaging measurements < I ratio > [%] Y r [%] Figure 6.5 Correlation of mean image intensity ratio to measured residual fraction for all 15 experiment conditions.

128 6.3. Correlation of Residual Gas Fraction to Image Intensity Variation 115 The relationship between the degree of mean spatial variation in the corrected data images and the bulk residual gas fraction was investigated across all 15 experiment conditions. Four image timings (30, 45, 60, and 99 btdc) were studied at each condition and a consistent trend of increasing charge stratification with increasing residual fraction was found for all timings. This trend invited further exploration of our image data based on residual fraction, independent of engine speed or load Cycle-Averaged Image Intensity COV Correlation Figures 6.6 through 6.9 present the correlation observed between measured residual gas fraction and the ensemble-mean pixel intensity variation, ( ) σ µ, captured in the skip-fired data images. Separate plots are prepared for each of the four image timings and each plot contains the appropriate reading for each of the 15 experiment conditions. Additionally, the spatial variations for the corresponding homogeneous image data are shown as a relative measure of the absolute shift in stratification when measuring the fired engine flow. The sequence in Figures includes the mean SNR values calculated from all data points at the individual image timings (Table 5.2). This value clearly decreases with the lower-charge density images at the advanced timings. This effect is also seen in the incremental upward shift in homogeneous image variation level at the advanced timings. y y

129 btdc 8 <(σ y / µ y )> [%] Fired Data Homogeneous Y r [%] Figure 6.6 Pixel intensity COV vs. residual gas fraction for all engine conditions at 30 btdc. Shot noise-limited maximum SNR was ~22:1 for this image timing btdc 8 <(σ y / µ y )> [%] Fired Data Homogeneous Y r [%] Figure 6.7 Pixel intensity COV vs. residual gas fraction for all engine conditions at 45 btdc. Shot noise-limited maximum SNR was ~20:1 for this image timing.

130 btdc 8 <(σ y / µ y )> [%] Fired Data Homogeneous Y r [%] Figure 6.8 Pixel intensity COV vs. residual gas fraction for all engine conditions at 60 btdc. Shot noise-limited maximum SNR was ~18:1 for this image timing btdc 8 <(σ y / µ y )> [%] Fired Data Homogeneous Y r [%] Figure 6.9 Pixel intensity COV vs. residual gas fraction for all engine conditions at 99 btdc. Shot noise-limited maximum SNR was ~15:1 for this image timing.

131 The most notable feature of Figures is the quasi-exponential growth in 118 ( ) σ µ at the highest recorded residual gas fractions. This trend, particularly the y y transition range of 35% to 40% residual fraction where both 600 RPM and 1200 RPM data points are located, indicates that there is a residual dilution level at which mixture composition inhomogeneity begins to rapidly increase, independent of speed or load. Since the engine could only exceed 40 % residual fraction at 1200 RPM, the maximum ( σ µ ) points naturally occur at 1200 RPM only. Figures 6.6 through 6.9 also demonstrate the absolute magnitude shift in the intensity variation metric from the motored flatfield condition. Corrected homogeneous images fall near 1-2% ( ) σ µ, depending primarily on the shot noise encountered at the image y y timing. The difference between the fired data points and the homogeneous points is a clear and consistent sign of the presence of residual gas unmixedness. y y Lower Residual Fraction Case-to-Case Comparison At the lower end of the measured scale in Figures , the nearest match between 600 RPM and 1200 RPM conditions was for the baseline overlap 1200 RPM set ( y r =27.3%) and the zero overlap 600 RPM low load set ( r y =28.7%). The ( ) σ µ data for these conditions are presented in Table 6.3. With the exception of the early 99 btdc timing, the 1200 RPM data show consistently lower variation than 600 RPM. Both data sets demonstrate an increasing image intensity variation at 30 btdc. y y

132 % y r 30 btdc 45 btdc 60 btdc 99 btdc 600 RPM, Zero Overlap RPM, Baseline OV Table 6.1 Comparison of lower-residual conditions at 600 and 1200 RPM. Development of σ y µ y [%] with crank angle. image ( ) 119 The data of Table 6.3 would seem to suggest an engine speed influence on residual gas mixing at this dilution level, with the higher engine speed doing a better job of mixing the residual gas with the homogeneous fresh charge in the imaging plane. Sample image data acquired for the two conditions at 45 btdc are presented in Figures 6.10 and Y r = 28.7%, <(σ y / µ y )> = 4.37% 600 RPM Low Load, Zero OV, 45 btdc 0.9 I max Y r = 27.3%, <(σ y / µ y )> = 3.99% 1200 RPM, Baseline OV, 45 btdc 0.9 I max I max 0.5 I max Figure 6.10 Sample data images for 600 RPM, low-residual condition. Figure 6.11 Sample data images for 1200 RPM, low-residual condition.

133 120 Visually comparing the 600 RPM data images in Figure 6.10 with the 1200 RPM set in Figure 6.11 indicates the difficulty in qualitatively distinguishing two conditions. The ( ) σ µ values for these are in fact similar, with the brighter and more numerous flow y y structures shown at 600 RPM likely accounting for the difference. The PDF is a more quantitative comparison between the two conditions, as shown in Figures Figures are an introduction to the nature of the residual gas data image series PDF, where a substantial deviation from the homogeneous mixture data is seen. At 600 RPM, where the mean intensity variation was lower and shows a slightly lower peak in the distribution, although the homogenous data was also less tightly grouped Homogeneous Fired Data 20 PDF btdc 600 RPM Low Load Zero Overlap Y r = 28.7 % <(σ y / µ y )> = 4.37 % I / I mean Figure image pixel intensity PDF for 600 RPM low-residual condition.

134 Homogeneous Fired Data 20 PDF btdc 1200 RPM Baseline Overlap Y r = 27.3 % <(σ y / µ y )> = 3.99 % I / I mean Figure image pixel intensity PDF for 1200 RPM low-residual condition Higher Residual Fraction Case-to-Case Comparison Since the engine was more tolerant of elevated residual fractions at 1200 RPM, comparison of image data between engine speeds was not possible for the maximum dilution levels. The closest match at the high-end occurred between the 600 RPM low load, symmetric increased overlap ( y r = 40.4%) and the 1200 RPM intake cam advance condition ( r y = 40.8%). Again, the development in ( ) σ µ is presented in Table 6.4. y y

135 % y r 30 btdc 45 btdc 60 btdc 99 btdc 600 RPM, Sym. Incr. OV RPM, Intake Advance Table 6.2 Comparison of higher-residual conditions at 600 and 1200 RPM. Development of ( σ y µ y) [%] with crank angle. 122 At the latest image timing, the variation was identical for the two conditions. However, the earlier development of intensity variation was much different, with the 600 RPM condition reaching an early minimum at 60 btdc and the 1200 RPM data decreasing steadily from a high initial level of 8%. At this dilution level, the influence of engine speed seems to be reversed, as the longer mixing time found at 600 RPM produced lower mixture variation in the ROI. Given the conflicting conclusions on engine speed, it would then appear that bulk residual gas fraction is a more applicable parameter in predicting the mixture inhomogeneity. Sample data images for the two conditions of this section are shown in Figures

136 123 Y r = 40.4%, <(σ y / µ y )> = 5.19% 600 RPM Low Load, Sym. Incr., 45 btdc 0.9 I max Y r = 40.8%, <(σ y / µ y )> = 7.09% 1200 RPM Intake Advance, 45 btdc 0.9 I max I max 0.5 I max Figure 6.14 Sample data images for 600 RPM, high-residual condition. 45 btdc. Figure 6.15 Sample data images for 1200 RPM, low-residual condition Prior-Cycle Effect on Image Intensity Variation Cylinder pressure was recorded for all engine cycles during the image acquisition periods and single-cycle IMEP data could be extracted in software. Since the experiment involved skip-firing, same-cycle pressure data was not relevant to the engine operating condition. The best correlation possible to the imaged engine flow would have to come from the previous engine cycle. It was proposed that strong and weak prior cycles would have some degree of influence over the residual gas fraction on the skip-fired cycle. Sample

137 results of the extracted prior-cycle IMEP and corresponding data image COV ( y y) n 124 σ µ are shown in Figures 6.16 and 6.17, with the axes scaled relative to the data-set mean. These figures display the prior-cycle IMEP data for all 100 images at 60 btdc. Figure 6.16 is for the 600 RPM low load, symmetric overlap condition and Figure 6.17 is the same overlap strategy at 1200 RPM. These engine conditions were chosen based on their high residual fraction and high pixel intensity variation [IMEP n-1 ] / [IMEP mean ] (σ y / µ y ) n / <(σ y / µ y )> Figure 6.16 Prior-cycle IMEP vs. image intensity COV. 600 RPM Low Load, Sym. σ y µ y =5.2%. Increase 60 btdc. Yr = 40.4%, IMEP=152 kpa, COVIMEP = 6.0%, ( )

138 [IMEP n-1 ] / [IMEP mean ] (σ y / µ y ) n / <(σ y / µ y )> Figure 6.17 Prior-cycle IMEP vs. image intensity COV ( RPM, Sym. Increase 60 σ ) y µ y btdc. Yr = 43.7%, IMEP=253 kpa, COVIMEP = 1.2%, n =7.3%. All engine conditions demonstrated a similar random data scatter and no correlation was found between prior-cycle IMEP and ( y y) σ µ. It was assumed that quantifying the large-scale cylinder mixture influence of the prior cycle combustion performance with a small ROI is poorly suited for the single-image analysis done here. Ensemble-averaged ROI data, such as presented in Section 6.3, are more likely to yield a better correlation. n

139 6.5. Engine Operating Conditions Effect on Data Image Intensity Variation 126 A major feature of this project was to establish engine conditions of varying residual gas dilution by means of valve overlap strategies. For this experiment, a baseline 20 overlap was established and three strategies for elevated residual fraction were explored intake cam advance, exhaust cam retard, and symmetric overlap increase. Finally, zero valve overlap was studied to establish a minimum residual fraction. Trends of ( y y) overlap strategies are shown in Figures , organized by speed/load. σ µ for the various Baseline Overlap Symmetric Increase Intake Advance Exhaust Retard Zero Overlap <(σ y / µ y )> [%] Low CAD atdc Figure 6.18 Mean image intensity variation vs. CA at 600 RPM low load, all overlaps.

140 Baseline Overlap Symmetric Increase Intake Advance Exhaust Retard Zero Overlap <(σ y / µ y )> [%] Mid CAD atdc Figure 6.19 Mean image intensity variation vs. CA at 600 RPM mid load, all overlaps <(σ y / µ y )> [%] Baseline Overlap 2 Symmetric Increase Intake Advance Exhaust Retard Zero Overlap 1200 Low CAD atdc Figure 6.20 Mean image intensity variation vs. CA at 1200 RPM, all overlaps.

141 128 The 600 RPM conditions in Figures 6.18 and 6.19 demonstrate a consistent hookup in ( y y) σ µ approaching 30 btdc. This trend is presumed to be indicative of a bulk in-cylinder flow component, mostly likely a tumble motion induced by the pent-roof geometry, consistently delivering pockets of unmixed fluid to the imaging plane. At the elevated load conditions in Figure 6.19, the intake cam advance set does not demonstrate the hook-up trend, indicating that the change in phasing of the induction process may have affected the timing of the flow at that intake manifold pressure. The most prevalent trend in the 1200 RPM plot of Figure 6.20 is the large magnitude shift in flow inhomogeneity from the low-residual baseline and zero overlap conditions to the increased overlap strategies. As with much of the combustion data in Chapter 4, the exhaust retard condition at 1200 RPM demonstrated the largest effect of residual dilution, with the highest image intensity variation levels (and also image-to-image variation) recorded in the experiment. In general, the 1200 RPM conditions are not as conclusive as to bulk flowfield influence on residual gas transport as the 600 RPM data Symmetric Overlap Increase At the 600 RPM low load condition, the symmetric 10 increase in valve overlap did not significantly shift ( y y) σ µ from the baseline values at either 99 or 60 btdc (6.8% and 5.2%, respectively). At the later timings, image variation became more pronounced, particularly at 45 btdc, where a 12% increase in ( y y) σ µ from the baseline was

142 129 calculated. In general, the level of variation was not distinguishable from the other overlap strategies. The mid load condition at 600 RPM demonstrated a flatter progression in Figure 6.19 through the compression stroke, roughly splitting the difference in ( y y) σ µ between the other two increased-overlap strategies. Similar behavior is seen in the 60 total overlap 1200 RPM data in Figure 6.20, although the level of variation is noticeably increased with the higher residual fraction (43.7%) at this condition Intake Cam Advance In Figure 6.18, the intake cam advance data points are indistinguishable in ( σ y µ y) from the exhaust retard condition and at the later timings also from the symmetric increased overlap. This plot seems to indicate that at 600 RPM low load the effect of cam phasing strategy is small, perhaps limited by the small increases in dilution level that were tolerable. At the higher IMEP conditions in Figures 6.19 and 6.20, there are clearer distinctions in the behavior of the three strategies. At both 600 RPM and 1200 RPM, the intake cam advance strategy delivered the lowest intensity variation at high residual fraction. The 1200 RPM data set is the only one in the experiment to exhibit continuously decreasing ( σ y µ y) throughout the four image timings. The improvements in image inhomogeneity can be partially attributable to the lower measured residual gas fraction relative to the symmetric increase and exhaust retard cases. Figures 6.21 through 6.24 compare intake advance and exhaust retard imaging at both speeds, followed by the PDF s for all 4 images.

143 130 Y r = 32.5%, <(σ y / µ y )> = 4.09% 600 RPM Mid Load, Int. Advance, 45 btdc 0.9 I max 190 Y r = 32.7%, <(σ y / µ y )> = 4.92% 600 RPM Mid Load, Exh. Retard, 45 btdc 0.9 I max I max I max 130 Figure 6.21 Intake advance data images at 600 RPM Mid Load. 45 btdc. Figure 6.22 Exhaust retard data images at 600 RPM Mid Load. 45 btdc. Y r = 40.8%, <(σ y / µ y )> = 7.09% Y r = 44.8%, <(σ y / µ y )> = 8.71% 1200 RPM Intake Advance, 45 btdc 0.9 I max RPM Exhaust Retard, 45 btdc 0.9 I max I max 0.5 I max 110 Figure 6.23 Intake advance data images at 1200 RPM. 45 btdc. Figure 6.24 Exhaust retard data images at 1200 RPM. 45 btdc.