The Pennsylvania State University. The Graduate School. Department of Energy and Mineral Engineering

Transcription

1 The Pennsylvania State University The Graduate School Department of Energy and Mineral Engineering IMPACTS OF FUEL FORMULATION AND ENGINE OPERATING PARAMETERS ON THE NANOSTRUCTURE AND REACTIVITY OF DIESEL SOOT A Dissertation in Energy and Mineral Engineering by Kuen Yehliu 2010 Kuen Yehliu Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy December 2010

2 The dissertation of Kuen Yehliu was reviewed and approved* by the following: André L. Boehman Professor of Fuel Science and Materials Science and Engineering Dissertation Co-Advisor Chair of Committee Randy L. Vander Wal Associate Professor of Energy and Mineral Engineering Dissertation Co-Advisor Angela D. Lueking Associate Professor of Energy and Geo-Environmental Engineering Thomas A. Litzinger Professor of Mechanical Engineering Octavio Armas-Vergel Associate Professor of Thermal Machines and Engines University of Castilla La Mancha Dissertation Co-Advisor Special Member R. Larry Grayson Professor of Energy and Mineral Engineering Graduate Program Officer of Energy and Mineral Engineering *Signatures are on file in the Graduate School

3 iii ABSTRACT This study focuses on the impacts of fuel formulations on the reactivity and nanostructure of diesel soot. A 2.5L, 4-cylinder, turbocharged, common rail, direct injection light-duty diesel engine was used in generating soot samples. The impacts of engine operating modes and the start of combustion on soot reactivity were investigated first. Based on preliminary investigations, a test condition of 2400 rpm and 64 Nm, with single and split injection strategies, was chosen for studying the impacts of fuel formulation on the characteristics of diesel soot. Three test fuels were used: an ultra low sulfur diesel fuel (BP15), a pure soybean methyl-ester (B100), and a synthetic Fischer-Tropsch fuel (FT) produced in a gas-to-liquid process. The start of injection (SOI) and fuel rail pressures were adjusted such that the three test fuels have similar combustion phasing, thereby facilitating comparisons between soots from the different fuels. Soot reactivity was investigated by thermogravimetric analysis (TGA). According to TGA, B100 soot exhibits the fastest oxidation on a mass basis followed by BP15 and FT derived soots in order of apparent rate constant. X-ray photoelectron spectroscopy (XPS) indicates no relation between the surface oxygen content and the soot reactivity. Crystalline information for the soot samples was obtained using X-ray diffraction (XRD). The basal plane diameter obtained from XRD was inversely related to the apparent rate constants for soot oxidation. For comparison, high resolution transmission electron microscopy (HRTEM) provided images of the graphene layers. Quantitative image analysis proceeded by a custom algorithm. B100 derived soot possessed the shortest mean fringe length and greatest mean fringe tortuosity. This suggests soot (nano)structural disorder correlates with a faster oxidation rate. Such results are in agreement with the X-ray analysis, as the observed fringe length is a measure of basal plane diameter. Moreover the relation between soot reactivity and structural disorder is consistent with past work

4 iv by Vander Wal and co-workers, but stands in contrast to past work by Boehman and co-workers which identified surface oxygen content as the primary explanation for increased oxidative reactivity. The characterization results of this study indicate that changing fuel formulation is a potential method to enhance soot reactivity, and thus diesel particulate filter (DPF) regeneration, through decreasing the degree of order in soot nanostructure. All soot samples were partially oxidized to investigate the structural and elemental surface changes during the oxidation process. The HRTEM image analysis of the B100 and BP15 soot at 50% burn-off shows the highly ordered soot nanostructure, coinciding with significant decreases in the apparent rate constants for soot oxidation. In contrast, for FT soot, no significant changes in the soot nanostructure is observed, coinciding with only a slight decrease in apparent rate constant for soot oxidation. The result of HRTEM image analysis and apparent rate constants for soot oxidation show a relationship between the lattice fringe parameters (the median fringe length and mean fringe tortuosity) and the apparent rate constant, coinciding with the trend observed among the initial soot samples generated by different fuels.

5 v TABLE OF CONTENTS Chapter 1 INTRODUCTION... 1 Chapter 2 LITERATURE REVIEW Diesel particulate matter Diesel particulate emission control Soot reactivity and properties Soot reactivity and soot nanostructure Soot reactivity and surface oxygen-containing functional groups Investigation of the evolution of soot nanostructure and reactivity with oxidation Impact of fuel formulations and engine operating parameters on soot properties Hypotheses and objectives Chapter 3 EXPERIMENTAL Technical approach Test engine Test fuels Engine operating mode and fuel injection parameters PM sample collection method Characterization method Thermogravimetric analysis (TGA) X-ray diffraction (XRD) Raman spectroscopy Transmission electron microscopy (TEM) X-ray photoelectron spectroscopy (XPS) Fourier transform infrared spectroscopy (FTIR) Chapter 4 SOOT CHARACTERIZATION METHOD Quantifying apparent rate constant from TGA isothermal oxidation test Quantifying crystalline parameters from the X-ray diffraction (XRD) data Transmission electron microscopy image analysis Image enhancement and image processing Negative transformation Region of interest (ROI) selection Contrast enhancement Gaussian lowpass filter Top-hat transformation Thresholding to obtain binary images Morphological opening and closing of fringes Clearing fringes on the ROI border Skeletonization Remove short length fringes Fringe characterization Fringe length Fringe tortuosity... 66

6 vi Fringe separation distance Demonstration of image processing and analysis algorithm Repeatability of HRTEM image analysis results Chapter 5 RESULTS AND DISCUSSION Impact of engine operating conditions and the start of injection on soot reactivity The impact of engine operating condition on soot reactivity The impact of the start of combustion on soot reactivity and nanostructure Impact of fuels on soot reactivity The initial surface oxygen-contained functional groups and reactivity Relation between the soot nanostructure and reactivity X-ray diffraction analysis High-resolution transmission electron microscopy (HRTEM) analysis Comparison of XRD and HRTEM analysis results Other characterization parameters and methods Comparison of partially oxidized soot generated by FT, BP15, and B100 fuels Comparison of soot reactivity of soot at initial state and 50% burn-off Comparison of surface oxygen content of soot at initial state and 50% burn-off Comparison of structural parameters of soot at initial state and 50% burnoff Changes in BP15 soot nanostructure at different soot oxidation stages Collection of soot at different oxidation stages Comparison of surface oxygen content of soot at different oxidation stages Comparison of structural parameters of soot at different oxidation stages A simplified oxidation progression model for BP15 soot Chapter 6 CONCLUSIONS AND RECOMMENDATIONS FOR FUTURE WORK Summary and conclusions Recommendations for future work Studying soot oxidation in diesel particulate filters Modeling soot formation process Improving HRTEM image analysis algorithm and other characterization methods BIBLIOGRAPHY Appendix A Using TGA to investigate the impact of the pretreatment method Appendix B The repeatability of isothermal oxidation tests Appendix C Uncertainty and repeatability of XPS Appendix D Uncertainty and repeatability of HRTEM image analysis algorithm Appendix E Analysis results of Raman spectra and oxygen chemisorptions capacity

7 vii LIST OF FIGURES Figure 2-1. Schematic representation of diesel particulate matter [1] Figure 2-2. Illustration of flow pattern in a wall-flow monolith filter. The black arrows represent exhaust gas with particulates, and the gray arrows represent the filtered exhaust gas. (Courtesy of Corning)... 6 Figure 3-1. Thermodynamic analysis results of four engine operating modes: (a) single injection, and (b) split injection. ( ) mode A, ( ) mode B, ( ) mode C, and ( ) mode D Figure 3-2 Thermodynamic diagnostic results of start of injection (SOI) variation using BP15. ( ) baseline, ( ) advanced, and ( ) retarded Figure 3-3 Thermodynamic diagnostic results obtained with a similar combustion process. Single injection. Without (a) and with (b) the adjusted SOI and fuel rail pressures. ( ) BP15 ( ) B100, and ( ) FT fuels Figure 3-4 Thermodynamic diagnostic results adjusting the SOI obtaining a similar combustion process. Split injection. Without (a) and with (b) adjusted SOI and fuel rail pressures. ( ) BP15 ( ) B100, and ( ) FT fuels Figure 3-5 Schematic representation of the crystalline parameters (d 002, L a, L c, and N) [90] Figure 4-1 Simulation results of two values of k c by Equation 4-2 assuming the oxidizer is air. ( ) k c = (1/Pa/min), and ( ) k c = (1/Pa/min) Figure 4-2 Demonstration of experimental and fitted curves of TGA isothermal oxidation tests. ( ) B100, Mode C, single injection, pretreated in nitrogen at 500 C for 60 minutes; ( ) FT, Mode C, single injection, pretreated in nitrogen at 500 C for 60 minutes; ( ) B100 fitted curve for the first 30 minutes (k c =1.30*10-6 1/Pa/min); ( ) FT fitted curve for the first 30 minutes (k c =6.33*10-7 1/Pa/min) Figure 4-3 A demonstrative XRD corrected pattern and the fitted curves of 002 and 10 peaks. Sample: PM generated at Mode C by FT fuel using split injection strategy with matched combustion phase. ( ) XRD corrected pattern, ( ) baseline for fitting 002 peak, ( ) fitted 002 peak, ( ) baseline for fitting 10 peak, and ( ) fitted 10 peak Figure 4-4 Magnification of (002) and (10) peaks (a) (002) peak fitted by Gaussian function, and (b) (10) peak fitted by Gaussian and Lorentzian functions. (10) peak was fitted better with Lorentzian function. ( ) XRD corrected pattern, ( ) baseline, ( ) Gaussian fitted peak, ( ) Lorentzian fitted peak Figure 4-5 Flow chart of the image processing program used to enhance and process HRTEM images

8 viii Figure 4-6 Illustration of the image enhancement and image processing steps Figure 4-7 Calculation of fringe length and tortuosity: (a) scheme in the continuous domain, and (b) approximation in the digitized image domain Figure 4-8 Calculation of fringe separation distance: (a) scheme in the continuous domain, and (b) approximation in the digitized image domain Figure 4-9 An HRTEM image analyzed for demonstration purpose Figure 4-10 The graphene layers extracted from Figure Figure 4-11 Fringe analysis results of Figure 4-9 using parameters listed in Table 4-2: (a) fringe length histogram (median: 1.05 nm), (b) fringe tortuosity histogram (mean: 1.16), and (c) fringe separation histogram (mean: nm) Figure 5-1 Normalized mass obtained from the isothermal oxidation test at 550 C of soot samples generated at four engine modes. ( ) Mode A: 1850 rpm, 64 Nm, ( ) Mode B: 1850 rpm, 110 Nm, ( ) Mode C: 2400 rpm, 64 Nm, and ( ) Mode D: 2400 rpm, 110 Nm Figure 5-2 Normalized mass obtained from the isothermal oxidation test at 550 C of soot samples generated at three SOIs at Mode C with BP15 fuelling. ( ) baseline: 4.83 BTDC, ( ) advanced 2 CAD: 6.83 BTDC, and ( ) retard 2 CAD: 2.83 BTDC Figure 5-3 TEM images of soot generated at Mode C with single injection by advanced and retarded SOI: (a)-(c) advanced 2 CAD, and (d)-(f) retarded 2 CAD Figure 5-4 Representative HRTEM images of soot generated at Mode C with single injection by advanced and retarded SOI: (a) image of soot generated at SOI with advanced 2 CAD, (b) ROI and extracted skeletons of (a), (c) image of soot generated at SOI with retarded 2 CAD, and (d) ROI and extracted skeletons of (c) Figure 5-5 Fringe length and tortuosity analysis of soot generated at Mode C with single injection by advanced and retarded SOI timings: (a) fringe length extracted from Figure 5-4(b), (b) fringe tortuosity extracted from Figure 5-4 (b), (c) fringe length extracted from Figure 5-4 (d), and (d) fringe tortuosity extracted from Figure 5-4 (d) Figure 5-6 Normalized weight vs. oxidation time curves of soots generated at Mode C by BP15, B100, and FT with matched combustion phasing: (a) single injection and (b) split injection. ( ) BP15 fuel, ( ) B100 fuel, ( ) FT fuel Figure 5-7 The derived VOF content versus the soot reactivity using the data in Table 5-1~ Table Figure 5-8 Surface oxygen content vs. apparent rate constant for soot oxidation

9 ix Figure 5-9 Transmission FTIR spectra of ( ) as received soot, and ( ) thermally pretreated soot, qualitatively indicating some functional groups was removed during the pretreatment process. (Peak assignment) 1450cm -1 : carbon-hydroxyl bond in aliphatic groups; 1600~1640 cm -1 : aromatic groups; 1710~1750 cm -1 : lactone, carboxyl and ketone acid groups, 2860~2950cm -1 : aliphatic groups Figure 5-10 High resolution scan of ( ) as received soot, and ( ) thermally pretreated soot Figure 5-11 Mass fraction profiles of isothermal oxidation test of PM samples pretreated at ( )100 C, ( ) 350 C, ( ) 500 C, and ( ) 650 C for 60 minutes under ultra high purity nitrogen Figure 5-12 The temperature and normalized weight profile of non-isothermal oxidation test. ( ) as received PM, ( ) PM pretreated at 500 C, ( ) Temperature ( C) Figure 5-13 The temperature and mass fraction profile of non-isothermal oxidation test. The mass fraction is normalized by the mass measured at 37.5 minutes of Figure ( ) As received PM normalized by weight at 400 C, ( ) PM pretreated at 500 C, ( ) Temperature ( C) Figure 5-14 The basal plane diameter, L a, versus the apparent rate constant for soot oxidation, indicating an inverse relation. single injection (02/12/09), single injection (02/18/09), split injection (03/05/09), and Δ split injection (03/12/09) Figure 5-15 Three representative HRTEM images of soot samples generated by three test fuels using single injection strategy at matched combustion phasing: (a) BP15, (b) FT, (c) B100 with marked ROI, and (d)~(f) show fringes extracted from images of (a)~(c) Figure 5-16 Fringe length and median values derived from the extracted fringes in Figure 5-15: (a) FT soot (median: 0.90nm), (b) BP15 soot (median: 0.79nm), and (c) B100 soot (median: 0.72nm) Figure 5-17 Fringe tortuosity histograms and mean values derived from the extracted fringes in Figure 5-15: (a) FT soot (mean: 1.14), (b) BP15 soot (mean: 1.17), and (c) B100 soot (mean: 1.37) Figure 5-18 Three representative HRTEM images of soot samples generated by three test fuels using split injection strategy at matched combustion phasing: (a) BP15, (b) FT, (c) B100 with marked ROI, and (d)~(f) show fringes extracted from images of (a)~(c) Figure 5-19 Fringe length and median values derived from the extracted fringes in Figure 5-18: (a) FT soot (median: 1.01nm), (b) BP15 soot (median: 0.96nm), and (c) B100 soot (median: 0.92nm)

10 x Figure 5-20 Fringe tortuosity histograms and mean values derived from the extracted fringes in Figure 5-18: (a) FT soot (mean: 1.12), (b) BP15 soot (mean: 1.18), and (c) B100 soot (mean: 1.31) Figure 5-21 Comparison of L a derived from XRD patterns and median fringe length derived from HRTEM images (BP15, FT, and B100 with single injection) versus apparent rate constant for soot oxidation. : L a from XRD (nm)(02/12/2009), : L a from XRD (nm)(02/18/2009), and : median fringe length from HRTEM (nm) Figure 5-22 Comparison of L a derived from XRD patterns and median fringe length derived from HRTEM images (BP15, FT, and B100 with split injection) versus apparent rate constant for soot oxidation. : L a from XRD (nm)(03/05/2009), : L a from XRD (nm)(03/12/2009), and : median fringe length from HRTEM (nm) Figure 5-23 The extracted skeletons in the ROIs of the representative HRTEM images of the three soot samples generated by single injection strategy at their initial states and 50% burn-off Figure 5-24 Fringe length histogram and median values derived from Figure 5-23: (a) FT soot (initial state), (b) BP15 soot (initial state), (c) B100 soot (initial state), (d) FT soot (50% burn-off), (e) BP15 soot (50% burn-off), and (f) B100 soot (50% burnoff) Figure 5-25 Fringe tortuosity histogram and mean values derived from Figure 5-23: (a) FT soot (initial state), (b) BP15 soot (initial state), (c) B100 soot (initial state), (d) FT soot (50% burn-off), (e) BP15 soot (50% burn-off), and (f) B100 soot (50% burn-off) Figure 5-26 The extracted skeletons in the ROIs of the representative HRTEM images of the three soot samples generated by split injection strategy at their initial states and 50% burn-offs Figure 5-27 Fringe length histogram and median values derived from Figure 5-26: (a) FT soot (initial state), (b) BP15 soot (initial state), (c) B100 soot (initial state), (d) FT soot (50% burn-off), (e) BP15 soot (50% burn-off), and (f) B100 soot (50% burnoff) Figure 5-28 Fringe tortuosity histogram and mean values derived from Figure 5-26: (a) FT soot (initial state), (b) BP15 soot (initial state), (c) B100 soot (initial state), (d) FT soot (50% burn-off), (e) BP15 soot (50% burn-off), and (f) B100 soot (50% burn-off) Figure 5-29 The median fringe lengths and mean fringe tortuosity versus the apparent rate constant of the soot samples at initial states and 50% burn-offs: (a) the median fringe lengths versus the apparent rate constant, and (b) the mean fringe tortuosity versus the apparent rate constant Figure 5-30 Isothermal oxidation curve of BP15 soot generated by split injection strategy with marked oxidation stages to be analyzed. 0% burn-off: 0 min, 25% burn-off: 17

11 min, 50% burn-off: 34 min, 75% burn-off: 56 min, oxidation completed: 110 min. ( ) normalized weight, and analyzed oxidation stages Figure 5-31 The extracted skeletons in the regions of interest (ROIs) of the representative HRTEM images of the BP15 soot samples at different oxidation stages: (a) initial state, (b) 25% burn-off, (c) 50% burn-off, and (d) 75% burn-off Figure 5-32 Fringe length histogram and median values derived from Figure 5-31: (a) BP15 soot (split injection, initial state), (b) BP15 soot (25% burn-off), (c) BP15 soot (50% burn-off), and (d) BP15 soot (75% burn-off) Figure 5-33 Fringe tortuosity histogram and mean values derived from Figure 5-31: (a) BP15 soot (split injection, initial state), (b) BP15 soot (25% burn-off), (c) BP15 soot (50% burn-off), and (d) BP15 soot (75% burn-off) Figure 5-34 The median fringe lengths and mean fringe tortuosity versus the burn-off rate of the soot samples at different oxidation stages: (a) the median fringe lengths versus the burn-off rate, and (b) the mean fringe tortuosity versus the burn-off rate Figure 5-35 Peak area ratios of (a) I D1 /I G, and (b) I D3 /I G at different burn-off stages Figure 5-36 Simplified oxidation progression model for BP15 soot generated by split injection strategy xi

12 xii LIST OF TABLES Table 3-1 Engine characteristics Table 3-2 Composition of test fuels (% wt) Table 3-3 Properties of test fuels Table 3-4 Operating modes and the timing of the start of injection Table 3-5 Exhaust gas temperature and injection pressure during each test with single injection Table 3-6 Exhaust gas temperature and injection pressure during each test with split injection Table 3-7 Engine operation mode and variation of the start of injection (SOI) Table 3-8 Start of injection and fuel rail pressures which produced a similar combustion process (single injection) Table 3-9 Start of injection and fuel rail pressures that produce similar combustion conditions (split injection) Table 3-10 Summary of TGA pretreatment and isothermal oxidation test procedures Table 3-11 Summary of TGA pretreatment and ASA measurement procedures Table 3-12 Functional groups on soot surface and their corresponding FTIR band assignments Table 4-1 Summary of the initial guess values and the termination tolerance for minimizing Equations 4-6 and Table 4-2 Summary of processing parameters for Figure Table 5-1 Analysis of VOF content of four engine operating modes using split injection with BP15 fuelling Table 5-2 Analysis of VOF content of the three SOIs using single injection at Mode C with BP15 fuelling Table 5-3 The derived VOF content and apparent rate constants for soot oxidation of BP15, B100, and FT soot generated at Mode C with matched combustion phasing using single and split injection strategy Table 5-4 Elemental analysis of soot surface using XPS Table 5-5 Comparison of apparent rate constant for soot oxidation of the PM pretreated at four different temperatures... 90

13 xiii Table 5-6 Crystalline parameters derived from XRD patterns Table 5-7 Summary of ASA and apparent rate constant for soot oxidation of BP15 soot with different SOI Table 5-8 The apparent rate constant for soot oxidation of the initial soot and after the 50% burn-off, and the percentage of reduction in the apparent rate constant Table 5-9 Surface elemental analysis of initial soot and 50% burn-off soot using XPS Table 5-10 Summary of the apparent rate constant, the median fringe length and mean fringe tortuosity at initial states and 50% burn-offs Table 5-11 Summary of the time segment of burn-off periods for four oxidation stages for BP15 soot (split injection) Table 5-12 Surface elemental analysis of the four oxidation stages of BP15 soot (split injection) Table 5-13 Summary of the median fringe length and mean fringe tortuosity at different oxidation stages

14 xiv ACKNOWLEDGEMENTS I want to express my sincere gratitude to my academic advisor, Dr. André L. Boehman, for his support, guidance, enthusiasm, patience and encouragement during this study. Thanks to his excellent direction, I could explore new field of knowledge and complete my Ph.D. study. I would also like to express my gratitude to Dr. Octavio Armas-Vergel, who advised the methodology in determining the engine operating parameters for matching the combustion phasing of different fuels. I would like to thank Dr. Randy L. Vander Wal for his advice on the implementation of the image analysis algorithms for high resolution transmission electron microscopy images. My appreciation and gratitude are extended to Dr. Angela D. Lueking and Dr. Thomas A. Litzinger for their valuable comments and suggestions, which greatly improved the quality of my research and dissertation. I also want to thank my group members at Diesel Combustion and Emissions Lab for their help. In particular, I am grateful to Vince Zello and Greg Lilik for their helps regarding the engine operation, Dr. Steve Kirby for his help in fuel management, and Dr. Hee Je Seong for many valuable discussions regarding soot characterization methods. Thanks are extended to Dr. Joe Kulik for TEM analysis, Nichole Wonderling for XRD analysis, Vince Bojan for XPS analysis, and Joe Stitt for Raman spectroscopy analysis in the Materials Characterization Lab. And, I am very thankful to Dr. Dania Alvarez Fonseca, Ronnie Wasco, and Joshua Heyne for the generous help in fuel characterization. I want to give my special thanks to my parents, parents-in-law and all family members who supported me. Finally and most importantly, I would like to thank my wife, Ko-Yu Chan, for

15 xv her love, support and encouragement along the journey. And, I want to express my loves and gratitude to our daughter, Karen, who was born while I was in graduate school. She has been bringing great happiness to my family. The financial support from National Science Foundation and General Electric Global Research Center are heartily acknowledged.

16 Chapter 1 INTRODUCTION The interest in researching alternatives to diesel fuel has been growing due to the increasing concern over the cost and supply of petroleum fuels. From this interest, there are two imperative issues that need to be solved. One is to find alternative fuel sources that can reduce the dependence on petroleum fuels. The other equally important issue is to evaluate the performance of these alternative fuels and whether they can reduce particulate matter (PM) and gaseous emissions, which are affected by the fuel injection strategy and fuel properties. An additional strategy for reducing PM emissions is the use of a diesel particulate filter (DPF) [1, 2]. When using a DPF, the trapped particulate matter must be intermittently combusted because the filter would become clogged if this particulate matter was not removed [3]. The combustion of the trapped particulate matter on the DPF, referred to as DPF regeneration, is significantly affected by soot reactivity, which is related to the physical and chemical properties of the particulates [4]. When chemical and physical properties of diesel particulate are considered, differences in diesel combustion characteristics and fuel formulation can influence these soot properties. Due to the increasing popularity of alternative fuels, the impacts of fuel formulation on soot emissions and soot characteristics have been studied by several researchers. Vander Wal et al. [5] showed that the soot particles generated by benzene, ethanol and acetylene have different structural order and reactivity. Benzene-derived soot is found to have a more amorphous structure and is more reactive than acetylene-derived soot. Song et al. [6] found that soot from soybean oilderived biodiesel is five times more reactive than soot from Fischer-Tropsch diesel fuel. The authors claimed that the relative amount of initial oxygen groups is an important factor

17 2 determining the oxidation rate. Lapuerta et al. [7] observed that PM emissions decrease when using bioethanol blended with conventional diesel. As shown widely in the literature, this observation is correlated with the presence of the bonded oxygen, reduction of aromatic content, lower C/H mass ratio, and a change in air/fuel ratio. Besides the impact of fuel composition on PM emissions and soot characteristics, the combustion process can also influence soot characteristics. The concentration and size distributions of particulate matter are affected by the dilution and cooling of exhaust, air/fuel ratio, and engine type [8]. Running an engine at low engine load, Choi and Reitz [9] showed that split injections were effective in reducing particulate emissions especially at advanced injection timings. Using a high temperature tube furnace, Vander Wal et al. [10] showed that the nanostructure of soot depends upon its formation conditions, such as residence time and temperature. Developing a numerical diesel engine model, Chan et al. [11] concluded that soot and nitrogen monoxide emissions were affected by engine operating conditions and fuel injection timing. Zhu et al. [12] found that the crystallite degree of soot increases with engine load and exhaust temperature. In addition, it is also found that soot particle size tends to decrease when the exhaust temperature increases because of particle oxidation at high temperature. Nevertheless, Neer et al. [13] observed an increase in soot spherule and aggregate sizes with an increase of the engine load and exhaust temperature. This observation was explained by the impact of the change of the air/fuel ratio. Meanwhile, it is also observed that soot particles are smaller at higher engine speed because of the shorter residence time. Due to the complicated nature of practical fuels and the engine control strategies, the impact of fuel composition on engine combustion has been studied extensively. When burning different fuels in a purely mechanical cam-driven type fuel injection system, the bulk modulus of

18 3 the fuel affects fuel injection timing, and thus combustion products [14]. Comparing a conventional diesel fuel and biodiesel fuels, Armas et al. [15] found that electronic control unit (ECU) determines an advance of the start of injection and a longer injection duration when using biodiesel fuels in order to maintain the same engine power and torque. This could be an additional reason for PM emission reduction for biodiesel fuels besides the chemical and physical property differences. Commanding ECU, Zhang et al. [16] fixed injection timing for ultra-low sulfur diesel fuel and soybean-derived biodiesel and obtained similar heat release rate profiles. To effectively evaluate the emissions from different fuels, Lapuerta et al. [17] suggested maintaining the engine at the same engine speed and torque to obtain a meaningful comparison. The present work addresses the impacts of fuel properties and engine operating parameters on the reactivity and nanostructure of diesel soot. In particular, controlling the injection parameters through ECU to obtain a similar heat release rate for different fuels improves upon the experimental setup adopted by Song et al. [6, 18, 19], where there was no control access to the fuel injection systems. This strategy, of comparing fuels when the heat release rates are similar, provides a clear evaluation of the impacts of fuel composition on the reactivity and nanostructure of diesel soot. Three test fuels are used in this work: an ultra low sulfur diesel (BP15), a pure soybean methyl-ester biodiesel (B100), and a Fischer-Tropsch fuel (FT), produced in a gas-to-liquid process. The operating condition (2400 rpm, 64 Nm) was selected in order to compare soot characteristics with previous work. And, an effort has been made to adjust injection timing for the different fuels to obtain the same combustion phasing, based on pressures traces and heat release rates. The matched engine combustion phasing for different fuels is used while collecting soot samples for characterization.

19 4 Motivated by the literature on diesel particulate matter and carbonaceous materials [6, 10, 18-39], the characterization of the collected soot samples addressed the relationship between the initial physical and chemical properties and the reactivity of diesel soot. Material analysis techniques were used, and data post-processing procedures were developed. The soot samples were partially oxidized to investigate the structural and surface elemental changes during oxidation process. A simplified oxidation progression model for one soot sample (BP15, split injection) was constructed using the characterization methods developed in this study.

20 Chapter 2 LITERATURE REVIEW 2.1 Diesel particulate matter Diesel particulate matter (PM) is a complex mixture of organic and inorganic compounds in solid and liquid phases as illustrated in Figure 2-1 [1]. PM is primarily made of spherical carbon particles. The spherical carbon particles are variously termed solid particulate, insoluble fraction of PM or soot [1-3]. A layer of unburned, condensed hydrocarbons is adsorbed onto the carbon particles. The origins of these unburned hydrocarbons are fuels or lube oils escaping oxidation and appearing as volatile or soluble organic compounds [8]. These adsorbed compounds are described variably as volatile organic fraction (VOF) [3] or soluble organic fraction (SOF) [1, 2, 8]. If the fuels contain sulfur, the particulate matter may contain SO 3, sulfuric acid or sulfate [1-3]. Additionally, nitrates and water can also be adsorbed to the carbon particles [2, 3, 40, 41]. Figure 2-1. Schematic representation of diesel particulate matter [1].

21 6 2.2 Diesel particulate emission control One of the control strategies is the use of diesel particulate filters (DPF) [1, 2]. A DPF is a device that captures diesel particulates from the exhaust gas of a diesel engine to prevent their release to the atmosphere. Figure 2-2 illustrates the flow pattern in a wall-flow monolith type DPF, the most common design of diesel particulate filters. The wall-flow monolith is an extruded, ceramic, porous honeycomb structure having alternatively blocked channel openings [42]. As shown in Figure 2-2, the exhaust gas flow through the DPF, and the PM is deposited in the inlet channel. The filtered exhaust gases exit from the surrounding channels. When using a DPF, the trapped particulate matter must be intermittently combusted, since the filter would become clogged if this particulate matter was not removed [3]. The combustion of the trapped particulate matter on the DPF is affected by soot reactivity, which is related to the physical and chemical properties of particulates [4]. Therefore, soot reactivity, nanostructure, and surface functionality have become significant research topics in research on diesel particulate control [32, 38, 43, 44]. Figure 2-2. Illustration of flow pattern in a wall-flow monolith filter. The black arrows represent exhaust gas with particulates, and the gray arrows represent the filtered exhaust gas. (Courtesy of Corning)

22 7 2.3 Soot reactivity and properties Soot reactivity and soot nanostructure The oxidative reactivity of soot is related to the initial soot nanostructure. Al-Qurashi showed that the average number of graphene layers in stacks decreased as the reactivity increased [20, 39]. It was also confirmed that the nanostructure of soot changes as the soot was oxidized [6, 20]. In addition, because a carbon atom at an edge site is much more reactive than one in the basal plane of a graphene layer, therefore, a soot particle with many populated edge sites has a higher reactivity [45, 46]. The term soot nanostructure refers to the size, orientation, and organization of the graphene layers. These properties are routinely studied by X-ray diffraction (XRD), Raman spectroscopy, and transmission electron microscopy (TEM) [47]. X-ray diffraction patterns have been used to derive crystalline parameters, such as interlayer spacing (d 002 ), crystalline basal plane diameter (L a ), and stacking height (L c ) [24, 36, 37, 48, 49]. In principle, an increase in basal plane diameter causes a decrease in the ratio of edge carbon atoms to basal carbon atoms, which affects the carbon reactivity [46]. Providing a measure of disorder relative to graphitic content, Raman spectroscopy provides complementary information [34, 50-53] to X-ray diffraction. Several methods using fundamental, overtone and combination band intensities and peak widths have been applied to various carbons. Relative to X-ray diffraction for which theory well defines extracted lattice parameters, Raman analysis is based upon semi-empirical relations for quantitative information in the field of soot characterization. Rigorous theory exists only for model carbons such as graphites and SWNTs, DWNTs, etc. In general use, Raman is used only for qualitative information.

23 8 X-ray diffraction has proven value in deriving crystalline parameters of a bulk material sample, but it lacks the ability to examine localized graphene layers within a sample. Raman spectroscopy reveals the disorder of carbon materials due to its sensitivity to short range order, but the data processing and the interpretation of Raman spectra remains a controversial issue in the literature [34]. To compensate for these deficiencies, high resolution transmission electron microscopy (HRTEM) has been widely used in conjunction with XRD or Raman spectroscopy in the study of carbonaceous materials [18, 20, 25, 47, 54-56]. In contrast to XRD and Raman spectroscopy, HRTEM provides direct information on the graphene layers. The application of HRTEM to the analysis of graphene layer structure of carbonaceous materials is not new [57]. Before the advent of computer-aided image processing, TEM images were often interpreted qualitatively [44]. With the advent of the modern image processing and pattern recognition techniques TEM image analysis of carbonaceous materials evolved further. Palotas et al. applied band-pass filtering in the Fourier space and binarized the images with different thresholding values [32]. Fringe parameters, such as length, circularity, and interplanar spacing, were measured after obtaining the binary images. A key feature of the processing was that the frequency band of the band-pass filter retained only those fringes having spacing within a certain range [32]. Similar strategies (image processing and fringe feature analysis) also appeared in other papers. Sharma et al. [58] utilized a step filter, instead of a band-pass filter, to extract more lattice fringes. Additionally they developed a series of steps to modify those fringes with T or Y shaped links. Based on a tolerance value of disorientation (±10 ), the parallelism of adjacent fringes was determined. The interlayer spacing was thereby evaluated. Another complete, semi-quantitative image analysis procedure was demonstrated by Shim et al [59]. Besides a series of image

24 9 processing operations (band-pass filter, thresholding, skeletonization, etc.), two-dimensional and three-dimensional parameters were used to describe the degree of orientational order. Galvez et al. applied top-hat transformation to compensate for intensity variations, and eliminated fringes smaller than the size of a single (benzene) aromatic ring (0.25nm) [60]. These authors also developed an algorithm to determine stacked layers based on the orientation angles, tortuosity, and interfringe spacings of adjacent fringes. A further advance by Goel et al. [61] defined a curvature parameter, the actual arc length divided by the circumference of a virtual circle (π arc diameter), to characterize the apparent circumference of the fullerenic nanostructure. Focusing upon a complete set of geometric parameters, Vander Wal et al. [56] developed an algorithm to extract fringe length, tortuosity and separation from HRTEM images because these three parameters are critical in determining the physical and chemical properties of carbonaceous materials. The algorithm was tested by application to a series of soot samples that were heated at different temperatures. The statistical results of the three parameters quantitatively differentiated carbon nanostructures [56]. Using recent commercial software, (Digital Micrograph, Gatan, Inc., Pleasanton, CA) Muller et al. [62] performed image processing functions, and to obtain the average fringe length and curvature from the HRTEM image of diesel engine soot. In this work, the development of the HRTEM image analysis method is an extension of previous work on diesel soot characterization [18-20, 63, 64]. Drawing upon the literature [45, 56], an HRTEM image analysis algorithm including image processing and lattice fringe analysis functions is developed and implemented [65, 66]. In addition to the methods mentioned, oxygen chemisorption capacity has a connection with the development of crystallinity. When the crystallite size (L a ) increases, it is suggested that the active site concentration (the ratio of the edge carbon atoms to total carbon atoms) decreases

25 [46]. Therefore, oxygen chemisorption, or active surface area, has been used as an indicator for the soot surface structure Soot reactivity and surface oxygen-containing functional groups Besides the nanostructure, surface functional groups may be another factor affecting soot reactivity [6]. In general, the edge carbons are much more reactive than the carbons in the basal planes, so the chemisorbed oxygen are predominantly located on the edge [46]. Therefore, The surface functional groups are closely related to the soot nanostructure. Smith and Chugtai reported that the reactivity of carbon black depends upon the nature of the surface functionality [38]. Boehm showed that the individual functional groups can be determined by characterization methods including acidimetric titration, IR spectroscopy, XPS, thermal desorption and electrokinetic measurements [26, 67]. Recently, by modifying soot surface, Seong [68] found a relation between the abundance of oxygen-containing functional groups and the soot oxidative reactivity. Nevertheless, no correlation between specific surface oxygen functional groups and soot oxidation reactivity was confirmed [68]. Fourier transform infrared spectroscopy (FT-IR) is useful in the analysis of the surface functional groups. However, the quantitative analysis of functional groups is difficult because of the controversies over the assignment of some bands [26, 27, 69, 70]. XPS can be used to determine both elemental content on soot surfaces, e.g. O/C ratio [20, 68], and to quantify surface functional groups [28]. Temperature-programmed desorption (TPD) is another technique used for the study of surface oxides. When using TPD to characterize functional groups, it is generally assumed that each type of functional group decomposes to a defined product, or a combination of products, over a specified range of temperature [27, 71].

26 Investigation of the evolution of soot nanostructure and reactivity with oxidation The structural evolution of soot primary particles during the oxidation process was investigated by several workers. Ishiguro et al. studied the structural changes of 25%, 50%, and 75% burn-off soot produced in an electric furnace at 500 C [44]. In the early oxidation stage, the SOF within the soot particles is released and the porous structure was formed. The layers were rearranged to form larger and turbostratic structure. Subsequently, the crystallites at the outermost shell of the soot particles strip from the outer surface, reducing the particle size. The infrared absorption spectra show a stronger absorption corresponding to C=C stretching vibration as burnoff increases. In addition, absorption corresponding to the C=O group decreases with increasing burn-off, attributable to the release of oxygen-containing functional groups [44]. In contrast to the stripping or disintegration of the outer surface layers that Ishiguro claimed, Hurt et al. proposed the oxidation-induced densification as an explanation for the size reduction during soot oxidation, and emphasized the importance of inner structural changes during soot oxidation [72]. Using XRD, Davis et al. [73] further confirmed that the mean crystallite diameter of a coal sample increased by a factor of 2 during oxidation, providing direct evidence of structural rearrangement. In addition, HRTEM images show evidence that the increase in the volume fraction of ordered material during oxidation, coinciding with the XRD results. Using soot samples generated by a spark ignition engine, Knauer et al. observed the change in structure by Raman spectrometry during the temperature programmed oxidation [54]. The structural arrangement is determined by a five band curve-fitting procedure (G, D1-D4) which indicates an increase of structural order upon soot oxidation. The soot reactivity and CO2

27 12 emissions measured by FTIR gas analyzer are in good agreement with the structural change. Investigating coal chars gasified in air with different conversion levels, Feng et al. observed large and highly ordered crystallites for a sample with a conversion level of 86%. But no significant change in separation distance between layers was observable. They claimed that the highly ordered graphene layers are probably formed during pyrolysis, and proposed a conceptual model of the gasification process [30]. Similarly, Palotas observed increasing order of carbon nanostructure as oxidation progresses by calculating the fractional coverage of a cross section of the particles with layer structures. In addition, they observed a decrease in the mean interlayer spacing [74]. Using TGA data to model the soot oxidation, Marcuccilli et al. observed a decrease in apparent activation energy above 730 C. The authors attributed the inherent reactivity decline to the change in microstructure associated with thermal annealing [75]. Hurt et al. observed the reactivity loss of coal char during the later stage of coal combustion. They related the reactivity loss with changes in the carbonaceous structure, which evolves over the course of combustion as a result of simultaneous oxidation and thermal treatment [76, 77]. Song examined structural changes during the oxidation process of Fischer-Tropsch soot (FT) and biodiesel soot (B100) generated by a 6-cylinder Cummins ISB 5.9L direct injection turbodiesel engine [6]. Because the engine had no access to control of the fuel injection system, the samples were not generated under identical conditions because changes in combustion phasing were unavoidable as a result of fuel property differences. The FT and B100 soot exhibit extremely different oxidation progression characteristics: FT soot shows surface oxidation while B100 soot exhibited internal burning. In addition, the order of the B100 soot decreases as the soot oxidizes from its initial state to 40% burn-off, and increases as the soot oxidizes to 75% burn-off.

28 13 In contrast, the order of FT soot decreases continuously as the soot oxidizes. Recently, using the peak area ratio (I D /I G ) derived from Raman spectra, Al-Qurashi observed increasing order for both 0% and 20% EGR soot as the soot oxidized [20, 39]. The soot was generated by a 2.5L, 4 cylinder, common rail, turbocharged direct injection diesel engine. The changes in the order of the soot nanostructure during oxidation were also examined by electron energy loss spectroscopy (EELS). However, the results obtained from EELS do not fully corroborate those results obtained from Raman spectra. 2.5 Impact of fuel formulations and engine operating parameters on soot properties The use of biodiesel or synthetic fuels as a diesel replacement poses a need for the study of fuel composition impacts on engine combustion and particulate matter (PM) emissions. Both the size and the concentration of the emitted particles influence the behaviour of particles in the environment. Due to the increasing popularity of biodiesel fuels, the impacts of fuel formulation on diesel particulate emissions and soot characteristics have been studied by several researchers. Many researchers have found that various oxygenated diesel fuel additives, such as methyl esters obtained from vegetable oils, lead to reductions in total particulate matter emissions [9, 78, 79]. Vander Wal and Tomasek [5] showed that the soot particles generated from pyrolysis of benzene, ethanol and acetylene have different nanostructural order and reactivity. Soot particles were examined by high-resolution transmission electron microscopy (HRTEM). Image analysis of the fringe length and fringe curvature from the TEM images showed that fuel composition influenced the degree of graphitization and soot reactivity. The concentrations and size distributions of particulate matter are related to the dilution and cooling of exhaust, equivalence ratio in the combustion process, and engine type [8]. Some

29 14 researchers have concluded that soot from various sources is structurally similar [4, 80], and similar to commercial carbon blacks [43]. More recently, however, many researchers have shown that the influence of engine conditions and fuel type on diesel soot properties is significant. Experimental results show that the nanostructure of soot depends upon its formation conditions, such as temperature and residence time [10]. Utilizing a numerical model for diesel combustion, Chan et al. [11] concluded that soot and nitrogen monoxide emissions were affected by engine operating conditions and fuel injection timing. Mathis et al. [81] demonstrated a reduction in primary soot particle diameter when increasing fuel injection pressure, or advancing the start of injection. Using laser induced incandescence and TEM, Oh and Shin [82] showed that the particle size and soot volume fraction abruptly decrease in the case of using CO 2 as a diluent instead of N 2. Neer and Koylu [13] reported that average spherule and aggregate sizes generally increase with the overall engine equivalence ratio at various engine loads and speeds. They also found that the particle sizes increased with exhaust temperature. On the other hand, Zhu et al. [12] found that the crystallinity of diesel soot increases with engine load and exhaust temperature. In addition, it is also found that soot particle size tends to decrease when the exhaust temperature increases because of particle oxidation at high temperature. The combined impacts of fuel composition and combustion process on combustion products have been studied. Armas et al. suggested that using biodiesel can reduce smoke opacity in both steady conditions and transient engine operation [15]. By analyzing the bulk modulus of fuels and needle lift sensor signals, Szybist et al. [14] showed that bulk modulus affects fuel injection timing in pump-line-nozzle type fuel injection systems. That investigation proposed that NOx, and soot properties, could be different because the injection timing shift resulted from fuel properties. Using an electronically controlled common rail direct injection system, the potential injection timing differences due to the different fuel properties can be eliminated. With the fixed

30 15 injection timing and similar heat release rate profiles in a cycle, Zhang et al. [16] showed that biodiesel blends produce different amounts of NOx compared to ultra low sulfur diesel fuel (BP15). In order to address the impacts of fuel properties on the combustion process and soot, PM, and gaseous emissions in a comprehensive manner, it is necessary to study the combustion process of different fuels. The PM and gaseous emissions from different fuels are also of interest because they are combustion products and provide information regarding the engine combustion process. In addition, Lapuerta et al. [17] concluded that a meaningful comparison of emissions and fuel consumption is only possible if tests are carried out under the same operation mode. The experimental comparison of diesel emissions produced by BP15, B100 and FT fuels, the three test fuels for generating soot in this study, were summarized [83, 84]. The study was carried out in a direct injection 2.5L common-rail turbodiesel engine working at four engine operation modes, spanning conditions of most interest in the engine map. In all modes the engine was tested with single and split injection (pilot and main), with constant start of injection (SOI), and without exhaust gas recirculation (EGR). It was found that the FT fuel can reduce both NOx and PM specific emissions in all modes under both single and split injection modes, bypassing the nitrogen oxides-particulate matter (NOx-PM) trade off. In addition, this work confirms that biodiesel can reduce the particle concentration [84]. The fuel impact on the combustion phasing was also eliminated by adjusting the start of injection and fuel rail pressure, while producing the same combustion phasing. The study has confirmed that the FT fuel can reduce all regulated diesel emissions under both single and split injection strategies. Finally, it has been confirmed that biodiesel can reduce particle mean diameter in comparison with BP15 [83].

31 Hypotheses and objectives The above literature survey shows that current understanding on the impact of fuels on soot properties is not yet complete, especially for the soot generated by a real diesel engine under a well-controlled combustion phasing. Existing results are limited to the usage of a real diesel engine with different fuels without matching the combustion phasing. The brief survey on diesel soot formation suggests that ignition delay and thus the premixing time can play a critical role in the formation of soot. And, the cylinder temperature and the rate of heat release may affect the formation and oxidation of soot. The present work involves a carefully designed test condition where all the soot samples undergo a similar combustion history before they were collected. The objective of this work is to study the impacts of fuel properties on soot oxidation and reactivity where fuel properties are decoupled from the ignition delay. The research hypotheses for this study are: Even where the phasing of combustion is matched for different fuels, B100 fuel increases soot oxidation reactivity to a measurable degree by forming a less ordered nanostructure or more surface oxygen-containing functional groups. Two subordinate hypotheses are tested: (1) The soot oxidative reactivity is dominated by the order of the soot nanostructure and the consequent increase of accessible carbons on the edge sites when the soot is attacked by air. (2) The order of soot nanostructure has a more dominant influence on the soot oxidative reactivity than the abundance of surface oxygen content.

32 17 It should be noted that Seong has recently shown that there is no correlation between specific oxygen functional groups and soot oxidative reactivity [68], although he found the surface oxygen content can be good indicators of soot oxidative reactivity. The second subordinate hypothesis above further assumes that the soot oxidative reactivity is dominated by the order of soot nanostructure to a greater degree than the surface oxygen content, contrast to the conclusion obtained by Song [6].

33 Chapter 3 EXPERIMENTAL 3.1 Technical approach The list of tasks to be undertaken includes: (1) examine impacts of engine operating modes and the timing of the start of injection (SOI) on soot properties; (2) determine an engine operating mode and the corresponding injection parameters (the SOI timing, fuel pressures) such that the three test fuels, biodiesel (B100), Fischer-Tropsch diesel (FT), and conventional lowsulfur diesel (BP15) all have similar combustion phasing; and (3) perform a comparison of the oxidation process, nanostructure and surface chemical compound between soot samples collected under the similar combustion phasing found in task (2). The suite of experiments to be conducted isolates the impacts of fuel chemistry on soot from other factors, such as the engine operating modes and fuel injection parameters. Although a diesel particulate filter (DPF) was not used directly in this study, this research can provide insights concerning the role of fuel and engine operating parameters on soot properties, which affect the regeneration of the DPF. The first task is to use a reference fuel (BP15) to investigate the impact of engine operating modes (engine speed, torque) and the SOI timing on soot properties. To study the impacts of engine operating modes, the soot samples were collected under four different operating conditions. In these four modes, split fuel injection strategy (pilot and main injections)

34 19 was used and the injection parameters were determined by the electronic control unit (ECU). To study the impacts of the SOI timing, three different starts of injection were used under an engine speed of 2400 rpm and an engine torque of 64 Nm. In this task, the collected particulate samples were pretreated and analyzed by a thermogravimetric analyzer (TGA). Kinetic constants were derived from the TGA results. Preliminary soot property analyses were performed using transmission electron microscopy (TEM) and X-ray photoelectron spectroscopy (XPS). The second task is to determine an engine operating mode and the suite of injection parameters for collecting particulate matter. Based on the exhaust temperatures of the four operating modes measured in the first task, the engine mode was selected such that the properties of PM are not altered by high exhust temperature. At that selected engine mode, with both single injection and split injection strategies, the combustion phases of the three test fuels were investigated first when the fuel injection parameters (start of injection and fuel pressure) were determined by ECU. Then, the fuel injection parameters were adjusted in order to match the combustion phasing for the three test fuels. Under the matched combustion phase, the PM was collected for analysis in reactivity and nanostructure. The third task is to investigate differences in particulate reactivity and nanostructure. A mass-based oxidation rate was determined using the thermogravimetric analyzer (TGA). To relate the differences in oxidation rate with initial soot properties and structural changes during particulate oxidation, various characterization methods were applied. X-ray diffraction (XRD) patterns were used to derive crystalline parameters from 002 and 10 peaks. Raman spectroscopy was used to obtain the disorder of the soot samples that is reflected by the peak area ratio of the defect peaks to the graphitic peak. High resolution transmission electron microscopy (HRTEM) was used to study the graphene layers of the primary soot particles. An HRTEM image analysis

35 20 method was developed for analyzing the configurations of the graphene layers. Oxygen content in the soot samples was measured and quantified by X-ray photoelectron spectroscopy (XPS). XPS and Fourier transform infrared spectroscopy were used to qualitatively investigate the surface functional groups. The active surface area was measured using the thermogravimetric analyzer. The third task also includes the investigation of the nanostructure and composition at different levels of oxidation. 3.2 Test engine An instrumented DDC/VM Motori 2.5L, 4-cylinder, turbocharged, common rail, direct injection light-duty diesel engine was used in steady-state testing. The test engine does not have a diesel particulate filter (DPF) fitted. The main engine characteristics are shown in Table 3-1. A 250HP Eaton eddy current water-cooled dynamometer was coupled to the engine to generate load. The engine and dynamometer were controlled by a Digalog Testmate control unit. Time-based data acquisition was managed using a custom programmed National Instruments LabView VI. Analog signals from pressure transducers, thermocouples, mass flow meters, and emissions data were read by a series of National Instruments FieldPoint modules. The data were collected by the FP modules every 1 second during a 3-minute period in each test. The fuel mass within the fuel tank was measured using a Sartorius electronic microbalance. A custom LabView program calculated the fuel mass consumption rates based on one hundred measurements of fuel tank weight, tracking the change in mass over sixty seconds [85]. An open electronic control unit (ECU) was used to control the main injection and pilot injection timings, as well as, EGR valve position, which was kept closed during all tests. The fuel

36 21 rail pressure was not externally controlled. The ECU was connected to an ETAS measurement and calibration (MAC 2) interface via an emulator test probe. The MAC 2 interface was connected to a PC running ETAS INCA v5.0 software. INCA managed the ECU modifications in real-time. Pressure traces were measured using AVL GU12P pressure transducers, which replaced the glow plug in each of the four cylinders. The pressure trace voltages from the pressure transducers were amplified by a set of Kistler type 5010 dual mode amplifiers. The amplified voltages were read by an AVL Indimodul 621 data acquisition system. Needle lift data were collected from a Wolff Controls Inc. Hall-effect needle lift sensor, which was installed in the fuel injector placed in cylinder 1. The needle lift signal was also collected by the Indimodul, which was triggered by a crank angle signal from an AVL 365C encoder placed on the crankshaft. The pressure traces and needle lift data were recorded at a resolution of 0.1 crank angle degrees, and were averaged over 200 cycles. The real-time Indimodul data were transferred to a PC, which ran AVL Indicom 1.3. The apparent rate of heat release was calculated from pressure trace data generated by cylinder 1, using a thermodynamic diagnostic model. The model combines both the first principle of thermodynamics and the state equation. For calculation of thermodynamic properties, the model takes three species into account: air, evaporated fuel and burned products [86, 87].

37 22 Table 3-1 Engine characteristics Engine code DDC 2.5L TD DI-4V Fuel injection system Bosch common rail injection system Number and relative position of injections 1 pilot injection before TDC 1 (optional) 1 main injection before or after TDC EGR system Disconnected Max. rated power 103 kw at 4000 min -1 Max. rated torque 340 Nm at 1800 min -1 Cylinders 4, in line Bore (mm) 92 Stroke (mm) 94 Swept volume (L) 2.5 Compression ratio 17.5 Valves per cylinder 4 1 Top Dead Center 3.3 Test fuels The current work was carried out using three different fuels: (1) an ultra low sulfur diesel fuel (BP15) was used as a baseline fuel; (2) a Fischer-Tropsch fuel (FT) produced by a gas-toliquids process; and (3) a pure biodiesel (B100), which is composed of methyl esters obtained from soybean oil. Table 3-2 and Table 3-3 show fuel composition and property information. The standards by which the fuel properties in Table 3-3 were measured are also included. The fuel temperature was controlled through a water cooling system and it was monitored during all tests. The mean fuel temperature of all test conditions (combination of different test fuels, operating conditions, injection strategies) is 25.3 C ± 1.0 C at 95% confidence [88]. Among the three test fuels, BP15 is the most similar to conventional diesel fuel. Therefore, BP15 was used as the reference fuel in this study.

38 23 Table 3-2 Composition of test fuels (% wt) Hydrocarbons BP15 FT Soybean methyl ester B100 Paraffins Palmitic acid 8.80 Olefins Palmitoleic acid 0.09 Aromatics Stearic acid 4.55 Oleic acid Linoleic acid Linolenic acid 7.74 Araquidic acid 0.39 Gadoleic acid 0.23 Behenic acid 0.41 Erucic acid 0.01 Linoceric acid 0.13 Nervonic acid 0.01

39 24 Table 3-3 Properties of test fuels BP15 B100 FT Standard Density (g/cm 3 ) a < 0.8 ASTM D1298 Kinematic viscosity (cst) b ASTM D446 Gross heating value (MJ/kg) EN Low heating value (MJ/kg) c ASTM D Acid number (mg KOH/g) EN % C (wt) d % H (wt) d % O (wt) d ppm S (wt) e < 2 - Molecular weight f Stoichiometric air-fuel ratio g IBP (ºC) Distillation T50 (ºC) EN3405 T90 (ºC) Iodine number h Derived cetane number ASTM D6890 a Measured at 15 ºC; b Measured at 40 ºC; c Calculated from composition and gross heating value d Obtained from elemental analysis; e Measured by the supplier f,g, h Obtained from composition 3.4 Engine operating mode and fuel injection parameters The engine was tested at four operating modes shown in Table 3-4. The engine mode C was chosen as the reference mode for comparing soot characteristics with the results of Song et al. [6, 18, 63]. The engine speed and torque of mode C were selected so as to obtain an exhaust gas temperature lower than the lowest break-even temperature for a typical DPF investigated by Boehman et al. [63]. Modes A, B and D were chosen in order to study the impact of engine speed, torque, and injection strategies on engine emissions. Meanwhile, the four modes form a speed-

40 torque zone that encloses a large portion of transient and steady states for a similar engine in a standardized driving cycle [78]. 25 The timing of the start of injection (SOI) at each operating mode in Table 3-4 were determined and recorded by the ECU when BP15 was used. When the engine was tested with B100 and FT, the timings of the start of injection (SOI) was fixed at the timings that were determined by the ECU when using BP15. Table 3-4 Operating modes and the timing of the start of injection Engine mode Engine speed (min -1 ) Torque (Nm) Single SOI (degree *) Pilot Split Main A B C D * (+) Before Top Dead Center, (-) After Top Dead Center The thermodynamic analysis results for the four engine modes for single injection and split injection strategies are shown in Figure 3-1. For the single injection strategy, the engine operating mode affects the combustion phasing significantly although the SOI timing is similar (Figure 3-1 (a)). On the contrary, for the split injection strategy, the diffusion combustion phasing (the rate of heat release after 10 ATDC) is very similar at the different operating modes even though the premixed combustion phasing (before 0 ATDC) is very different. The results show that engine operating condition has a more significant impact on the combustion phasing when using a single injection strategy.

41 Figure 3-1. Thermodynamic analysis results of four engine operating modes: (a) single injection, and (b) split injection. ( ) mode A, ( ) mode B, ( ) mode C, and ( ) mode D. 26

42 27 Table 3-5 and Table 3-6 show the exhaust gas temperature, measured with a thermocouple, and injection pressure, measured with a fuel rail pressure sensor with single and split injection, respectively. The gas temperature during mode C has been kept lower than the break even temperature for a typical diesel particulate filter for all the test fuels [63]. The exhaust temperature difference between fuels is less than 3%. Table 3-5 Exhaust gas temperature and injection pressure during each test with single injection Engine mode Exhaust gas temperature (ºC) 1 Injection pressure (bar) 2 BP15 B100 FT BP15 B100 FT A B C D Measured; 2 Obtained from the ECU Table 3-6 Exhaust gas temperature and injection pressure during each test with split injection Engine mode Exhaust gas temperature (ºC) 1 Injection pressure (bar) 2 BP15 B100 FT BP15 B100 FT A B C D Measured; 2 Obtained from the ECU; In order to study the impact of changing the start of injection (SOI) on PM properties, the engine was tested at Mode C as shown in Table 3-7. The engine was tested using BP15 fuel with single fuel injection, varying the timing of the SOI 2 degrees before and after the baseline SOI. The selected baseline SOI and fuel rail pressure was the suite defined in Table 3-8 (4.83 degree BTDC, 511 bar). The fuel pressure was kept the same when varying the SOI. Figure 3-2 shows

43 28 the thermodynamic diagnostic results of SOI variation. In Figure 3-2, the SOI timing differences can be verified by the needle lift signal traces. The impact of SOI variation on cylinder pressure and the rate of heat release can be observed distinctly. Table 3-7 Engine operation mode and variation of the start of injection (SOI) Engine mode Speed (min -1 ) Torque (Nm) Fuel injection parameters SOI, single injection (degree) Fuel pressure (bar) Advanced Baseline Delayed * (+) Before Top Dead Center, (-) After Top Dead Center Figure 3-2 Thermodynamic diagnostic results of start of injection (SOI) variation using BP15. ( ) baseline, ( ) advanced, and ( ) retarded.

44 29 To obtain the condition that isolates the impacts of fuel chemistry on soot from different combustion phasing, at mode C, the fuel injection parameters were adjusted further to achieve matched combustion phasing as shown in Table 3-8 and Table 3-9. Figure 3-3 and Figure 3-4 show the thermodynamic diagnostic results before and after adjusting the fuel injection parameters. The heat release results in Figure 3-3(a) and Figure 3-4(a) show different phasing, while the heat release results in Figure 3-3(b) and Figure 3-4(b) show similar phasing of heat release. The engine performance, fuel consumption, gaseous and PM emissions, and particle size distributions of the test modes listed in Table 3-4, Table 3-8, and Table 3-9 were measured and summarized in Ref. [83, 84]. The PM samples were collected under Modes A, B, C, D with split injection using BP15 fuel. In addition, the PM samples were collected under Mode C at matched combustion phasing with both single and split injection using BP15, FT, and B100 fuels. Table 3-8 Start of injection and fuel rail pressures which produced a similar combustion process (single injection) Injection parameters modified Fuel SOI (degree btdc) Fuel rail pressure (bar) BP (1 degree advanced from reference) 511 (90% of the reference) B (Baseline) 572 (Baseline) FT 1.83 (2 degree retarded from the reference) 595 (105% of the reference)

45 Table 3-9 Start of injection and fuel rail pressures that produce similar combustion conditions (split injection) Engine control parameters Fuel Pilot SOI (degree btdc) Main SOI (degree btdc) Fuel rail pressure (bar) BP (1.4 degree atdc) 572 (reference) B (1.4 degree atdc) 591 (reference) FT (1.4 degree atdc) 537 (95% of reference) 30

46 Figure 3-3 Thermodynamic diagnostic results obtained with a similar combustion process. Single injection. Without (a) and with (b) the adjusted SOI and fuel rail pressures. ( ) BP15 ( ) B100, and ( ) FT fuels. 31

47 Figure 3-4 Thermodynamic diagnostic results adjusting the SOI obtaining a similar combustion process. Split injection. Without (a) and with (b) adjusted SOI and fuel rail pressures. ( ) BP15 ( ) B100, and ( ) FT fuels. 32

48 33 In order to test the hypothesis in a comprehensive manner, the impact of the three factors below were investigated using the PM samples collected under the specified conditions: (1) The impact of the engine operating condition on soot reactivity: Mode A D using BP15 with split injection strategy (Table 3-4). (2) The impact of combustion phasing on soot reactivity and nanostructure: Mode C using BP15 with standard SOI, advanced SOI, and retarded SOI (Table 3-7), single injection. (3) The impact of fuel composition on soot reactivity and nanostructure: Mode C using adjusted injection parameters that yield similar combustion phasing for BP15, FT, and B100. Both single injection and split injection strategies were studied. 3.5 PM sample collection method Using a vacuum pump, particulate matter was collected on 47 mm Teflon filters for 30 to 60 minutes depending on the fuels and engine operating conditions. The collecting time was determined empirically by monitoring the saturation of the soot mass deposited on the Teflon filter. The particulate matter was subsequently scraped from the filters. Before characterization, the PM was thermally treated at 500 C for 60 minutes in a thermogravimetric analyzer (TGA) under ultra high purity nitrogen in order to remove volatile matter or adsorbed unburned hydrocarbons. Preliminary TGA tests, as shown in Appendix A, have confirmed that the pretreatment method does not affect the nanostructure or the reactivity of soot. And, the preliminary tests have confirmed that neither the pore structure nor the volatile organic fraction plays a significant role in the oxidative reaction.

49 Characterization method In order to test the hypothesis proposed in Chapter 2, thermogravimetric analysis (TGA), X-ray diffraction (XRD), Raman spectroscopy, transmission electron microscopy (TEM), and X- ray photoelectron spectroscopy (XPS) were used to characterize the pretreated PM samples. Fourier transform infrared spectroscopy (FTIR) was used to examine the PM pretreatment method. The characterization methods are briefly discussed in the subsequent sections, while the detail quantification algorithms will be described in Chapter Thermogravimetric analysis (TGA) A thermogravimetric analyzer (TA instruments, SDT Q600) with recording software was used to characterize the oxidative reactivity and VOF content of the PM samples. The steps of the isothermal oxidative reaction in conjunction with the pretreatment method are listed in Table Although the flow characteristics within a DPF during the regeneration period are complex, the TGA test captures two important properties. One is the VOF content of the PM samples that can be combusted by the diesel oxidation catalyst upstream of the DPF [3], while another is the representative rate of oxidation derived from an isothermal condition. The volatile organic fraction (VOF) content is determined by the weight of samples at the end of Step 4. According to XPS, no metal elements were found in any of the as-received soot samples, so the weight was assumed zero (without ash) at the end of an oxidation test. The traces recorded in step 8 are then normalized and used to derive the apparent rate constant for soot oxidation as discussed in Chapter 4. From suite of tests conducted for isothermal oxidation, the experimental uncertainty is ±4.4% error at 95% confidence, considering equipment perturbations, day-to-day variations of TGA, and systematic errors of the test engine (Appendix B). The oxidation temperature was set to

50 C because such a temperature will reliably initiate regeneration in a diesel engine aftertreatment system [89]. Unless specifically notated, all the samples analyzed by the characterization methods described in the following sections were pretreated and VOF free. Table 3-10 Summary of TGA pretreatment and isothermal oxidation test procedures Step Procedure 0 Start with nitrogen 1 Ramp 10 C/min to 30 C 2 Isothermal for 30 min (To stabilize the sample) 3 Ramp 10 C/min to 500 C 4 Isothermal for 60 min (To remove volatile matter) 5 Ramp 5 C/min to 550 C 6 Isothermal for 5 min (To stabilize the sample) 7 Select zero air 8 Isothermal for 150 min Additionally, TGA was used to investigate the active surface area (ASA) by oxygen chemisorption. The test procedures are listed in Table The ASA is calculated as follows: ASA N N m 0 0 A (Equation 3-1) i where N 0 is the number of moles of chemisorbed oxygen, σ 0 is the area occupied by each oxygen atom (0.083nm 2 ), N A is Avogadro s number, and m i is the initial mass of soot.

51 36 Table 3-11 Summary of TGA pretreatment and ASA measurement procedures Step Procedure 0 Start with nitrogen 1 Ramp 10 C/min to 30 C 2 Isothermal for 30 min (To stabilize the sample) 3 Ramp 10 C/min to 500 C 4 Isothermal for 60 min (To remove volatile matter) 5 Ramp 5 C/min to 200 C 6 Isothermal for 30 min (To stabilize) 7 Select zero air 8 Isothermal for 600 min 9 Select nitrogen 10 Isothermal for 90min X-ray diffraction (XRD) The crystalline parameters as the bulk sample properties were derived from XRD patterns. An X-ray powder diffractometer (PANanalytical X Pert PRO MRD) with Cu-Kα radiation (λ=0.154nm) was used to collect the XRD patterns of soot samples. The instrument was used with fixed slit incidence (0.5 divergence, 1.0 anti-scatter, specimen length 10 mm) and diffracted (0.5 anti-scatter, 0.02 mm nickel filter) optics. The powder sample was pressed into the cavity of a quartz low-background support. Data were collected at 45 kv and 40 ma from θ using a PIXcel detector in scanning mode with a PSD length of θ, and 255 active channels for a during time of about one hour. The patterns were calibrated for instrumental broadening and peak position using silicon as an external standard. The calibrated patterns were used to derive the crystalline structure of the soot samples to obtain interlayer spacing (d 002 ), stacking height (L c ), crystalline basal plane diameter (L a ) and average number of layers (N) in stacks, as illustrated in Figure 3-5. All samples have (002) peaks around 25 and (10) peaks

52 37 around 43. The crystalline parameters (d 002, L a, L c, and N) were derived from XRD patterns according to the procedures described in [24]. The interlayer spacing d 002 was calculated by Bragg s equation: where incident incident d 002 (Equation 3-2) 2sin 002 is the incident wavelength (0.154nm), and 002is the peak position of (002) peak. The crystalline basal plane diameter, L a, and the stack height, L c, were derived by using Scherrer equation: L 0.89* incident c, 002 cos 002 L a 1.84* cos incident (Equation 3-3) where 002and 10are the half maximum widths of (002) and (10) peaks, and 10 is the peak position of (10) peak. The number of layers per crystalline (N) is determined by L c and d 002 : L N c (Equation 3-4) d 002 Figure 3-5 Schematic representation of the crystalline parameters (d 002, L a, L c, and N) [90].

53 Raman spectroscopy A confocal Raman spectrometer (WITec CRM200) was used to determine the order of the soot samples. The excitation laser was an Argon ion laser (λ=514.5nm), focused onto the sample using a microscope objective lens (100x or 40x). The scattered light was collected by the same objective collected in a backscattering configuration (180 degree). The microscope is equipped with a XY stage, driven by piezo-electric actuators, for selecting the sample area of interest. The spectrum collection method is configured by setting the integration time (seconds), software accumulations, and hardware accumulations in the software WITec Project. All spectra were collected at the same source power level, and post-processed via a software package (Igor Pro 6.10, Wavemetrics Inc.). The software was used for the determination of spectral parameters. The Raman spectra were first corrected by a linear baseline. Then, the corrected spectra were curve fitted by three Lorentzian-shaped bands (1580, 1350, 1200 cm -1 ) and one Gaussian-shaped band (1500 cm -1 ) [68]. The band near 1580 cm -1 (G band) reflects the graphitic lattice, while the band near 1350 cm -1 (D1 band) reflects the disordered graphitic lattice. The bands near 1500 and 1200 cm -1 (D3 and D4 bands) also reflect amorphous structure and disordered graphitic lattice. The integrated intensity of I D /I G was used to investigate the disordering degree of the soot samples Transmission electron microscopy (TEM) To investigate the soot morphology and nanostructure, a 200-kV field-emission TEM (JEOL EM-2010F) was used to take high-resolution bright field images. Depending on the soot samples, the applied magnifications varied between 300k and 500k. For nanostructure analysis, the initial and partially oxidized soot samples were ultrasonically dispersed in methanol for 60

54 39 minutes [20]. Then, five drops of the solution were transferred to a lacey C/Cu TEM grid. For morphology analysis, the samples were collected on TEM grids directly via a thermophoretic sampling method [6]. Soot morphology was only considered qualitatively as one part of the preliminary study on engine injection parameters. Soot nanostructure (fringe length, tortuosity, and separation space) was investigated in a comprehensive way by developing the image analysis software. The software details are discussed in Chapter 4. For every analyzed sample, the images of soot aggregates at more than twenty locations were recorded. For each sample, 6 ~ 8 images of soot aggregates with clear configuration of graphene layers were selected from all the recorded images for further image analysis. For demonstration purpose, only representative images were shown X-ray photoelectron spectroscopy (XPS) The chemical compositions and oxygen-containing functional groups on the soot surface were investigated by XPS. XPS spectra were recorded in an Axis Ultra from Kratos Analytical using a monochromatic Al-Kα ( ev) X-ray source operated at 10-8 bar. The analysis area is roughly 1 mm * 1.5mm, and the sample surface was oriented normal to the analyzer entrance. The pass energy was set at 80eV for survey scans, and 20eV for high resolution scans. The soot samples were mounted on double sided conductive carbon tape which was fastened to a Cu substrate. The influence of the charge neutralizer was checked on each sample. If no peak shifts were observed when the change neutralizer was turned on, then the neutralizer was left off for the sample. Otherwise, the neutralizer was used to eliminate peak distortion. The XPS spectra were processed by commercial software (CASA XPS). A Shirley background was subtracted from the high resolution spectra. Then, the corrected spectra were curve-fitted for possible elements (e.g.,

55 40 carbon 1s: ev, oxygen 1s: ev). The element percentages were thereafter estimated by the software. In this study, the XPS spectra of all PM samples only contain carbon and oxygen, and do not show any signification amounts of metal elements Fourier transform infrared spectroscopy (FTIR) The oxygen-containing functional groups on the surface of soot particles were investigated by an FTIR spectrometer (Bruker IFS 66/S). To investigate the impact of various pretreatment methods, the FTIR spectra were collected using the standard transmission mode. In the transmission mode, the infrared radiation passed through a sample completely, and the extent of absorption was measured. On the other hand, to investigate any compositional changes during the heat pretreatment in inert gas, the data were collected using Diffuse Reflectance Infrared Fourier Transform Spectroscopy (DRIFTS). In this study, the FTIR results were qualitatively based on the band assignment of oxygen-containing function groups summarized in Table 3-12 [6, 70, 91].

56 41 Table 3-12 Functional groups on soot surface and their corresponding FTIR band assignments Range (cm -1 ) Surface functional groups C-H stretch mode in aromatic and aliphatic groups [70, 91] C=O stretching mode in lactone, carboxylic and ketone groups [6, 70, 91] C=C stretching mode in aromatic group and quinones [6, 70] 1450 C-H stretching mode in aliphatic group [6] C-O stretching mode of ethers, esters, acids, phenolic groups [6, 70]

57 Chapter 4 SOOT CHARACTERIZATION METHOD Six characterization techniques have been discussed in Section 3.6: (1) Thermogravimetric analysis (TGA), (2) X-ray diffraction (XRD), (3) Raman spectroscopy, (4) Transmission electron microscopy (TEM), (5) X-ray photoelectron spectroscopy (XPS), and (6) Fourier transform infrared spectroscopy (FTIR). Among the six techniques, the data from XPS and Raman spectroscopy were post-processed by commercial software, and the analytical methods were referred to those described in Ref. [68]. Additionally, because the spectra of FTIR were merely interpreted in a qualitative manner, the FTIR data analysis is not explained further. Therein, this chapter describes the methods for analyzing TGA, XRD, and TEM data in this study. 4.1 Quantifying apparent rate constant from TGA isothermal oxidation test In order to quantify the oxidative reactivity differences under the isothermal oxidation condition described in Section 3.6.1, a simplified rate expression for carbon oxidation was used to compute the apparent rate constant of diesel soot [6, 46, 92]: dm m c c k p dt (Equation 4-1) c O2 where m c is soot mass at time t, k c is apparent rate constant (1/Pa/min), and p O 2 is the partial pressure of oxygen (Pa). It should be noted that k c is a lumped parameter depending on the

58 43 concentration of carbon active sites [46, 93], rather than an intrinsic property [4, 94]. The apparent rate constant is used throughout this study because it is yet unclear how to accurately measure carbon active sites on soot, as discussed in section In this study, the regions used to derive the apparent rate constant are the region of a zero-order region as defined in [92]. Equation 4-1 assumes that k c and po 2 are independent of time. Integrating both sides of Equation 4-1 with respect to time yields: m m c c,0 k p ( t t ) c O2 0 e (Equation 4-2) where mc, 0 is the initial soot mass, t 0 is the time of starting oxidation. The left hand side of Equation 4-2 is the remaining weight fraction measured during the isothermal oxidation test. For demonstration purposes, Figure 4-1 plots the simulated mass fraction with two different values of k c by Equation 4-2 assuming that the oxidizer is ultra zero air (22% of oxygen). Figure 4-1 demonstrates that the model described by Equation 4-2 captures two important phenomena that were observed in the normalized soot mass during isothermal oxidation tests: (1) the slope of the initial period of oxidation is affected by k c ; and (2) the time needed to oxidize half of the sample mass depends on k c [6, 68]. The two phenomena justify the choice of Equation 4-2 as an oxidation model in this study.

59 44 Figure 4-1 Simulation results of two values of k c by Equation 4-2 assuming the oxidizer is air. ( ) k c = (1/Pa/min), and ( ) k c = (1/Pa/min). An algorithm was developed to curve fit a k c for each normalized isothermal oxidation curve. A discrepancy function was used to quantify how the model fits the normalized data obtained from an isothermal oxidation test: n i 1 k p ( 2 O t 2 i t0 ) f ( k) sum MF e (Equation 4-3) i where i is the data point indicator, n is the total number of data points to be fitted, k is the rate constant value under consideration, t i is the time corresponding to the i th states, and MF i is the mass fraction at t i. Then, the problem in curve fitting k c is simplified: finding a value of k such that f(k) defined by Equation 4.3 is minimized. The curve fitting algorithm was implemented in MATLAB. The program used a built-in functional minimization command based on a direct

60 45 search method for multidimensional nonlinear unconstrained optimization [95, 96]. In this study, the initial guess of k c is set to 9.2*10-7 (1/Pa/min), which is the decision after several trials. The termination tolerance on the function value is set to 10-6 (no unit). The repeatability study of TGA tests in Appendix B is also based on the rate constant analysis. For demonstration purposes, Figure 4-2 shows two experimental results with their fitted results for two different PM samples using Equation 4-3 with the minimization algorithm. The soot samples were generated from B100 and FT fuels at Mode C using the single injection strategy with matched combustion phasing. The experimental data shows that B100 soot reached 50% oxidation about 15 minutes earlier than the FT soot. The rate constants obtained by curve fitting for the first thirty minutes show a quantitative difference in the reactivity between the two samples. The apparent rate constant for the B100 soot is 1.30*10-6 1/Pa/min, which is about two times of that of the FT soot (6.33*10-7 1/Pa/min).

61 46 Figure 4-2 Demonstration of experimental and fitted curves of TGA isothermal oxidation tests. ( ) B100, Mode C, single injection, pretreated in nitrogen at 500 C for 60 minutes; ( ) FT, Mode C, single injection, pretreated in nitrogen at 500 C for 60 minutes; ( ) B100 fitted curve for the first 30 minutes (k c =1.30*10-6 1/Pa/min); ( ) FT fitted curve for the first 30 minutes (k c =6.33*10-7 1/Pa/min). The apparent rate constant analysis in this study does not consider the effects of oxygen diffusion. Song et al. recently reported that the rate constants do not change much (less than 1%) for both B100 and FT soot at 500 C when considering the external diffusion effects. At higher temperature (greater than 700 C), the oxygen diffusion has a more pronounced effect. In this study, the oxidation temperature is 550 C, so the external diffusion effect is considered to be insignificant (less than 5%) [97].

62 Quantifying crystalline parameters from the X-ray diffraction (XRD) data After correcting with silicon for instrumental broadening and peak shift, X-ray diffraction patterns were used to derive crystalline parameters, such as interlayer spacing (d 002 ), crystalline basal plane diameter (L a ), and stacking height (L c ) [24, 36, 48, 49]. For each pattern, the (002) peak was used to derive d 002 and L c, while the (10) peak was used to derive L a [24]. In order to use Equations 3.2 and 3.3, the peak positions ( 002 and 10 ) and the full widths at half maximum (FWHM, 002 and 10 ) of (002) and (10) peaks are needed. A curve-fitting algorithm was developed to find the peak positions and half maximum widths. The two peaks were modeled as a baseline with Gaussian or Lorentzian functions. The (002) peak was modeled as: 2 m ( ) w Intensity(2 ) a(2 ) b h e (Equation 4-4) 2 where a is the slope of the baseline, b is the constant term determining the latitude of the baseline, m is the peak position, w is the FWHM, and h is the height of the curve. The Gaussian function was modified from the normalized probability density function for the Gaussian random variable, which is described by its center and standard deviation [98]. In this study, Gaussian functions do not fit the (10) peaks as well as they fit the (002) peaks, because the (10) peaks generally have narrow FWHMs. To have a better fitting performance, the Lorentzian function was used to describe the (10) peak. The (10) peak was modeled as: A w Intensity(2 ) a(2 ) b (2 ) (2 m) w (Equation 4-5)

63 48 where a is the slope of the baseline, b is the constant term determining the latitude of the baseline, m is the peak position, w is the FWHM, and A is the area enclosed by the curve. The Lorentzian function was modified from the normalized probability density function for the Cauchy random variables, which is described by its center and FWHM [98]. An algorithm was developed to curve fit the parameters for each corrected XRD pattern. A discrepancy function was used to quantify how the model fits the (002) peak: n f ( h, m, w) sum PIi ( a (2 i) b) h e i 1 2 i m ( ) w 2 2 (Equation 4-6) where i is the data point indicator, n is the total number of data points to be fitted, 2θ i is the diffraction angle corresponding to the i th state, and PI i is the XRD pattern intensity corresponding to the i th state. The definitions of the variables to be determined (h, m and w) in Equation 4-6 are the same as those in Equation 4-4. Similarly, a discrepancy function was used to quantify how the model fits the (10) peak: n A w f ( A, m, w) sum PIi ( a (2 i) b) (2 ) i 1 4 (2 i m) w 2 2 (Equation 4-7) where i is the data point indicator, n is the total number of data points to be fitted, 2θ i is the diffraction angle corresponding to the i th state, and PI i is the XRD pattern intensity corresponding to the i th state. The definitions of the variables to be determined (A, m and w) in Equation 4-7 are the same as those in Equation 4-5. Therein, the problem in curve fitting the (002) or (10) peaks is simplified: within the preselected range of XRD patterns, giving the baseline parameters a and b, find the suite of values of (h, m, w) or (A, m, w) such that f(h, m, w) or g(a, m, w) is minimized. 2

64 49 The curve fitting algorithm was implemented in MATLAB. As with the TGA curve fitting, the program used a built-in functional minimization command based on a direct search method for multidimensional nonlinear unconstrained optimization [95, 96]. The program permits the user to assign the peak range to be fitted, and the baseline parameters, a and b as defined in Equation 4-4 and 4-5. The initial guess values and the termination tolerance on the function values are summarized in Table 4-1. The values listed in Table 4-1 were decided after several trials and the values do not affect the final results if the function (Equations 4-6 and 4-7) converges to a minimum value. Table 4-1 Summary of the initial guess values and the termination tolerance for minimizing Equations 4-6 and 4-7 Variables Equation 4-6 Equation 4-7 h (arbitrary unit) A (arbitrary unit) m (degree) w (degree) 4 5 Termination tolerance For demonstration purposes, Figure 4-3 shows one corrected XRD pattern with the curve fitting result for (002) and (10) peaks using Equations 4-6 and 4-7 with the minimization algorithm. The samples were generated by FT fuel at Mode C using a single injection strategy with matched combustion phasing. Figure 4-4 shows the magnified area of the (002) peak (16 32 ) and the (10) peak (30 80 ) of the XRD pattern in Figure 4-3. In Figure 4-4(a), the (002) peak was fitted with Gaussian function as described by Equation 4-6. In Figure 4-4 (b), two fitted curves were demonstrated for (10) peak using a Loretzian function (Equation 4-7) and a Gaussian function (Equation 4-6). Comparing the two fitting methods with the original XRD pattern in

65 Figure 4-4(b), the Lorentzian function is more representative than the Gaussian function. The result justifies the choice of using Equation 4-7 as a model for the (10) peak in this study. 50 Figure 4-3 A demonstrative XRD corrected pattern and the fitted curves of 002 and 10 peaks. Sample: PM generated at Mode C by FT fuel using split injection strategy with matched combustion phase. ( ) XRD corrected pattern, ( ) baseline for fitting 002 peak, ( ) fitted 002 peak, ( ) baseline for fitting 10 peak, and ( ) fitted 10 peak.

66 51 Figure 4-4 Magnification of (002) and (10) peaks (a) (002) peak fitted by Gaussian function, and (b) (10) peak fitted by Gaussian and Lorentzian functions. (10) peak was fitted better with Lorentzian function. ( ) XRD corrected pattern, ( ) baseline, ( ) Gaussian fitted peak, ( ) Lorentzian fitted peak. 4.3 Transmission electron microscopy image analysis An analytical method was implemented for quantitatively analyzing the TEM images of soot particles. The details of the method can be found in Refs. [65, 66]. The method is composed of two major parts: digital image processing and fringe characterization. The digital image processing is composed of the following operations: negative transformation, region of interest (ROI) selection, contrast enhancement, Gaussian lowpass filter, top-hat transformation, thresholding to obtain a binary image, morphological modification, clearing fringes on the ROI border, skeletonization, and removing short fringes that lack physical meaning. The fringe characterization generates statistics on fringe length, fringe tortuosity, and fringe separation based on the skeletons of the graphene layers. Fringe length and fringe tortuosity are obtained automatically from the features of the skeletons, while fringe separation permits the user to manually select fringe pairs. The method is implemented as a computer program.

67 52 In this study, in order to examine the impact of image processing parameters and have repeatable results, the algorithms were implemented exclusively in MATLAB (The Math Works, Inc., Natick, MA) without using other commercial image processing packages. MATLAB was chosen as a development tool for the following reasons: (1) a sequence of built-in commands can be encapsulated into functions, which simplifies the customization of algorithms; (2) using programming loops, image processing parameters can be calibrated and fined tuned recursively until the result meets the required standard; (3) the final image processing parameters can be saved and retrieved (With this feature, the image processing can be repeated with the same set of parameters); and (4) the image processing toolbox with MATLAB contains subroutines that readily execute spatial and frequency domain matrix operation, morphological image processing, and image segmentation [99]. Figure 4-5 shows the structure of the image analysis algorithm. The algorithm consists of four major subroutines: (1) processing of the HRTEM images using negative transformation, region of interest (ROI) selection, contrast enhancement, Gaussian lowpass filter, top-hat transformation, thresholding for binarization, morphological modification, clearing fringes on the ROI border, skeletonization, and removing short fringe that lack physical meaning, with the result being an enhanced, skeletonized image to be further analyzed for fringe properties; (2) analyzing fringe lengths and tortuosity; (3) analyzing fringe separation; and (4) Generating fringe properties (length, tortuosity, separation) histogram plots and saving data into a spreadsheet. In this section, the steps of image enhancement and image processing are discussed. Subsequently, the algorithms for calculating the fringe properties (length, tortuosity, separation space) are given.

68 53 (a) Main TEM image processing algorithm Start (b) Fringe length and tortuosity analysis Start Negative transformation (Reverse the intensity levels of all pixels) Select the Region(s) of Interest (ROI) Improve contrast (Yes: histogram equalization or histogram matching, No: ignore) Apply gaussian lowpass filter (Yes: filter size and deviation, No: ignore) Top-hat Transformation (Yes: dilation size, No: ignore) Load binarized images and data Obtain fringe length and tortuosity statistics Stop (c) Fringe separation analysis Start Load binarized images and data Save processed data and images Binarization (threshold by Otsu s method or threshold manually) Perform morphological opening and closing to obtain major shapes of carbon layers (Skip: disk size = 0) Select fringe pair Obtain separation distance Continue to select fringe pairs or not Yes Remove elements on the ROI border Skelentonize elements Break triple or quadruple joints Rule out artifacts by establishing a minimum length No Stop (d) Histogram generation and data output Start Load all processed data and images Save processed data and images No Save processed data and images The result image is satisfying to proceed Stop Yes Generate plot and histograms of fringe length, tortuosity and separation distance Stop Write image processing parameters and histogram data into a spreadsheet file Figure 4-5 Flow chart of the image processing program used to enhance and process HRTEM images.

69 Image enhancement and image processing The main image processing program is a loop and is illustrated in Figure 4-5(a). The program was developed to load and process eight-bit grayscale images. The algorithm can process images of higher grayscale depth (greater number of bits per pixels) by either sacrificing the spatial resolution or increasing the computational cost. However, the algorithm in the current work is not directly applicable to color images, such as RGB or CMYK color systems unless it is further revised. For demonstration purposes, Figure 4-6 used an exemplary image to illustrate each step listed in Figure 4-5 (a). Start Stop Further processed for fringe length, tortuosity, and fringe separation Negative transformation (Reverse the intensity levels of all pixels) Yes Save processed data and images Select the Region(s) of Interest (ROI) No The result image is satisfying to proceed Improve contrast (Yes: histogram equalization or histogram matching, No: ignore) Rule out artifacts by establishing a minimum length Apply gaussian lowpass filter (Yes: filter size and deviation, No: ignore) Top-hat Transformation (Yes: dilation size, No: ignore) Binarization (threshold by Otsu s method or threshold manually) Break triple or quadruple joints Skeletonize elements Remove elements on the ROI border Perform morphological opening and closing to obtain major shapes of carbon layers (Skip: disk size = 0) { { Figure 4-6 Illustration of the image enhancement and image processing steps.

70 Negative transformation In the bright-field TEM image, the carbon layer segments (fringes) appear as dark lines because the fringes block or scatter the incident electron beam. The blockage of the incident electron beam by the fringes forms the dark area at the detector, while the passage of incident electron beam forms the bright areas. However, in a grayscale digital image, a darker pixel has a lower intensity value and a brighter pixel has a higher intensity value. Therefore, a negative transformation [100] is applied after the program loads an eight-bit grayscale TEM image to be processed. The negative transformation can be expressed as I L 1 (Equation 4-8) negative I original where, in the equation above, L is the number of discrete intensity levels (256 for the exemplary image), I original the values of pixels before transformation, and I negative the values of pixels after transformation. After the negative transformation, the higher intensity pixels (the brighter area) corresponds to graphene layers while the lower intensity pixels (the darker area) corresponds to the background. Using the negative transformation in the beginning of the algorithm reduces the complexity in many of the following image processing steps, such as fringe morphological modification (opening and closing), top-hat transformation and fringe skeletoniztion Region of interest (ROI) selection The algorithm permits the user to select a region of interest (ROI) for further processing. The pixels outside the ROI were set to the lowest intensity level, 0. The ROI should consist of the regions that contain sufficient and clear fringe information. Meanwhile, selecting ROI for image

71 56 analysis can preclude regions that contain irrelevant information or objects that are difficult to identify. The scale bar, the lacey-carbon film of TEM grid, areas with fringes that weave together, and the background are often excluded during the ROI selection process. Mathematically, the histogram of the selected ROI can be obtained by two steps: h h n k ( Ik ), k=1 L-1 (Equation 4-9a) nroi n n non ROI ( Io) 0 (Equation 4-9b) nroi where I k is the kth intensity level in the intensity interval [0, L-1],n k is the number of pixels in the image whose value is I k, and n ROI is the number of pixels constituting the ROI. For k=0, the number of pixels that are outside of the ROI (n non-roi in Equation (4-9b)) is subtracted from the total number of pixels with the lowest intensity, 0, in the image. Because a large portion of the background pixels, ranging from intensity level 80 to 100, have been removed, the range of the histogram of the ROI is wider than the whole, gray-scale image. To summarize, the result of selecting the ROI is the extraction of the information regarding the fringes to be analyzed in a TEM image. The subsequent histogram processing for contrast enhancement applies to the selected ROI but not the whole image Contrast enhancement The contrast of an image is defined as the dynamic range of the pixels in an image. In general, the image has high contrast when the image has a large difference in intensity between the highest and lowest intensity levels. The proposed algorithm provides two options for enhancing image contrast of the ROI. According to the quality of the TEM image being processed, the user of the program can decide whether or not to do histogram equalization to

72 57 improve the contrast of the ROI. If the intensity histogram of ROI has a sufficiently large dynamic range (high contrast), the user may skip the histogram equalization and move to the next image processing step. Histogram equalization is used in order to make the intensity histogram of the image as a uniform probability density function. Performing histogram equalization on an image with a narrow histogram can usually improve contrast by spanning intensity to the full intensity scale. Drawing on the general histogram equalization method [100], the program builds a digitized histogram equalization transformation applied to the ROI. The algorithm can be expressed as: p n k input( Iinput, k ) k=0, 1, 2,, L-1 (Equation 4-10a) nroi I equalized, k ( L 1) k k ( L 1) pinput( Iinput, j ) n j k=0, 1, 2,, L-1 j 0 nroi j 0 (Equation 4-10b) Equation (4-10a) defines the probability of occurrence of intensity level I input,k in the ROI, where n k is the number of pixels of ROI that have intensity I input,k, L is the number of intensity levels in the image, and n ROI is the total number of pixels in the ROI. Equation (4-10a) is equivalent to the combination of Equations (4-9a) and (4-9b), which define the intensity histogram of the ROI. Equation (4-10b) is the discrete form of the histogram transformation. Using Equation (4-10b), each pixel with intensity I input,k in the input ROI is mapped into a corresponding pixel with level I equalized,k in the output ROI. It should be noted that a perfectly flat histogram, i.e. the uniform probability of occurrence of intensity level, is rarely obtained in digitized histogram equalization because a histogram of a digital image is an approximation of a probability density function in the continuous domain. If the intensity of a pixel is treated as a continuous and random variable in the interval [0, L-1], Equations (4-10a) and (4-10b) will be in

73 the form of probability density functions and integrals. Thereby, a uniform probability density function will be obtained after the transformation. 58 Regarding TEM analysis of carbon, improving contrast in the ROI has two advantages: (1) increases the visibility for the user to differentiate the fringes from the background; and (2) increase the intensity difference between a fringe and a background pixel, and reduce the complexity when performing image binarization and skeletonization Gaussian lowpass filter While improving contrast, the level of noise is also relatively elevated. The elevation of noise affects the robustness of the whole algorithm, especially those processes after the image binarization. Therefore, a lowpass filter is applied in order to attenuate the high-frequency noise [100]. If the parameters are designed appropriately, the lowpass filter can also be used to repair the TEM image corruptions that often appear on copied printed materials [65]. Following past work [45, 58, 59], a spatial lowpass filter was implemented in the frequency domain. While many types of lowpass filters are possible, Gaussian lowpass filter is used in the program because it does not have ringing effects like an ideal low pass filter or a Butterworth lowpass filter [100]. The relation between the input and output digital image can be expressed as: I H i input filter output ( u, v) [ i ( x, y)] (Equation 4-11a) input ( u, v) [ h ( x, y)] (Equation 4-11b) 1 filter ( x, y) [ H ( u, v) I ( u, v)] (Equation 4-11c) filter input

74 where and 1 are the Fourier transform and inverse discrete Fourier transform operators, I input (u,v) is the discrete Fourier transform of the input image, i input (x, y), i output (x,y) is the output image, and H filter (u,v) is the Gaussian lowpass filter transfer function. It should be noted that the product H filter ( u, v) Iinput( u, v) is the product multiplied on an element-by-element basis, but not the product of the matrix multiplication in mathematics. In this work, the Fourier transform pair of the Gaussian lowpass filter in the spatial and frequency domain is expressed as: 59 h 2 2 x y 2 (, ) 2 filter x y Ae (Equation 4-12a) H filter ( u, v) ( u v ) Ae (Equation 4-12b) A 1 ( fsize 1)/2 x, y ( fsize 1)/2 e 2 2 x y 2 2 (Equation 4-12c) where σ is the standard deviation, and x and y are horizontal and vertical index to the pixels relative to the center of the filter mask in the spatial domain. The indices u and v in equation (4-12b) are the horizontal and vertical indices relative to the center in the frequency domain. In Equation (4-12c), f size is the size of the filter mask in the spatial domain, which should be an odd number for obtaining symmetry about the center of the filter mask. The coefficient A, as defined in equation (4-12c), is a term to normalize h filter such that the sum of all values of h filter is one. In the program, the user assigns the filter mask size, f size, and the standard deviation, σ, according to the condition of the input image. Equation set (4-12) shows two characteristics: (1) the filter functions in spatial and frequency domain are both Gaussian and real. Thereby, the computation does not need to be concerned with complex numbers. In addition, the use of Gaussain lowpass filter avoids the ringing property that usually happens when using an ideal lowpass filter [100].

75 60 Because carbon material has stacks of lattice graphene layers, the existence of ringing effect causes artifacts when isolate the lattice fringes; and (2) The Fourier transform pair is reciprocal. When h filter (x,y) has a narrow profile, i.e. a small σ, H filter (u,v) has a broad profile, which implies a high cut-off frequency. On the contrary, as the standard deviation, σ, increases, the profile of h filter (x,y) becomes broader, which involves more pixels in the neighborhood when performing filtration. Meanwhile, the profile of H filter (u,v) becomes narrower and thus lowers the cut-off frequency Top-hat transformation In HRTEM imaging for diesel soot, uneven illumination across the image is common due to the inappropriate operation or the overlapping of multiple particles. Thereby, the correction of image illumination was usually performed before the TEM image analysis [33, 45, 60, 62]. The uneven illumination can be corrected by subdividing an image into multiple regions, followed by thresholding each region with one distinct threshold value [100]. The multiple thresholding based on subdivided regions is straightforward, but the implementation needs multistep processing and requires more computation resources. In this study, we applied top-hat transformation [33, 60, 100], a single step transformation, to correct the uneven illumination. The transformation is expressed as: I top hat Iinput ( Iinput SE) (Equation 4-13) where I input is the input image, is the operator of the morphological opening, SE is the structural element that is used to open I input, and I top-hat is the image after performing the top-hat transformation. In this study, a disk was used with a diameter assigned by the program user. The morphological opening [100] in the grayscale image suppresses the bright details (i.e., pixels with

76 61 a higher intensity) smaller than the structural element. Thereby, after subtracting the opened image from the input image, the output image has a reasonably even background. It should be noted that the diameter of the SE should be large enough so that it does not fit entirely within any graphene layers [100] Thresholding to obtain binary images Thresholding an image is needed in order to extract fringe area from the background [32, 45, 59, 60]. The thresholding function of an input image is expressed as: 1 I binary ( x, y) if 0 I I input input ( x, y) t ( x, y) t (Equation 4-14) Equation 4-14 is performed on a pixel-by-pixel basis, where I input (x,y) is the intensity of the pixel to be converted, and t is the intensity threshold value. When the intensity of a pixel is greater than the threshold value, it is converted to one (white). Otherwise, the pixel is converted to zero (black). Although the implementation of the thresholding function (Equation 4-14) is straightforward, choosing an appropriate threshold value given an input image is a complicated issue, which has only been discussed by a few researchers [32]. In general, larger (brighter) threshold values result in an image with thinner, shorter, and fewer fringes. As the threshold value decreases (becomes darker), the fringes become longer and thicker. Meanwhile, the number of identifiable fringes increases. As the threshold value decreases further, the fringes become more indistinguishable because the fringes begin to merge. A number of workers [58, 59] used a single threshold value for all images but have supplementary algorithms to update those area with merged or broken fringes. In this work, we applied the threshold selection method proposed by Otsu [100, 101] to the selected ROI of TEM images of carbon. Otsu s method reduces the threshold selection

77 62 process into an optimization problem to search for a threshold that maximizes the object function 2 that describes the goodness of the threshold at an intensity level [101]. The object function, σ B at an intensity level, k, is expressed as: 2 B 2 1( 1 T ) 2( 2 T 2 ) (Equation 4-15) where ω 1 is the probability of the first class (pixels with intensity less than k), ω 2 is the probability of the second class (pixels with intensity greater than k), μ 1 is the mean intensity level of the first class, μ 2 is the mean intensity level of the second class, and μ T is the total mean intensity level of the ROI. Evaluating Equation (4-15) is a relatively inexpensive computational process because only the gray-level histogram is required. In the program, a reference threshold value is generated based on Otsu s method, while the user can choose to adjust the value, if necessary, according to the condition of the ROI and the thresholding result Morphological opening and closing of fringes A number of workers [58, 59] developed algorithms to repair aggregate fringes after performing skeletonization (discussed in the following section). The procedures include two processes: (1) disconnecting joints with three connected neighbors, i.e., T and Y shape links, and (2) reconnecting the disconnected fringe branches that have a similar orientation. The second process is also a strategy for repairing fringes that have been bisected due to filtering or thresholding processes [45]. The two processes require the operation on a geometrical, pixel-bypixel base and are both expensive computational processes. In this study, morphological opening and closing [100] were used as a substitute for the pixel-by-pixel fringe repair algorithm. The morphological opening and closing of the input image, I input, by a structural element, SE, are defined as:

78 I I SE ( I SE) SE (Equation 4-16a) open input input I I SE ( I SE) SE (Equation 4-16b) close input input 63 where is the morphological erosion, and is the morphological dilation. In words, the morphological opening of an input image, I input, by an elemental structure, SE, is the erosion of I input followed by a dilation of the result. The closing of an input image, I input, by an elemental structure, SE, is the dilation of I input followed by an erosion of the result. Opening tends to smooth an image by breaking narrow joints and removing thin protrusions. As a complementary operation of opening, closing tends to smooth an image by fusing narrow breaks and eliminating small gaps and holes [100] Clearing fringes on the ROI border Removing fringes that are connected to the border of an image or selected ROI has not been discussed in the literature [32, 45, 58-60, 62]. Although the impacts of the incomplete fringes to the overall fringe analysis result may not be significant for all images and ROI, an algorithm was developed such that only complete fringes in the ROI remain for further analysis. The algorithm is expressed as: I output I I (Equation 4-17) input border In Equation (4-17), I input is the image with all fringes in the binary images, I border is the extraction of fringes that touch the border, I output is the image with only complete fringes in the ROI. I border is obtained by morphological reconstruction after identifying the pixels of the fringes that touch the ROI border [100].

79 Skeletonization The skeleton of each fringe is extracted for further analysis. In pattern recognition, the term skeleton has been used to denote a representation of the structural shape of a plane region by collecting thin arcs and curves [100, 102]. The skeleton of a plane region can be obtained via a thinning (skelentonizing) process. Depending on the choice of the thinning algorithm, the obtained skeleton of the same region may be different. In this study, we use a built-in function in MATLAB that implemented the parallel thinning algorithm [102]. The algorithm removes pixels from each fringe based on a pixel-deletion criterion iteratively until the fringe stops changing Remove short length fringes Discarding fringes whose length is less than a minimum length has been included in some TEM image analysis algorithms of carbon [33, 45, 59, 60]. The purpose is to separate the clearly recognizable fringes from smaller objects that do not have sufficient features to distinguish them from noisy or noncrystalline structures [59]. In the literature, the choice of the minimum fringe length varies. Galvez and coworkers [60] eliminated fringes that are shorter than nm because that is the size of a single aromatic ring. Shim et al. [59] chose 1.5 nm as the minimum length based on calibration procedures and the crystallite sizes that were measured by X-ray diffraction. Vander Wal et al. used 0.4 nm as the minimum length [5, 45, 56]. The choice of the minimum length value depends on the working definition of graphene layers that the user wants to analyze [59]. After considering and trying possible values listed in the literature [5, 45, 56, 59, 60], a value of 0.5 nm was chosen as the threshold value for all images in this study.

80 65 The result of removing the short fringes will be used as the input for further quantitative analysis. In addition to all the processes that were discussed, a program loop permits the user to calibrate the image processing parameters until the result satisfies the program user. Before the program stops, the calibrated processing parameters, results of intermediate steps, and the image for characteristic analysis are all saved. The data are retrieved when running the fringe characterization program Fringe characterization The skeletons of lattice fringes are further characterized in order to describe the carbon nanostructure. In this study, three representative parameters are extracted: fringe length, tortuosity, and fringe separation Fringe length Lattice fringe length is a measure of the physical extent of the atomic carbon layer planes as seen in the HRTEM image. The length reflects the dimension of the basal plane diameter, which can be characterized by X-ray diffraction pattern [20, 24, 36]. When the image analysis shows larger fringe lengths, the material has fewer crystallites or grain boundaries. And, Belekov et al. reported that the carbon-carbon bond distance approaches that in graphite as the crystallite size increases [90]. Thereby, the carbon material having larger fringe lengths is highly ordered, which resembles graphite. The length of the fringe (skeleton) in a digitized image is calculated by summing the distance between one point and the next point cumulatively. The distance between two pixels is either 1 pixel or 2 pixels, as illustrated by Figure 4-7 (b), which is an approximation of the scheme in the continuous domain (Figure 4-7(a)). The approximate fringe

81 66 length is affected by three factors: (1) image processing parameters, such as the filter coefficients, contrast improvement strategy, and binarization threshold value, (2) the skeletonization algorithm, and (3) spatial resolution. The image processing parameters and skeletonization algorithm have been discussed in the previous sections. Spatial resolution is a measure of the smallest distance that discerns detail in a digital image. For example, for a 2048*2048 pixel TEM image obtained by a JEOL 2010F, the spatial resolution is nm per pixel in an image with magnification of , while the spatial resolution is nm per pixel in an image with magnification of The difference in spatial resolution affects the interpretation of the curvature: small curvature is not recognizable when the spatial resolution is too low. Thereby, the spatial resolution thus affects the apparent fringe length derived from the digital image Fringe tortuosity Tortuosity is a measure of the curvature of the fringes. It reflects the extent of odd numbered 5- and 7-membered carbon rings with the material [56, 103, 104]. Tortuosity may indicate disorder within the material. Tortuous fringes can severely distort graphitic nanostructure by preventing the stacking of layers, and increase the fringe separation distance [56]. The tortuosity of a fringe is defined here as the ratio of the fringe length to the distance between the two endpoints. The scheme is shown in Figure 4-7. As with fringe length, tortuosity is affected by the image processing parameters, the skeletonizing algorithm, and the spatial resolution.

82 67 Fringe length End point distance End point distance (a) (b) Figure 4-7 Calculation of fringe length and tortuosity: (a) scheme in the continuous domain, and (b) approximation in the digitized image domain Fringe separation distance Lattice fringe separation is the mean distance between adjacent carbon layer planes. Although many workers have measured interlayer spacing (d 002 spacing) from TEM images [32, 58, 60], few elaborated on the algorithm for calculating this quantity. Palotas et al. [32] used the center of area and the orientation to calculate the distance between two adjacent fringes. To avoid errors in artificial pairing of adjacent fringes, the program in the present work permits the user to select each pair of fringes manually. The fringes can be trimmed in order to have an actual pair of fringes with similar lengths. The program first determines the shorter fringe as the reference fringe, and searches the closest distance between the fringe pair on a pixel-bypixel base. Figure 4-8 (a) illustrates the idea in the continuous domain, while Figure 4-8 (b) shows its implementation in a digital image. The dashed blue box in Figure 4-8 (a) illustrates the trimming function given the exemplary fringe pair. The closest distance from a point on the

83 68 reference fringe to the other fringe is perpendicular to the tangent line that passes the point on the second fringe. Although the two fringes are not geometrically parallel, averaging the closest distance from the reference fringe to the second fringe gives rise to the spacing level. In the digitized domain as shown in Figure 4-8 (b), the algorithm can be expressed as: d N min d, y ( x j, y j ),( x j, y j ) F xi i 2 i ave 1 N (Equation 4-18a) d 2 2 x, y ( x j, y j ) ( xi x j ) ( yi y j ) i i (Equation 4-18b) Equation (4-18b) defines the distance from a given pixel (x i, y i ) on the reference fringe to any pixel (x j, y j ) on the second fringe of a trimmed fringe pair. Equation (4-18a) computes the average of the minimum distance from the reference fringe to the second fringe in a fringe pair on a pixel-by-pixel basis, where N is the number of pixels in the reference fringe, and F 2 is the set of pixels that composes the second fringe. Reference fringe (x i,y i ) (a) 90 Second fringe (x j,y j ) (b) Figure 4-8 Calculation of fringe separation distance: (a) scheme in the continuous domain, and (b) approximation in the digitized image domain Demonstration of image processing and analysis algorithm The image processing and analysis algorithm was demonstrated using a negative HRTEM image of soot in Figure 4-9. The region of interest is the polygon enclosed by the white line segments. The soot was derived from ethanol, produced at 1650 C using 0.1 slpm total flow

84 69 rates. The TEM image was taken using a Phillips CM200 with Gatan image filter for digital imaging with live Fourier transforms with nominal resolution of 0.5 nm. Further TEM instrumental details can be found in Ref. [10]. The graphene layers (fringes) extracted from Figure 4-9 is shown in Figure The processing parameters demonstrated here are summarized in Table 4-2. These fringes have been screened such that all fringes have lengths above 0.5 nm. The number of fringes before and after the length screen operation is 1367 and 369, respectively. The fringe length, tortuosity, and fringe separation distributions for this processed image are shown in Figures 4-11(a), 4-11(b) and 4-11(c) respectively. The group spaces of fringe length (Figure 4-11(a)), tortuosity (Figure 4-11(b)), and fringe separation (Figure 4-11(c)) are 0.1 nm, 0.02, and 0.01nm, respectively. The values of the group spaces were chosen so as to observe the distribution shape. Notably the length distributions do not show a symmetric shape. Therefore, instead of arithmetic mean, the median of the fringe length was used to characterize the fringe length distribution, since it is a parameter that is more sensitive to the distribution tail. Figure 4-9 An HRTEM image analyzed for demonstration purpose.

85 70 Figure 4-10 The graphene layers extracted from Figure 4-9. Table 4-2 Summary of processing parameters for Figure 4-9. Processing parameter Value Contrast enhancement Histogram equalization Gaussian filter size 11 (pixels) Gaussian filter deviation 1.0 (1/pixel) Top-hat transformation disk shape with radius = 5 pixel structural element Binarization threshold value Morphological opening 2*2 pixel square structural element Morphological closing 2*2 pixel square structural element Fringe length screening 0.5 nm threshold Spatial resolution nm/pixel

86 % of fringes % of fringes % of fringes (a) 20 (b) 20 (c) fringe length (nm) fringe tortuosity (no unit) fringe separation (nm) Figure 4-11 Fringe analysis results of Figure 4-9 using parameters listed in Table 4-2: (a) fringe length histogram (median: 1.05 nm), (b) fringe tortuosity histogram (mean: 1.16), and (c) fringe separation histogram (mean: nm) Repeatability of HRTEM image analysis results In order to examine the repeatability of HRTEM image analysis algorithm, fifteen HRTEM images obtained from the soot generated by BP15 using single injection strategy with matched combustion phasing were analyzed as shown in Appendix D. The standard deviations of the median fringe length and mean fringe tortuosity are nm and 0.021, respectively. The standard deviations are used throughout Chapter 5. When the differences of the median fringe length and mean fringe tortuosity are smaller than the standard deviations, the differences are considered insignificant.

87 Chapter 5 RESULTS AND DISCUSSION 5.1 Impact of engine operating conditions and the start of injection on soot reactivity The impact of engine operating condition on soot reactivity The PM samples were collected at the four different engine conditions using split injection with BP15 fueling. The four engine modes with their injection parameters were summarized as Table 3-4. These samples were analyzed by TGA as described in Section The normalized mass traces during the oxidation tests were shown in Figure 5-1. The derived apparent rate constant for soot oxidation for the first 30 minutes, and the volatile organic fractions of the four samples are summarized in Table 5-1. The combined effect of the engine torque and speed results in difference in soot reactivity, in an increasing order of Mode B<Mode A<Mode D<Mode C. The greatest rate constant (7.66*10-7 1/Pa/min, Mode C) is 1.5 times the smallest rate constant (4.77*10-7 1/Pa/min, Mode B). Besides, the impact of engine speed at constant torque (Mode A vs. Mode C, Mode B vs. Mode D) is more pronounced than the impact of engine torque (equivalence ratio) at constant engine speed (Mode A vs. Mode B, Mode C vs. Mode D). Table 5-1 shows that higher engine speed is related to more reactive soot and higher VOF contents. Although the diffusion combustion phase is similar (Figure 3-1), the preliminary analysis of soot collected at Mode A-D has confirmed that it is necessary to collect samples at the same engine operating mode when investigating the impacts of fuels on soot. The impact of engine operating modes will be more significant when using single injection strategy, which results in more

88 diverse diffusion combustion phasing as shown in Figure 3-1. The VOF content has the similar trend as the rate constant, which will be summarized further with other samples in a later section. 73 Figure 5-1 Normalized mass obtained from the isothermal oxidation test at 550 C of soot samples generated at four engine modes. ( ) Mode A: 1850 rpm, 64 Nm, ( ) Mode B: 1850 rpm, 110 Nm, ( ) Mode C: 2400 rpm, 64 Nm, and ( ) Mode D: 2400 rpm, 110 Nm. Table 5-1 Analysis of VOF content of four engine operating modes using split injection with BP15 fuelling Engine operating mode Equivalence ratio ( ) VOF content (%) Apparent rate constant (1/Pa/min) A (1850 rpm, 64 Nm) *10-7 B (1850 rpm, 110 Nm) *10-7 C (2400 rpm, 64 Nm) *10-7 D (2400 rpm, 110 Nm) *10-7 * The uncertainty for the apparent rate constant is ±4.4% at 95% confidence (Appendix B)

89 The impact of the start of combustion on soot reactivity and nanostructure In order to investigate the impact of the start of combustion on soot reactivity and nanostructure, PM samples generated by BP15 at Mode C with single injection strategy using various start of injection (SOI) timings were analyzed by TGA and HRTEM. The engine operation mode, the three SOIs, and fuel pressure were summarized in Table 3-7. The thermodynamic analysis results were shown in Figure 3-2, where the combustion phases are consistent with the SOIs. These samples were analyzed by TGA as described in Section The normalized mass curves were shown in Figure 5-2. The VOF contents and the derived rate constants for the first 30 minutes were listed in Table 5-2. The rate constant of the sample for retarding SOI timing (retard 2 CAD) is 2.25 times that for advancing SOI timing (advanced 2 CAD). The difference in reactivity can be contributed by the different combustion phasing characterized by the rate of heat release and the mean temperature profile. A difference in combustion phase implies that the soot undergoes a different formation and oxidation history [105]. In particular, as shown in Figure 3-2, the soot generated at advanced SOI timing experienced longer period at high temperature in engine cylinder than the soot generated at retarded SOI timing. Therein, the soot generated at advanced SOI timing may oxidize more and form higher crystalline order, thus results in a low reactivity [73].

90 75 Figure 5-2 Normalized mass obtained from the isothermal oxidation test at 550 C of soot samples generated at three SOIs at Mode C with BP15 fuelling. ( ) baseline: 4.83 BTDC, ( ) advanced 2 CAD: 6.83 BTDC, and ( ) retard 2 CAD: 2.83 BTDC. Table 5-2 Analysis of VOF content of the three SOIs using single injection at Mode C with BP15 fuelling Start of injection VOF content (%) Apparent rate constant (1/Pa/min) Advanced 2 CAD *10-7 Baseline *10-7 Retard 2 CAD *10-6 The surface oxygen contents of the soots from advanced and retarded SOI timings were both around 5.5%, according to the survey scans of XPS. Therein, the surface oxygen content cannot explain the increased oxidative reactivity for soot from retarded SOI timing. To further study the impact of the SOI timing and consequently, the start of combustion on soot, the PM samples were analyzed by TEM at magnifications of 30,000X and 500,000X. To investigate the morphology, the samples were collected onto the TEM grid with a thermophoretic sampling

91 76 device [6] at Mode C with advanced and retarded SOI timings. Figure 5-3 shows three representative TEM images for each sample. The particles from advanced SOI timing tend to have grapelike structures [13]. On the contrary, the particles in the PM samples generated at retarded SOI timing appear as stretched chain-like shapes, as shown in Figure 5-3 (d)-(f). The soot with chain-like shapes has a higher fractal dimension [12]. The active surface areas of the soots from the advanced and retarded SOI timings are 12.2±1.2 and 29.4±3.3 m 2 /g, respectively. Although somewhat speculative, the higher fractal dimension may indicate a higher surface area and result in a higher oxidation rate constant for soot oxidation.

92 77 (a) Single Injection: advanced 2 CAD (d) Single Injection: retarded 2 CAD (b) Single Injection: advanced 2 CAD (e) Single Injection: retarded 2 CAD

93 78 (c) Single Injection: advanced 2 CAD (f) Single Injection: retarded 2 CAD Figure 5-3 TEM images of soot generated at Mode C with single injection by advanced and retarded SOI: (a)-(c) advanced 2 CAD, and (d)-(f) retarded 2 CAD. Two representative HRTEM images at a magnification of 500,000X are shown in Figure 5-4. As described in Section and 3.6.4, the soot samples were deposited directly onto Teflo filters using a vacuum pump, pretreated to remove VOF, and sonicated in methanol for one hour. The HRTEM image of the soot sample generated at advanced SOI timing has a higher degree of order (Figure 5-4 (a) and (b)). The soot sample generated at retarded SOI timing (Figure 5-4(c) and (d)), however, shows signs of disorder configuration. The observation was further confirmed by the fringe length and tortuosity analysis shown in Figure 5-5. Comparing with the soot of advanced SOI (Figure 5-5 (a) and (b)), the soot of retarded SOI timing (Figure 5-5 (c) and (d) has narrower distribution of fringe length, and a wider distribution of fringe tortuosity. These are evidence that retarded SOI timing produced soot that is less ordered, which results in a high reactivity [5].

94 79 (a) (c) (b) Figure 5-4 Representative HRTEM images of soot generated at Mode C with single injection by advanced and retarded SOI: (a) image of soot generated at SOI with advanced 2 CAD, (b) ROI and extracted skeletons of (a), (c) image of soot generated at SOI with retarded 2 CAD, and (d) ROI and extracted skeletons of (c). (d)

95 percentage (%) percentage (%) percentage (%) percentage (%) (a) 25 (c) 20 median: 0.91 nm 20 median: 0.76 nm fringe length (nm) (b) fringe length (nm) (d) 16 mean: mean: tortuosity (no unit) tortuosity (no unit) Figure 5-5 Fringe length and tortuosity analysis of soot generated at Mode C with single injection by advanced and retarded SOI timings: (a) fringe length extracted from Figure 5-4(b), (b) fringe tortuosity extracted from Figure 5-4 (b), (c) fringe length extracted from Figure 5-4 (d), and (d) fringe tortuosity extracted from Figure 5-4 (d). The preliminary results shown here confirm that combustion phasing can affect the soot nanostructure and reactivity. Under the fixed SOI condition, as shown in Figure 3-3(a) and Figure 3-4(a), where ignition delay is coupled with fuel formulation, the characteristics of soot are a combined result of different combustion phasing and other fuel properties, such as fuel volatility,

96 81 density, and formulation. To study soots that undergoes a similar formation process, the SOI and injection pressure were adjusted to obtain a similar combustion phasing for the different test fuels. The parameters were summarized in Table 3-8 and Table 3-9 in Chapter Impact of fuels on soot reactivity The PM samples generated by BP15, FT and B100 fuels were collected at Mode C with both single and split injection strategy using the injection parameters listed in Table 3-8 and Table 3-9. The start of injection (SOI) and fuel rail pressures were adjusted such that the three test fuels had similar combustion phasing. The thermodynamic analysis results were shown and discussed in Section 3.4. These samples were analyzed by TGA as described in Section Figure 5-6 shows the normalized weight versus time curves of the three soot samples using single and split injection strategies.

97 82 (a) (b) Figure 5-6 Normalized weight vs. oxidation time curves of soots generated at Mode C by BP15, B100, and FT with matched combustion phasing: (a) single injection and (b) split injection. ( ) BP15 fuel, ( ) B100 fuel, ( ) FT fuel. The VOF contents and the apparent rate constants for soot oxidation for the first 30 minutes are summarized in Table 5-3. The calculation of VOF was discussed in Section 3.6.1, while the derivation of the apparent rate constant for soot oxidation was discussed in Section 4.1.

98 83 Compared with BP15 and FT samples (both single injection and split injection), the B100 soot has more derived VOF content, as expected [78]. According to Table 5-3, B100 soot exhibits the fastest oxidation rate on a mass basis with BP15 and FT soot following, in order of oxidation rate. The calculation shows that B100 soot has two times higher apparent rate constant than FT soot at 550 C when using single injection. When using split injection, although not as significantly as for single injection, B100 still has the highest oxidation rate constant among the three soot samples. Table 5-3 The derived VOF content and apparent rate constants for soot oxidation of BP15, B100, and FT soot generated at Mode C with matched combustion phasing using single and split injection strategy. Sample Derived VOF content (%) Apparent rate constant for soot oxidation (1/Pa/min) BP15 (single injection) *10-7 B100 (single injection) *10-6 FT (single injection) *10-7 BP15 (split injection) *10-7 B100 (split injection) *10-7 FT (split injection) *10-7 Figure 5-7 summarizes the derived VOF content versus the soot reactivity using the data in Table 5-1 and Table 5-2. A relationship between the VOF (adsorbed hydrocarbons) and soot reactivity is observable. However, because the VOF does not participate in the soot isothermal oxidation test in this study (Appendix A), the VOF content was not claimed as an effective index representing soot reactivity. A possible explanation for the results is that thermal treatment under inert atmosphere generally is accompanied by a rapid initial rise in pore area, which indicates the

99 Derived VOF content (%) blocked pores are being opened up to the oxidizer [4, 44, 106]. The size of the opened-up pore area may be related to the amount of VOF. 84 In order to explain the factors that contribute to the difference in soot reactivity, the soots were further analyzed for surface oxygen content and the structural order. 35 y = x R= Apparent rate constant (10-7 /Pa/min) Figure 5-7 The derived VOF content versus the soot reactivity using the data in Table 5-1~ Table The initial surface oxygen-contained functional groups and reactivity Past work by Song [6] identified surface oxygen content as the primary explanation for increase oxidative reactivity for B100. The elemental content on the surface of the soot sample is summarized in Table 5-3. The apparent rate constants for soot oxidation derived from TGA are

100 also included. As shown in Appendix C, the preliminary study of the possible errors in the determination of elemental content by XPS is 3%. 85 The surface oxygen contents in atomic percentage and the apparent rate constants in Table 5-4 are plotted as Figure 5-8. No clear trend is concluded from the result. Therefore, the impact of the abundance of the surface oxygen contents, and thus the oxygen containing functional groups, on the soot reactivity was not confirmed in this study. Table 5-4 Elemental analysis of soot surface using XPS Sample Oxygen content Carbon content Apparent rate constant for soot (atom %) (atom %) oxidation (1/Pa/min) BP15 (single injection) *10-7 B100 (single injection) *10-6 FT (single injection) *10-7 BP15 (split injection) *10-7 B100 (split injection) *10-7 FT (split injection) *10-7

101 Oxygen atomic % BP15 (split inj) 4 FT (split inj) FT (single inj) BP15 (single inj) B100 (split inj) B100 (single inj) Apparent rate constant (10-7 1/Pa/min) Figure 5-8 Surface oxygen content vs. apparent rate constant for soot oxidation. The effect of the pretreatment on the oxygen-containing functional groups was investigated by FTIR and XPS. Figure 5-9 shows the transmission FTIR spectra of the soot without thermal pretreatment ( soot as received ) and the soot after thermal pretreatment in nitrogen at 500 C for 60 minutes. In the spectra of soot as received, aliphatic groups (2860~2950cm -1, 1450cm -1 ), aromatic groups (1600~1640 cm -1 ), and oxygen-containing functional groups are present [6, 91]. In the spectra of the soot after thermal pretreatment in nitrogen, the peaks for aliphatic groups and oxygen-containing functional groups disappeared, indicating an effect of thermal pretreatment.

102 87 Figure 5-9 Transmission FTIR spectra of ( ) as received soot, and ( ) thermally pretreated soot, qualitatively indicating some functional groups was removed during the pretreatment process. (Peak assignment) 1450cm -1 : carbon-hydroxyl bond in aliphatic groups; 1600~1640 cm - 1 : aromatic groups; 1710~1750 cm -1 : lactone, carboxyl and ketone acid groups, 2860~2950cm -1 : aliphatic groups. The reduction of surface oxygen-containing functional groups was confirmed by XPS. Figure 5-10 shows the XPS high resolution scan of carbon (C1s) and oxygen (O1s) of the as received soot and thermally pretreated soot. The soot was generated by BP15 using split injection with matched combustion phasing at Mode C. The high-resolution scan of the C1s spectra of the as received soot reveal the presence of sp2 and sp3 non-functionalized carbon of the polyaromatic ring structures ( ev), carbon present in alcohol or ether groups ( ev), and carboxyl or ester groups ( ev) [107, 108]. However, the C1s spectra of the thermally pretreated soot using nitrogen only reveals the non-functionalized carbon ( ev). The peaks related to the oxygen-containing functional groups were not present in the C1s spectra of the thermally pretreated soot, indicating the effect of pretreatment process.

103 88 Compared with soot as received, the O1s spectrum of thermally pretreated soot shows less significant peaks. These peaks correspond to oxygen bonded to carbon by a double bond (~532 ev) or oxygen in C-O-C or C-O-H groups ( eV) [107, 108]. Qualitatively, the XPS result shown in Figure 5-10 is consistent with trend observed through the FTIR result shown in Figure 5-9. The oxygen atomic percentage derived from the survey scan of as received soot and thermally pretreated soot are 16% and 6.3%, respectively. Figure 5-10 High resolution scan of ( ) as received soot, and ( ) thermally pretreated soot. The impact of the reduction of oxygen-containing functional groups caused by thermal pretreatment on soot reactivity was further investigated by the following TGA tests: (1)

104 89 investigating isothermal soot reactivity with different soot pretreatment temperature in nitrogen; and (2) performing non-isothermal oxidation on as received soot and thermally pretreated soot. The experimental details are included in Appendix A. PM samples collected at Mode C with split injection strategy were used for the two tests. Figure 5-11 shows the normalized weight profiles from isothermal oxidation. The apparent rate constants for soot oxidation of the curves in Figure 5-11 are summarized in Table 5-5. The difference in apparent rate constants caused by the pretreatment temperature is less than 2%. The isothermal oxidation tests indicate that the reactivity of the soot particles was not altered by the pretreatment process even the surface oxygen-containing functional groups reduces during the pretreatment process as illustrated by Figure 5-9 and Figure The result corroborates the second subordinate hypothesis that surface oxygen content does not dominate the soot oxidative reactivity. Additionally, the result implies that the soot nanostructure, which affects the soot reactivity, was not altered by the pretreatment temperature, 500 C, in this study. Figure 5-11 Mass fraction profiles of isothermal oxidation test of PM samples pretreated at ( )100 C, ( ) 350 C, ( ) 500 C, and ( ) 650 C for 60 minutes under ultra high purity nitrogen.

105 90 Table 5-5 Comparison of apparent rate constant for soot oxidation of the PM pretreated at four different temperatures Pretreatment temperature ( C) Apparent rate constant (1/Pa/min) As shown in Table 5-3, the PM mass contains as much as 30% volatile organic fraction (VOF) (B100, single injection). One may suspect that the VOF and the pore area covered by VOF plays a significant role when measuring the oxidation reactivity. To clarify this issue, nonisothermal oxidation tests were performed for the as received PM sample and the sample pretreated at 500 C. The programmed temperature and mass fraction curves are shown in Figure During the test, the zero air was injected from the start of the test in order to examine the effect of PM on the oxidation process. In Figure 5-12, the mass fraction of the thermally pretreated PM did not change until the temperature reached 400 C. On the contrary, the mass fraction of the as received PM reduced from the beginning of the test until the soot burned out. For both the as received PM and thermally pretreated PM, the mass fraction reduced to 50% at 600 C. The curves beyond 50% of mass fraction are very similar for both samples. To further study the impact of porous area covered by VOF, in Figure 5-13, the mass curve of the as received PM was normalized based on the mass at 400 C, the estimated temperature for initial soot oxidation using the mass fraction of the thermally pretreated PM in Figure Except for the mass loss due to the continuous removal of VOF in as received PM,

106 91 the two mass fraction curves in Figure 5-13 are similar, confirming the lack of impact of pretreatment process (or adsorbed hydrocarbons) on soot reactivity. Figure 5-13 also coincides with the results of Yezerets et al., who claim that adsorbed hydrocarbons provided limited contribution to the overall reactivity of the un-pretreated particulate matter [35]. Figure 5-12 The temperature and normalized weight profile of non-isothermal oxidation test. ( ) as received PM, ( ) PM pretreated at 500 C, ( ) Temperature ( C).

107 92 Figure 5-13 The temperature and mass fraction profile of non-isothermal oxidation test. The mass fraction is normalized by the mass measured at 37.5 minutes of Figure ( ) As received PM normalized by weight at 400 C, ( ) PM pretreated at 500 C, ( ) Temperature ( C). 5.4 Relation between the soot nanostructure and reactivity X-ray diffraction analysis The thermally pretreated PM samples of three test fuels were analyzed by XRD. The data analysis method for extracting the crystalline parameters (d 002, L a, L c, N) is described in Section and 4.2. Table 5-5 summarized these crystalline parameters obtained from different test dates and the rate constant of each sample. Figure 5-14 plots L a versus the rate constant in Table 5-6, indicating an inverse relation. The results show that the trend of an inverse relation of L a and the apparent rate constant is repeatable for different test dates. In addition, Figure 5-14 indicates that the difference between derived apparent rate constants depends on the difference of L a. For example, the three samples (FT, BP15, B100) generated by the single injection strategy have

108 93 larger discrepancy in L a compared to those generated by the split injection strategy. The larger discrepancy in L a coincides with the more discernible difference in the apparent rate constants (Figure 5-14, single injection). Table 5-6 Crystalline parameters derived from XRD patterns k c,, Apparent rate XRD test date Sample d 002 (nm) L c (nm) L a (nm) N constant (10-7 /Pa/min) FT (single injection) /12/09 BP15 (single injection) B100 (single injection) FT (single injection) /18/09 BP15 (single injection) B100 (single injection) FT (split injection) /05/09 BP15 (split injection) B100 (split injection) FT (split injection) /12/09 BP15 (split injection) B100 (split injection)

109 L a obtained from XRD (nm) 94 3 FT FT BP BP15 B100 2 B Apparent rate constant (10-7 /Pa/min) Figure 5-14 The basal plane diameter, L a, versus the apparent rate constant for soot oxidation, indicating an inverse relation. single injection (02/12/09), single injection (02/18/09), split injection (03/05/09), and Δ split injection (03/12/09). In addition to L a, compared with BP15 and FT soot for both single and split injection strategies in Table 5-6, B100 soot has the smallest crystallite height (L c ) and smallest average number of layers per crystallite (N), indicating a less ordered nanostructure [109]. Although the differences were small, the results were repeatable from the XRD analysis on different test dates. Moreover, the small N of B100 soot is consistent with its high tortuosity, which will be discussed in section The crystalline parameters for the bulk samples, L a and N, derived from XRD results are consistent with the subordinate hypothesis stating that the relation between soot nanostructure and

110 reactivity (Section 2.6). The hypothesis was further tested by examination of the graphene layer (carbon fringe) configuration derived from the HRTEM image analysis algorithm High-resolution transmission electron microscopy (HRTEM) analysis The graphene layers in the primary soot particles were studied by HRTEM image analysis algorithm as described in Sections and 4.3. Figure 5-15 shows three representative HRTEM images and their graphene layer (fringe) images of the soot samples generated by FT, BP15, and B100 fuels using the single injection strategy with the selected regions of interest (ROI). The differences in fringe length and tortuosity of the graphene segments for FT soot compared to B100 soot are readily apparent upon visual comparison of the fringes extraction images. A qualitative observation shows that FT soot has longer fringes, while B100 has more tortuous fringes (Figure 5-15(d) and (f)).

111 96 (a) FT, single injection (b) BP15, single injection (c) B100, single injection (d) Fringes extracted from (a) (e) Fringes extracted from (b) (f) Fringes extracted from (c) Figure 5-15 Three representative HRTEM images of soot samples generated by three test fuels using single injection strategy at matched combustion phasing: (a) BP15, (b) FT, (c) B100 with marked ROI, and (d)~(f) show fringes extracted from images of (a)~(c). Figure 5-16 shows the fringe length histograms resulting from the HRTEM image analysis of Figure As seen in Figure 5-16, the fringe length histogram for the FT soot extends to much longer fringe length, indicating a larger graphene layer dimensions. Relative to the fringe length histogram for the B100 soot with 85% of the lamella smaller than 1 nm in length, only 65% of the fringe length histogram of FT soot is less than 1 nm. Alternative only 1% of the B100 soot fringe length histogram is greater than 2 nm while 10% of the lamella in the FT soot is greater than 2nm. The median of the fringe length histograms shows an order of FT (0.90nm) > BP15 (0.79nm) > B100 (0.72nm). The uncertainty, based on the results in Appendix D, is ± 0.03 nm.

112 % of fringes % of fringes % of fringes fringe length (nm) (a) fringe length (nm) (b) fringe length (nm) Figure 5-16 Fringe length and median values derived from the extracted fringes in Figure 5-15: (a) FT soot (median: 0.90nm), (b) BP15 soot (median: 0.79nm), and (c) B100 soot (median: 0.72nm). (c) Figure 5-17 shows the fringe tortuosity histograms obtained from the extracted fringes of FT, BP15 and B100 soot. As expected from a visual comparison, B100 soot contains a high degree of tortuosity among the lamella. Relative to the distribution for the FT and BP15 soot with 1.7% and 3.2% of the lamella greater than 1.5 in tortuosity ratio, more than 14% of the B100 soot fringe tortuosity histogram is greater than 1.5. Alternatively the BP15 and FT soot contain fringes that have a lower level of tortuosity, with 91% and 85% of the measured fringes having a tortuosity ratio of less than 1.2. As a comparison, B100 soot only contains 58% of fringes with a fringe tortuosity ratio of less than 1.2. The mean of the tortuosity ratio for B100 soot is 1.37, greater than the mean values, 1.14 and 1.17, for the FT and BP15 soot. The uncertainty, based on the results in Appendix D, is

113 % of fringes % of fringes % of fringes fringe tortuosity (a) fringe tortuosity (b) fringe tortuosity Figure 5-17 Fringe tortuosity histograms and mean values derived from the extracted fringes in Figure 5-15: (a) FT soot (mean: 1.14), (b) BP15 soot (mean: 1.17), and (c) B100 soot (mean: 1.37). (c) The HRTEM images of soot generated with the split injection strategy were analyzed in the same methods. Figure 5-18 shows three representative HRTEM images and their graphene layer extraction images within the ROI of the soot samples generated by FT, BP15 and B100 fuels using split injection strategy at matched combustion phasing. Unlike the case of single injection, for split injection, the differences in fringe length and tortuosity are not so apparent using visual comparison of the fringe extraction images (Figure 5-18 (d) ~ (f)). Figure 5-19 shows the fringe length and fringe tortuosity histograms resulting from the HRTEM image analysis of Figure Visual comparison of the shapes of the fringe length histograms does not indicate any trend. However, the fringe length histogram for the B100 soot has 64% of the lamella smaller than 1 nm in length, while 55% of the fringe length histogram of FT soot is less than 1 nm. Alternatively 7.6% of the B100 soot fringe length histogram is greater than 2 nm while 12% of the lamella in the FT soot are greater than 2 nm. The median of the fringe length histograms shows an order of FT (1.01 nm) > BP15 (0.96nm) > B100 (0.92nm). The uncertainty, based on the results in Appendix D, is ± 0.03 nm. While the trend is the same, the

114 differences in the median fringe length in the case of split injection are not as significant as seen in the case of single injection. 99 (a) FT, split injection (b) BP15, split injection (c) B100, split injection (d) Fringes extracted from (a) (e) Fringes extracted from (b) (f) Fringes extracted from (c) Figure 5-18 Three representative HRTEM images of soot samples generated by three test fuels using split injection strategy at matched combustion phasing: (a) BP15, (b) FT, (c) B100 with marked ROI, and (d)~(f) show fringes extracted from images of (a)~(c).

115 % of fringes % of fringes % of fringes fringe length (nm) (a) fringe length (nm) (b) fringe length (nm) Figure 5-19 Fringe length and median values derived from the extracted fringes in Figure 5-18: (a) FT soot (median: 1.01nm), (b) BP15 soot (median: 0.96nm), and (c) B100 soot (median: 0.92nm). (c) Figure 5-20 shows the fringe tortuosity histograms obtained from the extracted fringes in Figure Visual comparison of the histograms indicates that B100 soot has fringes that contain a wide range of tortuosity, implying high degree of curvature among the lamella of the soot [56]. Relative to the distribution of the FT and BP15 soot with 94% and 83% of the lamella smaller than 1.2 in tortuosity ratio, only 78% of the fringe tortuosity of B100 soot fringe tortuosity histogram is smaller than 1.2. The mean tortuosity ratio of B100 is 1.31, greater than the mean values 1.12 and 1.18, of the FT and BP15 soot, respectively. The uncertainty of tortuosity ratio, 0.021, obtained in Appendix D indicates that the difference in the mean tortuosity ratios between B100 and the other two soot samples is statistically significant.

116 % of fringes % of fringes % of fringes fringe tortuosity (a) fringe tortuosity (b) fringe tortuosity Figure 5-20 Fringe tortuosity histograms and mean values derived from the extracted fringes in Figure 5-18: (a) FT soot (mean: 1.12), (b) BP15 soot (mean: 1.18), and (c) B100 soot (mean: 1.31). (c) Comparison of XRD and HRTEM analysis results In the previous two sections, both the XRD and HRTEM analysis results indicate a relation between soot reactivity and nanostructure. The L a derived from XRD patterns and the median fringe lengths derived from HRTEM images were compared directly. Figure 5-21 shows L a, median fringe length versus apparent rate constant for soot oxidation of BP15, FT and B100 samples generated with single injection strategies at matched combustion phasing. Figure 5-21 indicates a qualitative agreement between the results from XRD and HRTEM analysis. The numerical discrepancy between the L a obtained by XRD pattern analysis and the median fringe length obtained from the HRTEM image analysis method has been observed by Sharma et al. [25]. The crystallite dimension derived from XRD pattern tends to shift toward larger values, because a small quantity of these will increase the peak height [110]. Using Diamond s empirical formula may reduce the discrepancy [25, 110], and is suggested as a future work.

117 L a from XRD (nm) Median fringe length (nm) FT BP B Apparent rate constant (10-7 /Pa/min) Figure 5-21 Comparison of L a derived from XRD patterns and median fringe length derived from HRTEM images (BP15, FT, and B100 with single injection) versus apparent rate constant for soot oxidation. : L a from XRD (nm)(02/12/2009), : L a from XRD (nm)(02/18/2009), and : median fringe length from HRTEM (nm). Figure 5-22 compares the XRD and HRTEM analysis results of the soot samples generated by the three test fuels using the split injection strategy at matched combustion phasing. It should be noted that the scales of the three axes in Figure 5-22 are different from those in Figure 5-21 in order to compare effectively. The median fringe lengths qualitatively agree with the trend of the L a. As indicated in Figure 5-14, the differences in crystalline parameter values (L a and median fringe length) and soot reactivity for the soot samples generated by the split injection strategy are not as significant as those for the soot samples generated by the single injection strategy. In Figure 5-22, the overlap between the error bars of the median fringe lengths indicates

118 L a from XRD (nm) Median fringe length (nm) 103 that characterizing structural differences in the soot samples is less certain when the difference in soot reactivity is less significant. Figure 5-21 and Figure 5-22 shows that L a obtained from XRD patterns and median fringe length obtained from TEM images are both indicators of the ratio of edge (active) carbon atoms to basal (inactive) carbon atoms. The ratio directly affects the soot reactivity but is more difficult to evaluate by measuring active surface area [111, 112] FT BP B Apparent rate constant (10-7 /Pa/min) Figure 5-22 Comparison of L a derived from XRD patterns and median fringe length derived from HRTEM images (BP15, FT, and B100 with split injection) versus apparent rate constant for soot oxidation. : L a from XRD (nm)(03/05/2009), : L a from XRD (nm)(03/12/2009), and : median fringe length from HRTEM (nm). Using the graphene dimension (L a ) and lattice fringe length rather than active surface area, Figure 5-21 and Figure 5-22 also indirectly clarifies that the concentration of carbon active sites dominates the apparent rate constant [46]. The relation can be further investigated by

119 correlating L c (XRD), median fringe length (TEM), and the ratio of the edge carbon atoms to total carbon atoms [20, 90]. 104 Besides Figure 5-21 and Figure 5-22, the highest tortuosity of B100 soot generated by single and split injection is consistent with the smallest crystallite height (Lc) and average number of layers per crystallite (N) derived from XRD pattern. Tortuosity measures the undulation of carbon lamella, arises from 5- and 7-membered ring structures within the aromatic framework. Therein, high tortuosity in soot nanostructure prevents development of stacked layers [56]. Although the variations in L c and N shown in Table 5.6 are small, B100 soot shows a consistent trend of having smaller valves of Lc and N, indicating smaller crystallites with fewer stacked layers. These observations are confirmed by the fringe tortuosity derived from the TEM images Other characterization parameters and methods The hypothesis regarding the relationship between soot nanostructure and soot reactivity has been confirmed by XRD (L a, N) and HRTEM (fringe length and fringe tortuosity). The tortuosity may affect the fringe separation distance [45], however, the d 002 from XRD patterns and fringe separation distances from HRTEM image analysis do not show a relation with the soot reactivity. Both the XRD and HRTEM image analyses yields separation distances between 0.36 ~ 0.38 nm. The results in this study indicate that the differences in soot reactivity are not dominated by the initial separation distances between fringes. In addition, the hypothesis was tested by Raman spectroscopy and oxygen chemisorption tests (TGA), using the procedures described in Section 3.6. The results, attached as Appendix E,

120 do not show a relation between the structural disorder and the soot reactivity. The possible explanations are as follows. 105 Curve fitting of Raman spectra has been widely used to distinguish different carbon materials. Many carbonaceous materials show two characteristic peaks appearing at ~1350 cm -1 (D peak) and ~1580 cm -1 (G peak) for first-order Raman spectra. The G peak is a stretching mode at sp 2 sites. The D peak has been proven to arise from the existence of edge sites and the relative position of the laser spot with respect to the edge [21, 22]. However, most successful demonstrations of using Raman D/G peak ratio to discriminate carbonaceous materials compared inherently different carbons. For example, Tuinstra and Koenig showed the spectral intensity differences among stress annealed pyrolite graphite, a commercial graphite, and activated charcoal [23]. Sadezky et al. demonstrated that the D1 band (1350 cm -1 ) FWHM can distinguish a highly ordered carbon from soot. But, the standard deviation ranged up to 100% for the D 1 /G band intensity ratios. The high statistical variability of D 1 /G does not enable clear discrimination of the different diesel soot samples and various carbon black materials [34]. Casiraghi et al. [21] developed a real space theory for Raman scattering to analyze the general case of disordered edges. Besides amount of edge disorder, polarization, relative position of the laser spot with respect to the edge also affect the D/G peak intensity ratio. For real samples, it is concluded that the D/G peak intensity ratio does not always show a significant dependence on the microscopic edge orientation [21]. In this study, although generated by different test fuels, all the soot samples were generated at a similar combustion phasing. The inconsistency of the D/G rations of Raman spectra analysis and the results in Section coincides with the results mentioned in Ref. [21, 34].

121 106 The active surface area (ASA) measured by oxygen chemisorption depends on the edge versus basal plane carbon atoms [46, 93, 113]. In principle, if the crystallite height does not increase, an increase in crystallite diameter (obtained from XRD or HRTEM analysis) decreases the ratio of edge carbons to basal carbon atoms [46, 90]. In the literature, the ASA values were compared using carbonaceous materials with very different textural and structural properties. Arenillas et al. presented a relationship between the initial reactivity and active surface area. The samples were active carbons treated at several temperatures (1400, 1800, 2000, 2200, 2400 C) for 120 minutes [29, 111]. Laine et al. shows that ASA increased from 0.24% to 2.12% of the BET area when the high-temperature treated carbon was oxidized from 0% to 35% [113]. In this study, however, the active surface areas derived from the oxygen chemisorption of soot samples generated by the three test fuels disagree with the trend of the soot reactivity and the crystallite parameters derived from XRD and HRTEM. The explanation of the discrepancy lies with two possible reasons: (1) the differences in ASA of soot samples in this study are not significant, because the samples were all generated at similar combustion phasing and did not undergo much different heat treatment or oxidation procedures described in Ref. [29, 111, 113]; and/or (2) the adopted experimental method does not yield representative ASA values during soot oxidation. The discussion of these two reasons are given as follows. (1) In order to corroborate the first reason, the ASA were measured for the soot samples that were generated by BP15 using single injection strategy at different combustion phasing (Figure 3-2 and Table 3-7). The results are summarized in Table 5-7. Although no rigorous agreement between ASA and apparent rate constant of the three samples was established, the ASA of the soot of 2 retarded SOI timing is more than two times larger than the ASA of the soot generated by 2 advanced SOI timing (i.e. 4 SOI difference in the engine cycle when generating

122 107 the two samples), coinciding with the clear difference in the apparent rate constant. On the contrary, the ASA difference between the soot of 2 advance and the standard SOI is not significant, a reflection of the close values of apparent rate constants. In addition, the large difference (~ 22% for BP15 generated at standard SOI) between the two tests indicates that ASA should not be used as indices representing reactivity. Table 5-7 Summary of ASA and apparent rate constant for soot oxidation of BP15 soot with different SOI. Sample Apparent rate constant (1/Pa/min) BP15 (single injection, SOI 2 advanced) BP15 (single injection, standard) BP15 (single injection, SOI 2 retarded) 8.58* * *10-6 ASA (m 2 /g) (11/2009) ASA (m 2 /g) (04/2010) (2) Teng and Hsieh conducted oxygen chemisorption measurements at different temperatures to demonstrate that the active sites derived from low-temperature chemisorption cannot be applied to the reaction at higher temperatures [112]. In this study, the ASA was measured after the sample was chemisorbed in zero air for ten hours at 200 C, which was lower than the reaction temperature [114]. But, apparent rate constant for soot oxidation was derived from the isothermal oxidation test at 550 C. The disagreement between the ASA and the apparent rate constants may result from the discrepancy of the oxygen chemisorption temperature and isothermal oxidation temperature. In addition, the ASA values measured by TGA were affected by physisorption [115, 116], which was not eliminated in this study. To further investigate the ASA in future work, the oxygen chemisorption needs to be tested by a different method [111], or to include additional apparatus such as an in vacuo device [117].

123 Comparison of partially oxidized soot generated by FT, BP15, and B100 fuels The initial structural parameters, L a (XRD) and lattice fringe parameters (TEM) of soot nanostructure have been identified as the factors governing soot reactivity. This section continued to investigate whether the structural parameters would affect the later stages during the oxidation process. Although section implies a dependence on active carbon concentration, however, the relation between the concentration of active site and lattice dimensions was not yet clarified in this study. Therein, the same oxidation model (zero-order) was consistently used to derive apparent rate constant to represent the soot reactivity. The goal was to examine whether or not similar relation exists during soot oxidation between the apparent rate constants and lattice fringe length as in Figure 5-21and Figure It should be noted again that the apparent rate constant in this study is not an intrinsic property of soot, and expected to be a function of other material properties Comparison of soot reactivity of soot at initial state and 50% burn-off Soot samples generated by FT, BP15, and B100 fuels using both single and split injection strategies at matched combustion phasing were used to investigate the structural and reactivity changes during soot oxidation. All the soot particles at their 50% burn-off conditions were investigated to compare with the soot particles at their initial states presented in Section 5-1 to Section 5-4. The apparent rate constants for soot oxidation were characterized using the same data as shown in Figure 5-6. The apparent rate constants of the first 30 minutes, and the 30 minutes after the 50% burn-off are derived based on the procedures in Section 4-1, and are summarized in

124 109 Table 5-8. For all soot samples, compared to their initial states, the apparent rate constants at 50% burn-off decreased, indicating a decrease in soot reactivity during the soot oxidation. The percentage decrease in apparent rate constant from initial state to 50% burn-off are calculated and included in Table 5-8. Compared with BP15 and B100 soot for single and split injection strategies, FT soot has the least amount of decrease in soot reactivity during oxidation process. The decrease in soot reactivity during the soot oxidation of BP15 and B100 are greater than 20%. For BP15 soot generated by single injection strategy, the soot reactivity decrease 37% at 50% burn-off. Table 5-8 The apparent rate constant for soot oxidation of the initial soot and after the 50% burnoff, and the percentage of reduction in the apparent rate constant. Sample Initial apparent rate constant (1/Pa/min) Apparent rate constant after 50% burn-off (1/Pa/min) Apparent rate constant decrease in percentage (%) BP15 (single injection) 9.5* * B100 (single injection) 1.3* * FT (single injection) 6.3* * BP15 (split injection) 7.6* * B100 (split injection) 9.1* * FT (split injection) 6.0* * Comparison of surface oxygen content of soot at initial state and 50% burn-off To investigate soot reactivity during soot oxidation, the oxidized soot samples were analyzed by XPS, HRTEM and Raman spectroscopy. Because of sample quantity limitation, the oxidized samples were not analyzed by XRD. Table 5-9 shows the elemental contents on the

125 110 surface of the initial soot samples (as Table 5-4) and the 50% burn-off soot samples. The uncertainty is ±3% (Appendix C). In Table 5-9, the differences in surface oxygen contents at 50% burn-off are not significant. Nevertheless, the oxygen contents after 50% burn-off of all soot samples increased from their initial states. The increase is attributed by the chemisorption and the formation of oxygen-containing functional groups [27, 70], or oxygen complexes [116], during the oxidation process. The TGA test presented by Figure 5-11 and Table 5-5 has shown that the surface oxygen content does not alter the soot reactivity. Therein, although the consistent increase of surface oxygen content of all samples indicates the progress in soot oxidation, it does not provide an explanation for the decrease in soot reactivity at 50% burn-off. Table 5-9 Surface elemental analysis of initial soot and 50% burn-off soot using XPS Sample Oxygen content at the initial state (atom %) Oxygen content after 50% burn-off (atom %) BP15 (single injection) B100 (single injection) FT (single injection) BP15 (split injection) B100 (split injection) FT (split injection) Comparison of structural parameters of soot at initial state and 50% burn-off The structural parameters of all 50% burn-off soot samples were investigated by HRTEM image analysis. Figure 5-23 shows the extracted skeletons in the regions of interest (ROIs) of the representative HRTEM images of the three soot samples generated by single injection strategies at their initial states and at 50% burn-off. The initial states of the samples were presented in Figure 5-15 (d) (f), and included in Figure 5-23 to enable visual comparison. For BP15 and

126 111 B100 soot, the development of longer fringes and highly ordered nanostructure at 50% burn-offs are readily apparent upon visual comparison of the HRTEM images in Figure For FT soot, the visual comparison of the initial state and the 50% burn-off does not yield any apparent conclusions. The quantitative analysis results of the extracted skeletons (carbon layers) are needed in order to justify the contention based on visual comparison or investigate the nonnoticeable difference. (a) FT soot (initial), single injection (b) BP15 soot (initial), single injection (c) B100 soot (initial), single injection (d) FT soot (50% burn-off), single injection (e) BP15 soot (50% burn-off), single injection (f) BP15 soot (50% burn-off), single injection Figure 5-23 The extracted skeletons in the ROIs of the representative HRTEM images of the three soot samples generated by single injection strategy at their initial states and 50% burn-off. Figure 5-24 shows the histograms of fringe length for the three soot samples at their initial states and at the 50% burn-off. It should be noticed that the uncertainty, according to

127 % of fringes % of fringes % of fringes % of fringes % of fringes % of fringes 112 Appendix D, of the median fringe length is ± 0.03nm. Compared with their initial states, the fringe length histograms at 50% burn-off for BP15 and B100 soot extend to much longer graphene layer plane dimensions. Relative to the fringe length for the B100 soot at initial state with 85% of the lamella smaller than 1 nm in length, only 55% of the fringe length histogram of the 50% burn-off B100 soot is less than 1 nm. Alternative only 1% of the fringe length histogram of the B100 soot at initial state is greater than 2nm while 12% of the lamella in the 50% burn-off B100 soot is greater than 2nm. The fringe length histograms of BP15 soot before and after oxidation (Figure 5-24(b) and (e)) show the similar trend as B100 soot. In contrast, for FT soot, no significant difference can be observed between the fringe length histograms before and after the oxidation. The changes in median fringe lengths are consistent with the direct observation of the fringe length histograms FT initial state (median: 0.90nm) 25 BP15 initial state (median: 0.79nm) 25 B100 initial state (mean: 0.72nm) fringe length (nm) (a) fringe length (nm) (b) fringe length (nm) (c) 25 FT 50% burn-off (median: 0.94nm) 25 BP15 50% burn-off (median: 1.00nm) 25 B100 50% burn-off (mean: 1.02nm) fringe length (nm) (d) fringe length (nm) (e) fringe length (nm) Figure 5-24 Fringe length histogram and median values derived from Figure 5-23: (a) FT soot (initial state), (b) BP15 soot (initial state), (c) B100 soot (initial state), (d) FT soot (50% burnoff), (e) BP15 soot (50% burn-off), and (f) B100 soot (50% burn-off). (f)

128 % of fringes % of fringes % of fringes % of fringes % of fringes % of fringes 113 Figure 5-25 shows the tortuosity histograms obtained from the three soot samples at their initial states and at 50% burn-off. Relative to the distribution for the B100 soot at initial state with 58% less than 1.2 in tortuosity ratio, more than 87% of the B100 soot fringe tortuosity histogram is less than 1.2. Alternatively the B100 soot at initial state has 14% of the measured fringes having a tortuosity ratio of greater than 1.5, while only 3.8% of the fringe tortuosity histogram of the 50% burn-off B100 soot is greater than 1.5. For BP15 soot, the change in tortuosity histogram before and after oxidation is consistent with the trend of B100. In contrast, for FT soot, no significant difference can be observed between the fringe tortuosity histograms before and after oxidation. The changes in mean fringe tortuosity are consistent with the direct observation of the fringe tortuosity histograms FT initial state (mean: 1.14) BP15 initial state (mean: 1.17) B100 initial state (mean: 1.37) fringe tortuosity (a) fringe tortuosity (b) fringe tortuosity (c) FT 50% burn-off (mean: 1.13) BP15 50% burn-off (mean: 1.09) B100 50% burn-off (mean: 1.19) fringe tortuosity (d) fringe tortuosity (e) fringe tortuosity Figure 5-25 Fringe tortuosity histogram and mean values derived from Figure 5-23: (a) FT soot (initial state), (b) BP15 soot (initial state), (c) B100 soot (initial state), (d) FT soot (50% burnoff), (e) BP15 soot (50% burn-off), and (f) B100 soot (50% burn-off). (f)

129 114 The changes in soot nanostructure during oxidation were also investigated for the three soot samples generated by split injection strategy. Figure 5-26 shows the extracted skeleton in the regions of interest (ROIs) of the representative images of the three soot samples generated by split injection strategies at their initial states and at 50% burn-off. The initial states of the samples were presented in Figure 5-18 (d) (f), and included in Figure 5-26 to enable visual comparison. It should be noticed that the magnification for the initial soot (0.034 nm/pixel) and 50% burn-off soot (0.021 nm/pixel) are different. The discrepancy in the image magnification does not affect the quantification results of lattice fringes, according to the repeatability analysis in Appendix D. For BP15 and B100 soot, the development of a more ordered nanostructure at 50% burn-off is readily apparent upon visual comparison of the HRTEM images in Figure The quantitative analysis results of the extracted skeletons are used for further support.

130 115 (a) FT soot (initial), split injection (b) BP15 soot (initial), split injection (c) B100 soot (initial), split injection (d) FT soot (50% burn-off), split injection (e) BP15 soot (50% burn-off), split injection (f) B100 soot (50% burn-off), split injection Figure 5-26 The extracted skeletons in the ROIs of the representative HRTEM images of the three soot samples generated by split injection strategy at their initial states and 50% burn-offs. Figure 5-27 shows the histograms of fringe length derived from the extracted skeleton images shown in Figure No conclusions can be drawn from the comparison of the histogram shapes. The uncertainty of the median fringe length is ± 0.03nm (Appendix D). The differences in the median fringe length calculated from the fringe length histogram before and after the soot oxidation are not significant enough for all the soot samples. Therein, neither the fringe length histogram nor the median fringe lengths were used to comment on the changes in nanostructure during soot oxidation for the soot generated by the split injection strategy.

131 % of fringes % of fringes % of fringes % of fringes % of fringes % of fringes FT initial state (median: 1.01nm) 25 BP15 initial state (median: 0.96nm) 25 B100 initial state (median: 0.92nm) fringe length (nm) (a) fringe length (nm) (b) fringe length (nm) (c) 25 FT 50% burn-off (median: 1.01nm) 25 BP15 50% burn-off (median: 1.02nm) 25 B100 50% burn-off (median: 0.96nm) fringe length (nm) (d) fringe length (nm) (e) fringe length (nm) Figure 5-27 Fringe length histogram and median values derived from Figure 5-26: (a) FT soot (initial state), (b) BP15 soot (initial state), (c) B100 soot (initial state), (d) FT soot (50% burnoff), (e) BP15 soot (50% burn-off), and (f) B100 soot (50% burn-off). (f) Figure 5-28 shows the tortuosity histogram obtained from the three soot samples at their initial states and the 50% burn-offs. Relative to the distribution for the initial B100 soot with 78% less than 1.2 in tortuosity ratio, more than 94% of the B100 soot fringe tortuosity histogram is less than 1.2 after 50% burn-off. BP15 soot underwent similar changes in tortuosity during the oxidation process. In contrast, for FT soot, no significant difference was observed in the fringe tortuosity histograms before and after the oxidation. The changes in the mean fringe tortuosity are consistent with the direct observation of the fringe tortuosity histograms.

132 % of fringes % of fringes % of fringes % of fringes % of fringes % of fringes FT initial state (mean: 1.12) BP15 initial state (mean: 1.18) B100 initial state (mean: 1.31) fringe tortuosity (a) fringe tortuosity (b) fringe tortuosity (c) FT 50% burn-off (mean: 1.11) BP15 50% burn-off (mean: 1.09) B100 50% burn-off (mean: 1.22) fringe tortuosity (d) fringe tortuosity (e) fringe tortuosity Figure 5-28 Fringe tortuosity histogram and mean values derived from Figure 5-26: (a) FT soot (initial state), (b) BP15 soot (initial state), (c) B100 soot (initial state), (d) FT soot (50% burnoff), (e) BP15 soot (50% burn-off), and (f) B100 soot (50% burn-off). (f) The HRTEM image analysis results of the samples at their initial states and 50% burn-off coincide with the hypothesis regarding the relationship between the soot nanostructure and reactivity. The apparent rate constant for soot oxidation, the median fringe length and mean fringe tortuosity at initial states and at 50% burn-off are summarized in Table For BP15 and B100 soot generated by the single injection strategy, the development in fringe lengths and the reduction in fringe tortuosity at 50% burn-off indicate the formation of more ordered soot nanostructure, which coincides with the significant decrease in the apparent rate constant for soot oxidation. For split injection strategy, the fringe tortuosity of BP15 and B100 soot at 50% burnoff shows a significant decrease, coinciding with the decrease in the apparent rate constant for soot oxidation. In contrast, for FT soot generated by either single injection or split injection, the

133 changes in fringe length and fringe tortuosity were not significant, associated with a smaller decrease in the apparent rate at 50% burn-off. 118 Table 5-10 Summary of the apparent rate constant, the median fringe length and mean fringe tortuosity at initial states and 50% burn-offs Sample Apparent rate constant (10-7 /Pa/min) Initial state Median fringe length (nm) Mean fringe tortuosity Apparent rate constant (10-7 /Pa/min) 50% burn-off Median fringe length (nm) Mean fringe tortuosity FT (single injection) BP15 (single injection) B100 (single injection) FT (split injection) BP15 (split injection) B100 (split injection) * The uncertainty for the apparent rate constant is ±4.4% at 95% confidence. ** The uncertainty (standard deviation) is 0.03nm for the median fringe length, and for the mean fringe tortuosity. Based on Table 5-10, Figure 5-29 shows the median fringe lengths and mean fringe tortuosity versus the apparent rate constant for the soot samples at initial states and 50% burn-off. The data was fitted by a linear model calculated for a minimized vector norm of error. The scattering pattern and the fitted curve indicate a relationship between the lattice fringe parameters (the median fringe length and mean fringe tortuosity) and the apparent rate constant. The result, compared with the XPS result summarized in Section 5.5.2, corroborates the subordinate hypothesis regarding the more dominant effect of soot nanostructure on the soot oxidative reactivity than the abundance of surface oxygen content.

134 119 y=-0.035x+1.20, R 2 =0.64 (a) y=0.035x+0.91, R 2 =0.77 Figure 5-29 The median fringe lengths and mean fringe tortuosity versus the apparent rate constant of the soot samples at initial states and 50% burn-offs: (a) the median fringe lengths versus the apparent rate constant, and (b) the mean fringe tortuosity versus the apparent rate constant. (b) The peak area ratios of the D 1, D 3, and D 4 peaks to G peak derived from Raman spectra were compared for soot at initial states and 50% burn-off. However, among the three indicated ratios derived from the Raman spectra, no consistent results were found to corroborate the

135 120 relationship between the soot nanostructure and reactivity as clearly shown in Figure Possible reasons for explaining the inconsistency have been discussed in Section The data were summarized and attached in Appendix E (Section E3). When the samples are considered separately, however, some D/G peak area ratios coincide with the analysis results of HRTEM. For example, for BP15 soot generated by both single and split injection strategies, I D1 /I G and I D3 /I G decreases as the soot oxidizes from initial states to 50% burn-off. Moreover, the I D3 /I G ratio of FT soot samples generated by both single injection and split injection strategies decreases after 50% burn-off. In contrast, the changes in the peak area ratios of B100 soot from initial states and 50% burn off do not indicate increasing order in nanostructure during soot oxidation as discovered by HRTEM image analysis. 5.6 Changes in BP15 soot nanostructure at different soot oxidation stages The changes in soot nanostructure at initial state, 25% burn-off, 50% burn-off, and 75% burn-off were investigated for the soot sample generated by BP15 using split injection strategy at matched combustion phasing (Mode C). The goal is to use the characterization methods developed in this study to refine the progression model of nanostructure during soot oxidation constructed by Song [6], who observed a drastic realignment of graphene layers Collection of soot at different oxidation stages Figure 5-30 shows the normalized weight curve and the collecting points of the four oxidation stages for the nanostructure analysis.

136 121 Figure 5-30 Isothermal oxidation curve of BP15 soot generated by split injection strategy with marked oxidation stages to be analyzed. 0% burn-off: 0 min, 25% burn-off: 17 min, 50% burnoff: 34 min, 75% burn-off: 56 min, oxidation completed: 110 min. ( ) normalized weight, and analyzed oxidation stages. In Table 5-11, the time of each burn-off period increases as soot oxidizes. The burn-off curve indicated a similar trend of a multiple-region curve as observed in coal oxidation [92]. In the first quarter (0-17 min) of oxidation, the reaction speed increase slowly. The second and third quarter (17-34, min) of oxidation indicate a zero-order reaction, in which the increase in burn-off is proportional to the increase in time. The last quarter (after 56 min) shows a reduced rate. The soot collected at each oxidation stage was investigated by surface oxygen content and lattice fringe parameters.

137 Table 5-11 Summary of the time segment of burn-off periods for four oxidation stages for BP15 soot (split injection) Oxidation stage Time segment of burn-off Burn-off time of each period period (min) (min) 0% burn-off 0 17 min 17 min 25% burn-off min 17 min 50% burn-off min 22 min 75% burn-off min 54 min Comparison of surface oxygen content of soot at different oxidation stages The variation in soot properties at different oxidation stages in Table 5-11 were investigated using XPS, HRTEM, and Raman spectroscopy. Table 5-12 shows the elemental content on the surface of the soot samples at the four oxidation stages. The uncertainty for XPS surface elemental measurement is ±3% (Appendix C). The surface oxygen content increased from 6.3% to 13.4% after 25% burn-off. The increase in the oxygen content is associated with the buildup of oxygen complexes on the carbon s surface during the first quarter of oxidation. However, the oxygen contents at 25%, 50%, and 75% burn-off do not show a significant difference. Therein, the surface oxygen content reflects the initiation and progress in soot oxidation, but does not indicate any changes during soot oxidation.

138 123 Table 5-12 Surface elemental analysis of the four oxidation stages of BP15 soot (split injection). Oxidation stage Oxygen content (atom %) 0% burn-off (initial state) % burn-off % burn-off % burn-off Comparison of structural parameters of soot at different oxidation stages The lattice fringe length and tortuosity of the BP15 soot samples (split injection) were investigated by HRTEM image analysis. Figure 5-31 shows the extracted skeletons in the regions of interest (ROIs) of the representative HRTEM images of the BP15 soot samples at the initial states, 25% burn-off, 50% burn-off, and 75% burn-off. In should be noted that the magnification of the HRTEM image for the soot at initial state (0.034 nm/pixel) is different from that of the images for the oxidized soot (0.021 nm/pixel). The magnification discrepancy does not affect the quantification results of lattice fringes, according to the repeatability analysis in Appendix D. Visual comparison shows that the lattice fringes in oxidized soot (Figure 5-31 (b) (d)) are less tortuous than those fringes in the initial soot (Figure 5-31 (a)). The change in lattice fringe length at different oxidation stages is not readily observable, and must rely on the HRTEM image analysis.

139 124 (a) Initial state (b) 25% burn-off (c) 50% burn-off (d) 75% burn-off Figure 5-31 The extracted skeletons in the regions of interest (ROIs) of the representative HRTEM images of the BP15 soot samples at different oxidation stages: (a) initial state, (b) 25% burn-off, (c) 50% burn-off, and (d) 75% burn-off. Figure 5-32 shows the fringe length histograms and the median lengths for the BP15 soot samples at four different oxidation stages. The uncertainty, according to Appendix D, of the median fringe length is ± 0.03nm. Visual comparison of the fringe length histograms shows that much longer fringes were developed as the soot oxidized from 25% to 50% and 75% burn-offs.

140 125 The median fringe lengths of 25%, 50% and 75% burn-off soot samples are 0.88nm, 1.02 nm and 1.11 nm. It should be noted that the median fringe length (0.88 nm) of the 25% burn-off soot are shorter than the initial soot (0.96 nm), indicating the formation of a less ordered nanostructure during the first quarter of the soot oxidation.

141 % of fringes % of fringes % of fringes % of fringes BP15 initial status (median: 0.96nm) 25 BP15 25% burn-off (median: 0.88nm) fringe length (nm) (a) Initial state fringe length (nm) (b) 25% burn-off 25 BP15 50% burn-off (median: 1.02nm) 25 BP15 75% burn-off (median: 1.11nm) fringe length (nm) (c) 50% burn-off fringe length (nm) (d) 75% burn-off Figure 5-32 Fringe length histogram and median values derived from Figure 5-31: (a) BP15 soot (split injection, initial state), (b) BP15 soot (25% burn-off), (c) BP15 soot (50% burn-off), and (d) BP15 soot (75% burn-off). Figure 5-33 shows the tortuosity histogram obtained from the soot samples at different oxidation stages. Compared to the initial soot, the oxidized soot samples (25%, 50% and 75% burn-off) contain a narrower range of tortuosity, indicating low degree of curvature among the

142 % of fringes % of fringes % of fringes % of fringes 127 lamella of the soot. The mean fringe tortuosity decreased as the soot oxidized from initial state to 25% burn-off, while the mean fringe tortuosity does not show significant changes when the soot is oxidized further BP15 initial status (mean: 1.18) 25 BP15 25% burn-off (mean: 1.11) fringe tortuosity (a) Initial state fringe tortuosity (b) 25% burn-off BP15 50% burn-off (mean: 1.09) 25 BP15 75% burn-off (mean: 1.12) fringe tortuosity (c) 50% burn-off fringe tortuosity (d) 75% burn-off Figure 5-33 Fringe tortuosity histogram and mean values derived from Figure 5-31: (a) BP15 soot (split injection, initial state), (b) BP15 soot (25% burn-off), (c) BP15 soot (50% burn-off), and (d) BP15 soot (75% burn-off).

143 128 The median fringe length and mean fringe tortuosity at different oxidation stages are summarized in Table 5-13and Figure The evolution of fringe properties during soot oxidation also supports the concept of equilibrium nanostructure [118, 119]. Because the basal plane energies are lower than edge plane energies, the mature soot tends to have a more ordered structure, the equilibrium configuration [118]. The longer and less tortuous fringes observed in the burn-off soot may be attributed to the formation of the equilibrium configuration during the oxidation process. Table 5-13 Summary of the median fringe length and mean fringe tortuosity at different oxidation stages Characteristic parameters Sample Median fringe length Mean fringe tortuosity (nm) 0% burn-off (initial state) % burn-off % burn-off % burn-off * The uncertainty (standard deviation) is 0.03nm for the median fringe length, and for the mean fringe tortuosity.

144 129 (a) Figure 5-34 The median fringe lengths and mean fringe tortuosity versus the burn-off rate of the soot samples at different oxidation stages: (a) the median fringe lengths versus the burn-off rate, and (b) the mean fringe tortuosity versus the burn-off rate. (b) The peak area ratios of the D 1, D 3, and D 4 peaks to the G peak derived from Raman spectra were compared for soot at different oxidation stages. Although the deviations are large, the average peak area ratios of D 1 and D 3 peaks to G peak decreased as the soot oxidized, while the ratio of D 4 peak to G peak does not indicate a decreasing trend. Figure 5-35 shows the ratio I D1 /I G and I D3 /I G at different oxidation stages. The decreasing trend is consistent with the decrease in tortuosity and the increase in fringe length along the course of oxidation. However, it should be

145 130 noted that the results from Raman spectra are suspicious because: (1) the deviations obtained from 10 different locations are large and overlapped much to each other, and (2) such a relation was not confirmed by the initial soot samples and 50% burn-offs generated by different fuels (Section 5.4.4, 5.5.1, and Appendix E). Therein, whether or not Raman spectra indicate structural difference between soot under a similar formation process (similar combustion phasing) should be investigated further. The ratios obtained from Raman spectra are not used as major evidence to develop the simplified oxidation progression model described as follows. (a) Figure 5-35 Peak area ratios of (a) I D1 /I G, and (b) I D3 /I G at different burn-off stages. (b) A simplified oxidation progression model for BP15 soot Using HRTEM image analysis results at different oxidation stages, a simplified oxidation progression model for BP15 soot is proposed as shown in Figure The soot at its initial state contains various kinds of lattice fringes, including short, long, straight, and tortuous fringes. The first quarter (initial state to 25% burn-off) of oxidation removed those long but tortuous fringes, while the shorter but straight fringes remained as observed for the 25% burn-off soot. The increase in shorter fringes during the first quarter may explain an initial slope drop in Figure 5-30 (near 0 min), because the shorter lattice fringe implies a higher active carbon concentration [46]. From the 25% burn-off to 50% and 75% burn-offs, the soot formed higher ordered nanostructure,

146 131 as shown by the longer median fringe length obtained from HRTEM image analysis. The long, straight fringes observed at 75% burn-off soot indicate a decrease in active surface area [46], leading to a reduced reaction speed in the last quarter [92]. Initial status 25% burn-off 50% burn-off 75% burn-off Mixed with long, short, straight, tortuous fringes Burning long but tortuous fringes, remaining short fringes Developing longer fringes Developing longer fringes Figure 5-36 Simplified oxidation progression model for BP15 soot generated by split injection strategy. The proposed BP15 oxidation progression model based on the changes in lattice fringes can be connected with the explanation of a burn-off curve of heat-pretreated coals inspected by Jenkins et al. [92]. In the beginning of the carbon oxidation, there is a buildup of oxygen complexes on the carbon s surface, and the carbon undergoes activation, leading to the increase of curve slope (initial state to 25% burn-off rate). The oxidation then undergoes a stable period (around 50% burn-off shown in Figure 5-30). In the final period, the surface area per unit weight decreases, leading to a reduced curve slope. The surface oxygen content measured by XPS shows a significant increase when the soot oxidizes from initial state to 25% burn-off, indicating a buildup of oxygen complexes by chemisorption on the soot surface. In contrast, the similar surface oxygen content of the 25%, 50% and 75% burn-off soot indicate that the abundance of surface oxygen-containing functional groups are not correlated with the lattice fringe parameters and active surface area.

147 Chapter 6 CONCLUSIONS AND RECOMMENDATIONS FOR FUTURE WORK 6.1 Summary and conclusions This study focuses on the impacts of fuels and engine operating parameters on the nanostructure and reactivity of diesel soot. Using an ultra low sulfur diesel fuel (BP15), both the engine operating mode (engine speed and torque) and injection parameters (start of injection and fuel rail pressure) are confirmed to have impacts on soot reactivity and nanostructure. BP15, B100, and FT fuels are selected to investigate the effect of fuel formulation on soot reactivity and properties. For the three test fuels, the soot samples are generated in steady state conditions using an instrumented DDC/VM Motori 2.5L, 4-cylinder, turbocharged, common-rail, direct injection light-duty diesel engine. The operating condition (2400 rpm, 64 Nm) is selected in order to compare soot characteristics with previous work [6, 18, 19]. The fuel injection parameters were adjusted for the different fuels to obtain the same combustion phasing, based on pressures traces and heat release rates. The matched engine combustion phasing for different fuels is used while collecting soot samples for characterization. Even where the combustion phasing is matched for the different fuels, it was hypothesized that B100 fuel would increase soot oxidative reactivity to a measurable degree by forming a less ordered nanostructure or more surface oxygen-containing function groups. In order to examine the hypotheses, the soot samples are characterized using TGA (pretreatment, oxidation, active surface area), XRD, HRTEM, Raman spectroscopy, XPS, and

148 133 FTIR. The results of TGA, XRD and HRTEM strongly support the claim that fuel formulation affects the soot reactivity by forming different (nano)structural order. On the other hand, the results of XPS do not indicate that initial surface oxygen groups cause the different reactivity of soot generated by different fuels. Finally, the soot characterization results are summarized into four conclusions: (1) Even when the combustion phasing is matched for different fuels, fuel formulation such as oxygen content and aromatics has an influence on soot reactivity. B100 soot exhibits the fastest oxidation on a mass basis with BP15 and FT soot in order of apparent rate constant. When using single injection, B100 soot has two times higher apparent rate constant than FT soot. (2) The soot oxidative reactivity is dominated by the disorder of carbonaceous nanostructure and consequent increase of accessible carbons on the edge sites. An inverse relation between apparent rate constant and lattice fringe parameter (L a of XRD and median fringe length of TEM image analysis) was confirmed. Nevertheless, the soot oxidative reactivity is not dominated by the abundance of surface oxygen content. (3) Compared with their initial states, both B100 and BP15 soot samples after 50% burn-off become highly ordered. On the other hand, no significant difference in structural order can be observed for FT soot samples before and after the oxidation. For the soot generated by single injection strategy, when the soot oxidized 50% from its initial state, the median fringe length of B100 soot increased from 0.79 nm to 1.0 nm, while that of FT soot only changed from 0.9 nm to 0.94 nm.

149 134 (4) HRTEM image analysis of BP15 soot at different oxidation states shows the removal of tortuous but long lattice fringes from the initial state to 25% burn-off, causing an increased slope in the normalized weight curve of soot oxidation. From the 25% burn-off to 50% and 75% burnoffs, the soot became more ordered, leading to reduced active carbon concentration and thus a reduced slope in normalized weight curve. 6.2 Recommendations for future work Studying soot oxidation in diesel particulate filters Although the soot reactivity and nanostructure have been investigated thoroughly, these properties do not truly resemble the soot regeneration characteristics occurring in a catalytic DPF substrate. For example, it is yet to be confirmed whether or not the differences in the apparent rate constants for soot generated by different fuels (Section 5.3) would significantly affect the DPF regeneration process. Therein, it is recommended to measure the break even temperatures of the three test fuels using the same operation condition with matched combustion phasing [6, 18, 19]. Besides using the whole DPFs, single wall [120], single layer [121], or a disk DPF substrate [122] can also be considered in the experimental design in order to simplify the model and data interpretation. Moreover, a laboratory reactor is recommended where PM can be directly deposited in the DPF substrate under a constant flow condition [123]. Combining the experimental setup with an infra-red camera enables the visual observation of soot oxidation initiated by thermal activation [121]. In addition, an extensive set of well-controlled experiments can be designed with modeling to describe soot oxidation kinetics in diesel particulate filters [124].

150 Modeling soot formation process In this study, the impacts of fuel formulation on soot reactivity and nanostructure have been confirmed under matched combustion phasing. Nevertheless, the reaction mechanisms causing the difference in soot properties were not addressed in this study. Therein, testing surrogate fuels in a motored engine is recommended for understanding the role of the fuel-bound oxygen in biodiesel in producing soot with less ordered nanostructure [125, 126]. Besides, rapid compression machines [127, 128], shock tubes [ ], variable pressure flow reactors [131], and jet-stirred reactors [131, 132] can be considered to investigate the soot formation using surrogate fuels. The soot formation and oxidation process can be further investigated by chemical modeling [11, ]. In particular, the effects of the fuel formulation on rich premixed ignition and the formation of unsaturated species, known as precursors of soot, need to be investigated [137, 138]. The exploration of reaction pathways and the estimation of thermodynamic and transport properties can be accomplished through the combination of molecular dynamics and ab initio methods [130, 139, 140] Improving HRTEM image analysis algorithm and other characterization methods The image analysis algorithm has been developed to obtain fringe length, fringe tortuosity and fringe separation distance distributions from the HRTEM images. The repeatability and consistency of the algorithm has been investigated and summarized as shown in Appendix D. A thorough investigation is recommended to eliminate the subjective choice of the image processing parameters, such as the choice of region of interest and the size of the Gaussian low-

151 136 pass filter. Tilting angle of TEM grids is also recommended to verify the two-dimensional curvature projected from a three-dimensional structure. In addition, the sensitivity of image processing parameters on the fringe properties distributions and characteristic values (median for fringe length and mean for fringe tortuosity) can be tested over a range of values. Moreover, in order to reduce the quantitative discrepancies between the lattice fringe lengths (L a ) derived from XRD and HRTEM, as shown in Figure 5-21 and 5-22, the Diamond s empirical formula can be used to extract the lattice fringe parameters [25, 110]. In section 5.5 and 5.6, although the crystallite growth and coalescence of graphene layers during soot oxidation were consistent with the results observed by others [6, 44], the data acquired in this study was insufficient to provide a clear explanation. To address the reasons for structural changes during soot oxidation, in situ TEM can be used [141]. In this study, the active surface area measured by TGA was not correlated with the soot reactivity. The explanations have been proposed in Section It is either because the soot samples in this study are not significantly different in ASA, or the adopted experimental method does not yield representative ASA values. In order to measure ASA effectively, the ASA need to be measured by a different method [111], or to include additional apparatus such as an in vacuo device [117]. In addition, because C/H ratio depends on the evolution of soot particles [142, 143], the C/H ratio in soot may correlate with the growth of crystallite and reactivity difference that were observed in this study. This study examined the general relations among the soot reactivity, multiple nanostructure and surface oxygen-containing functional groups. The relations were investigated by linear regression fitting or direct inspection on variables in pairs, e.g. reactivity versus L a. In

152 137 order to further quantify the relationships, more sophisticated multivariate regression analysis [144] can be used to describe the correlation and dependence among multiple variables and their impact on soot reactivity.

153 BIBLIOGRAPHY [1] Johnson JH, Bagley ST, Gratz LD, Leddy DG. A review of diesel particulate control technology and emissions effects Horning memorial award lecture. SAE International Congress and Exposition; 1992: SAE International [2] Neeft JPA, Makkee M, Moulijn JA. Diesel particulate emission control. Fuel Processing Technology 1996; 47(1):1-69. [3] Walker AP. Controlling particulate emissions from diesel vehicles. Topics in Catalysis 2004; 28(1-4): [4] Stanmore BR, Brilhac JF, Gilot P. The oxidation of soot: A review of experiments, mechanisms and models. Carbon 2001; 39(15): [5] Vander Wal RL, Tomasek AJ. Soot oxidation: Dependence upon initial nanostructure. Combustion and Flame 2003; 134(1-2):1-9. [6] Song JH. Effect of fuel formulation on soot properties and regeneration of diesel particulate filters. Ph.D. Thesis, Fuel Science. University Park: The Pennsylvania State University; [7] Lapuerta M, Armas O, Herreros JM. Emissions from a diesel-bioethanol blend in an automotive diesel engine. Fuel 2008; 87(1): [8] Kittelson DB. Engines and nanoparticles: A review. Journal of Aerosol Science 1998; 29(5-6): [9] Choi CY, Reitz RD. An experimental study on the effects of oxygenated fuel blends and multiple injection strategies on DI diesel engine emissions. Fuel 1999; 78(11):

154 139 [10] Vander Wal RL, Tomasek AJ. Soot nanostructure: Dependence upon synthesis conditions. Combustion and Flame 2004; 136(1-2): [11] Chan TL, Cheng XB. Numerical modeling and experimental study of combustion and soot formation in a direct injection diesel engine. Energy & Fuels 2007; 21(3): [12] Zhu JY, Lee KO, Yozgatligil A, Choi MY. Effects of engine operating conditions on morphology, microstructure, and fractal geometry of light-duty diesel engine particulates. Proceedings of the Combustion Institute 2005; 30: [13] Neer A, Koylu UO. Effect of operating conditions on the size, morphology, and concentration of submicrometer particulates emitted from a diesel engine. Combustion and Flame 2006; 146(1-2): [14] Szybist JP, Kirby SR, Boehman AL. NOx emissions of alternative diesel fuels: A comparative analysis of biodiesel and FT diesel. Energy & Fuels 2005; 19(4): [15] Armas O, Hernandez JJ, Cardenas MD. Reduction of diesel smoke opacity from vegetable oil methyl esters during transient operation. Fuel 2006; 85(17-18): [16] Zhang Y, Boehman AL. Impact of biodiesel on NOx emissions in a common rail direct injection diesel engine. Energy & Fuels 2007; 21(4): [17] Lapuerta M, Armas O, Rodriguez-Fernandez J. Effect of biodiesel fuels on diesel engine emissions. Progress in Energy and Combustion Science 2008; 34(2): [18] Song JH, Alam M, Boehman AL. Impact of alternative fuels on soot properties and DPF regeneration. Combustion Science and Technology 2007; 179(9): [19] Song JH, Alam M, Boehman AL, Kim U. Examination of the oxidation behavior of biodiesel soot. Combustion and Flame 2006; 146(4): [20] Al-Qurashi K, Boehman AL. Impact of exhaust gas recirculation (EGR) on the oxidative reactivity of diesel engine soot. Combustion and Flame 2008; 155(4):

155 140 [21] Casiraghi C, Hartschuh A, Qian H, Piscanec S, Georgi C, Fasoli A, et al. Raman spectroscopy of graphene edges. Nano Letters 2009; 9(4): [22] Ferrari AC, Robertson J. Interpretation of Raman spectra of disordered and amorphous carbon. Physical Review B 2000; 61(20): [23] Tuinstra F, Koenig JL. Raman spectrum of graphite. Journal of Chemical Physics 1970; 53(3): [24] Chen HX, Dobbins RA. Crystallogenesis of particles formed in hydrocarbon combustion. Combustion Science and Technology 2000; 159: [25] Sharma A, Kyotani T, Tomita A. Comparison of structural parameters of PF carbon from XRD and HRTEM techniques. Carbon 2000; 38(14): [26] Boehm HP. Surface oxides on carbon and their analysis: A critical assessment. Carbon 2002; 40(2): [27] Figueiredo JL, Pereira MFR, Freitas MMA, Orfao JJM. Modification of the surface chemistry of activated carbons. Carbon 1999; 37(9): [28] Langley LA, Villanueva DE, Fairbrother DH. Quantification of surface oxides on carbonaceous materials. Chemistry of Materials 2006; 18(1): [29] Arenillas A, Rubiera F, Pevida C, Ania CO, Pis JJ. Relationship between structure and reactivity of carbonaceous materials. Journal of Thermal Analysis and Calorimetry 2004; 76(2): [30] Feng B, Bhatia SK, Barry JC. Variation of the crystalline structure of coal char during gasification. Energy & Fuels 2003; 17(3): [31] Goel A, Howard JB, Vander Sande JB. Size analysis of single fullerene molecules by electron microscopy. Carbon 2004; 42(10):

156 141 [32] Palotas AB, Rainey LC, Feldermann CJ, Sarofim AF, Vander Sande JB. Soot morphology: An application of image analysis in high-resolution transmission electron microscopy. Microscopy Research and Technique 1996; 33(3): [33] Rouzaud JN, Clinard C. Quantitative high-resolution transmission electron microscopy: A promising tool for carbon materials characterization. Fuel Processing Technology 2002; 77: [34] Sadezky A, Muckenhuber H, Grothe H, Niessner R, Poschl U. Raman microspectroscopy of soot and related carbonaceous materials: Spectral analysis and structural information. Carbon 2005; 43(8): [35] Yezerets A, Currier NW, Eadler HA, Suresh A, Madden PF, Branigin MA. Investigation of the oxidation behavior of diesel particulate matter. Catalysis Today 2003; 88(1-2): [36] Iwashita N, Inagaki M. Relations between structural parameters obtained by X-ray powder diffraction of various carbon materials. Carbon 1993; 31(7): [37] Iwashita N, Park CR, Fujimoto H, Shiraishi M, Inagaki M. Specification for a standard procedure of X-ray diffraction measurements on carbon materials. Carbon 2004; 42(4): [38] Smith DM, Chughtai AR. The surface structure and reactivity of black carbon. Colloids and Surfaces a-physicochemical and Engineering Aspects 1995; 105(1): [39] Al-Qurashi K. The impact of carbon dioxide and exhaust gas recirculation on the oxidative reactivity of soot from ethylene flames and diesel engines. Ph.D. Thesis, Energy and Geo-Environmental Engineering. University Park: The Pennsylvania University; [40] Duran A, Carmona M, Ballesteros R. Competitive diesel engine emissions of sulphur and nitrogen species. Chemosphere 2003; 52(10): [41] Lapuerta M, Hernandez JJ, Ballesteros R, Duran A. Composition and size of diesel particulate emissions from a commercial European engine tested with present and future fuels.

157 142 Proceedings of the Institution of Mechanical Engineers Part D-Journal of Automobile Engineering 2003; 217(D10): [42] Van Setten B, Makkee M, Moulijn JA. Science and technology of catalytic diesel particulate filters. Catalysis Reviews-Science and Engineering 2001; 43(4): [43] Neeft JPA, Nijhuis TX, Smakman E, Makkee M, Moulijn JA. Kinetics of the oxidation of diesel soot. Fuel 1997; 76(12): [44] Ishiguro T, Suzuki N, Fujitani Y, Morimoto H. Microstructural changes of diesel soot during oxidation. Combustion and Flame 1991; 85(1-2):1-6. [45] Vander Wal RL. Soot nanostructure: Definition, quantification and implications. SAE World Congress; 2005; Detroit, Michigan: SAE International. Paper # [46] Radovic LR, Walker PL, Jenkins RG. Importance of carbon active sites in the gasification of coal chars. Fuel 1983; 62(7): [47] Oberlin A. Carbonization and graphitization. Carbon 1984; 22(6): [48] Fujimoto H, Mabuchi A, Tokumitsu K, Kasuh T, Akuzawa N. Effect of crystallite size on the chemical compositions of the stage 1 alkali metal-graphite intercalation compounds. Carbon 1994; 32(2): [49] Franklin RE. The interpretation of diffuse X-ray diagrams of carbon. Acta Crystallographica 1950; 3(2): [50] Jawhari T, Roid A, Casado J. Raman spectroscopic characterization of some commercially available carbon black materials. Carbon 1995; 33(11): [51] Rosen H, Novakov T. Raman scattering and characterization of atmospheric aerosol particles. Nature 1977; 266(5604): [52] Nemanich RJ, Solin SA. First- and second-order Raman scattering from finite-size crystals of graphite. Physical Review B 1979; 20(2):

158 143 [53] Katagiri G, Ishida H, Ishitani A. Raman spectra of graphite edge planes. Carbon 1988; 26(4): [54] Knauer M, Schuster ME, Su DS, Schlogl R, Niessner R, Ivleva NP. Soot structure and reactivity analysis by Raman microspectroscopy, temperature-programmed oxidation, and highresolution transmission electron microscopy. Journal of Physical Chemistry A 2009; 113(50): [55] Vander Wal RL, Tomasek AJ, Street K, Hull DR, Thompson WK. Carbon nanostructure examined by lattice fringe analysis of high-resolution transmission electron microscopy images. Applied Spectroscopy 2004; 58(2): [56] Vander Wal R, Tomasek AJ, Pamphlet MI, Taylor CD, Thompson WK. Analysis of HRTEM images for carbon nanostructure quantification. Journal of Nanoparticle Research 2004; 6(6): [57] Heidenreich RD, Hess WM, Ban LL. A test object and criteria for high resolution electron microscopy. Journal of Applied Crystallography 1968; 1:1-19. [58] Sharma A, Kyotani T, Tomita A. A new quantitative approach for microstructural analysis of coal char using HRTEM images. Fuel 1999; 78(10): [59] Shim HS, Hurt RH, Yang NYC. A methodology for analysis of 002 lattice fringe images and its application to combustion-derived carbons. Carbon 2000; 38(1): [60] Galvez A, Herlin-Boime N, Reynaud C, Clinard C, Rouzaud JN. Carbon nanoparticles from laser pyrolysis. Carbon 2002; 40(15): [61] Goel A, Hebgen P, Vander Sande JB, Howard JB. Combustion synthesis of fullerenes and fullerenic nanostructures. Carbon 2002; 40(2): [62] Muller JO, Su DS, Wild U, Schlogl R. Bulk and surface structural investigations of diesel engine soot and carbon black. Physical Chemistry Chemical Physics 2007; 9(30):

159 144 [63] Boehman AL, Song J, Alam M. Impact of biodiesel blending on diesel soot and the regeneration of particulate filters. Energy & Fuels 2005; 19(5): [64] Song JH, Wang JG, Boehman AL. The role of fuel-borne catalyst in diesel particulate oxidation behavior. Combustion and Flame 2006; 146(1-2): [65] Yehliu K, Vander Wal R, Boehman AL. Development of an HRTEM image analysis method to quantify carbon nanostructure. Combustion and Flame (submitted). [66] Yehliu K, Vander Wal RL, Boehman AL. Quantitative comparison of two HRTEM image analysis methods, X-ray diffraction, and Raman spectroscopy. Carbon (submitted). [67] Boehm HP. Some aspects of the surface chemistry of carbon blacks and other carbons. Carbon 1994; 32(5): [68] Seong HJ. Impact of oxygen enrichment on soot properties and soot oxidative reactivity. Ph.D. Thesis, Energy and Geo-Environmental Engineering. University Park: The Pennsylvania University; [69] Painter PC, Snyder RW, Starsinic M, Coleman MM, Kuehn DW, Davis A. Concerning the application of FTIR to the study of coal - a critical assessment of band assignments and the application of spectral analysis programs. Applied Spectroscopy 1981; 35(5): [70] Fanning PE, Vannice MA. A DRIFTS study of the formation of surface groups on carbon by oxidation. Carbon 1993; 31(5): [71] Haydar S, Moreno-Castilla C, Ferro-Garcia MA, Carrasco-Marin F, Rivera-Utrilla J, Perrard A, et al. Regularities in the temperature-programmed desorption spectra of CO2 and CO from activated carbons. Carbon 2000; 38(9): [72] Hurt RH, Sarofim AF, Longwell JP. Gasification induced densification of carbons - from soot to form coke. Combustion and Flame 1993; 95(4):430-2.

160 145 [73] Davis KA, Hurt RH, Yang NYC, Headley TJ. Evolution of char chemistry, crystallinity, and ultrafine structure during pulverized-coal combustion. Combustion and Flame 1995; 100(1-2): [74] Palotas AB, Rainey LC, Sarofim AF, VanderSande JB, Ciambelli P. Effect of oxidation on the microstructure of carbon blacks. Energy & Fuels 1996; 10(1): [75] Marcuccilli F, Gilot P, Stanmore B, Prado G. Experimental and theoretical study of diesel soot reactivity. Symposium (International) on Combustion 1994; 25(1): [76] Hurt R, Sun J-K, Lunden M. A kinetic model of carbon burnout in pulverized coal combustion. Combustion and Flame 1998; 113(1-2): [77] Hurt RH, Davis KA. Near-extinction and final burnout in coal combustion. Symposium (International) on Combustion 1994; 25(1): [78] Lapuerta M, Armas O, Ballesteros R, Fernandez J. Diesel emissions from biofuels derived from Spanish potential vegetable oils. Fuel 2005; 84(6): [79] Miyamoto N, Ogawa H, Nurun NM, Obata KA, T. Smokeless, low NOx, high thermal efficiency, and low noise diesel combustion with oxygenated agents as main fuel. SAE Technical Paper 1998; [80] Palmer HB, Cullis CF. The chemistry and physics of carbon. New York: Marcel Dekker [81] Mathis U, Mohr M, Kaegi R, Bertola A, Boulouchos K. Influence of diesel engine combustion parameters on primary soot particle diameter. Environmental Science & Technology 2005; 39(6): [82] Oh KC, Shin HD. The effect of oxygen and carbon dioxide concentration on soot formation in non-premixed flames. Fuel 2006; 85(5-6): [83] Armas O, Yehliu K, Boehman AL. Effect of alternative fuels on exhaust emissions during diesel engine operation with matched combustion phasing. Fuel 2010; 89(2):

161 146 [84] Yehliu K, Boehman AL, Armas O. Emissions from different alternative diesel fuels operating with single and split fuel injection. Fuel 2010; 89(2): [85] Lilik G. Hydrogen assisted diesel combustion. M.S. Thesis, Energy and Geo- Environmental Engineering. University Park: The Pennsylvania State University; [86] Lapuerta M, Armas O, Bermudez V. Sensitivity of diesel engine thermodynamic cycle calculation to measurement errors and estimated parameters. Applied Thermal Engineering 2000; 20(9): [87] Lapuerta M, Armas O, Hernandez JJ. Diagnosis of DI diesel combustion from in-cylinder pressure signal by estimation of mean thermodynamic properties of the gas. Applied Thermal Engineering 1999; 19(5): [88] Moffat RJ. Describing the uncertainties in experimental results. Experimental Thermal and Fluid Science 1988; 1(1):3-17. [89] Johnson TV. Review of diesel emissions and control. International Journal of Engine Research 2009; 10(5): [90] Belenkov EA. Formation of graphite structure in carbon crystallites. Inorganic Materials 2001; 37(9): [91] Mawhinney DB, Yates JT. FTIR study of the oxidation of amorphous carbon by ozone at 300 k - direct COOH formation. Carbon 2001; 39(8): [92] Jenkins RG, Nandi SP, Walker PL. Reactivity of heat-treated coals in air at 500 degrees c. Fuel 1973; 52(4): [93] Ehrburger P, Louys F, Lahaye J. The concept of active sites applied to the study of carbon reactivity. Carbon 1989; 27(3): [94] Smith IW. Intrinsic reactivity of carbons to oxygen. Fuel 1978; 57(7): [95] Nelder JA, Mead R. A simplex-method for function minimization. Computer Journal 1965; 7(4):

162 147 [96] Lagarias JC, Reeds JA, Wright MH, Wright PE. Convergence properties of the Nelder- Mead simplex method in low dimensions. Siam Journal on Optimization 1998; 9(1): [97] Song J, Jeon CH, Boehman AL. Impacts of oxygen diffusion on the combustion rate of in-bed soot particles. Energy & Fuels 2010; 24: [98] Leon-Garcia A. Probability and random processes for electrical engineering. Second ed: Addison-Wesley Publishing Company, Inc [99] Gonzalez RC, Woods RE, Eddins SL. Digital image processing using MATLAB. Upper Saddle River, New Jersey, US: Pearson Prentice Hall [100] Gonzalez RC, Woods RE. Digital image processing. Upper Saddle River, New Jersey, US: Pearson Prentice Hall [101] Otsu N. Threshold selection method from gray-level histograms. Ieee Transactions on Systems Man and Cybernetics 1979; 9(1):62-6. [102] Lam L, Lee SW, Suen CY. Thinning methodologies - a comprehensive survey. Ieee Transactions on Pattern Analysis and Machine Intelligence 1992; 14(9): [103] Pope CJ, Marr JA, Howard JB. Chemistry of fullerenes c-60 and c-70 formation in flames. Journal of Physical Chemistry 1993; 97(42): [104] Grieco WJ, Howard JB, Rainey LC, Vander Sande JB. Fullerenic carbon in combustiongenerated soot. Carbon 2000; 38(4): [105] Heywood JB. Internal combustion engine fundamentals McGraw-Hill Book Company [106] Collura S, Chaoui N, Azambre B, Finqueneisel G, Heintz O, Krzton A, et al. Influence of the soluble organic fraction on the thermal behaviour, texture and surface chemistry of diesel exhaust soot. Carbon 2005; 43(3): [107] Biniak S, Szymanski G, Siedlewski J, Swiatkowski A. The characterization of activated carbons with oxygen and nitrogen surface groups. Carbon 1997; 35(12):

163 148 [108] Papirer E, Lacroix R, Donnet JB, Nanse G, Fioux P. Xps study of the halogenation of carbon black - part 1. Bromination. Carbon 1994; 32(7): [109] Takagi H, Maruyama K, Yoshizawa N, Yamada Y, Sato Y. XRD analysis of carbon stacking structure in coal during heat treatment. Fuel 2004; 83(17-18): [110] Short MA, Walker PL. Measurement of interlayer spacings and crystal sizes in turbostratic carbons. Carbon 1963; 1(1):3-9. [111] Arenillas A, Rubiera F, Parra JB, Pis JJ. A comparison of ASA values determined by different methods. Carbon 2002; 40(8): [112] Teng H, Hsieh CT. Activation energy for oxygen chemisorption oil carbon at low temperatures. Industrial & Engineering Chemistry Research 1999; 38(1): [113] Laine NR, Vastola FJ, Walker PL. The importance of active surface area in carbon oxygen reaction. Journal of Physical Chemistry 1963; 67(10): [114] Miura K, Hashimoto K, Silveston PL. Factors affecting the reactivity of coal chars during gasification, and indices representing reactivity. Fuel 1989; 68(11): [115] Allardice DJ. The adsorption of oxygen on brown coal char. Carbon 1966; 4(2): [116] Donnet JB. Chemical reactivity of carbons Carbon 1968; 6(2): [117] Causton P, McEnaney B. Determination of active surface areas of coal chars using a temeperature programmed desorption technique. Fuel 1985; 64(10): [118] Hurt RH, Crawford GP, Shim HS. Equilibrium nanostructure of primary soot particles. Proceedings of the Combustion Institute 2000; 28: [119] Inagaki M. Discussion of the formation of nanometric texture in spherical carbon bodies. Carbon 1997; 35(5): [120] Yang J, Stewart M, Maupin G, Herling D, Zelenyuk A. Single wall diesel particulate filter (DPF) filtration efficiency studies using laboratory generated particles. Chemical Engineering Science 2009; 64(8):

164 149 [121] Martirosyan KS, Chen K, Luss D. Behavior features of soot combustion in diesel particulate filter. Chemical Engineering Science 2010; 65(1):42-6. [122] Ohara E, Mizuno Y, Miyairi Y, Mizutani T, Yuuki K, Noguchi Y, et al. Filtration behavior of diesel particulate filters (1). SAE World Congress & Exhibition, April ; SAE [123] Herz RK, Sinkevitch RM. Reactors for investigation of soot combustion on filter surfaces. Carbon 1986; 24(4): [124] Konstandopoulos AG, Kostoglou M, Lorentzou S, Pagkoura C, Papaioannou E, Ohno K, et al. Soot oxidation kinetics in diesel particulate filters. SAE World Congress & Exhibition, April ; [125] Zhang Y, Boehman AL. Experimental study of the autoignition of C8H16O2 ethyl and methyl esters in a motored engine. Combustion and Flame 157(3): [126] Zhang Y, Yang Y, Boehman AL. Premixed ignition behavior of C-9 fatty acid esters: A motored engine study. Combustion and Flame 2009; 156(6): [127] Walton SM, Wooldridge MS, Westbrook CK. An experimental investigation of structural effects on the auto-ignition properties of two C-5 esters. Proceedings of the Combustion Institute 2009; 32: [128] Dooley S, Curran HJ, Simmie JM. Autoignition measurements and a validated kinetic model for the biodiesel surrogate, methyl butanoate. Combustion and Flame 2008; 153(1-2):2-32. [129] Metcalfe WK, Dooley S, Curran HJ, Simmie JM, El-Nahas AM, Navarro MV. Experimental and modeling study of C5H10O2 ethyl and methyl esters. Journal of Physical Chemistry A 2007; 111(19): [130] Farooq A, Davidson DF, Hanson RK, Huynh LK, Violi A. An experimental and computational study of methyl ester decomposition pathways using shock tubes. Proceedings of the Combustion Institute 2009; 32:

165 150 [131] Gail S, Thomson MJ, Sarathy SM, Syed SA, Dagaut P, Dievart P, et al. A wide-ranging kinetic modeling study of methyl butanoate combustion. Proceedings of the Combustion Institute 2007; 31: [132] Gail S, Sarathy SM, Thomson MJ, Dievart P, Dagaut P. Experimental and chemical kinetic modeling study of small methyl esters oxidation: Methyl (e)-2-butenoate and methyl butanoate. Combustion and Flame 2008; 155(4): [133] Westbrook CK, Pitz WJ, Curran HJ. Chemical kinetic modeling study of the effects of oxygenated hydrocarbons on soot emissions from diesel engines. Journal of Physical Chemistry A 2006; 110(21): [134] Blanquart G, Pitsch H. Analyzing the effects of temperature on soot formation with a joint volume-surface-hydrogen model. Combustion and Flame 2009; 156(8): [135] Kazakov A, Frenklach M. Dynamic modeling of soot particle coagulation and aggregation: Implementation with the method of moments and application to high-pressure laminar premixed flames. Combustion and Flame 1998; 114(3-4): [136] Pitsch H, Riesmeier E, Peters N. Unsteady flamelet modeling of soot formation in turbulent diffusion flames. Combustion Science and Technology 2000; 158: [137] Fisher EM, Pitz WJ, Curran HJ, Westbrook CK. Detailed chemical kinetic mechanisms for combustion of oxygenated fuels. Proceedings of the Combustion Institute 2000; 28: [138] Westbrook CK. Chemical kinetics of hydrocarbon ignition in practical combustion systems. Proceedings of the Combustion Institute 2000; 28: [139] Huynh LK, Violi A. Thermal decomposition of methyl butanoate: Ab initio study of a biodiesel fuel surrogate. Journal of Organic Chemistry 2008; 73(1): [140] Blanquart G, Pepiot-Desjardins P, Pitsch H. Chemical mechanism for high temperature combustion of engine relevant fuels with emphasis on soot precursors. Combustion and Flame 2009; 156(3):

166 151 [141] Hernandez E, Meunier V, Smith BW, Rurali R, Terrones H, Nardelli MB, et al. Fullerene coalescence in nanopeapods: A path to novel tubular carbon. Nano Letters 2003; 3(8): [142] Stein SE, Fahr A. High temperature stabilities of hydrocarbons. Journal of Physical Chemistry 1985; 89(17): [143] Harris SJ, Weiner AM. Chemical kinetics of soot particle growth. Annual Review of Physical Chemistry 1985; 36: [144] Friedman JH. Multivariate adaptive regression splines. Annals of Statistics 1991; 19(1):1-67.

167 152 Appendix A Using TGA to investigate the impact of the pretreatment method As described in Section 3.5, before characterization, the PM was thermally treated at 500 C for 60 minutes in a thermogravimetric analyzer (TGA) under ultra high purity nitrogen in order to remove volatile matter or adsorbed unburned hydrocarbons. To investigate the impact of the pretreatment method, two TGA tests were performed: (1) performing isothermal (550 C) oxidation tests on PM samples pretreated at 100 C, 350 C, 500 C, and 650 C for 60 minutes; and (2) performing non-isothermal oxidation tests on as received PM samples and PM samples pretreated at 500 C for 60 minutes. The consistency of TGA tests and soot sample production is covered in Appendix B. Therefore, the discussion here is limited to the impact of pretreatment temperature on soot reactivity. (1) Performing isothermal (550 C) oxidation test on PM samples pretreated at 100 C, 350 C, 500 C, and 650 C for 60 minutes The pretreatment and oxidation processes were combined as listed below: 0: Start with Nitrogen 1: Ramp C/min to C 2: Isothermal for min (To stabilize the sample) 3: Ramp C/min to the pretreatment temperature 4: Isothermal for min (To remove volatile matters)

168 Temperature ( o C) 153 5: Ramp C/min to C, then ramp 5.00 C/min to C (ramp 5.00 C/min to C if the pretreatment temperature is 500 C, ramp 10 C to 550 C) 6: Isothermal for 5.00 min (Stabilize) 7: Select gas Zero Air 8: Isothermal for min PM samples collected at Mode C with split injection strategy were used for all tests. Figure A1 shows the temperature profile of the four isothermal oxidation tests with different PM pretreatment temperatures. Due to the different pretreatment process, the start of oxidation is different for the PM pretreated at 100 C, 350 C, 500 C, and the PM pretreated at 650 C. Figure A2 shows the normalized mass profile of the four isothermal oxidation tests Time (minutes) Figure A1. Temperature profiles of pretreatment and isothermal oxidation test. The start of oxidation is 153 minutes for PM pretreated at ( ) 100 C, ( ) 350 C, ( ) 500 C, and 167 minutes for PM pretreated at ( ) 650 C.

169 154 Figure A2. Mass fraction profiles of isothermal oxidation test of PM samples pretreated at ( )100 C, ( ) 350 C, ( ) 500 C, and ( ) 650 C for 60 minutes under ultra high purity nitrogen. To quantitatively investigate the impact of the four pretreatment temperatures on the soot reactivity, the oxidative rate constant of each cases was derived from 0 30 minutes and summarized in Table A1. If using the PM sample pretreated at 500 C as a reference, the difference in oxidative reactivity caused by the pretreatment temperature is less than ± 2%. The impact of pretreatment temperature on soot reactivity is much less significant than the difference caused by the start of injection or fuel composition as discussed in Chapter 5. Table A1. Comparison of oxidative rate constant of the PM pretreated at four different temperatures Pretreatment temperature ( C) Apparent rate constant (1/Pa/min)

170 155 The isothermal oxidation tests have indicated that the reactivity of the soot particles was not altered by the pretreatment process. Additionally, the result implies that the soot nanostructure, which affects the soot reactivity, was not altered by the pretreatment temperature, 500 C, in this study. (2) Performing non-isothermal oxidation test on PM sample as received and PM sample pretreated at 500 C for 60 minutes. According the Chapter 5, for a sample generated by BP15 at Mode C with split injection, heat pretreatment at 500 C under nitrogen generally removes 13-15% of the total PM mass. The removed part is defined as the volatile organic fraction (VOF). One can suspect that the volatile organic fraction and the pore area that it covers play significant roles when measuring the oxidation reactivity. To clarify this issue, non-isothermal oxidation tests were performed for the as received PM samples and pretreated at 500 C. The non-isothermal oxidation processes were combined as listed below: 0: Start with Nitrogen 1: Ramp C/min to C 2: Isothermal for min (To stabilize the sample) 3: Select gas Zero Air 4: Isothermal for 5.00min 5: Ramp C/min to C Figure A3 shows the temperature and normalized weight profiles of the two PM samples. For the PM sample as received, the mass decreases faster than the PM sample pretreated at 500 C

171 156 for 60 minutes. This observation is attributed to the loss of VOF for the PM sample as received, while the VOF of the PM sample pretreated has been removed during its pretreatment. But, it should be noted that the time to reach the 50% of their total weight were very similar. Additionally, the mass profiles after 50% mass loss were very similar, and the 100% burnout time for the two samples were very close. The non-isothermal oxidation result has two implications: (1) the VOF in the PM does not play a signification role in soot oxidation. Otherwise, the oxidation reaction of the PM as received would be completed earlier than the PM pretreated at 500 C; and (2) whether the pore structure covered by VOF or not is not a dominant factor in soot oxidation. Otherwise, the oxidation reaction of the PM pretreated at 500 C would be completed earlier than the PM as received. Figure A3. The temperature and normalized weight profile of non-isothermal oxidation test. ( ) as received PM, ( ) PM pretreated at 500 C, ( ) Temperature ( C).

172 Appendix B The repeatability of isothermal oxidation tests This section concerns the repeatability of the isothermal oxidation tests. The study is separated into two parts: (1) performing isothermal oxidation tests on different days for the same PM sample collected from the same experiment; and (2) performing isothermal oxidation tests of two PM samples collected from the same engine operating condition in two experiments. The two experiments wer sixth months apart. The samples were generated using BP15 under the engine condition and injection parameters listed in Table B1. Table B1. Engine operating modes and values of the start of injection for studying repeatability of oxidation tests SOI (degree *) Engine Engine speed Torque mode Split (min -1 ) (Nm) Pilot Main C * (+) Before Top Dead Center, (-) After Top Dead Center The pretreatment and oxidation processes were combined as listed below: 0: Start with Nitrogen 1: Ramp C/min to C 2: Isothermal for min (To stabilize the sample) 3: Ramp C/min to 500 C 4: Isothermal for min (To remove volatile matters) 5: ramp 5.00 C/min to C 6: Isothermal for 5.00 min (Stabilize)

173 158 7: Select gas to zero air 8: Isothermal for min The three samples presented for TGA data repeatability are listed as Table B2. Table B2. Summary of sampling date, analysis date, and equipment locations of five different soot isothermal oxidation tests Sample Collection date Analysis date Equipment (TGA) Apparent rate ID location * constant (1/Pa/min) 1 01/23~24/ /10/2008 Steidle 7.58* /23~24/ /15/2008 Steidle 7.10* /23~24/ /28/2009 AA 7.66* /23~24/ /18/2010 AA 7.96* /11/ /19/2010 AA 7.30*10-7 * Steidle: The Department of Material Science and Engineering (Steidle Building), The Pennsylvania State University; AA: Academic Activities Building, The Pennsylvania State University.

174 159 Figure B1. The normalized weight of five different samples during the isothermal oxidation tests. The five samples are summarized in Table B2. ( ) PM collected in 01/2008, Steidle (1), ( ) PM collected in 01/2008, Steidle (2), ( ) PM collected in 01/2008, AA (1), ( ) PM collected in 01/2008, AA (2), ( ) PM collected in 07/2008, AA Figure B1 overlays the results of the five different TGA tests. The apparent rate constant of each test can be obtained using the curve fitting algorithm described in Section 4.1. The apparent rate constants are listed in Table B-2. The mean of the apparent rate constants of all the five tests is 7.52*10-7 1/Pa/min. The standard deviation is 3.32*10-8 1/Pa/min, which is 4.4% of the mean. Therefore, in this suite of tests, it is concluded that the isothermal oxidation tests has ±4.4% uncertainty at 95% confidence [88].

175 160 Appendix C Uncertainty and repeatability of XPS The XPS of soots generated by BP15, B100, and FT generated by single injection strategy at Mode C with matched combustion phasing were repeated for two times. Both the elemental analysis by survey scan and high-resolution scan were used to investigate the repeatability and error range. Repeating the tests to a sufficient number for commenting on the confidence level (~30 times) was not the major focus of this study. Therefore, the repeatability was merely discussed based on the two tests of each sample for roughly estimating the repeatability and error range of the analysis. The oxygen content does not show any repeatable order for the three test samples. And, the standard deviation from the mean is as large as 2.2% (for FT samples). In order to comment on XPS data effectively in this work, any difference of the elemental content with 3% derived from XPS is considered insignificant.

176 161 Table C1. Summary of uncertainty and repeatability of XPS. Sample and type of scan BP15 (single injection, first test, survey scan) BP15 (single injection, first test, high resolution scan) BP15 (Single injection, second test, survey scan) BP15 (single injection, second test, high resolution scan) B100 (single injection, first test, survey scan) B100 (single injection, first test, high resolution scan) B100 (single injection, second test, survey scan) B100 (single injection, second test, high resolution scan) FT (single injection, first test, survey scan) FT (single injection, first test, high resolution scan) FT (single injection, second test, survey scan) FT (single injection, second test, high resolution scan) Oxygen content (%) Carbon content (%) Average oxygen content (%) Standard deviation (%)

177 162 Appendix D Uncertainty and repeatability of HRTEM image analysis algorithm In order to examine the uncertainty and repeatability of HRTEM image analysis algorithm, HRTEM images of soot particles generated by BP15 using single injection strategy at matched combustion phasing were analyzed. While the same grid was used, fifteen HRTEM images taken on different dates and magnitudes (0.021 nm/pixel and nm/pixel) were analyzed. Figure D1 shows the original fifteen HRTEM images of the soot with the region of interest (ROI). Figure D2 and D3 shows the fringe length and tortuosity histograms, respectively, derived from the ROIs of each image in Figure D1. According to visual comparison, the differences of the property histograms derived from different images are not significant. The media fringe length and the mean fringe tortuosity are attached on each histogram, and summarized in Table D1. The standard deviations of the median fringe length and mean fringe tortuosity are nm and 0.021, respectively. Table D2 summarizes the suite of images processing parameters for each image in Figure D1. The values in Table D2 were determined based on the condition of each images. In Chapter 5, the first image was used as a representative image for BP15 soot using single injection at matched combustion phasing (Figure 5-13(b)). The standard deviations, nm and 0.021, resulted by the repeatability analysis here were used throughout Chapter 5. When the differences of the median fringe length and mean fringe tortuosity are smaller than the standard deviations, the differences are considered insignificant.

178 163 (1) (2) (3) (4) (5) (6) (7) (8) (9)

179 164 (10) (11) (12) (13) (14) (15) Figure D1. HRTEM images of soot particles generated by BP15 using single injection strategy at matched combustion phasing. (1) (12): images taken at the magnitude of 500k (02/27/2009). (13)-(15): images taken at the magnitude of 400k (02/06/2009).