LIBRARY
Mlchl an State
Un versity

 

 

 

This is to certify that the
thesis entitled

ASSOCIATION AND DISCRIMINATION OF IGNITABLE
LIQUIDS FROM MATRIX INTERFERENCES USING
CHEMOMETRIC PROCEDURES

presented by

JAMIE MELISSA BAERNCOPF

has been accepted towards fulfillment
of the requirements for the

MS. degree in Criminal Justice

 

 

@d

Major Professor’s Signature

15 TH JUNE, 2009

 

Date

MSU is an Affirmative Action/Equal Opportunity Employer

— «- -.-.--‘- -.—-'-«¢--.--a-

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 K:IProj/Acc&Pres/CIRC/DatoDue.indd

ASSOCIATION AND DISCRIMINATION OF IGNITABLE LIQUIDS FROM
MATRIX INTERFERENCES USING CHEMOMETRIC PROCEDURES

By

Jamie Melissa Baemcopf

A THESIS

Submitted to
Michigan State University
In partial fulfillment of the requirements
For the degree of

MASTER OF SCIENCE
Criminal Justice

2009

ABSTRACT

ASSOCIATION AND DISCRIMINATION OF IGNITABLE LIQUIDS FROM
MATRIX INTERFERENCES USING CHEMOMETRIC PROCEDURES

By
Jamie Melissa Baemcopf

Ignitable liquid residues (ILRs) are extracted from fire debris and the resultant
extract is analyzed by gas chromatography-mass spectrometry. Weathering of the
ignitable liquid during burning and contributions from household matrices from which
ILRs are extracted complicate the identification of an ILR by visual assessment. As such,
this research aims to develop an objective method for identifying ILRs extracted from
fire debris to remove the subjectivity associated with visual assessment. The objective
method involves Pearson product moment correlation (PPMC) coefficients and principal
components analysis (PCA) to associate an ILR back to a neat ignitable liquid in the
presence of matrix interferences.

Firstly, the effect of GC temperature program on the association and
discrimination of diesel samples was investigated. This was assessed using PPMC
coefficients and PCA. Similar results were obtained for all temperature programs
investigated indicating that GC temperature program had minimal effect on the
association and discrimination of samples. In the second study, ILRs were extracted from
burned carpet to assess the association of the ILR to the neat liquid. Six liquids (gasoline,
diesel, lamp oil, adhesive remover, torch fuel, and paint thinner) were each spiked onto
carpet and burned. Both light and heavy burning conditions were investigated. Each
simulated ILR was successfully associated to the corresponding neat liquid using PPMC

coefficients and PCA, regardless of the extent of burning.

ACKNOWLEDGMENTS

I’d like to thank all of the people who have supported me during my time at
Michigan State while completing my Master’s degree. Thank you to Dr. Ruth Smith who
has challenged me every step of the way in this project. Without you, I know that my
research would be complete rubbish! I’ve become such a confident chemist thanks to
you!

I must also acknowledge the people in the Forensic Science Program, specifically
the chemists who have shown me so much support over the past two years! Thanks to
Patty Joiner, John McIlroy, Ruth Udey, Melissa Bodnar, Christy Hay, Tiffany Van de
Mark, Seth Hogg, and Kari Anderson who have become such great friends!

Thank you to my committee members. Thanks to Dr. Steve Dow for taking the
time to sit on my committee. Thanks to Dr. Victoria McGuffin for being on my
committee and for all the advice and expertise you’ve given during my research.

I also must thank my family for always providing support and love throughout my
education. Thanks for being there for me and encouraging me to strive towards

excellence!

iii

TABLE OF CONTENTS

 

 

 

 

 

iv

List of Tables vi
List of Figures - viii
Chapter 1 — Introduction -- - - - --l
1.1 Classification of Ignitable Liquids __________________________________________________________________________ l
1.2 Extraction of Ignitable Liquid Residues from Fire Debris ____________________________________ 3
1.3 Analysis of Ignitable Liquid Residues ____________________________________________________________________ 4
1.4 Problems in Identifying Ignitable Liquid Residues in Fire Debris ________________________ 5
1.5 Literature Review ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 7
1.5.1 Statistical and Chemometric Analysis of Ignitable Liquids _______________________________ 7
1.5.2 Factors Influencing Data Analysis Procedures __________________________________________________ 10
1.5.3 Matrix Effects ________________________________________________________________________________________________________ 11
1.6 Research Objectives _________________________________________________________________________________________________ 14
1.7 References _________________________________________________________________________________________________________________ 16
Chapter 2 — Theory - - - - - ..... - 18
2.1 Gas Chromatography-Mass Spectrometry ______________________________________________________________ 18
2.2 Data Pre-treatment Procedures ________________________________________________________________________________ 26
2.2.1 Retention Time Alignment — Peak-Matching Algorithm ___________________________________ 26
2.2.2 Retention Time Alignment — Correlation Optimized Warping Algorithm ________ 28
2.2.3 Normalization and Mean-Centering _____________________________________________________________________ 31
2.3 Pearson Product Moment Correlation _____________________________________________________________________ 32
2.4 Principal Components Analysis _______________________________________________________________________________ 33
2.5 References _________________________________________________________________________________________________________________ 36
Chapter 3 — Effect of GC Temperature Program on the Association and
Discrimination of Diesel Samples -_ 37
3.1 Introduction _______________________________________________________________________________________________________________ 37
3.2 Materials and Methods _____________________________________________________________________________________________ 38
3.2.1 Sample Collection _________________________________________________________________________________________________ 38
3.2.2 GC-MS Analysis ___________________________________________________________________________________________________ 39
3.2.3 Data Analysis _________________________________________________________________________________________________________ 41
3.3 Results and Discussion _____________________________________________________________________________________________ 42
3.3.1 Association and Discrimination of Diesel Samples Based on Total Ion
Chromatogram ________________________________________________________________________________________________________________ 45
3.3.2 Association and Discrimination of Diesel Samples Based on Alkane

' Extracted Ion Profile _______________________________________________________________________________________________________ 52
3.3.3 Association and Discrimination of Diesel Samples Based on Aromatic
Extracted Ion Profile ....................................................................................................... 55
3.3.4 Effect of Increased MS Scan Rate on Association and Discrimination of
Diesel Samples ________________________________________________________________________________________________________________ 58
3.3.5 Effect of Automated Injection Technique on Association and
Discrimination of Diesel Samples _________________________________________________________________________________ 61

3.4 Conclusions 64

...............................................................................................................

3.5 References 66

-----------------------------------------------------------------------------------------------------------------

Chapter 4 - Discrimination of Ignitable Liquid Residues from Matrix

 

 

 

Interferences Using Chemometric Procedures,“ ______________________________ 6 7
4.1 Introduction _______________________________________________________________________________________________________________ 67

4.2 Materials and Methods _____________________________________________________________________________________________ 68
4.2.1 Sample Collection _________________________________________________________________________________________________ 68
4.2.2 Neat Ignitable Liquids ___________________________________________________________________________________________ 69
4.2.3 Minimal Matrix Interferences _______________________________________________________________________________ 70
4.2.4 Increased Matrix Interferences _____________________________________________________________________________ 70
4.2.5 GC-MS Analysis ___________________________________________________________________________________________________ 71
4.2.6 Data Analysis _________________________________________________________________________________________________________ 71

4.3 Results _______________________________________________________________________________________________________________________ 72
4.3.1 Neat Ignitable Liquids ___________________________________________________________________________________________ 72
4.3.2 Minimal Matrix Interferences _______________________________________________________________________________ 73
4.3.3 Increased Matrix Interferences _____________________________________________________________________________ 84

4.4 Conclusions _______________________________________________________________________________________________________________ 90

4.5 References _________________________________________________________________________________________________________________ 92
Chapter 5 - Conclusions and Future Work ..............93
5.1 Effect of GC Temperature Program on the Association and Discrimination

of Diesel Samples ___________________________________________________________________________________________________________ 93

5.2 Discrmination of Ignitable Liquid Residues from Matrix Interferences Using
Chemometric Procedures _______________________________________________________________________________________________ 95

5.3 Future Work ______________________________________________________________________________________________________________ 96

5.4 References _________________________________________________________________________________________________________________ 98
Appendix A - PPMC Coefficients for TIC, Alkane EIP, and Aromatic EIP for
Diesel Samples Analyzed by Each Temperature Program ,,,,,,,,,,,,,,,,,,,,,,, 99
Appendix B — Total Ion Chromatograms of Neat Ignitable Liquids ,,,,,,,,,,,,,, - ,,,,,,,,,,, 121

LIST OF TABLES

Table 1.1 ASTM classification of ignitable liquids ........................................................... 2
Table 1.2 Common EIPs used to identify an ignitable liquid ............................................. 4
Table 3.1: GC temperature programs investigated ............................................................ 40

Table 3.2 PPMC coefficients between Diesels 2 and 4 for each temperature program....45
Table 3.3 Comparison of manual and pulsed autosampler injection precision ................. 61
Table 4.1 Ignitable liquids investigated ............................................................................. 69

Table 4.2 Mean PPMC coefficients i standard deviation for neat ignitable liquids
replicates (n=3) and ILR replicates (n=3) as well as between neat ignitable liquid and the
corresponding ILR (n=9) in the presence of minimal matrix interferences ...................... 80

Table 4.3 Mean PPMC :t standard deviation coefficients indicating correlations of
gasoline ILR with neat i gnitable liquids (n=9) ................................................................. 82

Table 4.4 Mean PPMC coefficients 1 standard deviation for neat ignitable liquid
replicates (n=3) and ILR replicates (n=3) as well as between neat ignitable liquid and the
corresponding ILR (n=9) in the presence of increased matrix interferences .................... 89

Table A.l PPMC Coefficients for the TIC of Diesel Samples Analyzed by
Method A ......................................................................................................................... 100

Table A2 PPMC Coefficients for the TIC of Diesel Samples Analyzed by
Method B ......................................................................................................................... 101

Table A3 PPMC Coefficients for the TIC of Diesel Samples Analyzed by
Method C ......................................................................................................................... 102

Table A4 PPMC Coefficients for the TIC of Diesel Samples Analyzed by
Method D ......................................................................................................................... 103

Table A5 PPMC Coefficients for the TIC of Diesel Samples Analyzed by
Method B ......................................................................................................................... 104

Table A6 PPMC Coefficients for the TIC of Diesel Samples Analyzed by
Method F ......................................................................................................................... 105

Table A7 PPMC Coefficients for the TIC of Diesel Samples Analyzed by
Method D Using a Pulsed Automated Injection Technique ............................................ 106

vi

Table A8 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed
by Method A .................................................................................................................... 107

Table A9 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed
by Method B .................................................................................................................... 108

Table A.10 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed
by Method C .................................................................................................................... 109

Table A.11 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed
by Method D ................................................................................................................... 110

Table A.12 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed
by Method B .................................................................................................................... 111

Table A.13 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed
by Method F .................................................................................................................... 112

Table A.14 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed
by Method D Using a Pulsed Automated Injection Technique ....................................... 113

Table A.15 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed
by Method A ................................................................................................................... 114

Table A.l6 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed
by Method B ................................................................................................................... 115

Table A. l 7 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed
by Method C ................................................................................................................... 116

Table A. l 8 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed
by Method D ................................................................................................................... 117

Table A.19 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed
by Method B ................................................................................................................... 118

Table A.20 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed
by Method F ................................................................................................................... 119

Table A2] PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed
by Method D Using a Pulsed Automated Injection Technique ...................................... 120

vii

LIST OF FIGURES

Figure 2.1 Schematic of a gas chromatograph .................................................................. 19
Figure 2.2 Schematic of a mass spectrometer ................................................................... 22
Figure 2.3 Diagram of a quadrupole mass analyzer .......................................................... 23
Figure 2.4 A) TIC, B) EIC of m/z 57, and C) alkane EIP of a diesel sample .................... 25

Figure 2.5: Diagram depicting the alignment of the last segment in a sample (m=5) to the
corresponding segment in the target (m=5) with a slack of 1 (t=1). { Indicates interpolated
data points .......................................................................................................................... 30

Figure 2.6: Diagram depicting the alignment of the penultimate segment ........................ 31

Figure 3.1 TIC of one diesel sample analyzed by A) Method A, B) Method D, and C)
Method F ........................................................................................................................... 43

Figure 3.2: A) Alkane EIP and B) aromatic EIP of one diesel sample analyzed by
Method D .......................................................................................................................... 44

Figure 3.3: A) Well aligned hexadecane peak in the TIC of diesel samples analyzed by
Method D, and B) poorly aligned hexadecane peak in the TIC of diesel samples analyzed
by Method F ...................................................................................................................... 46

Figure 3.4: Scores plot of TIC for A) Method A, B) Method D, and C) Method F. Diesel
1 (6), Diesel 2 (I), Diesel 3 (A), Diesel 4 (0), Diesel 5 (O) ............................................ 48

Figure 3.5: Loadings plot of the first principal component of diesel samples analyzed by
Method D, identifying the components contributing the most to the variance in the
sample set: 1)1-ethyl, 3-methylbenzene, 2) 1, 3, 5-trimethylbenzene, 3) undecane,
4) dodecane, and 5) tridecane ........................................................................................... 50

Figure 3.6: A) Loadings plot of the first principal component of diesel samples analyzed
by Method F, B) a magnified portion of the loadings plot from A, which demonstrates

derivative-shaped curves indicative of misalignments ..................................................... 51

Figure 3.7: Scores plot of the alkane profile for Method D. Diesel 1 (9), Diesel 2 (I),
Diesel 3 (A), Diesel 4 (0), Diesel 5 (O) ........................................................................... 53

Figure 3.8: Loadings plot of the first principal component based on the alkane EIP of
diesel samples analyzed by Method D ............................................................................. 54

viii

Figure 3.9: Scores plot of the aromatic profile for Method D. Diesel 1 (0), Diesel 2 (I),
Diesel 3 (A), Diesel 4 (0), Diesel 5 (O) ........................................................................... 56

Figure 3.10: Loadings plot of the first principal component based on the aromatic EIP of
diesel samples analyzed by Method D, identifying the components contributing the most
to the variance in the sample set: 1)1-ethyl-3-methylbenzene and 2) 1, 3, 5-

trimethylbenzene .............................................................................................................. 58

Figure 3.11: A) Section of aligned TICs of diesel samples analde by Method F with a
scan rate of 2.91 scans per second showing poor alignment. B) Loadings plot showing the
derivative-shaped curve indicative of the misalignment in A. C) Same section of the TIC,
showing the aligned chromatograrns of diesel samples analyzed by Method F with a scan
rate of 5.51 scans per second. D) Loadings plot without any derivative-shaped curves
indicating an improvement in alignment .......................................................................... 60

Figure 3.12: Scores plot of TIC for A) manual injection and B) pulsed autosampler
injection. Diesel 1 (0), Diesel 2 (I), Diesel 3 (A), Diesel 4 (0), Diesel 5 (O) ................. 63

Figure 4.1: Chromatograms of A) lightly burned carpet, B) gasoline spiked onto carpet
then lightly burned, and C) neat gasoline. Major components are labeled: 1) Cn-Clz
isoparaffinic/naphthenic components, 2) C13, C14 normal alkanes, 3) diester related to
adipic acid, 4) Cz-alkylbenzenes, 5) C3—alkylbenzenes .................................................... 74

Figure 4.2: A) Full view and B) magnified view of the scores plot for six ignitable liquids
and the corresponding ILRs in the presence of minimal matn'x interferences. Neat liquids
are indicated by filled symbols and lLRs are indicated by open symbols. (I) Diesel, (O)
Gasoline, (A) Adhesive Remover, (D) Lamp Oil, (V) Paint Thinner, (O) Torch Fuel,
('73?) Burned Carpet (unsp1ked)76

Figure 4.3: Loadings plots for A) PCI and B) PC2 for six ignitable liquids and the
corresponding ILR in the presence of minimal matrix interferences. Major components
are labeled: 1) ethylbenzene, 2) o-xylene, 3) p-xylene, 4) C12, 5) C13, and 6) C14..........78

Figure 4.4: Chromatograms of A) neat diesel, B) diesel ILR, and C) neat torch fuel,
illustrating the similar chemical composition between the ignitable liquids. Major
components are labeled: 1) aromatic components, 2) C“, 3) C12, 4) C13, and 5) C14 ...... 83

Figure 4.5: Chromatograms of A) heavily burned carpet, B) gasoline spiked onto carpet
then heavily burned, and C) neat gasoline. Major components are labeled: 1) styrene,
2) benzaldehyde, 3) Cz-alkylbenzenes, 4) C3-alkylbenzenes ........................................... 85

Figure 4.6: A) Full view and B) magnified view of scores plot for six ignitable liquids
and the corresponding ILRs in the presence of increased matrix interferences. Neat
liquids are indicated by filled symbols and ILRs are indicated by open symbols. (I)
Diesel, (0) Gasoline, (A) Adhesive Remover, (>) Lamp Oil, (V) Paint Thinner, (O)
Torch Fuel, (if) Burned Carpet ....................................................................................... 86

ix

Figure B] Total ion chromatogram of gasoline with major components labeled .......... 122
Figure B.2 Total ion chromatogram of diesel with major components labeled .............. 122
Figure B.3 Total ion chromatogram of lamp oil with major components labeled .......... 123

Figure B.4 Total ion chromatogram of adhesive remover with major components
labeled ............................................................................................................................. 123

Figure B.5 Total ion chromatogram of torch fuel with major components labeled ........ 124

Figure B.6 Total ion chromatogram of paint thinner with major components labeled...124

CHAPTER 1

INTRODUCTION

Ignitable liquids are often used as accelerants in arson cases to increase the rate
and spread of the fire and cause maximum damage. Thus, the presence of an ignitable
liquid in fire debris can indicate a deliberate, rather than accidental, fire. Ignitable liquids
used as accelerants cover a wide range of products that include gasoline, diesel, and
kerosene. Generally, the goal of a fire debris analyst is to identify any ignitable liquid
present in fire debris. With this goal in mind, research into developing a successful

method for correctly identifying ignitable liquids in fire debris is essential.

1.1 Classification of Ignitable Liquids

The American Society for Testing and Materials (ASTM) has defined eight
classes of ignitable liquids based on chemical composition: gasoline (including gasohol),
petroleum distillates (including de-aromatized distillates), isoparaffinic products,
aromatic products, naphthenic-paraffinic products, normal alkanes products, oxygenated
solvents, and others-miscellaneous [l]. The eight classes are divided into three sub-
classes based on carbon content: light (C4-C9), medium (Cg-C13), or heavy (Cg-C204.) The
sub-classes are flexible so that an ignitable liquid may be classified as “light to medium”
or “medium to heavy” as necessary. The eight classes are described further in Table 1.1

with several examples of products from each class given [1].

Table 1.1: ASTM classification of ignitable liquids.

 

088.: 338% 32:5

083: 038% 85>

800.60% Beam 080m

005820

385822

 

 

 

 

 

 

 

 

 

 

 

 

 

. £00005 30:05 050m £00005 80:05 050w
flosvoi 80:05 080m 50:00.:— 0:::0&:._. .3038.— ::0:on::00 035m on 0015050 8:50 :05 02:3..— -8050
80.9080 80 80:8 :0> 0 :0. 8.000 00: 0:03:
.fifiuom _wwm:v:~ ”WHOWE fl.8>_:_m_.cv<.w_0:m .Hgmm :0 0:020: 0020.: 98:09:00 wager—om
. . . . . . . 8.3.: ”8:300. .828 63:00? nounswg
805:5. 00:000..— 080m 8:003 080m 8:202 3:002 $60—05 30:00:: 0000:8090
80:0,: 8.000
. a 5300 . on
.950: 0830wa .0 ”Such :0: ”W0 .m§:0m.000“MH0W wESSthM 004:” WWW0 n: h : $260.5 853—2-:
£5 06:00 080m _.0 8:00 m m _ m . . = . 0:: _ O
. 0:0 :53 0:03: :0 Haggai 850:0:
£NM>wwm EMZMHMWHRWHWS .8_o£0> 0200008— 060m Mwsvwénfiofiflm 8050:: 320355005 0:96 8:355 0MMEEA
_ .:0> v. . u m .8235 1:00:55 080m 0 3 :88: _ U 0:: 0:580:03 50:0 858m 2:05: :2
8>u€0< .25 .8_0£0> 30:00.5 0080
3:038 $2005 .0389: 02000005 080w .m:0>_0m 0:03.09 .8:0_>X 80000—0 33.— EScEmmmE :0 «:03: 8:009:00 m s o 0 :8
.m0_0E0> 02050005 080m mafia—U £3025 800005 0208003. 080m .8055 0320:: ”00:00:58 03:80.3 3 u .5 «0 o._.<
05m 0>.:0::3:< 080w 5:55 0:: “Eu.— 080m
3:023 .80:0H 8300 080m 8:025 3:803 “:03: :0 0:02.33me 850:0: mu 0 300 .E
3300.5 38:08:50 080m 80:55. :55 080m .80 :0::_>< 0:: 853—: ABE—«5:005 0:: «Hanan
. . .thflm 1:00:30 080m . . 0:08 ”3:580:00: 55.0 00:0:Em . a u
m: :00: 0:5 :0> 0 w:.::0 b 080 m 0: 35:8 050 35:09:00
3m...” ”w. m a a”. . gm. h. m . __._ .s. as. we... ._.§.........a.§s seas
. 0: _ 0 "— Badcommb .80 Ewnwo .am 0 w 0:. a.” .Hohuo .U m 68560000: 0:03:90 ago—chum
_ n— _ 8.5 v— tfim _ a :U E m 5m. _ :0.— 82 ”8:00:00 «0 8t?" mama—080m
2 4 w a m 8:000:80 0220:: 80:? m _0:00.0w
U- U 0 :8 0:: 5 =83»: w: 0::08 :8i 0:: 8880.53-00 0:: .mo £030.: mac—85
=a - 2:85
080000 Es: 30-000 :58: E 20-000 Em: 828:8: as:

 

 

 

1.2 Extraction of Ignitable Liquid Residues from Fire Debris

Several procedures for the extraction of ignitable liquid residues (ILRs) from fire
debris are outlined by ASTM [2, 3, 4]. Three extraction procedures are commonly used
by fire debris analysts: dynamic headspace extraction, passive headspace extraction, and
solvent extraction. In both dynamic and passive headspace extraction, the fire debris is
placed in a sealed, unlined paint can or nylon bag and heated in an oven at 50-80 °C for
2-24 hours [2]. For dynamic headspace, a sample of the headspace is taken from the
heated container with a gas-tight syringe and directly analyzed by gas chromatography-
mass spectrometry (GC-MS) [3]. For passive headspace, an activated carbon strip is
suspended in the container before heating and removed after heating [2]. With heating,
the headspace of the container becomes saturated with the volatile components of an
ignitable liquid and the volatile components adsorb onto the carbon strip. After heating,
the components are eluted from the strip with 50-1000 “L of organic solvent, such as
carbon disulfide, and the resulting extract is analyzed by GC-MS [2]. Dynamic and
passive headspace extraction are appropriate for very volatile ignitable liquids such as
gasoline. However, this extraction technique discriminates against less volatile ignitable
liquids that may contain components with boiling points above the extraction
temperature. These components do not volatilize or adsorb effectively onto the carbon
strip; hence, the resulting extract and chromatogram are not representative of the
ignitable liquid and may lead to misidentification.

In solvent extraction, the fire debris, or a representative sample if the debris is too

large, is immersed in an organic solvent such as carbon disulfide, pentane, or diethyl

ether. The fire debris is generally extracted for less than a minute, and then the solvent is
decanted, filtered if necessary, and concentrated to approximately one milliliter for
subsequent analysis by GC-MS [4]. Solvent extraction can be used for any ignitable
liquid, but it is most suitable for less volatile ignitable liquids because the extraction does

not depend on the volatility of the sample.
1.3 Analysis of Ignitable Liquids Residues

Ignitable liquid residues extracted from fire debris are routinely analyzed by GC-
MS to generate total ion chromatograms (TIC) as well as extracted ion profiles (EIPs) of
characteristic compound classes. Common EIPs compiled by fire debris analysts include
profiles of alkane, aromatic, indane, olefin/cycloparaffin, and polynuclear aromatic
compound classes. Table 1.2 lists the mass-to-charge (m/z) ratios used to compile each
EIP.

Table 1.2 Common EIPs used to identify an ignitable liquid.

 

 

Characteristic Class Mass-to-Charge (m/z) Ratios
Alkane 57 + 71 + 85 + 99
Aromatic 91 +105 +119 +133
Indane 117+131+145+159
Olefin/Cycloparaffin 55 + 69 + 83 + 97
Polynuclear Aromatic 128 + 142 + 156

 

The TIC and EIPs are visually compared to a reference collection to identify the
class and type of ignitable liquid present. For comparisons of ILRs, fire debris analysts
often use in-house reference collections or national databases, such as the Ignitable
Liquids Reference Collection maintained by the National Center for Forensic Science [5].

The visual comparison of an ILR to a chromatogram in the reference collection is based

on pattern recognition, which involves identifying components of an ILR in relation to
other components, rather than the presence of a specific component [6]. For example,
trimethylbenzene (TMB) is contained in gasoline, but the presence of TMB in a sample
does not exclusively indicate gasoline because TMB is contained in other ignitable
liquids and also matrices from which ILRs are extracted. The identification of gasoline is
based on the relationship of TMB to other aromatic components in gasoline, such as C2-

alkylbenzenes, C3-alkylbenzenes, and methylnaphthalenes.

1.4 Problems in Identifying Ignitable Liquid Residues in Fire Debris

Visual assessment of chromatograms and comparison to a reference collection is
notoriously difficult and often subjective, for a number of reasons. The two main
problems associated with visual assessment are weathering or evaporation of the ignitable
liquid during the fire and contributions from the fire debris matrix to the overall
chromatogram of the extracted ILR.

Volatile components of the ignitable liquid, such as aromatic components, are lost
during the burning process, making the TIC different from the chromatogram of the neat
liquid. The loss of volatile components is reflected in the chromatogram by the absence
or decreased abundance of the peaks at the beginning of the chromatogram. This also
leads to differences in peak ratios, as some components are lost or diminished and other
components are unaffected by the burning process. Loss of volatile components makes it
difficult to successfully identify ignitable liquids by pattern recognition. In the previous

example for gasoline, the ratios of alkylbenzenes and methylnaphthalenes are assessed to

identify a gasoline ILR. However, a change in the ratios of these components as a result
of burning can cause a weathered gasoline to be misidentified by only visual assessment
of the TIC.

Fire debris matrices from which ILRs are routinely extracted, such as clothing,
carpet or upholstery, often contain hydrocarbons that are also present in ignitable liquids.
Chromatograms of unburned and burned matrices can be similar to that of an ignitable
liquid due to peaks from inherent hydrocarbons, thermal degradation, or pyrolysis
products [7]. Inherent hydrocarbons are generally due to the raw material of the matrix or
are introduced during the manufacturing process. For example, the chromatogram of
unburned polyester carpet contains normal alkane components decane, undecane,
dodecane, and tridecane (Clo-C13), which are also present in ignitable liquids in the
petroleum distillate or alkane classes such as diesel or lamp oil. Thermal degradation and
pyrolysis products result when components of the matrix are exposed to heat and begin to
break down with and without oxygen, respectively [8]. The chromatogram of burned
polyester carpet contains thermal degradation and pyrolysis products such as styrene,
benzaldehyde, and branched alkenes. The presence of such interfering components
complicates the chromatogram of an ILR, which can make identification difficult or even
mask the presence of the ILR. As such, it is important for a fire debris analyst to analyze
control samples of the matrix that do not contain any ignitable liquid to identify any
inherent hydrocarbons. Furthermore, an objective method for identifying an ILR in the
presence of matrix interferences could potentially remove the problems associated with
matrix interferences. The objective method could make the identification of an ILR

simpler, less subjective and minimize the risk of false positive or negative identifications.

1.5 Literature Review

Studies concerning fire debris analysis are widespread in the literature. Much of
the research has addressed the subjective nature of fire debris analysis and focused on
developing more objective ways for identifying ILRs in fire debris. Several studies have
concentrated on association and discrimination of ignitable liquids by statistical and
Chemometric procedures to remove the subjectivity inherent in identifying ILRs. These
studies have mainly focused on gasoline or diesel because they are commonly
encountered in arson cases or environmental applications [9-15]. With the frequent use of
Chemometric procedures, the importance of data pretreatment prior to Chemometric
procedures has been explored [16-20]. Studies have also investigated the potential
interferences from both unburned and burned matrices and their effect on the
interpretation of ILR chromatograms in fire debris analysis [7, 13, 21-24]. However, no
statistical or Chemometric procedures have been applied to differentiate ILRs from matrix

interferences, which may make the interpretation of ILR chromatograms more objective.

1.5.] Statistical and Chemometric Analysis of Ignitable Liquids

Studies in the literature have used statistical and Chemometric procedures to
associate and discriminate ignitable liquids, both within and between classes of liquids,
often to identify the source. However, these studies have focused on specific
characteristic components, peak ratios of characteristic components, or small sections of
the chromatogram to identify or differentiate ignitable liquids, rather than the entire

chromatogram which may offer more discrimination [9-12]. In a study by Barnes and

coworkers, 16 gasoline samples that were 50% evaporated were successfully associated
to the neat counterpart using six ratios of aliphatic components in the isobutene to
methylnaphthalene region of the chromatogram [9]. Ten gasoline samples that were 75%
evaporated were successfully associated to the neat counterpart using only four ratios of
aliphatic components in the same region. Ratios of components were selected based on
their reproducibility within gasoline samples at different levels of evaporation and
variability between gasoline samples. Sandercock and Du Pasquier used the Co- to C2-
naphthalene compounds present in gasoline to chemically fingerprint unevaporated
samples [10]. By applying principal components analysis (PCA) to the chromatographic
range, the authors found that the Co- to Cz-naphthalene compounds could be used to
group 35 gasoline samples of different grades and from different service stations into 32
groups. Of the 32 groups, 30 groups contained only one gasoline sample, one group
contained two gasoline samples, and one group contained three gasoline samples. Further
studies by the Sandercock and Du Pasquier used PCA to classify 35 gasoline samples at
different levels of evaporation (25, 50, 75, and 90% evaporated by weight) from 24
service stations into 18 unique groups [11]. Of the 18 groups, 11 contained only one
gasoline sample at all evaporation levels and the remaining 24 gasoline samples were
spread among 7 groups. A study by Gaines et al. used PCA to determine that as few as
nine components of diesel, those that contribute most to the variance, could be used to
differentiate 14 diesel samples from different refineries [12]. However, none of these
studies incorporated the entire chromatogram, which offers more discrimination among

samples.

Studies have also applied other Chemometric procedures to associate and
discriminate ignitable liquids [13-14]. Tan and coworkers employed soft independent
modeling of class analogy (SIMCA), a supervised learning technique related to PCA, to
correctly predict the classification of 51 ILRs extracted from unburned wood and carpet
matrices [13]. In the study, the ignitable liquids were evaporated at various temperatures
and for different lengths of time. Uneven evaporation of components in the ignitable
liquid and low spike volumes affected successful classification. Doble et al. used PCA as
well as artificial neural networks (ANN) to classify 88 premium and regular gasoline
samples according to grade [14]. Both PCA and ANN were successful in classifying each
gasoline sample as premium or regular, but ANN was able to further classify 97% of the
gasoline samples by the season in which they were collected (summer or winter).

Previous research concerning the Chemometric analysis of ignitable liquids has
shown that important association and discrimination information is not only contained in
the TIC, but also in extracted ion chromatograms (EICs) or EIPs of characteristic
compound classes [15]. Hupp et al. showed that 25 diesel samples from different service
stations could be separated into four groups according to brand by applying PCA to the
TICs. More groups were observed when PCA was applied to the EICs corresponding to
alkane (m/z 57) and aromatic components (m/z 91), suggesting further discrimination of
the diesel samples. Hupp et al. also examined PC loadings plots to identify the chemical
components that were contributing most to the variance among samples. The normal
alkane and aromatic components provided the greatest discrimination among diesel

samples. This work indicated that additional characteristic compound classes and EIPs

should be investigated to determine the potential for further association and

discrimination of ignitable liquids.

1. 5. 2 Factors Influencing Data Analysis Procedures

Statistical and Chemometric procedures are employed to remove the subjectivity
in comparing chromatograms, but it is also important to ensure that only chemical
variation is being compared, rather than artificial variation introduced by an instrument or
analyst. Two important factors that can affect subsequent data analysis procedures are
instrumental parameters and data pretreatment procedures.

Studies have focused on pretreatment of chromatographic data; however, no
studies have investigated the effect of GC temperature program. The GC temperature
program affects total analysis time and chromatographic resolution. For crime labs, it is
important to have a fast analysis time to avoid case backlogs, but the loss of
chromatographic resolution as a result of fast GC temperature ramps may compromise
the potential discriminatory information obtained from a chromatogram. As such, it is
important to find the optimum GC temperature program that offers a compromise
between analysis time and resolution.

Several data pretreatment procedures exist for chromatographic data after the data
have been collected and prior to data analysis. The most commonly investigated data
pretreatment procedure is the use of a retention time alignment algorithm [16-20]. In
chromatographic analyses, several different retention time alignments are available to
correct normal instrumental drift in retention times, including peak-matching, correlation

optimized warping (COW), and dynamic time warping (DTW) algorithms [16-18]. Each

10

algorithm aligns chromatograms using a different method which should be considered
when choosing an appropriate alignment algorithm. For example, some aligmnents may
be more appropriate for chromatograms of very complex samples containing many
chemical components such as the peak matching algorithm developed by Johnson et al.
which was developed using diesel samples. The COW algorithm can also be used to align
complex samples and is capable of aligning chromatograms that have different numbers
of data points. Other data pretreatment procedures that are often applied to
chromatographic data include normalization and mean-centering [19-20]. Normalization
is a scaling method to ensure that no particular sample in a data set is weighted more
heavily in subsequent data analysis. A commonly used normalization method is a peak
area normalization, which corrects variation in injection volume to ensure that all
chromatograms in a data set have similar abundances and are not weighted differently.
Mean-centering is another data pretreatment procedure used prior to PCA. Mean-
centering redefines the mean of the data set as zero to visualize how the data vary from

zero .

1. 5. 3 Matrix Interferences

Matrix interferences cause difficulties for fire debris analysts and studies
concerning matrix interferences are well represented in the literature, but no statistical or
Chemometric procedures have been used to discriminate an ILR from matrix interferences
[7, 13, 21-24]. Lentini et al. demonstrated potential matrix interferences in a wide range
of household products (e. g. furniture polish, terry cloth towel, printed T-shirt, shoes,

newspaper, magazines, etc.) that produced chromatographic patterns similar to those of

11

an ignitable liquid or ILR [7]. For example, several matrices and household solvents such
as spandex shorts, newspaper, and lemon oil furniture polish contained heavy normal
alkanes that could complicate the identification of petroleum distillates such as diesel or
kerosene. However, the matrices were not burned as would be the case in a real fire. Tan
et al. investigated potential interferences from three types of wood and three types of
carpet on the identification of 51 ignitable liquids from five ASTM classes [13]. All
ignitable liquids extracted from spiked matrices were correctly classified using SIMCA
and PCA. However, similar to the study by Lentini et al., the matrices were not burned,
so potential interferences from pyrolysis and degradation products were not explored.

Borusiewicz et al. conducted a study examining several factors affecting the
detection of an ignitable liquid in fire debris, such as type of ignitable liquid, type of
burned matrix, burn time, and air availability [21]. The study investigated five ignitable
liquids including gasoline, kerosene, and diesel and three burned matrices (carpet,
deciduous wood, and chipboard), which had varying matrix contributions to the
chromatogram. Through visual assessment of chromatograms, the authors found that the
most important factor in detecting the presence of an ignitable liquid residue was the type
of burned material, due to the absorbent nature of the matrix. Traces of ignitable liquids
were identified in nearly all of the samples of carpet, regardless of burn time, whereas
traces were identified in only a few of the samples of deciduous wood and none of the
samples of chipboard.

Combustion and pyrolysis products extracted from burned matrices have also
been reported in the literature [22-24]. Almirall and Furton studied a variety of burned

matrices including carpet, wallpaper, synthetic flooring, and packaging materials [22].

12

The authors showed that target compounds for the identification of ILRs such as
alkylbenzenes, naphthalenes, and various hydrocarbons were often found as combustion
or pyrolysis products of burned matrices and hence, could potentially interfere with the
identification of an ILR. However, the chromatographic patterns of the target compounds
were different for the matrices and the ILRs. Femandes and coworkers investigated the
presence of pyrolysis products in 15 different burned household items [23]. The study
showed that unburned and burned newspaper, adhesives, and items finished with lacquer
or polish contained some components including alkanes and alkylbenzenes that could
lead to a false identification of an ILR. However, the matrices in the studies by Femandes
et al. and Almirall et al. were not spiked with an ignitable liquid, so the effect of matrix
interferences on the identification of an ILR was not demonstrated. Bertsch evaluated the
potential for pyrolysis products obtained from carpet samples to interfere with the
identification of an ignitable liquid [24]. In the study, visual assessment of
chromatograms showed pyrolysis products such as alkylbenzenes, naphthalenes, and
other aromatic hydrocarbons in simulated and actual fire debris samples that could
interfere with the identification of gasoline. However, no statistical or chemometric
procedures were used in this study.

From these studies, the potential for interferences from inherent hydrocarbons,
thermal degradation products, and pyrolysis products in matrices is clear. The purpose of
these studies has mainly been to identify interfering components in unburned and burned
matrices. Some studies, such as the study by Tan and coworkers, have applied statistical
and chemometric procedures to differentiate an unburned ignitable liquid from matrix

interferences [13]. However, none of these studies have investigated the potential

13

association of ILRs to neat ignitable liquids in the presence of matrix interferences or

approached the development of an objective method for use in fire debris analysis.

1.6 Research Objectives

The overall aim of this research was to develop an objective method to associate
ILRs to neat ignitable liquids while differentiating from matrix interferences. Statistical
and chemometric procedures, including Pearson product moment correlation (PPMC)
coefficients and PCA were applied to the full chromatogram to remove the subjectivity
associated with visual comparisons. The first step in this research was to investigate the
effect of GC temperature program on the association and discrimination of diesel
samples. One type of ignitable liquid was chosen for this initial study because successful
association and discrimination of samples with very similar chemical composition
indicated that this procedure would be successful with ignitable liquids from different
classes. The optimum GC temperature program was determined and then used for all
further analyses. In a second study, six ignitable liquids were each spiked on to samples
of carpet then burned and analyzed. Light and heavy burning conditions were
investigated to determine the effect of increased matrix interferences. Both PPMC
coefficients and PCA were used to assess the association of an ILR back to its neat liquid
in the presence of matrix interferences.

In the future, the objective method developed in this research may be
implemented in a forensic setting to aid fire debris analysts in the identification of ILRs.

With more work, the objective method can be made into commercially available software

14

for easy integration into forensic laboratories. It is hoped that, with the implementation of
this new data analysis method for fire debris, the occurrence of false positives and

negatives in the identification of ILRs will be reduced.

15

1.7

10.

11.

12.

13.

References

American Society for Testing and Materials, ASTM E 1618-06e1. Annual Book of
AS TM Standards 14. 02.

American Society for Testing and Materials, ASTM E 1412-07. Annual Book of
ASTM Standards 14. 02.

American Society for Testing and Materials, ASTM E 1388-00. Annual Book of
ASTM Standards 14.02.

American Society for Testing and Materials, ASTM E 1386-00(2005). Annual Book
of ASTM Standards 14. 02.

National Center for Forensic Science, Ignitable Liquids Reference Collection.
http://ilrc.ucf.edu/ (Accessed August 2008).

Newman, R. Interpretation of laboratory data. In: Nic Daeid N, editor. Fire
investigation. Boca Raton: CRC Press, 2004; 155-190.

Lentini JJ, Dolan JA, Cherry C. The petroleum-laced background. J Forensic Sci
2000; 45:968-89.

Stauffer, E. Sources of interferences in fire debris analysis. In: Nic Daeid N, editor.
Fire investigation. Boca Raton: CRC Press, 2004; 155-190.

Barnes AT, Dolan JA, Kuk RJ; Siegel JA. Comparison of gasolines using gas
chromatography-mass spectrometry and target ion response. J Forensic Sci 2004;
49:1018-1023.

Sandercock PML, Du Pasquier B. Chemical fingerprinting of unevaporated
automotive gasoline samples. Forensic Sci Int 2003; 134:1-10.

Sandercock PML, Du Pasquier E. Chemical fingerprinting of gasoline 2. Comparison
of unevaporated and evaporated automotive gasoline samples. Forensic Sci Int 2004;
140: 43-59.

Gaines RB, Hall GJ, Frysinger GS, Gronlund WR, Juaire KL. Chemometric
determination of target compounds used to fingerprint unweathered diesel fuels.

Environ Forensics 2006; 7:77-87.

Tan B, Hardy JK, Snavely RE. Accelerant classification by gas chromatography/mass
spectrometry and multivariate pattern recognition. Anal Chim Acta 2000; 42:37-46.

16

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

Doble P, Sandercock M, Du Pasquier E, Petocz P, Roux C, Dawson M. Classification
of premium and regular gasoline by gas chromatography/mass spectrometry, principal
component analysis and artificial neural networks. Forensic Sci Int 2003; 132:26-39.

Hupp AM, Marshall LJ, Campbell DI, Waddell Smith R, McGuffin VL.
Chemometric analysis of diesel fuel for forensic and environmental applications.
Anal Chim Acta 2008; 606:159-71.

Johnson KJ, Wright BW, Jarman KH, Synovec RE. High-speed peak matching
algorithm for retention time alignment of gas chromatographic data for chemometric
analysis. J Chromatogr A 2003; 996: 141-55.

Nielsen NPV, Carstensen JM, Smedsgaard J. Aligning of single and multiple
wavelength chromatographic profiles for chemometric data analysis using correlation
optimized warping. J Chromatogr A 1998; 805: 17-35.

Tomasi G, van den Berg F, Andersson C. Correlation optimized warping and
dynamic time warping as preprocessing methods for chromatographic data. J
Chemometr 2004; 18:231-241.

Brereton, RG. Applied chemometrics for scientists. Hoboken, NJ: John Wiley &
Sons, Ltd, 2007.

Miller JN, Miller JC. Statistics and chemometrics for analytical chemistry. 4th ed.
Harlow, England: Pearson Education Limited, 2000.

Borusiewicz R, Zieba-Palus J, Zadora G. The influence of the type of accelerant, type
of burned material, time of burning and availability of air on the possibility of
detection of accelerants traces. Forensic Sci Int 2006; 160:115-126.

Almirall JR, F urton KG. Characterization of background and pyrolysis products that
may interfere with forensic analysis of fire debris. J Anal Appl Pyrol 2004; 71 :51-67.

Femandes MS, Lau CM, Wong WC. The effect of volatile residues in burnt householf
items on the detection of fire accelerants. Sci Justice 2002; 42:7-15.

Bertsch W. Volatiles from carpet: a source of frequent misinterpretation in arson
analysis. J Chromatogr A 1994; 674:329-333.

17

CHAPTER 2

THEORY

2.1 Gas Chromatography-Mass Spectrometry

Chromatography is a technique used to separate chemical components of a sample
mixture [1]. Gas chromatography (GC) is commonly used in the forensic analysis of
several different classes of evidence including controlled substances, questioned
documents, and fire debris. Gas chromatography coupled with a mass spectrometer
detector (GC-MS) is the universally accepted technique for the analysis of ignitable
liquids and fire debris and therefore is the focus of this section [2-3].

Chromatography techniques use a mobile phase and a stationary phase to separate
the components of a sample mixture. The components or analytes are carried through the
stationary phase by the mobile phase. As the analytes interact with the stationary phase,
they are each retarded to a different degree based on their affinity for the stationary
phase, causing separation.

The separation is based on different factors depending on the type of
chromatography. In GC, the separation is based on differences in boiling points of the
analytes as well as differences in affinity for the mobile and stationary phases. The
mobile phase is an inert gas, commonly helium, and the stationary phase is a thin liquid
film which coats the inside of a capillary column. Several types of detectors are available
to detect the analytes after separation, including a mass spectrometer which is used in

GC-MS analyses.

18

A GC-MS consists of several components: the carrier gas (mobile phase),
injection port, column (containing stationary phase), and detector, which are depicted in

Figure 2.].

Injection port

Ii F

A Regulator

 

 

 

 

 

l

Detector

 

 

 

 

Column Oven

 

 

 

 

Carrier Gas Supply

Figure 2.1: Schematic of a gas chromatograph.
Generally, one microliter of liquid or gaseous sample is drawn up into a syringe followed
by one microliter of air, then introduced into the heated injection port. The plug of air is
necessary to avoid volatilization of the sample in the needle before injection onto the
column. The injection port is heated (150-300 °C) to ensure that liquid samples are
vaporized upon injection. For any sample analyzed by GC, it is important that the
samples are volatile and thermally stable so that they are vaporized upon injection
without thermal degradation. The injection port contains a split valve to allow samples to
be injected in split or splitless mode. For a split injection, the analyst sets a ratio to
determine the amount of sample that goes onto the column compared to the amount that
is vented to waste. For example, a split ratio of 50:1 indicates that one part of the sample
is directed onto the column for every 50 parts that are diverted to waste. This is

commonly used for concentrated or highly contaminated samples to avoid overloading

19

the column, which subsequently results in poor chromatography. For a splitless injection,
the split valve is closed and the entire sample is directed onto the column for separation
and detection. This mode of injection is generally used for dilute samples or to detect
trace levels of an analyte of interest. A pressure pulse can also be used at the time of
injection. In a pulsed injection, the pressure in the inlet is increased to a specified
pressure for a specified amount of time. A pulsed injection is sometimes employed to
ensure that a sample that contains components with a wide range of boiling points enters
the column in a small band to prevent peak splitting or band broadening and make the
injection more reproducible.

After the sample is vaporized in the injection port, the flow of the carrier gas
sweeps the sample onto the column. The column has a thin liquid film of stationary phase
coating the interior, which is typically a few tenths of a micrometer thick. The thickness
and polarity of the stationary phase can vary depending on the analytes to be separated.
The polarity of the stationary phase can greatly affect the separation and the retention of
analytes. The stationary phase commonly used in crime labs is polydimethylsiloxane,
which is a nonpolar phase that is good for general use [1].

The column is stored in a temperature-controlled oven. For some samples, the
temperature of the oven can be maintained at a constant temperature for the entire
analysis, known as an isothermal analysis. It is important that the oven temperature be
kept above the boiling point of the least volatile analyte to prevent the condensation and
loss of the analyte inside the column. Isothermal analyses cannot be used for sample
mixtures containing components with a wide range of boiling points. In such cases, the

oven temperature can be ramped at different rates throughout the analysis such as 2

2O

°C/min or 10 oC/min to increase the speed of the analysis while still allowing sufficient
interaction with the column to achieve separation. In analyses involving a temperature
ramp, slower ramp rates result in better resolution, which is important for analytes with
similar boiling points or for a sample containing a complex mixture of components.
Faster ramp rates may be used to reduce analysis time, though this causes a loss of
resolution and the components of the mixture may start to coelute, which can complicate
subsequent identification of components.

The retention time of each analyte (the time taken to elute from the column and
reach the detector) is affected by several factors. The most important factor is the boiling
point of the analyte. Analytes with lower boiling points will elute from the column and
enter the detector first, and hence will have shorter retention times. Also, the interaction
of the analyte with the stationary phase affects the retention time. A greater affinity of the
analyte for the stationary phase, which is based on the polarity of the analyte compared to
that of the stationary phase, results in longer retention on the column. For example, a
polar analyte passing through a column coated with a polar stationary phase will have a
longer retention time than a non-polar analyte with the same boiling point since it will
have a greater affinity for the stationary phase. Other factors that increase retention time
are slower flow rate of the carrier gas, longer capillary column, or increased film
thickness of the stationary phase; however, these factors also increase the overall analysis
time.

After separation, the analytes are detected. For GC-MS, the analytes pass from the
column, through a heated transfer line into the mass spectrometer. The transfer line is

heated (~300 °C) to prevent condensation of the analyte in the transfer line before

21

reaching the detector. The main components of a mass spectrometer are the ionization
source (where ions are formed), mass analyzer (where ions are separated), and detector
(where ions are detected) as shown in Figure 2.2. The mass spectrometer is stored under
vacuum at 10'7 to 10'8 torr to increase the mean free path (or distance between collisions)
of analyte molecules as well as prevent neutralization of the newly formed ions by ion-

ion and ion-molecule interactions.

.....................................................................................................................................

 

 

 

Ion Source Mass Detector
—- Analyzer -—

 

 

 

 

 

 

 

 

 

 

 

Ion production Ion separation Ion detection
Sample _________ _________ Data
Introduction Vacuum pump System
5 5 Generation of mass
From 0c 5._-(_1.9:?-!9-19:*.t_9r_r)_..§ spectrum

 

 

 

 

 

 

Figure 2.2: Schematic of a mass spectrometer.
Mass spectrometers in forensic laboratories often use electron ionization (E1), quadrupole
mass analyzers, and electron multiplier detectors. In El, electrons emitted from a hot
tungsten filament are accelerated across an electric potential (70 eV). The GC column
feeds into the ion source and the separated analytes are directed through the beam of
electrons. Molecular ions are formed by the ejection of an electron from the analyte
molecule through electron-molecule interactions. Fragment ions are formed by further
interactions with other ions, molecules, or electrons. The fragmentation of each analyte is

unique, allowing for definitive identification of an analyte of interest.

22

After the ions are formed, they are focused into quadrupole mass analyzer through
a series of electrostatic lenses. The quadrupole consists of four parallel conducting rods

arranged to define the comers of a square as shown in Figure 2.3.

 

 

UblLLIUK

 

 

/

 

  

 

ION
SOURCE

 

 

 

Figure 2.3: Diagram of a quadrupole mass analyzer.
The ions are directed into the space between the rods. Adjacent rods are oppositely
charged and radio frequency (RF) and direct current (DC) voltages are applied to the
rods. As the positive ions are attracted to the negative rods, the signs of the rods are
continually alternated, giving the ions a corkscrew trajectory as they pass through the
quadrupole. At each RF/DC voltage, ions in a narrow m/z range can pass through the four
rods with a stable trajectory (resonant ions). All other ions have unstable trajectories
(non-resonant) and will collide with the rods, causing them to be neutralized and
therefore, not detected. Each m/z has a specific RF/DC voltage at which the trajectory of
the ion through the rods is stable, allowing each m/z ratio to be detected. Hence, the
voltage across the rods is scanned, keeping the ratio of RF/DC ratio constant, to detect

the full mass range (e.g. m/z 50-500).

23

After the ions successfully pass through the quadrupole, they are detected,
generally by an electron multiplier. A continuous dynode electron multiplier is a hom-
shaped detector made of glass that is coated with a material of a low work function. A
potential of 1.8 to 2 kV is applied across the detector and ions from the mass analyzer
entering the detector strike the surface, causing secondary electrons to be ejected. This
starts an electron cascade as secondary electrons travel down the length of the detector,
ejecting more electrons on each collision with the surface. This type of detector results in
a signal amplification of 108 [1].

The output of an analysis by GC-MS is a total ion chromatogram (TIC). The TIC
is the sum of the signals at each m/z over time. Figure 2.4A shows the TIC of a diesel
sample. Each peak in the TIC corresponds to one or more chemical components in the
sample mixture and has an associated mass spectrum. The mass spectrum of a particular
component contains m/z peaks for the molecular ion and fragments ions. Because each
compound has a unique fragmentation pattern, the mass spectrum can be used to identify
each component in the mixture. Components of a specific compound class can be
targeted by using the chromatographic and mass spectral data together to generate an
extracted ion chromatogram (EIC). For an EIC, a specific m/z peak characteristic of a
compound class of interest is chosen and a new chromatogram is plotted to include only
the chromatographic peaks that contain the chosen m/z in their corresponding mass
spectrum. Figure 2.4B shows the EIC for m/z 57, which is characteristic of the alkane
compound class, to highlight the alkane content of the diesel sample. Furthermore, an

extracted ion profile (EIP) can be created by summing the signals of several EICs.

24

'15E06““”

 

 

 

 

 

 

 

 

 

 

 

 

 

L4EO6+WW--MWW-WWW

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

3.5E06i

 

 

 

 

 

 

 

 

 

 

 

 

 

Ll

Retention Time (min)

 

Figure 2.4: A) TIC and B) EIC of m/z 57 and C) alkane EIP of a diesel sample.

25

For the alkane EIP, m/z 57, 71, 85, and 99 are summed to provide a greater signal and a
better representation of the alkane content in a sample as shown in Figure 2.4C for the
diesel sample. Several different EIPs can be compiled to characterize the chemical

content of a complex sample such as the aromatic or indane content.

2.2 Data Pretreatment Procedures

In GC-MS analyses, there is typical run-to-run variation caused by several factors
including slight changes in injection volume (particularly with manual injection),
pressure, or flow rate or column degradation over time. These variations cause
discrepancies in retention time that can introduce differences in a sample set for
subsequent chemometric procedures. Hence, data pretreatment is necessary before further
data analysis procedures to correct for any variation in chromatograms that is not due to
chemical composition. A retention time alignment algorithm can be used to account for

this type of variation. Two alignment algorithms were used for this research.

2. 2. 1 Retention Time Alignment — Peak-Matching Algorithm

A peak-matching algorithm from the literature that was developed using diesel
samples analyzed by gas chromatography was used for the first part of this research [4].
The first step of the algorithm is to perform a baseline correction to account for baseline
drift and remove any sources of variation that do not relate to chemical composition. The
baseline correction is calculated using a best-fit line through the first and last two seconds

of the chromatogram, where the signal is only caused by background noise. This portion

26

of the alignment algorithm can be substituted for another baseline correction or removed
if desired.

For aligmnent, the user must define a target chromatogram to which the sample
chromatograms will be aligned. It is important to choose a target chromatogram that
contains nearly all of the peaks contained in the sample chromatograms. The peaks in the
target chromatogram should also be near the center of the distribution of peaks in the set
of chromatograms. The chromatogram that allows for the best alignment of all the sample
chromatograms is generally chosen as the target chromatogram.

After a target has been chosen, the alignment algorithm identifies peaks in the
target and then each sample chromatogram. To do so, the algorithm calculates the first
derivative at each point in the chromatogram, which is estimated as the difference
between the signal strength at the current point and the previous point. When this value
increases beyond a threshold value, defined by the user as five times the standard
deviation of the baseline noise, the algorithm detects the leading edge of a peak. The
algorithm then finds the zero crossing, or sign change, in the first derivative which
indicates the peak maximum. The next zero crossing is used to identify the tailing edge of
the peak. The algorithm then interpolates the retention time of the zero crossing, rounded
to the next integer time point, and adds this peak to a list of peaks being generated for
that chromatogram. This procedure is carried out for the target chromatogram then each
sample chromatogram in turn.

After the peaks have been identified in each chromatogram, the algorithm
compares the target to each sample chromatogram in turn. The algorithm examines each

peak in the target chromatogram and finds the peak in the sample chromatogram that

27

most closely matches in retention time. If the peak in the sample chromatogram is within
a certain user—defined window of the peak in the target, the peaks are considered a match.
The retention time axis of the sample chromatogram is adjusted through interpolation so
that the time point on either side of the zero crossing in the first derivative is aligned. The
size of the peak matching window is an important user-defined variable. If the window
size is not large enough to account for normal retention time drift, a corresponding peak
may not be identified. As such, it is important that the average distance that two
corresponding peaks are offset is not larger than the average distance between two peaks
to ensure alignment.

Although the alignment algorithm was designed using diesel chromatograms, it
does have limitations. Peaks with a small signal-to-noise ratio may fall below the baseline
noise threshold as defined by the user and therefore, would not be identified. The
alignment is also limited by the scan rate of the mass spectrometer. With fewer data
points, it is difficult to align shouldered peaks with the alignment algorithm. The
alignment of these peaks depends on the scan rate of the mass spectrometer and the size

of the shoulder in relation to the parent peak.

2.2.2 Retention Time Alignment — Correlation Optimized Warping Algorithm

In the second part of this research, LineUpTM (Infometrix, Bothell, WA) was used
to align a set of chromatograms of ignitable liquids with highly varied components.
LineUpTM employs a correlation optimized warping (COW) algorithm, which aligns
chromatograms in a piece-wise manner where segments in a sample chromatogram are

aligned to the corresponding segment in a target chromatogram [5-6]. Like the peak-

28

matching algorithm, the target chromatogram should contain most of the peaks that are in
each sample chromatogram to be aligned. In this research, the chosen target was a
mixture of all liquids to be aligned analyzed by GC-MS.

The two main user-defined variables for the COW algorithm are the segment size,
denoted as m, and the slack parameter, or warp, denoted as t. The segment size
determines the number of segments into which the chromatogram will be divided. The
segment size should be defined as the approximate number of data points across a peak,
which varies depending on the detector and data collection parameters. The warp
determines how much a segment can be stretched or compressed to achieve alignment
with the corresponding segment in the target chromatogram.

For alignment, the last segment in the chromatogram is considered first. Every
possible warp is examined from -t to +t to determine the best warp for alignment. As a
simple example, consider a sample and target chromatogram with m=5 and t=1. In this
case, there are three possibilities for alignment (-1, 0, +1): compression by 1 data point
(-1), no compression or stretching (0), or stretching by 1 data point (+1). This example is
depicted in Figure 2.5. The resulting segment in each of these cases would be 4, 5, or 6,
respectively. For a segment that has been compressed or stretched, the data points are
interpolated, much like a curve fitting, so that the aligned segment includes the same
number of data points as the target segment. Alignment of the segment is assessed by

calculating a local correlation coefficient (p) between the sample and target segment.

29

Sample segment I .[I I I I I]

Target segment 0 o [I O I O 0]

Segment compressed by 1 point I .[I I I I] la
(i=4)

Segment unchanged I II:- I I I I] 1b
(t=0)

Segment stretched by I point IE- I I I I I] 1c

(t = +1)
Redraw segments Calculate local
(via interpolation, correlation coefficient (p),
as necessary) for each possible segment

p(la)lII{IIIII13

p(lb) IIEIIIII] “3

p(lc) I{I I I I I] 1c

Figure 2.5: Diagram depicting the alignment of the last segment in a sample (m=5) to the
corresponding segment in the target (m=5) with a slack of 1 (t=1).
{ Indicates interpolated data points.
[Adapted from reference 6]

After each warp is considered for the last segment, the penultimate segment is
considered. Each possible warp for the penultimate segment is examined in conjunction
with each possible warp for the last segment. For the previous example with t=1, there
are three possible warps for the last segment, each of which is paired with the three
possible warps for the penultimate segment (-1, 0, +1), resulting in 9 permutations
(Figure 2.6). This procedure continues for each segment, examining each permutation in
turn. The permutation that offers the optimal alignment is determined by calculating a

global correlation coefficient (P), which is a Stun of local correlation coefficients (p) for

each segment in each permutation.

30

Samplesegment II:- I I I I [l I I B

Targetsegment 0 IE. 0 O. O [O 0 O O O

..__\

Segment compressed by 1 point ll:- I I .— ,- u I: 1a,2a
(1:4) a :

Segmentunchanged II IIIIIjI "a It 1a,2b

(t=0) — ~‘

_.._‘r-

"\r‘

 

 

Segment stretched bylpoint IE- I I I I I- i: n a; la, 2c
(t=+1) - l -
Redraw segments Calculate local
(via interpolation, correlation coefficient
as necessary) (p), for each possible
segment
_,
P=p(1a)+p(2a) I I{I I I I EH. I I l 13,23
P=p(1a)+p(2b) IIEIIIIEHIIIIIE 1a,2b
P=p(1a)+p(20) I{IIIII:HII“ ii 13,26

 

* Continue for segments 1b and lc (Figure 2.5), resulting in 9 permutations.
P=p(1a)+ p(2a) P=p(1b)+p(2a) P=p(10)+p(2a)
P=p(1a)+ p(2b) P=p(1b)+p(2b) P=p(1c)+ p(2b)
P=p(1a)+p(2c) P=p(1b)+p(2c) P=p(1c)+p(2c)

Figure 2.6: Diagram depicting the alignment of the penultimate segment.
[Adapted from reference 6]

2. 2.3 Normalization and Mean-Centering

The retention time alignment algorithm addresses slight fluctuations in GC-MS
conditions; however further data pretreatments are necessary to address other sources of
variation that are not chemical such as random error introduced by the analyst [7]. The
pretreatment steps used in this research are normalization and mean-centering. Peak area
normalization of chromatograms is used to account for slight differences in injection
volume. Peak area normalization scales each chromatogram to the average total area of

all of the chromatograms. With normalization, it is easier to compare the concentration of

31

a particular component across samples. Mean-centering is a data-pretreatment step that is
performed before principal components analysis (PCA), to transform the data into
deviations from the average. Mean-centering redefines the average of the data as zero to
visualize the deviation from the average at each retention time. For this research, mean-
centering was incorporated into the calculation of the principal components, so a separate

step to mean-center the data was not necessary.
2.3 Pearson Product Moment Correlation (PPMC) Coefficients

Pearson product moment correlation (PPMC) coefficients are used to determine
the degree of linear correlation between two variables. For chromatographic data, a
PPMC coefficient can be calculated between two chromatograms to determine how
closely they are related, which is especially useful for forensic applications in comparing
a known and a questioned sample. The PPMC coefficient (r) is calculated according to

Equation 2.1,

ice.- —3c')<y. — E)

 

r _

xy _ n — n — o
205,- —.7C)2 2(yi _y)2 EquationZJ
i=1 i=1

 

where the x variables correspond to data points in one chromatogram and the y variables
correspond to the data points in the chromatogram to which it is being compared [7].

PPMC coefficients can range from -1 to +1. PPMC coefficients between +0.8 and 1

32

indicate strong correlation, +0.5 to 0.8 indicate moderate correlation, and coefficients
between 0 and +0.5 indicate weak correlation [8].

Applying this procedure to chromatographic data, replicates of the same sample
should have PPMC coefficients very close to 1 as the chromatograms should be nearly
identical. As such, PPMC can be used as an indication of precision of analysis as well as
an assessment of similarity between samples. For this research, PPMC coefficients are
also used to indicate the similarity between different ignitable liquids. Ignitable liquids
from the same class with similar components should have higher PPMC coefficients than

ignitable liquids from different classes with differing components.

2.4 Principal Components Analysis (PCA)

Principal components analysis (PCA) is a multivariate statistical procedure used
to reduce the dimensionality of large data sets. PCA assumes that the data do not have
any particular distribution and is an unsupervised procedure so no prior knowledge of the
data is required. PCA is used to assess the association and discrimination of samples
within the data set, thus indicating relationships between samples that could not have
been observed otherwise [9]. Generally, in PCA, a new axis is drawn through a multi-
dimensional data set in the direction of maximum variance. A second axis is drawn
perpendicular to the first in the direction of next greatest variation. As a result, a data set
of n dimensions is now described by two dimensions, or principal components; hence, the

relationships between samples become more obvious.

33

The first step in performing PCA is to calculate the covariance matrix of the
normalized data. It is important that the data be pretreated to ensure that the variation of
the data set that is being analyzed is variation among the samples rather than variation
introduced by the analyst or instrumental drift. The covariance is calculated according to

Equation 2.2,

:(x. — 35x)».- — E)

cov(x, y) = M

 

n _ 1 Equation 2.2

where x and y correspond to the data points in the x and y dimensions and n represents the
total number of dimensions [9]. For example, a data set with x, y, and 2 dimensions would
yield the following covariance matrix:

Pcov(x,x) cov(x, y) cov(x,z)_

cov(y,x) cov(y, y) cov(y,z)
_cov(z,x) cov(z, y) cov(z,z)_

 

 

Since cov(x, y)=cov( y, x), this matrix is symmetric about the diagonal, which is

important for principal components analysis. If PCA is applied to a data matrix that is not
square, additional rows or columns of zeroes are added as necessary. For
chromatographic data, each retention time is a dimension, so the total number of
dimensions is given by the number of data points in the chromatogram.

The principal components (PCs) are then calculated as the eigenvectors of the
covariance matrix. An eigenvector is a unit vector that, when multiplied by the data
matrix, gives a resulting vector that is a multiple of the original unit vector. The number

that can be factored out of the resulting vector to give the original vector is called the

34

eigenvalue. A nx n square matrix has n associated orthogonal eigenvectors, each of
which has a corresponding eigenvalue that gives the amount of variance that is explained
by that PC. The eigenvector with the largest associated eigenvalue is considered to be the
most significant because it accounts for the most variance in the data set. This
eigenvector is known as the first principal component, or PC1. The remaining principal
components are ranked in descending order based on their eigenvalues. The amount of
variance accounted for by each eigenvector is determined by the proportion of its
eigenvalue to the sum of the eigenvalues.

Two important outputs of PCA that can be useful in analyzing multivariate data
are scores and loadings plots. A score is calculated for each sample by multiplying the PC
eigenvector by the original data to give a single number. Often, the score on PC1 is
plotted versus the score of PC2 to generate a scatter plot in which clustering patterns of
similar samples can be visualized. Three-dimensional scores plots are also used to plot
the projections of the data on PC1, PC2, and PC3 in a scatter plot; however, association
and discrimination may be more difficult to discern in these plots. In applying PCA to
chromatographic data, the loadings plot, or a plot of a PC against retention time, can be
used to identify the components in the chromatogram that are contributing the most to the
variance among the data set. Such components are observed as those that deviate the
most from zero in the loadings plot. Identifying the components that contribute most the
variance in chromatographic data can give an indication as to which compound classes or
EIPs may offer more discrimination between samples in the data set. The loadings plot
may also help explain the positioning of samples in the scores plot and determine the

components that are responsible for discrimination between samples.

35

2.5

References

. Skoog DA, Holler FJ, Crouch SR. Principles of instrumental analysis. 6tlrl ed.

Belmont, CA: Thomson, 2007.

American Society for Testing and Materials, ASTM E 1387-01. Annual Book of
ASTM Standards 14. 02.

American Society for Testing and Materials, ASTM E l618-06e1. Annual Book of
AS TM Standards 14. 02.

Johnson KJ, Wright BW, Jarman KH, Synovec RE. High-speed peak matching
algorithm for retention time alignment of gas chromatographic data for chemometric
analysis. J Chromatogr A 2003; 996:141-55.

. Nielsen NPV, Carstensen JM, Smedsgaard J. Aligning of single and multiple

wavelength chromatographic profiles for chemometric data analysis using correlation
optimized warping. J Chromatogr A 1998; 805: 17-35.

Tomasi G, van den Berg F, Andersson C. Correlation optimized warping and
dynamic time warping as preprocessing methods for chromatographic data. J
Chemometr 2004; 18:231-241.

Brereton, RG. Applied chemometrics for scientists. Hoboken, NJ: John Wiley &
Sons, Ltd, 2007.

Hupp AM, Marshall L], Campbell DI, Waddell Smith R, McGuffin VL.
Chemometric analysis of diesel fuel for forensic and environmental applications.

Anal Chim Acta 2008; 606:159-71.

Miller JN, Miller JC. Statistics and chemometrics for analytical chemistry. 4th ed.
Harlow, England: Pearson Education Limited, 2000.

36

CHAPTER 3
EFFECT OF GC TEMPERATURE PROGRAM ON THE ASSOCIATION AND
DISCRIMINATION OF DIESEL SAMPLES

3.1 Introduction

Several gas chromatographic (GC) temperature programs exist for analyzing
ignitable liquids which range from 30 minutes to two hours [1-3]. A longer temperature
program is not ideal since fast analysis times are preferred in forensic labs to prevent or
minimize case backlogs. However, with a shorter temperature program, chromatographic
resolution is lost, which may decrease inherent variation that is observed among ignitable
liquids, especially those that are similar in chemical composition. The loss of such
discriminatory information may lead to misclassification or even misidentification of an
ignitable liquid.

The research reported herein focuses on minimizing the analysis time while still
retaining sufficient chemical information for the association and discrimination of
ignitable liquids. Five diesel samples were analyzed in triplicate using six different GC
temperature programs. Temperature programs varied in the ramp rate as well as the
number of ramp steps (one- and two-step ramps). One type of ignitable liquid was chosen
for this initial study because successful association and discrimination of samples with
very similar chemical composition indicate that this procedure will be sufficiently robust
to differentiate ignitable liquids from different classes. Diesel was chosen as the ignitable
liquid to analyze due to its complex chemical composition and wide range of boiling

points. The total ion chromatogram (TIC) and extracted ion profiles (EIPs) of

37

characteristic alkane and aromatic compound classes were generated for all samples.
Specific EIPs were studied to determine the class of compounds that varied most among
diesel samples and, therefore, would be useful for discrimination.

Prior to data analysis procedures, the TIC and EIPs were subjected to two data
pretreatment steps: retention time alignment and normalization [3]. Pearson product
moment correlation (PPMC) coefficients were calculated for the aligned data. Principal
components analysis (PCA) was performed on the TIC and EIPs to determine the extent
of association and discrimination among the diesel samples afforded by each temperature
program. PPMC coefficients were used for a pair-wise comparison of chromatograms,
which may be directly applicable in crime labs for the comparison of a known ignitable
liquid to a questioned ignitable liquid residue (ILR). PCA was performed on the complete
data set to assess the variation among the diesel samples, which is more useful for

determining the class of an unknown ILR by comparing it to a reference collection.

3.2 Materials and Methods

3. 2. 1 Sample Collection

Diesel samples were collected from four different brands of service station in the
Lansing, Michigan area. For one brand, samples were collected at two locations, giving a
total of five diesel samples. Approximately one-half gallon of each sample was collected
and stored in an acid-washed amber bottle at 4 0C until analysis. Samples were diluted

10:1 in dichloromethane (spectrophotometric grade, Sigma-Aldrich, St. Louis, MO) prior

38

to GC-MS analysis.

3.2.2 GC-MS Analysis

All analyses were performed using an Agilent 6890 gas chromatograph interfaced
with an Agilent 5975 mass spectrometer (Agilent Technologies, Santa Clara, CA),
equipped with an Agilent HP-SMS capillary column (30 m x 0.25 mm x 0.25 pm). The
inlet temperature was 250 oC and 1 pL of the diluted diesel sample was drawn into the
syringe followed by 1 [1L of air for manual injection. The syringe was held in the
injection port for two seconds before and after injection to ensure volatilization and
transfer of the full sample onto the column. Six different GC temperature programs were
investigated. Each temperature program had an initial temperature of 40 °C and a final
temperature of 280 °C, with the ramp rate varying in each program. The programs were
developed based on a two-step temperature program available in the literature [3] and a
one-step temperature program used by the National Center for Forensic Sciences (NCFS)
for their Ignitable Liquids Reference Collection [1]. The ramp rates of the two-step
program were increased from 2 °C/min and 3 °C/min (Method A) to 5 °C/min and 10
°C/min (Method B). A single-step temperature ramp consistent with the NCFS
temperature program parameters except with a slower ramp rate of 5 °C/min (Method C)
was investigated. The temperature ramp rate of the NCFS program was increased from 10
°C/min (Method D) to 20 °C/min (Method B) and to 30 °C/min (Method F). Faster ramp
rates were not investigated as the limit of the GC oven was 30 °C/min. Each of the five
diesel samples was analyzed in triplicate using each of the six GC temperature programs

which are summarized in Table 3.1.

39

Table 3.1: GC temperature programs investigated.

 

 

Initial Initial FinaFinal Total
Method Temp Hold Ramp Rate Temp Hold Time
(°C) (min) (°C) (min) Quin)

A 40 0 2 °C/min to 150 °C, 3 °C/min 280 15 113

B 40 0 5 °C/min to 150 °C, 10 °C/min. 280 15 50

C 40 3 5 °C/min 280 4 55

D 40 3 10 °C/min 280 4 31

E 40 3 20 °C/min 280 4 19

F 40 3 30 °C/min 280 4 15

 

The transfer line between the GC column and the mass spectrometer was
maintained at a temperature of 300 °C. The mass spectrometer was equipped with an
electron ionization source operating at 70 eV and a quadrupole mass analyzer operating
in full scan mode (m/z 50-550) at a scan rate of 2.91 scans/s. Diesel samples were
reanalyzed by Methods E and F, increasing the scan rate to 5.51 scans/s. This was
deemed necessary as initial data analysis indicated that more data points were necessary
to define each peak for alignment.

An automated injection technique was also investigated with and without a
pressure pulse at the time of injection to determine the effect of manual injection on
further data analysis procedures. The five diesel samples were reanalyzed by Method D
using an Agilent 7683B series automated liquid sampler (ALS), with both pulsed and
unpulsed injections. For unpulsed injections, pre- and post-dwell times were 0.03 minutes
with no pressure pulse to imitate the manual injection technique. Previous studies were
completed to determine the optimum pre- and post-injection dwell times as well as the

optimum pulse conditions for pulsed injections (data not shown). For pulsed injections,

4o

the optimum parameters were as follows: pre-injection dwell time 0.02 minutes, post-

injection dwell time 0.05 minutes, and pressure pulse 15.0 psi for 0.25 minutes.

3. 2.3 Data Analysis

For each diesel sample analyzed by each GC temperature program, alkane EIPs
(m/z 57 + 71 + 85 + 99) and aromatic EIPs (m/z 91 + 105 + 119 + 133) were generated in
ChemStation Software (Agilent Technologies, Santa Clara, CA). The TIC and EIPs for
diesel samples were compiled into individual data sets in Microsoft Excel (Microsoft
Corp., Redmond, WA), resulting in three files for each of the six GC temperature
programs. The chromatograms of the diesel samples were retention time aligned using a
peak-matching algorithm [5]. Matlab (version 7.4.0.287, MathWorks, Natick, MA) was
used to align the chromatograms as well as to calculate PPMC coefficients (full table of
PPMC coefficients available in Appendix A). After alignment, each set of
chromatograms was peak area normalized in Microsoft Excel. To normalize the
chromatograms, the total area under the chromatogram was calculated for each
chromatogram. The average of the total area under all sample chromatograms was taken
and a normalized chromatogram was generated by dividing each data point in the
chromatogram by the total area under the chromatogram, then multiplying it by the
average total area.

Principal components analysis was performed by eigenanalysis of the covariance
matrix of the data set using Matlab. Scores and loadings plots were generated for the TIC

and alkane and aromatic EIPs in Microsoft Excel for each GC temperature program.

41

3.3 Results and Discussion

Figure 3.1 shows the TICs of one diesel sample analyzed by Methods A, D, and
F. These chromatograms represent slow two-step (2 °C/min, 3 °C/min), medium single-
step (10 °C/min), and fast single-step (30 °C/min) temperature ramps. Figure 3.2 shows
the alkane and aromatic EIPs for Method D to highlight the alkane and aromatic content
of the diesel. The normal alkanes (C9-C24) are visible in a distribution that is
characteristic of a diesel chromatogram (Figure 3.2A). However, a loss of resolution is
apparent as ramp rate increases, which is evident from the rising baseline in Figures 3.1B
and 3.1C. The aromatic content of the diesel is observed at the beginning of the
chromatogram (Figure 3.28). The aromatic content is observed with better resolution in
Figure 3.1A because of the slower ramp rate in Method A, which could affect association

and discrimination of diesel samples.

42

4 .0506 j * -~~ , ~ , , ,. v . .

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6 j 113

 

 

 

 

 

 

 

 

 

 

1.0E07" "

 

 

 

 

 

 

 

 

 

Retention Time (min)

Figure 3.1: TIC of one diesel sample analyzed by A) Method A, B) Method D, and
C) Method F.

43

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1.2E06 l

 

 

 

 

I

ll

5
.l j l
t l‘ , . I i
I .11. Il ‘ j
r : T

I I I

 

 

 

 

 

Retention Time (min)

Figure 3.2: A) Alkane EIP and B) aromatic EIP of one diesel sample analyzed by
Method D.

44

3. 3.] Association and Discrimination of Diesel Samples Based on Total Ion

Chromatogram

Figure 3.3A shows a section of the TIC for a set of five well-aligned diesel
samples analyzed by Method D after retention time alignment. As the ramp rate increased
to 30 °C/min, chromatograms did not align as well (Figure 3.3B), indicating that the
increased ramp rate and resulting loss of resolution negatively affected the alignment.
The poor alignment was reflected in subsequent PPMC and PCA results, which will be
discussed later.

PPMC coefficients calculated between pairs of chromatograms were consistently
higher between replicates of one diesel than among different diesel samples, as expected,
regardless of ramp rate or number of ramp steps. PPMC coefficients tended to increase
for faster GC temperature programs. For example, the mean PPMC coefficient between
Diesels 2 and 4 for Method D was 0.9126 + 0.0026 (Table 3.2). The mean PPMC
coefficient increased to 0.9309 + 0.0021 for Method E, and finally to 0.9489 + 0.0053 for
Method F. With faster ramp rates, the diesels appeared more similar due to the
corresponding loss of chromatographic resolution. Overall, PPMC coefficients were
affected by the loss of resolution resulting from faster ramp rates rather than the number
of ramp steps.

Table 3.2: Mean PPMC coefficients + standard deviation between Diesels 2 and 4 (n=9)
for each temperature program.

Method Ramp Rate Mean PPMC Coefficient

 

 

A 2, 3 °C/min 0.8691 :1: 0.0063
B 5, 10 °C/min 0.9111 + 0.0006
C 5 °C/min 0.9136 + 0.0021
D 10 °C/min 0.9126 + 0.0026
E 20 °C/min 0.9309 + 0.0021
F 30 °C/min 0.9489 + 0.0053

 

45

 

 

 

 

 

8.9 é.1

Retention Time (min)

Figure 3.3: A) Well aligned hexadecane peak in the TIC of diesel samples analyzed by
Method D, and B) poorly aligned hexadecane peak in the TIC of diesel samples analyzed
by Method F.

46

The clustering of diesel samples in the PC scores plots based on the TIC was
similar for all GC temperature programs. Replicates of each diesel were clustered closely
while different diesels were well separated (Figure 3.4). No difference in association and
discrimination of diesel samples was observed for one- or two-step temperature ramps.
Exemplar scores plots for Methods A, D, and F are shown in Figure 3.4. Slight spread in
the second principal component was observed for the replicates, which may be due in
part to lack of precision as a result of manual injection technique. However, there was no
overlap among different diesel samples, allowing for discrimination of the five samples,
even with the fastest ramp rate. The scores plots for Methods B and F showed greater
spread among replicate samples. Visual assessment of the aligned chromatograms for
these two temperature programs indicated peak misalignments (Figure 3.3), which may
have contributed to the spread among replicate samples. Hence, poor alignment may
limit the utility of PCA in differentiating diesel samples for faster temperature ramp rates.

The PPMC coefficients were consistent with the association and discrimination of
diesel samples in the PC scores plots. For Method D, the highest PPMC coefficients were
between Diesels 3 and 5 (0.9893 + 0.0020), which corresponded to the diesel samples
that were clustered closest together in the scores plot. The lowest PPMC coefficients
were between Diesels 2 and 4 (0.9126 + 0.0026), which corresponded to the samples that

were furthest apart in the scores plot.

47

 

L4E07

 

PC2 (23.67%)
C

 

 

 

 

—1 .4507
-1.6E07 o 1.6EO7
PC1 (63.49%)

 

I4EO7

PC2 (27.01%)
0

 

 

-1 .4507 :
-1.6E07 o 1.6E07
per (55.30%)

 

 

I4E07

 

%»

PC2 (32.77%)
C

to

 

 

 

-1 .4507 .
-l.6E07 0 1.6507
PC1 (50.21%)

Figure 3.4: Scores plot of TIC for A) Method A, B) Method D, and C) Method F. Diesel
1 (9), Diesel 2 (I), Diesel 3 (A), Diesel 4 (0), Diesel 5 (o)

 

48

The PC loadings plots for the first principal component for each GC temperature
program were generated to identify chemical components contributing most to the
variance among samples. For all temperature programs, the loadings plots were
dominated by the CH-C13 normal alkanes loading negatively with some contributions
from aromatic compounds loading positively, such as trimethylbenzene and other C3-
alkylbenzenes (Figure 3.5). The loadings plots for Methods E and F showed a high
occurrence of derivative-shaped curves (Figure 3.6). Visual assessment of the
corresponding retention time aligned chromatograms confirmed that derivative-shaped
curves in the loadings plots were indicative of peak misalignments (Figure 3.3B).

From the loadings plot, it was determined that the positioning of the diesel
samples on the first principal component was mainly dependent on the aromatic content
of the diesel samples. Diesels 2 and 5 had positive scores on the first principal component
which was a result of higher aromatic content, such as 1-ethyl-3-methylbenzene and 1, 3,
5-trimethylbenzene, as these components were positive in the loadings plot of the first

principal component.

49

 

 

 

 

 

 

 

 

 

2
._. 1
U
a.
8
go 0 , "“1”, I], . ...n.u.u....."'r1 -l J L .
a 1
‘8
(U
o
_r
3 4 5
0 Retention Time (min) 31

Figure 3.5: Loadings plot of the first principal component of diesel samples analyzed by
Method D, identifying the components contributing the most to the variance in the
sample set: 1)1-ethyl-3-methylbenzene, 2) 1, 3, 5-trimethylbenzene, 3) undecane,

4) dodecane, and 5) tridecane.

50

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

£3
a.
{-1 . 1
E00 _1. 14‘. a“. l‘r'l ”Mill. lLl-
:3 1 . '1‘” 3 3
°° l
O
”J
l
l
g
-0.3 r ___ 3
0 15
0.2 ‘“ 3
B :
;
6 l
a.
0
3’6 0 .
fl
‘8
Cd
0
._1 9
l
9 Retention Time (min) 11

Figure 3.6: A) Loadings plot of the first principal component of diesel samples analyzed
by Method F, B) a magnified portion of the loadings plot from A, which demonstrates
derivative-shaped curves indicative of misalignments.

51

3. 3.2 Association and Discrimination of Diesel Samples Based on Alkane Extracted Ion

Profile

Several of the trends observed for the TIC were also observed for the alkane EIP.
Peak alignment was increasingly worse as the GC temperature ramp rate increased.
PPMC coefficients between replicates of the same diesel were generally lower for the
alkane EIP than the TIC for all temperature programs, which would indicate less
similarity among replicates. Visual examination of the retention time aligned alkane EIP
showed greater misalignments than for the TIC, which may have resulted in lower PPMC
coefficients. Nevertheless, PPMC coefficients for the alkane EIP were generally higher
between replicates of the same diesel and lower between different diesels, as expected.

Principal components analysis based on the alkane EIP did not offer
discrimination among four of the diesel samples as shown in the scores plot for Method
D (Figure 3.7). This lack of discrimination was observed in all temperature programs,
regardless of ramp rate or the number of ramp steps. Since the alkane content is so
similar among diesels, PCA is forced to find differences among replicates, causing
greater spread in the scores plots. Although manual injection may have contributed to
some of the spread in the scores plot, it is clearly not the sole cause of the spread
observed for the alkane EIP when compared to the TIC (Figure 3.4B).

Again, the association and discrimination of the diesel samples in the scores plot
were similar for all temperature programs and some overlap was observed among
different diesel samples. For example, the mean PPMC coefficient between Diesel 3 and
5 was 0.9782 + 0.0087. However, in the scores plot, one replicate of Diesel 5 was closer

to a replicate of Diesel 3 (indicated by * in Figure 3.7) than the other Diesel 5 replicates.

52

 

 

 

 

 

 

1.01307
1; CD
9)
<1- ‘10»
V I
8 . '-
Q1
-l.0E07
-1.6E07 0 1.6E07

PC1 (68.87%)
Figure 3.7: Scores plot of the alkane profile for Method D. Diesel 1 (0), Diesel 2 (I),
Diesel 3 (A), Diesel 4 (0), Diesel 5 (O)
The PPMC coefficient between these two individual samples was 0.9933. This was
higher than the mean PPMC coefficient for the replicates of each diesel sample (0.9901 +
0.0053 for replicates of Diesel 3 and 0.9871 + 0.0085 for Diesel 5) and reinforces the
similarity of alkane content among diesel samples.

As with the TIC, PPMC coefficients for the alkane EIP were consistent with
association and discrimination of the diesel samples in the scores plots. For Method D,
Diesels 2 and 4 had the lowest PPMC coefficients (0.8555 + 0.0154) and were furthest
apart in the scores plot, whereas Diesels 1 and 3 had the highest PPMC coefficients

(0.9864 + 0.0042) and were closest. The two diesels with the highest mean PPMC

53

coefficient varied by temperature program. For Methods A, B, C, and D, Diesels l and 3
had the highest mean PPMC coefficient. For Methods E and F, Diesels 3 and 5 had the
highest mean PPMC coefficient. The pair of diesels with the highest mean PPMC
coefficient changed with the fastest temperature programs, which may have been a result
of the loss of resolution observed for Methods E and F.

In the loadings plot for the first principal component, more volatile hydrocarbons
contributed most to the variance among the diesel samples (Figure 3.8). This was
consistent for all temperature programs. Normal alkanes C3-C14 dominated the variance,
while the less volatile alkanes (C15-C24) contributed less to the variance. Like the TIC,

loadings plots for the alkane EIP showed derivative-shaped curves indicative of peak

 

 

 

 

 

 

 

 

 

 

misalignments.

“ AL

o

:30 O . l-l I l l 1

:5.

o

._1

-0.35 -—-- ..........
0 Retention Time (min) 31

Figure 3.8: Loadings plot of the first principal component based on the alkane EIP of
diesel samples analyzed by Method D.

54

From the loadings plot of PC1, it was determined that the position of the diesel
samples on PC1 depended on the relative abundance of the more volatile (Cs-CM) and
less volatile alkanes (C15-C24). In the loadings plot, the more volatile alkanes loaded
negatively while the less volatile alkanes loaded positively. Diesel samples with a higher
relative abundance of the more volatile alkanes tended to have negative scores on PC1,
such as Diesel 4. Conversely, Diesel 2 had a lower relative abundance of more volatile
alkanes and had a positive score on PC 1.

Overall, the alkane profile did not provide any further information in
distinguishing the diesel samples than the TIC for any of the temperature programs. The
similar alkane content among diesel samples caused spread in the replicates in the scores

plot, making association and discrimination of the diesel samples difficult.

3. 3.3 Association and Discrimination of Diesel Samples Based on Aromatic Extracted

Ion Profile

Many of the same trends observed for the TIC and alkane EIP were also observed
for the aromatic EIP. PPMC coefficients between replicates of a diesel were consistently
higher than the coefficients between different diesels. PPMC coefficients between
replicates of the same diesel for the aromatic EIP were higher than the alkane EIP, though
lower than the TIC. However, PPMC coefficients between different diesel samples were
lowest for the aromatic EIP and also had a greater range of values than the alkane EIP or
TIC. A greater range of PPMC coefficients indicated a greater variability in aromatic
content among the diesel samples and suggested that the aromatic EIP could be more

useful than the alkane EIP for discriminating diesel samples.

55

Scores plots for the aromatic EIP showed the same general pattern for all GC
temperature programs. The scores plot of the aromatic EIP for Method D is shown in
Figure 3.9. Again, replicates of each diesel were associated while different diesels were
discriminated in the scores plots. This association and discrimination were similar for all

temperature programs, irrespective of ramp rate or the number of ramp steps.

 

1.5E06

 

 

 

 

5: 1 .1-

q I

e 0 ‘

N

U

D.

3 3
-1.5E06 3
-5.0E06 0 5.0506

PC1 (81.68%)
Figure 3.9: Scores plot of the aromatic profile for Method D. Diesel 1 (6), Diesel 2 (I),
Diesel 3 (A), Diesel 4 (0), Diesel 5 (O)
The PPMC coefficients were consistent with the clustering observed in the PC
scores plot. The highest PPMC coefficients corresponded to the diesels that were

clustered closest together on the first principal component. For Method D, Diesels 3 and

5 had the highest PPMC coefficients (0.9894 + 0.0026) and were closest on PC1. Diesels

56

2 and 5 had the lowest PPMC coefficients for Method D (0.8481 + 0.0051) and were
furthest apart. For comparison, Diesels 2 and 4 had the lowest mean PPMC coefficients
for the TIC and alkane EIP. Since different information was obtained from the aromatic
EIP than from the TIC and alkane EIP, the use of different EIPs was valuable to further
associate and discriminate the diesel samples.

The loadings plots for the first principal component for the aromatic EIP showed
the same general pattern for all GC temperature programs (Figure 3.10). The compounds
that contributed to the variance were Cz- and C3-alkylbenzenes. The most variable
components were identified as 1-ethyl-3-methylbenzene and 1, 3, 5-trimethylbenzene,
which were the same aromatic components contributing to the variance in the first
principal component based on the TIC. Again, derivative-shaped curves were prevalent in
the loadings plots for the fastest temperature programs (Methods E and F), indicating
peak misalignments, which was confirmed by visual assessment of the aligned EIPs.

Examination of the loadings plot for PC1 based on the aromatic EIP showed that
the diesel samples were mainly discriminated based on the 1-ethyl-3-methylbenzene and
1, 3, 5-trimethylbenzene content. These two aromatic components were positive in the
loadings plot. Accordingly, the only diesel sample to load positive on the first principal
component was Diesel 2, which had a greater abundance of these two components than
the other diesel samples.

From the PPMC and PCA results, the aromatic content among diesels was more
variable than the alkane content. As a result, the aromatic EIP offered additional
discriminatory information to supplement the information gained from the TIC in

differentiating the diesel samples.

57

 

 

 

 

 

 

 

 

 

 

0.3 “"""""
1
2

6
a“ i
G
° 1
Ba 0 32.2% -
r:
‘8
N
O
._1

-O.3 r- W l

0 Retention Time (min) 31

Figure 3.10: Loadings plot of the first principal component based on the aromatic EIP of
diesel samples analyzed by Method D, identifying the components contributing the most
to the variance in the sample set: 1)1-ethyl-3-methylbenzene and
2) 1, 3, 5-trimethylbenzene.

3. 3.4 Effect of Increased MS Scan Rate on Association and Discrimination of Diesel

Samples

Peak misalignments were prevalent in the TIC and EIPs for the fastest
temperature programs. Upon closer inspection of the aligned chromatograms of Methods
B and F, it was observed that there were insufficient data points defining a peak in order
to align the chromatograms satisfactorily [6]. Therefore, the scan rate of the mass
spectrometer was increased from 2.91 scans/s to 5.51 scans/s resulting in a change from

approximately 6 points across a peak for the slower scan rate to approximately 12 points

across a peak for the faster scan rate. In Figure 3.11A, a poorly aligned section of the TIC

58

for Method F is shown with the corresponding derivative-shaped curve in the loadings
plot (Figure 3.1 18). Using the faster scan rate, the alignment of these peaks is improved
(Figure 3.11C), which is reflected in the loadings plot by the improvement of the
derivative-shaped curve (Figure 3.11D). Overall, there was little change in the loadings
plots with a faster scan rate, though fewer derivative-shaped curves were apparent,
indicating some improvement in the alignment. This was further verified by visual
comparison of the aligned chromatograms. However, similar association and
discrimination of diesel samples were observed in the scores plots for the TIC, alkane,
and aromatic EIPs as described previously. Spread was still observed in the replicates for
the TIC, alkane, and aromatic EIPs indicating that an increase in the scan rate did not
correct all misalignments. The manual injection technique may explain other

misalignments not corrected by the increased scan rate.

59

     

 

 

 

 

 

 

__ In,
. 05
E
E
v
D
.E
E-‘
G
O
'5
t:
8
0
1 a:
_ l
A In Q l ' 0'1
e 3 3 . ON
C; 13d uo sflurpeoq C? 0 13d uo sfiurpeoq 0,
i" ' 05
3, \
\7
.1}
7
(“3.5" .E
““:> S
w”. j ' o
4“N‘H‘ H
\ c:
:3 .2
‘ 3
j o
.1} , , m
‘.\‘.\.‘| \ .V
..//"/ :U (I (”'1'
1\ l\ ox
O O
LL} LIJ
“l N.
v—1 '—

Figure 3.11: A) Section of aligned TICs of diesel samples analyzed by Method F with a
scan rate of 2.91 scans per second showing poor alignment. B) Loadings plot showing the
derivative-shaped curve indicative of the misalignment in A. C) Same section of the TIC,
showing the aligned chromatograms of diesel samples analyzed by Method F with a scan

rate of 5.51 scans per second. D) Loadings plot without any derivative-shaped curves
indicating an improvement in alignment.

6O

3.3.5 Eflect of Automated Injection Technique on Association and Discrimination of

Diesel Samples

From the results of PPMC coefficients and PCA, Method D was determined to be
the optimum GC temperature program, which offered a compromise between analysis
time and resolution. Consequently, each diesel sample was reanalyzed by Method D
using an automated injection technique to determine the effect of manual injection. The
automated injection technique was investigated with and without a pressure pulse at the
time of injection. The abundances of the diesel samples analyzed with the unpulsed
autosampler injection were lower than those of the diesel samples analyzed with the
pulsed autosampler and manual injections. Therefore, only the pulsed autosampler
injection (15 psi for 0.25 min) was compared to the manual injection.

As with manual injection, PPMC coefficients of samples analyzed with the pulsed
autosampler injection were higher for replicates of the same diesel than they were
between diesel samples for the TIC, alkane EIP, and aromatic EIP. For the pulsed
injection, mean PPMC coefficients for replicates tended to be greater with lower standard
deviations than PPMC coefficients for the manual injection, indicating a more precise
injection. Table 3.3 shows a comparison of PPMC coefficients for manual and pulsed
autosampler injection to indicate the improved precision achieved with the autosampler.

Table 3.3 Comparison of mean PPMC coefficients + standard deviation of replicates for
manual and pulsed autosampler injections (n=3).

 

 

Replicates Manual Injection Pulsed Autosampler Injection
Diesel 1 0.9943 + 0.0010 0.9979 + 0.0012
Diesel 2 0.9940 + 0.0017 0.9976 + 0.0011
Diesel 3 0.9930 + 0.0016 0.9983 + 0.0002
Diesel 4 0.9952 + 0.0023 0.9977 + 0.0004
Diesel 5 0.9965 + 0.0005 0.9982 + 0.0007

 

61

In the PC scores plot based on the TIC of the pulsed injection, replicates of each
diesel sample were closely clustered while different diesel samples were well separated.
The PC scores plot showed less discrimination on the first principal component indicating
that some of the variance described by PC1 previously was due to the manual injection
technique (Figure 3.12). Visual examination of the loadings plot for PC 1 showed that
normal alkanes C1 1-C13, the most abundant peaks in the chromatograms, contributed less
to the variance for the autosampler injection than the manual injection. Compared to an
automated injection technique, manual injection introduces greater differences in sample
volume injected and hence differences in peak abundance. Normalization is employed to
account for differences in injection volume, although it may not remove all variation in
peak abundance. With an automated injection technique, a more precise volume is
injected, so it follows that differences in peak abundance would be less than with manual
injection. Less variation in peak abundance may explain the reduced contribution of the
most abundant peaks and slight loss of discrimination on the first principal component for
the automated injection technique.

For the alkane EIP, the PC scores plot for the pulsed injection did not show
considerable improvement in the association and discrimination over the manual
injection. For the pulsed injection, the scores plot showed better association of the
replicates on the first principal component than manual injection, but spread was still
observed in the second principal component. Comparison of the PC loadings plots for
both injection techniques showed that similar components contributed to the variance. As
with the manual injection, the autosampler did not offer more discrimination for the

alkane EIP than the manual injection due to the spread in the replicates.

62

 

1.4507 A

 

PC2 (27.01%)
C

u

 

 

4 .4507 1

-1.6E07 0 1.6507
PC1 (55.30%)

 

 

1 .4E07

O

 

PC2 (35.91%)
0

 

 

 

-1 .4E07

-1.6EO7 0 1.6507
PC1 (51.38%)

Figure 3.12: Scores plot of TIC for A) manual injection and B) pulsed autosampler
injection. Diesel 1 (0), Diesel 2 (I), Diesel 3 (A), Diesel 4 (0), Diesel 5 (O)

 

63

The scores plot based on the aromatic EIP for the pulsed injection showed similar
association and discrimination to the manual injection. In the scores plot, replicates were
more closely clustered than manual injection, especially in PC2, suggesting improved
precision. As with the TIC and alkane EIP, the loadings plot for the pulsed injection was
very similar to the manual injection.

Overall, the automated injection offered increased precision in injections over the
manual technique. Little change was observed in the association and discrimination of the
diesel samples based on the TIC, alkane EIP, and aromatic EIP. The advantage of the
autosampler is that it reduces the amount of time and involvement for an analyst and

makes the analyses more time efficient.

3.4 Conclusions

The association and discrimination of five diesel samples were not greatly
affected by GC temperature ramp rate or number of ramp steps. The mean PPMC
coefficients between replicates of the same diesel were generally higher than those
between different diesel samples and PPMC results were consistent with the results of
PCA. The PC scores plots showed similar association and discrimination of the five
diesel samples and PC loadings plots identified the same components (C ”-Cls normal
alkanes and alkylbenzenes) as contributing to the variance among diesel samples for each
GC temperature program. Both the TIC and aromatic EIP offered valuable discriminatory
information, whereas the alkane profile was not useful in distinguishing diesels due to the

similar alkane content among diesels. The fastest GC ramp rates (Methods E and F)

64

showed an obvious loss of resolution. However, the clustering of the diesel samples in the
scores plot and components contributing the most to the variance in the loadings plots
were comparable to the slowest GC ramp rate (Method A). Although the slowest GC
ramp rates offered improved resolution, the total analysis time (113 minutes) is not
desirable for fire debris analyses. For Methods E and F (the fastest ramp rates), it was
necessary to increase the scan rate of the mass spectrometer to sufficiently improve
alignment of chromatograms. Therefore, Method D was determined to be the most
appropriate program for fire debris analysis. This method offered adequate association
and discrimination without any change in mass spectrometer instrument parameters from
the method currently used by the NCFS. Furthermore, the use of a pulsed automated
injection technique minimized analyst involvement and did not considerably affect the
association or discrimination of the diesel samples for the TIC, alkane EIP, or aromatic
EIP. Slight improvement was observed in the association of replicates in PC scores plots
due to increased precision over manual injection. Overall, the automated injection
procedure was advantageous because it improved precision of analyses and reduced the

time required by an analyst.

65

3.5

References

. National Center for Forensic Science, Ignitable Liquids Reference Collection.

http://ilrc.ucf.edu/ (Accessed August 2008).

. American Society for Testing and Materials, ASTM E1618. Annual Book of

AS TM Standards 14. 02.

. Hupp AM, Marshall LJ, Campbell DI, Waddell Smith R, McGuffin VL.

Chemometric analysis of diesel fuel for forensic and environmental applications.
Anal Chim Acta 2008; 606: 159-71.

. Johnson KJ, Wright BW, Jarman KH, Synovec RE. High-speed peak matching

algorithm for retention time alignment of gas chromatographic data for
chemometric analysis. J Chromatogr A 2003;996:141-55.

. Chesler SN, Cram SP. Effect of peak sensing and random noise on the precision

and accuracy of statistical moment analysis from digital chromatographic data.
Anal Chem 1971 Dec;43 : 1922-1933

66

CHAPTER 4
DISCRIMINATION OF IGNITABLE LIQUID RESIDUES FROM MATRIX
INTERFERENCES USING CHEMOMETRIC PROCEDURES

4.1 Introduction

Visual assessment of gas chromatograms for the identification of ignitable liquid
residues (ILRs) is often difficult or subjective due to weathering, or evaporation, of the
ignitable liquid and contributions from the fire debris matrix. In this study, these
problems were addressed by developing an objective method to associate an ILR to a neat
liquid while differentiating from matrix interferences. This objective method builds on
previous work by Hupp et al. in which 25 diesel samples from 25 service stations
analyzed by gas chromatography-mass spectrometry (GC-MS) were differentiated using
Pearson product moment correlation (PPMC) coefficients and principal components
analysis (PCA) [1]. In the work by Hupp and coworkers, as well as this work, the entire
chromatogram was investigated rather than specific regions of the chromatogram or peak
ratios, as in other studies. The entire chromatogram will potentially offer more
discrimination between samples.

Firstly, six ignitable liquids from six classes outlined by the American Society for
Testing and Materials (ASTM) were each spiked onto samples of nylon carpet and lightly
burned to assess the association of an ILR to a neat liquid in the presence of minimal
matrix interferences using PPMC coefficients and PCA. The PPMC coefficients were
used for a pair-wise comparison of chromatograms, which was useful for assessing the

correlation of an ILR with the corresponding neat ignitable liquid. Additionally, PCA was

67

applied to the entire data set to assess association and discrimination of all samples. In a
second study, the nylon carpet was more heavily burned to increase the contribution of
the matrix to the chromatogram of the ILR. Ignitable liquids were also spiked before and
after burning to assess the effect of weathering of the ignitable liquid on the association

of the ILR to the neat liquid.

4.2 Materials and Methods

4. 2.1 Sample Collection

Six ignitable liquids from six ASTM classes were investigated for this research.
Gasoline and diesel samples were collected from service stations in the Lansing,,
Michigan area. Adhesive remover, lamp oil, torch fuel, and paint thinner were collected
from local grocery stores, hardware stores, and online sources. The ignitable liquids and
the class to which each belong are listed in Table 4.1. Unused, beige nylon carpet (source
unknown, fiber type identified by infrared spectroscopy) was used as the matrix in all

studies.

68

Table 4.1: Ignitable liquids investigated.

 

 

Ignitable Liquid ASTM Class Major Components
Gasoline . C9-C15, alkylbenzenes,
Ga 1
(Shell) so me naphthalenes
Diesel . . C9-C20, alkylbenzenes,
(Sunoco) Petroleum Drstrllates naphthalenes
Adhesive Remover Aromatic ethylbenzene,
(Goof Off) 0 -xylene, p -xylene
Lamp Oil normal alkanes
(Lamplight Farms) Alkane Cll'CIS
Torch Fuel . . normal alkanes Cn-CM,
. . N hth P ff
(T1k1) ap enrc ara rnrc branched/cyclic alkanes
Odorless Paint Thinner Iso mic branched alkanes
(Sunnyside) pm a“ c743,2

 

4. 2.2 Neat Ignitable Liquids

Each of the six neat ignitable liquids was extracted by passive headspace with
activated carbon strips, and the resulting extract was analyzed by GC—MS to identify the
major components. Five microliters of neat liquid were spiked onto a 2 x 2 cm piece of
Kimwipe (Kimberly-Clark, Irving, TX), which was placed in an unlined paint can
(Arrowhead Forensics, Lenexa, KS). One-fourth of an activated carbon strip (Albrayco
Technologies, Inc., Cromwell, CT) was suspended in the headspace of the can and the
sealed can was heated at 80 °C for 4 hours. After extraction, the activated carbon strip
was removed and eluted with 200 11L of carbon disulfide (spectrophotometric grade,

Sigma-Aldrich, St. Louis, MO). The resultant extract was then analyzed by GC-MS.

69

4. 2.3 Minimal Matrix Interferences

Nylon carpet was analyzed using the same passive headspace extraction as the
neat liquids to identify potential interferences present in the burned carpet. Three 5 x 5
cm pieces of unused nylon carpet were burned by applying a blow torch (Benzomatic,
Medina, NY) for 10 seconds and allowing the carpet to self-extinguish. Then, the three
burned pieces of carpet were separately extracted by the same passive headspace
extraction, eluted with carbon disulfide, and analyzed by GC-MS.

A 5 x 5 cm piece of nylon carpet was spiked with 750 uL of ignitable liquid then
burned by applying a blow torch for 10 seconds. The resulting ILR was extracted by the
same passive headspace extraction, eluted with carbon disulfide, and analyzed by GC-
MS. This was repeated in triplicate for each ignitable liquid, resulting in a total of 18

simulated ILRs.

4. 2. 4 Increased Matrix Interferences

In a second study, the carpet was burned to a greater extent to increase the
contribution of matrix interferences to the chromatogram of the ILR. The carpet was
heavily burned by applying a blow torch to the carpet for 20 seconds, then allowing the
carpet to burn further for 1 minute. The carpet was then extinguished, by covering it with
a beaker, and allowed to cool. The unspiked, heavily burned carpet was analyzed in
triplicate using the previously described passive headspace extraction to identify the
change in components from the lighter burning conditions.

A 5 x 5 cm piece of heavily burned carpet was spiked with 1 uL of ignitable

liquid to assess the effect of matrix interferences without weathering the ignitable liquid.

70

A 5 x 5 cm piece of unburned carpet was also spiked with 750 pL of ignitable liquid and
then heavily burned to simulate the identification of an ILR in the presence of increased
matrix interferences. In both cases, the samples were extracted by the same passive
headspace extraction as in the previous study, eluted with carbon disulfide, and the
resultant extract analyzed by GC-MS. Again, each ignitable liquid was investigated in

triplicate, resulting in a total of 18 simulated ILRs.

4. 2.5 GC—MS Analysis

All analyses were performed using an Agilent 6890 gas chromatograph interfaced
with an Agilent 5975 mass spectrometer (Agilent Technologies, Santa Clara, CA) and
equipped with an Agilent HP-SMS capillary column (30 m x 0.25 mm x 0.25 um). The
inlet temperature was 250 °C and 1 (LL of the sample was injected in splitless mode using
an Agilent 7683B series automated liquid sampler (ALS). The GC temperature program
had an initial temperature of 40 0C for 3 minutes, was ramped to 280 °C at 10 °C/min, and
held at 280 °C for 4 minutes. The transfer line between the GC column and the mass
spectrometer was maintained at 300 °C. The mass spectrometer was equipped with an
electron ionization source operating at 70 eV and a quadrupole mass analyzer operating

in full scan mode (m/z 50-550) at a scan rate of 2.91 scans/s.

4. 2.6 Data Analysis
Data analysis procedures were performed on three separate data sets: the data for
the ILRs extracted from lightly burned carpet, the unweathered ignitable liquid extracted

from heavily burned carpet, and the ILRs extracted from heavily burned carpet. For each

71

data set, total ion chromatograms (TICs) of the neat ignitable liquids, burned carpet, and
ILRs of each liquid were retention time aligned using LineUpTM (Infometrix, Bothell,
WA). Matlab (version 7.4.0.287, MathWorks, Natick, MA) was used to calculate PPMC
coefficients of aligned chromatograms. PPMC coefficients above 0.8 indicated a strong
correlation, between 0.5 and 0.8 indicated a moderate correlation, and below 0.5
indicated a weak correlation [1]. After retention time alignment, chromatograms were
peak area normalized to account for variations in injection volume. For normalization, the
total area under each chromatogram as well as an average total area was calculated. Then,
a normalized chromatogram was generated by dividing each data point in the
chromatogram by the total area under the chromatogram, then multiplying it by the
average total area. PCA was performed by an eigenanalysis of the covariance matrix of
each data set using Matlab. Scores plots were generated in Origin 8 (OriginLab Corp.,
Northampton, MA) and loadings plots were generated in Microsoft Excel (Microsoft

Corp., Redmond, WA).

4.3 Results

4. 3.1 Neat Ignitable Liquids

Total ion chromatograms for each ignitable liquid are shown in Appendix B to

illustrate the chemical content of each liquid.

72

4. 3.2 Minimal Matrix Interferences

The TIC of the burned carpet contained mainly isoparaffinic (branched and
unsaturated Cn-Clz) and alkane (C13-C16) components with some minor aromatic
components (Figure 4.1A). Chromatograms of spiked then burned samples (ILRs) were
dominated by the presence of the ignitable liquid, as expected due to the high spike
volume and light burning conditions. Despite the major contribution of the ignitable
liquid to the ILR chromatogram, some chromatograms showed matrix interferences. For
example, the chromatogram of the gasoline ILR showed a slight rise in the baseline
around 10 minutes (Figure 4.1B), which was consistent with isoparaffinic components
from carpet (Figure 4.1A).

The gasoline and diesel ILRs showed a decrease in the abundance of volatile
aromatic components such as Cz- and C3-alkylbenzenes when compared to the neat
ignitable liquids (Figure 4.1B, 4.1C). Chromatograms of ILRs of liquids without volatile
aromatic components, such as torch fuel and paint thinner, or with a limited number of
alkane or aromatic components, such as lamp oil and adhesive remover, were very

similar to their neat counterparts.

73

7.5E05 T‘ ‘"““ _“_.._.., Wm“ ._-___,_,

 

 

 

 

 

2.5E06 ~——--~—~—----+

 

 

 

 

 

 

 

 

 

14>
J

 

 

 

 

 

 

 

 

Jurlllllhrl .

0 Retention Time (min) 31
Figure 4.1: Chromatograms of A) lightly burned carpet, B) gasoline spiked onto carpet
then lightly burned, and C) neat gasoline. Major components are labeled: 1) Cu-Clz
isoparaffinic/naphthenic components, 2) C13, C14 normal alkanes, 3) diester related to
adipic acid, 4) Cz-alkylbenzenes, 5) C3-alkylbenzenes.

 

74

In the PC scores plot, three main groups of samples were observed (Figure 4.2A).
Neat lamp oil and the lamp oil ILR were positioned negatively on both PC1 and PC2.
Neat adhesive remover and the adhesive remover ILR were positioned positively on PC1
and negatively on PC2. The remaining four ignitable liquids, the corresponding ILRs, and
the burned carpet (unspiked) were positioned around zero on PC1 and positively on PC2.
The PC scores plot showed close association of replicates of all neat ignitable liquids and
replicates of most ILRs (Figure 4.2A). Spread was observed in the replicates of the lamp
oil ILR and adhesive remover ILR, which was caused by retention time misalignments
and greater variability in the burning process than other ILRs. For paint thinner, torch
fuel, and diesel, the ILR was positioned closely to the corresponding neat ignitable liquid
(Figure 4.2B). The gasoline ILR was positioned between unspiked burned carpet and neat
gasoline. The position of samples in the scores plot can be further explained by
interpreting the loadings plot to identify components contributing most to the variance

among samples.

75

 

 

 

 

 

 

 

 

 

 

 

 

2.51307 A
A 5L
B 5’ o
5r.
a 0
N
0
°- A
a
[> I
>
D
-2.5E07 ’
-4.0E07 0 4.0E07
PC1(51.00%)
15E07
° B
:3?
g o
0
C11.
.._,\° 0
\D
5r.
5 0
N
L)
a,
-l.5E07
4.5507 0 1.5507

PC1 (51.00%)

Figure 4.2: A) Full view and B) magnified view of the scores plot for six ignitable liquids
and the corresponding ILRs in the presence of minimal matrix interferences. Neat liquids
are indicated by filled symbols and ILRs are indicated by open symbols. (I) Diesel, (0)
Gasoline, (A) Adhesive Remover, (P) Lamp Oil, (V) Paint Thinner, (O) Torch Fuel,
(it?) Burned Carpet (unspiked).

76

The loadings plots of PC1 and PC2 are shown in Figure 4.3. The first principal
component discriminated the samples based on two groups of components: aromatics
(ethylbenzene, o-xylene, and p-xylene) and alkanes (Cu-C14) (Figure 4.3A). The
aromatic components loaded positively and the alkane components loaded negatively on
PC1. Some minor C3-alkylbenzenes were also shown to contribute positively to the
variance, while minor isoparaffinic and naphthenic-paraffinic components were shown to
contribute negatively on the first principal component.

The presence, absence, or relative ratio of the groups of components determined
the position of the samples on the first principal component in the scores plot. For
example, adhesive remover and gasoline loaded positively on PC1 (Figure 4.2). Adhesive
remover contained only ethylbenzene, o-xylene, and p-xylene, which were the aromatic
components contributing most to the variance described by PC1, while gasoline
contained a higher content of the aromatic components than the alkane components.
Lamp oil, diesel, and torch firel loaded negatively on PC1 (Figure 4.2). Lamp oil
contained only normal alkanes C12-C14, which were the alkane components contributing
most to the variance described by PC1. Diesel contained a higher content of the alkane
components than the aromatic components, and torch fuel contained the alkane
components as well as other minor isoparaffinic or naphthenic components that loaded
negatively on PC1. Paint thinner contained none of the major components identified in
the loadings plot, but was positioned negatively on PC1 (Figure 4.2) because it contained
some of minor isoparaffinic or naphthenic components. The burned carpet, like torch
fuel, contained normal alkanes (C 1 3-C 14) as well as other minor isoparaffinic or

naphthenic components that caused it to load negatively on PC 1.

77

 

 

 

I lqu _ l
V l

Loadings on PC1
O
A
(It ————-%1

 

 

 

 

-03 - - v w .,
0.2 .._ ' “ ' r
B
Isoparaffinic/naphthenic
N ts
U lcomponen 1
a,
::
° 1
3’0 0 -L was --— A A -
.r: f
'6
N
o
..J

 

 

 

 

 

6

 

5

0‘] I I If I Retention Time (min) i r I 31

Figure 4.3: Loadings plots for A) PC1 and B) PC2 for six ignitable liquids and the
corresponding ILR in the presence of minimal matrix interferences. Maj or components
are labeled: 1) ethylbenzene, 2) o-xylene, 3) p-xylene, 4) Cu, 5) C13, and 6) C14.

-0.2

 

78

The second principal component further discriminated the ignitable liquids based
on isoparaffinic and naphthenic-paraffinic content (Figure 4.3B). Ignitable liquids that
contained isoparaffinic or naphthenic-paraffinic components (gasoline, diesel, paint
thinner, torch fuel, burned carpet) had positive scores on PC2 whereas liquids without
those components (lamp oil, adhesive remover) had negative scores on PC2.

The chromatograms of the adhesive remover and lamp oil ILRs were very similar
to the corresponding neat ignitable liquids, showing only a slight decrease in overall
abundance. The adhesive remover ILR shifted slightly less positively on PC1 compared
to the neat adhesive remover. This is consistent with a loss of the aromatic components
that loaded positively on PC1 relative to the neat liquid. The position of the lamp oil ILR
did not shift considerably on PC1 compared to the neat liquid due to the lack of volatile
aromatic components in the ignitable liquid. However, both the adhesive remover and
lamp oil ILRs showed considerable spread in the replicates making association back to
the neat liquid difficult.

In comparing the chromatogram of the gasoline ILR to the neat gasoline, a
decrease in the abundance of aromatic components such as toluene and Cz-alkylbenzenes
was observed in the gasoline ILR (Figure 4.1B, 4.1C). The gasoline ILR was positioned
less positively on the first principal component than the neat gasoline, which is consistent
with the loss of the aromatic components that loaded positively on the first principal
component. However, with the shift in the scores plot, the gasoline ILR was positioned
close to the paint thinner, diesel, and burned carpet samples, which complicated the

association of the gasoline ILR back to the neat gasoline in the scores plot.

79

As with gasoline, the chromatogram of the diesel ILR showed a loss of aromatic
components compared to the neat diesel. As such, the diesel ILR was positioned slightly
more negatively on the first principal component than the neat diesel. Despite the slight
shift, the diesel ILR was still closely associated with the neat diesel in the scores plot due
to the high content of less volatile components that remained unchanged after burning.

To further assess the association of an ILR to a neat ignitable liquid in the
presence of matrix interferences, PPMC coefficients were also investigated. The mean
PPMC coefficients for replicates of the neat ignitable liquids (n=3) and replicates of the
ILRs (n=3) are given in Table 4.2. The mean PPMC coefficients for replicates of neat
ignitable liquids indicated a strong correlation (0.9871-0.9956), as expected. The mean
PPMC coefficients for replicates of ILRs tended to be lower than the coefficients for
replicates of the neat liquids; however, a strong correlation was still observed (0.8422-
O.9933). The three samples with the lowest PPMC coefficients were adhesive remover
ILR (0.8422 + 0.1339), lamp oil ILR (0.9285 + 0.0353), and gasoline ILR (0.9585 :5
0.0219). Visual examination of the aligned ILR chromatograms showed that the low
PPMC coefficients were partly a result of poor alignment of replicates and variability in
the burning process, as noted previously.

Table 4.2: Mean PPMC coefficients + standard deviation for neat ignitable liquids

replicates (n=3) and ILR replicates (n=3) as well as between neat ignitable liquid and the
corresponding ILR (n=9) in the presence of minimal matrix interferences.

 

 

__Ignitable Liquid Neat ILR Neat vs. ILR
Gasoline 0.9875 + 0.0054 0.9585 + 0.0219 0.7331 + 0.0570
Diesel 0.9956 + 0.0032 0.9930 + 0.0020 0.8704 + 0.0070
Adhesive Remover 0.9897 + 0.0082 0.8422 + 0.1339 0.9088 + 0.1143
Lamp Oil 0.9982 + 0.0012 0.9285 + 0.0353 0.9468 + 0.0446
Paint Thinner 0.9871 + 0.0046 0.9882 + 0.0029 0.8921 + 0.0147
Torch Fuel 0.9923 + 0.0057 0.9907 + 0.0011 0.9609 + 0.0102

 

80

All correlation coefficients calculated between ILRs and corresponding neat
liquids (n=9) indicated a strong correlation (0.8704-0.9609), except gasoline which
indicated a moderate correlation (0.7331 + 0.0570) (Table 4.2). Because gasoline has a
relatively high aromatic content, the chromatogram of the gasoline ILR was different
than the neat ignitable liquid, causing a moderate correlation. Diesel, which showed a
strong correlation with the diesel ILR (0.8704 + 0.0070), was also affected by the loss of
aromatic components, but not to the same degree as gasoline due to the greater alkane
content than aromatic content in diesel. Mean PPMC coefficients were highest for
ignitable liquids with no volatile aromatic content to be lost in the burning process, such
as torch fuel (0.9609 + 0.0102) or lamp oil (0.9468 + 0.0446).

PPMC coefficients were not only calculated for replicates, but also calculated for
all pair-wise comparisons of samples (n=9, each liquid analyzed in triplicate). Neat paint ‘
thinner showed a moderate degree of correlation with the burned carpet (0.4983 +
0.0093) due to the similar isoparaffinic content in both samples. Neat diesel and neat
torch fuel showed a moderate degree of correlation (0.5665 + 0.0069) because both
liquids contained similar isoparaffinic and naphthenic-paraffinic components. The major
difference between diesel and torch fuel was the aromatic content and extended range of
normal alkanes in diesel which is illustrated in Figure 4.4. Due to the minimal change in
the chromatograms of neat torch fuel and the torch fuel ILR (0.9609 + 0.0102), a
moderate correlation was also observed for diesel and the torch fuel ILR (0.5563 +
0.0076). An increased correlation was observed between the diesel ILR and neat torch

fuel (0.6698 + 0.0187) as well as the diesel ILR and the torch fuel ILR (0.6886 + 0.0184).

81

This increased correlation was due to the loss of aromatic components in the diesel ILR
compared to the neat diesel, making the diesel ILR more similar to torch fuel.

The PPMC coefficients were useful to assess correlation among samples that were
difficult to associate in the PC scores plot. For example, the gasoline ILR was positioned
close to burned carpet, paint thinner, diesel, and neat gasoline in the scores plot, making
it difficult to associate the gasoline ILR back to neat gasoline (Figure 4.2). However, the
mean PPMC coefficient for the gasoline ILR and the neat gasoline (0.7331 + 0.0570)
showed a moderate correlation, whereas PPMC coefficients for the gasoline ILR and all

other samples showed a weak correlation, with PPMC coefficients below 0.5 (Table 4.3).

Table 4.3: Mean PPMC coefficients + standard deviation indicating correlations of
gasoline ILR with neat ignitable liquids (n=9).

 

 

Comparison Mean PPMC Coefficient Degree of Correlation
Gasoline ILR vs. Gasoline 0.7331 + 0.0570 Moderate
Gasoline ILR vs. Diesel 0.2323 + 0.0287 Weak
Gasoline ILR vs. Paint Thinner 0.4065 + 0.0163 Weak
Gasoline ILR vs. Burned Carpet 0.3463 + 0.0123 Weak

 

82

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2.2E06 "
A 4 5
3
2
1
l I
3
2
l 3
I3 1
- Ir
3.2E06 ' 4
C
3
" 5
L... - ;
0 Retention Time (min) 31

Figure 4.4: Chromatograms of A) neat diesel, B) diesel ILR, and C) neat torch fuel,
illustrating the similar chemical composition between the ignitable liquids. Major
components are labeled: 1) aromatic components, 2) C1 1, 3) C12, 4) C13, and 5) C14.

83

4.3.3 Increased Matrix Interferences

The chromatogram of the heavily burned carpet showed different interferences
than the lightly burned carpet (Figure 4.1A compared to 4.5A). The TIC of the heavily
burned carpet contained common interferences such as styrene and benzaldehyde. These
components were easily identified in the chromatogram of ILRs spiked before and after
burning (Figure 4.5B). Interferences from the carpet complicated the chromatogram of
several ILRs. For example, in the chromatogram of the gasoline ILR (Figure 4.5B),
styrene and benzaldehyde coeluted with C2- and C3-alkylbenzenes, making the
chromatogram of the ILR different than that of the neat gasoline (Figure 4.5C). However,
ignitable liquids containing a complex mixture of components, such as diesel or torch
fuel, masked the presence of many of the interfering components from the heavily burned
carpet.

The results of PCA and PPMC were similar for the ignitable liquids spiked before
burning and after burning, despite the weathering of the ignitable liquids that were spiked
before burning. As such, only the results for the spiked then burned carpet samples will
be discussed to also account for weathering of the ignitable liquids. The PC scores plot
for the spiked then heavily burned carpet was similar to the spiked then lightly burned
carpet (Figure 4.6), which indicates that the samples were discriminated based on

differences in chemical content among liquids rather than matrix interferences.

84

6.0E05 V

 

 

 

 

 

 

 

 

 

 

 

l
l
l
1
111,1. J

1.5E06 3 "

l
l
l

74>

 

 

 

7w

 

 

 

 

 

Jr. ._

0 Retention Time (min) 31
Figure 4.5: Chromatograms of A) heavily burned carpet, B) gasoline spiked onto carpet

then heavily burned, and C) neat gasoline. Major components are labeled: 1) styrene,
2) benzaldehyde, 3) Cz-alkylbenzenes, 4) C3-alkylbenzenes.

 

85

 

2 .5EO7

 

PC2 (27.09%)
C

-2 .5E07
-3 .0E07 0 3.0E07

PC1 (48.02%)

 

 

 

 

 

1.2E07

 

PC2 (27.09%)
C

 

 

 

-1 .2E07

4.2507 0 1.2507
PC1 (48.02%)

Figure 4.6: A) Full view and B) magnified view of scores plot for six ignitable liquids
and the corresponding ILRs in the presence of increased matrix interferences. Neat
liquids are indicated by filled symbols and ILRs are indicated by open symbols. (I)

Diesel, (O) Gasoline, (A) Adhesive Remover, (F) Lamp Oil, (V) Paint Thinner,

(O) Torch Fuel, (79?) Burned Carpet (unspiked).

 

86

Again, three main groups of samples were observed in the PC scores plot: two
groups containing a single neat ignitable liquid and the corresponding ILR (lamp oil and
adhesive remover) and a third group containing the remaining four neat liquids,
corresponding ILRs, and burned carpet. Close association was observed for replicates of
most samples. Slight spread was observed for the adhesive remover ILR due to poor
retention time alignment, but the ILR was still positioned close to neat adhesive remover
in the scores plot. Also, the adhesive remover ILR and corresponding neat liquid were
well separated from other samples in both PC1 and PC2 to simplify the association of the
ILR to neat counterpart. Replicates of the lamp oil ILR were clustered closely and
positioned negatively on PC1 and PC2 like neat lamp oil; however the ILR was
positioned more positively in PC1 and PC2 than neat lamp oil. Despite the distance
between neat lamp oil and the corresponding ILR, no other ignitable liquids were
positioned negative in both PC1 and PC2 to complicate the association of the ILR to the
neat liquid. Again, ILRs for paint thinner, diesel, and torch fuel were positioned near the
corresponding neat ignitable liquid and the gasoline ILR was positioned between neat
gasoline and burned carpet.

In the loadings plot of PC1, the same alkane and aromatic components varied the
most among both heavily and lightly burned samples. However, there was an increase in
contribution of the aromatic components relative to the alkane components for the
heavier burning conditions. Again, the loadings plot for PC2 showed that samples were
discriminated based on isoparaffinic and naphthenic-paraffinic content. The similarity

between the loadings plots for both studies indicates that the variance among samples

87

was dominated by chemical differences in the ignitable liquids, irrespective of the extent
of burning or prevalence of matrix interferences in these studies.

Correlation coefficients of replicates of neat liquids and replicates of ILRs
showed a high degree of correlation (0.9871-0.9982 for neat replicates, 0.8807-0.9932 for
ILR replicates) (Table 4.4). Lower PPMC coefficients were observed for the paint thinner
ILR due to poor alignment and greater variability in the burning process than other ILRs.
The PPMC coefficients between ILRs and corresponding neat ignitable liquids also
showed a moderate to strong degree of correlation (05697-09694) (Table 4.4). The
gasoline ILR (0.5697 + 0.0647) had the lowest average PPMC coefficient because of the
high volatile aromatic content in gasoline that was most affected by burning. Coelution of
components of the carpet with aromatic components of gasoline also contributed to lower
PPMC coefficients. For example, styrene and benzaldehyde from the carpet coeluted with
p-xylene and 1-ethyl-3-methylbenzene, respectively, from gasoline (Figure 4.5B). Neat
ignitable liquids that had the strongest correlation with corresponding ILRs were those
containing a complex mixture of components that masked many of the matrix
interferences, such as diesel (0.9140 + 0.0053) and torch fuel (0.9694 + 0.0124).
Compared to light burning conditions, ignitable liquids that contained few components,
such as lamp oil (0.8022 + 0.0241) and adhesive remover (0.8412 + 0.0244), showed
lower mean PPMC coefficients due to an increased contribution of matrix interferences to
ILR chromatograms.

Again, PPMC coefficients were calculated for all samples (n=9) and a moderate
degree of correlation was observed between the following samples: neat diesel and neat

torch fuel (0.5665 + 0.0069), neat diesel and torch fuel ILR (0.5714 + 0.0076), diesel ILR

88

and neat torch fuel (0.6977 + 0.0248), and diesel ILR and torch fuel ILR (0.7332 :1:
0.0171). The diesel and torch fuel ILRs showed greater similarity compared to the lightly

burned samples due to the addition of the same matrix interferences to the chromatogram

of both ILRs.

Table 4.4: Mean PPMC coefficients + standard deviation for neat ignitable liquid
replicates (n=3) and ILR replicates (n=3) as well as between neat ignitable liquid and the
corresponding ILR (n=9) in the presence of increased matrix interferences.

 

 

Ignitable Liquid Neat ILR Neat vs. ILR
Gasoline 0.9875 + 0.0054 0.9641 + 0.0169 0.5697 + 0.0647
Diesel 0.9956 + 0.0032 0.9932 + 0.0036 0.9140 + 0.0053
Adhesive Remover 0.9897 + 0.0082 0.9884 + 0.0074 0.8022 + 0.0241
Lamp Oil 0.9982 + 0.0012 0.9693 + 0.0136 0.8412 + 0.0244
Paint Thinner 0.9871 + 0.0046 0.8807 + 0.0730 0.7991 + 0.0528
Torch Fuel 0.9923 + 0.0057 0.9840 + 0.0106 0.9694 + 0.0124

 

A moderate degree of correlation was also observed between neat gasoline and
the adhesive remover ILR (0.6174 + 0.0107). However, in examining the PC scores plot,
these two samples were well separated in both PC1 and PC2. Comparison of the
chromatograms indicated the presence of peaks at similar retention times, but with
different abundances. This demonstrates the ability of PCA to discriminate samples based
on differences in peak abundance.

Conversely, PPMC coefficients were useful in associating the gasoline ILR back
to neat gasoline, which was difficult based solely on the PC scores plot (Figure 4.6). The
mean correlation coefficient for the gasoline ILR and neat gasoline (0.5697 + 0.0647)
was much greater than the mean correlation coefficient calculated between the gasoline
ILR and diesel (0.2452 + 0.0249), paint thinner (0.3609 + 0.0290), or burned carpet

(0.4152 + 0.0825), all of which indicated weak correlations.

89

The two previous examples illustrate the utility of using PCA and PPMC
coefficients together. PCA is useful for associating and discriminating a large number of
samples, whereas PPMC coefficients offer a pair-wise comparison that can be used to
associate and discriminate individual samples. When used together, PPMC coefficients
can support the association or discrimination of samples that are positioned closely in the
scores plot. However, PPMC coefficients are independent of peak abundance, whereas

PCA considers differences in peak abundances in associating and discriminating samples.

4.4 Conclusions

By using PPMC coefficients in conjunction with PCA, each ILR was associated
to the neat ignitable liquid and discriminated from matrix interferences under both light
and heavy burning conditions. With heavy burning, ILRs showed a greater loss of
aromatic components compared to light burning. However, the increased weathering of
the ignitable liquids with heavy burning had a minimal effect on the successful
association of the ILR to the neat liquid. Based on the position of the samples in the
scores plot for both light and heavy burning, most ILRs were successfully associated to
the neat liquid using only PCA. However, due to positioning of the gasoline ILR relative
to neat gasoline, gasoline required the use of PPMC as well as PCA to associate the ILR
back to the neat liquid.

Successful discrimination from matrix interferences was achieved regardless of
the extent of burning. With light burning, isoparaffinic and naphthenic components from

the carpet were prevalent, but were generally masked by the components from the

90

ignitable liquid due to the high spike volume. With heavy burning, more abundant matrix
interferences were observed, such as styrene and benzaldehyde. Several components from
the carpet coeluted with components in the ignitable liquids, making the chromatogram
of the ILR visually different from that of the neat liquid. However, the increased matrix
interferences did not affect the successful association and discrimination of ILRs. To
further test the utility of the method, more ignitable liquids and matrices with different
interfering components should be investigated. A lower spike volume should also be
investigated to evaluate the robustness of the method.

The results from the two studies reported herein demonstrate the potential that
PCA and PPMC coefficients may be a useful method for overcoming the two main
problems associated with visual assessment of ILR chromatograms. Ignitable liquid
residues were successfully associated back to the corresponding neat liquid after

weathering of the ignitable liquid and the introduction of matrix interferences.

91

4.5 References

1. Hupp AM, Marshall LJ, Campbell DI, Waddell Smith R, McGuffin VL.
Chemometric analysis of diesel fuel for forensic and environmental applications.
Anal Chim Acta 2008; 606:159-71.

92

CHAPTER 5

CONCLUSIONS

This research involved two major studies with the overall aim of developing an
objective method for associating an ignitable liquid residue (ILR) to a corresponding neat
liquid in the presence of matrix interferences. Previous work has laid the foundation of
the objective method in discriminating diesel samples [1]; however, the method was not
applied to discriminate ILRs from matrix interferences. The first study in this research
was conducted to determine the effect of gas chromatographic (GC) temperature program
on the discrimination of different diesel samples. The second study in this research
addressed the problem that matrix interferences present in the identification of an ILR

extracted from fire debris.

5.1 Effect of GC Temperature Program on the Association and Discrimination of

Diesel Samples

This aim of this study was to determine a GC temperature program that offered a
fast analysis time while still retaining sufficient chemical information for association and
discrimination of ignitable liquids. The effect of GC temperature program on the
association and discrimination of diesel samples was assessed to determine the optimum
temperature program for use in forensic laboratories. Several GC temperature programs
exist for the analysis of ignitable liquids, but a fast GC temperature program is ideal for

implementation in a forensic laboratory to avoid or minimize case backlogs. With faster

93

temperature programs, resolution is lost which may make it difficult to discriminate
samples. As such, this study investigated the potential loss of discrimination among
samples as a result of a loss of resolution.

Pearson product moment correlation (PPMC) coefficients and principal
components analysis (PCA) were used to evaluate the differences in discrimination
among the diesel samples afforded by each of six temperature programs. Temperature
programs ranged from 15 minutes to nearly two hours. Overall, the association and
discrimination of the five diesel samples was not greatly affected by GC temperature
program. With the fastest temperature programs, poor retention time alignment was
observed and replicates were not well associated in PCA scores plots. Because alignment
is important for PCA to ensure that discrimination of samples is based on chemical
content, it is necessary to choose a temperature program that offers adequate resolution
for alignment. As such, the optimum temperature program was identified as the fastest
program that did not result in poor alignment.

The program determined to be most appropriate for fire debris analysis had a total
analysis time of 31 minutes. This program offered association of replicates,
discrimination among the five diesel samples, and a fast analysis time. This temperature
program was based on one used by the National Center for Forensic Science which is
used by many forensic laboratories. Therefore, these forensic labs may not need to
change their standard protocols to incorporate a temperature program to successfully

associate and discriminate ignitable liquids.

94

5.2 Discrimination of Ignitable Liquid Residues from Matrix Interferences Using

Chemometric Procedures

The purpose of this study was to use an objective method to remove the
subjectivity inherent in visual assessment of ILR chromatograms. The two main problems
associated with visual assessment of ILR chromatograms are weathering of the ignitable
liquid due to the burning process and matrix interferences. The objective method
involved PPMC coefficients and PCA to associate and discriminate ILRs in the presence
of matrix interferences.

In this study, pieces of carpet were spiked with an ignitable liquid. Six ignitable
liquids from six classes defined by the American Society for Testing and Materials were
investigated. The carpet was then burned to simulate fire debris. Ignitable liquid residues
were extracted from the burned carpet using a passive headspace procedure and analyzed
by gas chromatography-mass spectrometry (GC-MS). Chemometric procedures were
used to assess the association of the ILR to the corresponding neat liquid in the presence
of matrix interferences. Heavy burning conditions were investigated in addition to light
burning conditions to introduce increased matrix interferences. By using PPMC
coefficients and PCA together, all ILRs were successfully associated to the corresponding
neat liquid under both light and heavy burning conditions. This demonstrates the ability
of the objective method to successfully associate and discriminate ILRs regardless of the
extent of burning. For some samples, such as gasoline, the use of PCA alone was not
sufficient to associate the ILR to the corresponding neat liquid, which demonstrates the

utility of the two chemometric procedures together. The results show the potential for

95

PPMC coefficients and PCA to overcome the problems associated with visual assessment
of ILR chromatograms.

With more work, these data analysis procedures may be implemented in a
forensic laboratory to aid in the identification of an ILR extracted from fire debris.
Because PPMC coefficients offer a pair-wise comparison of samples, they may be
directly applicable for fire debris analysts in comparing an unknown ILR to a neat liquid.
Furthermore, PCA can be used to assess the variation among an entire data set, which
may be useful in determining the class of an unknown ILR by comparing it to a reference

collection of ignitable liquids.

5.3 Future Work

There are many directions that this research could take in the future. Firstly, more
ignitable liquids and matrices should be investigated and compiled into a reference
collection. The reference collection of ignitable liquids should also contain liquids at
many levels of evaporation. More matrices, such as upholstery, clothing, newspaper, and
other household items should be examined to identify potential interferences. In addition,
reference collections of ignitable liquids and matrices could be made available to forensic
laboratories as a resource for fire debris analysts in identifying ILRs.

Regular communication with fire debris analysts has introduced the problem of
mixed ignitable liquids in forensic labs. The presence of more than one ignitable liquid
further complicates the identification of an ILR extracted from fire debris. Studies should

be conducted to investigate the association of mixed ILRs to neat liquids. Also, the

96

ignitable liquids in a mixture may be contained at different levels of evaporation. As
such, various levels of evaporation of one or more of the ignitable liquids in the mixture
should be examined. The incorporation of multiple ignitable liquids may demonstrate the
robustness of the objective method developed for this research.

Another problem suggested for further research is the extraction of ILRs from
mixed matrices. At the scene of a fire, law enforcement officers rarely collect discreet
pieces of fire debris, but rather a mixture of debris that may include carpet, soil, and
various paper products. The matrix interferences contributed from each type of matrix
may further complicate the identification of an ILR, but could be overcome with the
objective method used to discriminate ILRs from a single matrix in this research.

A major consideration of this research was firture implementation in a forensic
laboratory. As such, the studies reported herein focused on the application of
chemometric procedures to fire debris analysis and considered common problems
encountered in forensic labs. With further research, the objective method developed in
the course of this research could be very powerful tool for fire debris analysts. Rather
than replace the analyst, this method could remove the subjectivity associated with the
identification of an ILR to help an analyst. Further work needs to be done to make the
objective method suitable for a forensic setting. To be implemented in a forensic lab, the
method needs to be time efficient, simple, and easy to use. Currently, the method is very
time and labor intensive; however, if the method were automated and easily integrated
into commercial software, it could prove to be invaluable for fire debris analysts in

identifying an ILR extracted from very complex matrices.

97

5.4 References

1. Hupp AM, Marshall LJ, Campbell DI, Waddell Smith R, McGuffin VL.
Chemometric analysis of diesel fuel for forensic and environmental applications.
Anal Chim Acta 2008; 606:159—71.

98

APPENDIX A

PPMC Coefficients for TIC, Alkane EIP, and Aromatic EIP for Diesel Samples Analyzed
by Bach Temperature Program

99

 

Table A] PPMC Coefficients for the TIC of Diesel Samples Analyzed by Method A.

 

 

 

 

 

 

 

 

 

 

 

 

coco; 336 Good

mcood coco; $436

34306 9:36 coco.—

_ .mod N _ mod 52:. coco; 336 3006

Nomad nomad nomad mwgd oooo.. ovoad

Nomad E. mod .336 good ovaod coco.—

oonod m _ M36 33.0 «306 vocad macad oooo._ owamd omoad

03.06 vanod 34de 5306 2.36 236 owaad 925.— mead

thod cowod 3436 good «Food owood owood mood coco.—

nowod 336 20de 25d Shad thwd wmmod ammod mmmad coco; voaad 543.0

ommad ”god 9.96 Send ooowd 386 3.56 23.0 oflod good coco; 3596

$36 N586 .amod Nhomd toad mound NNNod nomad 236 >356 amoad coco.—

Zmad mead wooad 835 Quad mmwad 0236 mawod vwmod @200 @256 336 coco; @306 good
3486 836 9306 335 Nomad o-wod wwwad oomod @306 mde wooad N306 $606 0250 200°
336 good wmcod oohod 336 mmwmd wwwod good 2336 N506 @806 «cm—ad N236 good coco.—
Um~0 020 <m~0 Uv~0 mv~0 <30 Um~0 mmm0 <mm0 UNNO mmm0 <NNO U _ N0 0 _ N0 < 0 N0

02.0
03.0
<mm0
Uv~0
9&0
<v~0
Um~0
mmm0
<mm0
UN~0
0~N0
<Nm0
U_ m0
0 _ m0
< _ N0

100

Table A2 PPMC Coefficients for the TIC of Diesel Samples Analyzed by Method B.

 

 

 

 

 

 

 

 

 

 

 

 

 

dddd. _ «wodd whood

good 82: Nnodd

«hood good coed.—

wdvod howdd on mod dddd._ Sadd dmood

9. mad Sad Smod good dddd._ dmodd

3w mod NA. mod Nomad dmddd dmood deed.—

omdod w _ odd mmddd owndd 33d bmmdd dddd._ «road mhdod

emood bmddd vmood Ahmad wmmod gmdd good ddddg madd

mmdod good 3 odd 0. _ cod dwmdd memdd m3dd Stood code.—

0dnod edgd owvad dmad cdad Nmad mdmod mmmod cdmdd dddd._ moodd wwwdd

nomad Nomad Nvmdd dead Adm—dd 23d nomad owmdd 53d nomad dddd._ modod

Fmod cnmdd hmmdd mtdd 310d mmad mnmod 33d owmmd wwwdd moood coco.—

ndgd dbhod dmhod omhdd good dwodd ddddd mdddd ddodd Ndmdd 33d Scmdd dddd._ vmddd Nwomd
mesod wmnod vmnod hmnod Noodd omodd mdood odwdd good oovod mmmdd hvmdd good dddd._ mcood
Endd Shad Shad mocdd mmcdd Snood Naod «mood mwdod mdmdd hmmdd ommdd Nwood moodd code.—
Umm0 mm~0 <m~0 Uv~0 030 <30 030 030 <m~0 U-0 m-0 <-0 U _ N0 0 _ N0 < _ N0

030
030
<m~0
Ovm0
03.0
<¢N0
030
020
<m~0
UNNO
m-0
<NNO
U_~0
m _ N0
<_ N0

101

Table A3 PPMC Coefficients for the TIC of Diesel Samples Analyzed by Method C.

 

 

 

 

 

 

 

 

 

 

 

 

dddd._ mwood Sood

mwood dddd._ good

good good code.—

wmmod Smod ammod dado; good voood

Nmmod demod w _ mod good dddd._ dwood

unmod evmod w_ mod veood dwood deed.—

vMood wmood wmood oomod wood wood dddo._ mood good

vvood mvood .mood w _ nod edbod deod good dddd._ wwood

mvood wvood VMood v _ bod edhod wood good good dddd._ .

odmod wdmod Emod N~_od mZod oZod admod hovod Ndnod dddd._ Good :ood

o_mod Emod Smod cd_od od_od od_od novod movod edmod Eood dddd._ woood

mmmod Smod NVmod od_od md_od wd_od Xmod o_mod dmmod Food woood coed.—

vonod Shod cvnod vvwod mmwod vaod odood ddood Nowod vaod NNvod vaod dddd._ Hood woood
emnod ovhod vmnod wewod dowod vowod mwwod obwod thod envod Evod mmvod good dddd._ voood
monod Rnod ovnod mvwod mmwod mmwod mdood mdood oowod Evod nmvod oovod woood voood dado.—
Umm0 03.0 <m N0 030 mv~0 <30 Um~0 030 <mm0 UNNO m~N0 <NNO U _ m0 0 _ N0 < _ m0

Umm0
mmN0
<m~0
Uv~0
mvm0
<v~0
Umm0
mmm0
<m~0
UNNO
0~N0
<NNO
0_ N0
0 _ N0
< _ N0

102

Table A.4 PPMC Coefficients for the TIC of Diesel Samples Analyzed by Method D.

 

 

 

 

 

 

 

 

 

 

 

 

dddd. _ vcood ddood

good dddd. _ doood

ddood dhood dddd. _

wvvod movod wovod dddd._ cmood good

mood vvvod Svod cmood dddd._ whood

omvod dcvod covod Good woood deed.—

omwod mowod mowod «wood good good dddd._ wvood N. _ ood

vcwod wowod oowod cwood Noood Stood wvood coed; mNood

ddood _ .ood o_ood good wmood cmood :ood mNood coco.—

vvvod cnvod dvvod mfod dd_od mead nmvod mmvod mvvod dddd._ hmood v~ood

vcvod wwvod wovod oSod Shad _m_od vaod Evod vovod smood dddd._ ovood

odmod vmmod ”.mod v~_od md_od wodod dwvod geod Emod vmood ovood cede.—

Noood Ngod dFod :wod doood Emod .vwod mmwod vvwod N_vo.d vmvod ngod coco; mmood vMood
ooood Ehod odood owood donod good omwod omwod Ewod Evod envod vmvod mmood dddd._ Nvood
ddood AZnod Cood wmwod deod mmwod Ncwod Swod dhwod Evod vaod .vvod vmood Nvood coed.—
Umm0 mmm0 <m~0 U¢N0 0:0 <30 UMNO 0mm0 <m~0 UNNO m-0 <NNO 0_ N0 0 _ m0 < _ N0

UmN0
030
<m~0
03.0
030
<30
020
030
<m~0
UNNO
0NNO
<NNO
0_ m0
050
<_N0

103

Table A5 PPMC Coefficients for the TIC of Diesel Samples Analyzed by Method E.

 

 

 

 

 

 

 

 

 

 

 

 

dddd._ wmood mvood

wmood dddd._ evood

mvood ovood dddd. _

whvod _ovod ommod dddd._ dwood ohood

:vod mwvod Nvmod dwood dddd._ mwood

edmod 2 mod mo mod ooood mwood code.—

Nwwod vcwod m _ood wwood good o _ nod .52: hcood odood

vwwod cowod o _ ood odnod donod dmnod heood dddd._ avood

v _ ood Nowod vmood mccod mvood moood odood wvood coco.—

Evod ovnod wcmod obmod mmmod mmmod :mod memod o_mod dddd._ moood wmood

mwvod Smod mwmod Nomod mmmod Nmmod momod $3.0 memod moood dddd._ ooood

envod vmmod hmmod oomod Emod Emod dmmod Emod Smod wmood ooood coco.—

cdood «mood owood .mwod Ewod cmwod Ewod cowod dmwod dwwod momod oOmod dddd._ wmood owwod
omcod dgod mmbod oonod wmwod :Hwod vaod mmwod mwbod flood :cod Sood «wood dddd._ mnood
nomod mwcod mdood vmbod doood ccbod cowod smood coood vomod Ndood oomod nwwod mmood coco.—
Um m0 030 <m~0 030 030 <v~0 020 030 <mm0 UNNO 0NNO <NNO U _ N0 0 _ N0 < _ N0

Um~0
030
<m~0
03.0
030
<v~0
020
030
<m~0
020
030
<NNO
U _ N0
0 _ N0
< _ N0

104

Table A.6 PPMC Coefficients for the TIC of Diesel Samples Analyzed by Method F.

 

 

 

 

 

 

 

 

 

 

 

 

dddd. _ woood voood

woood dddd._ moood

voood moood dddd. _

omcod Sood good dddd._ moood doood

good dmeod omood moood dddd._ ooood

dcood :bod woood doood ohood coed.—

doood good good Nmnod MMhod unhod deco; Nwood voood

wmood Noood hmood ~anhod hobod cmhod Nwood dddd._ dwood

dhood Noood cdood Food dhood hobod voood dwood odd?—

mwood owcod vwcod o_vod omvod mmvod ooood mucod Nwood deco; nmwod mowod

modod ohood Vmcod Ndmod mmmod wwvod onood :3d owood bmwod dddd._ :ood

good mood Sood ammod onod mmmod mwcod _o©od moood Nowod good deed.—

nvwod waod mmwod dcwod mnwod ovwod Cood ovood vmood mmood oKod ooood deco; good good
vaod ovwod .mwod cowod vowod mowod dmood dNood mmood dmood ooood vocod boood dddd._ mmood
Nonod oonod ccnod mwwod mdood hmwod dcwod good mhwod dvcod wmood flood Nwood mmood coed.—
me0 030 <m~0 Uv~0 mvm0 <v~0 030 080 <mm0 UNNO 0~N0 <NNO 0 _ N0 0 _ N0 <_ m0

030
030
<m~0
030
5&0

<vm0.

Umm0
0R0
<m~0
UNNO
0N~0
<NNO
0. N0
0 _ N0
<_ N0

105

Table A.7 PPMC Coefficients for the TIC of Diesel Samples Analyzed by Method D
Using a Pulsed Automated Injection Technique.

 

 

 

 

 

 

 

 

 

 

 

 

dddo._ wwood good

wwood dddd._ mwood

good mwood code.—

.wmod oomod Synod dddd._ dwood moood

mmmod hdmod odmod dwood dddd._ Rood

cnmod Nomod cvmod moood hoood dado.—

Nvood mvood Nmood vmood odood omnod dddd._ vwood mwood

meood good vmood good odood wmood vwood dddd. _. .wood

dnood wvood Vmood _ :od ooood mmhod mwood good coed.—

momod womod ommod cdvod domod momod domod nowod Emod dddd._ ndood owood

comod momod memod oGod :vod .mvod o_©od osod Cood moood dddd._ onood

Nwmod vomod dvmod odvod _omod Svod mwmod Nowod _omod owood coood deco.—

Ndwod mdwod dohod .dwod cmwod Nowod Nowod Nowod mowod dvcod omood :Vdod dddd._ Noood Kood
Ewod Bwod dwood wmwod vmwod cmwod mowod ddood wowod mndod wncod owdod Noood dddd._ mnood
Swod Ndwod Kood .mwod omwod waod wwwod nowod dowod mmood oncod ovood Kood moood coed.—
Umm0 020 <mm0 “3&0 030 <v~0 UMNO mmm0 <mm0 UNNO mmm0 <Nm0 050 m _ N0 < _ N0

020
030
<m~0
Uv~0
030
<vm0
080
m MNO
<m~0
UNNO
0NNO
<NNO
U _ N0
0 _ N0
< _ N0

106

Table A.8 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed by

Method A.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ food ommod

flood oooo._ flood

ommod flood oooo._

voood mnmod ~“ammod oooo._ Vmood ooood

ooood N. mod w _ Nod vmood oooo._ owwod

NNood vwvod wvvod ooood owwod oooo._

onood m Sod flood mNood mmnod :wod oooo._ Hood woood

flood anod onmod omood wmnod good flood oooo._ ovood

mm _ od mmood Nmood mmood good :wod woood ovood oooo._

meowd owmod movod owmwd onmwd vmmwd ~o_od ooood _o_o.o oooo._ owood oowod

~m_o.o oomod N~vod nomad :mwd ogwd moowd vwwwd good owhod oooo._ Nvood

food womod memod mmmwd vommd wovwd flood ovowd good oowod Nvood oooo._

mmood ovmod memod ooood mmnod omwod ovood ovood ooood Eood ”wood Synod oooo._ food Noood
ovood Smod ommod wood wood omwod Qmood oNood «mood emowd food mwwwd food oooo._ good
mmood momod wmmod .mood good oowod Snood vmood o_ood good Nomad omwwd Noood Good oooo._
030 030 <m~0 030 mvm0 <30 Um~0 mm~0 <mN0 UN~0 0-0 <~N0 0:0 m_~0 <_N0

030
020
<m~0
0:0
030
<v~0
Um~0
0mm0
<mm0
UNNO
m-0
<-0
0_ N0
0 _ N0
<_~0

107

Table A9 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed by

Method B.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ omood .mood

omood oooo. _ woood

Good woood oooo._

mmood mvood omood oooo._ ovood moood

m Sod noood oomod ovood oooo._ oowod

owmod wNood onod moood oowod oooo._

omood owwod good omood o _ ood owood oooo._ o _ ood mvood

vowod m _ood mowod N _ wod ooood vwood o _ ood oooo._ vMood

moood wwwod Swod mwood owood voood mvood vmood oooo._

momod momod womod :owd Nomad mowwd oovod oovod vvvod oooo._ mmood ooood

vmood oomod oomod ooowd nomad vwwwd wovod oovod oovod mmood oooo._ Eood

ovmod .mnod NNmod wmwwd Good onwd wgod mnvod Novod ooood :ood oooo._

ooood ooood Nowod Nowod o_wod woood wwwod _owod .owod o_mod mmmod oNNod oooo._ _owo.o omood
good good Nwood Nowod omwod wowod Good Eood ooood ommod oomod owNod _owod oooo._ ovood
omwod omwod o_wo.o vowod vaod mmwod woood onood wNood womod _ovod ovmod omood ovood oooo._
030 020 <m~0 Uv~0 mvm0 <30 Umm0 mm~0 <mm0 Umm0 0NNO <NNO U _ N0 0 _ N0 < _ N0

UmN0
020
<n~0
UVNO
0v~0
<v~0
Um~0
030
<m~0
UNNO
m-0
<NNO
0_ N0
0 _ N0
<_ N0

108

Table A. 10 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed by

Method C.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ wmood omood

wnood oooo._ omood

onood omood oooo._

ovood ooood voood oooo._ good NNood

owood owood o _ ood good oooofl _ooo.o

mwood owood ooood NNood _ooo.o oooo._

ooood _~oo.o o_oo.o wowod wwood woood oooo._ nmood ovood

vMood oNood vmood owood vowod wowod n mood oooo._ Nvood

mmood Nmood flood .owod NNwod ooood ovood Nvood oooofl

ovood o_oo.o flood ovowd Nvowd wvowd onod ommod omnod oooo._ flood ooood

oonod anod wwmod .owwd ooowd owwwd oovod oovod wovod flood oooo._ vmood

voood Nomod womod ovwwd oowwd wowwd .ovod vovod owvod ooood emood oooofl

oowod Nwwod wowod waod wwwod mowod Vnood good good vaod nomod momod oooo._ wvood omood
vaod vowod good omwod oowod mowod _Noo.o omood flood oomod mmmod :Hmod wvood oooofl mmood
mwwod enwod Ewod omwod oowod ovwod flood mvood _noo.o wmvod _Nvo.o vomod omood mmood oooo._
Uw~0 020 <n~0 030 030 <30 Um~0 03.0 <m~0 UN~0 0-0 <N~0 U_ N0 0 _ N0 < _ N0

030
020
<n~0
030
03.0
<30
030
0.80
<m~0
U~N0
m-0
<-0
U_ N0
0 _ N0
< _ N0

109

Table A] 1 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed by

Method D.

 

 

 

 

 

 

 

 

 

 

 

 

oooofl ooood .vood

ooood oooo._ voood

good voood oooo._

vomod ~o_od momod oooo._ ooood Noood

omvod envod flflmod ooood oooo._ Noood

.ovod vovod Nomod Noood Noood oooo._

mnood omood mmood momod _moo.o voood oooofl ovood o_oo.o

ooood ooood moood voood good _mood ovood oooofl mvood

omood moood mmood _~mod ooood ooood o_ood nvood oooo._

flmod ommod ommod ozod Eood ovood Nomod Nflod Emod oooo._ oNood voood

ovvod onvod _ovod .omod vmood omood Nomod momod nmvod omood oooo._ flood

oomod Nvmod _omod moNod movod o_nod mo_od voood momod voood flood coco.—

momod ommod Nvood ooood omood ovood ooood flood flood moood fl_o.o omood oooo._ Vnood omood
ooood o_ood oNood oomod ooood _ooo.o omood Noood o_ood oflod nmmod ooood «mood oooo._ ooood
noood Noood omood :ood Noood flood ooood flood noood oo_od vomod oo_o.o omood ooood oooo._
020 03.0 <m~0 Uv~0 mv~0 <v~0 Um~0 mmm0 <m~0 UNNO 0~N0 <~N0 U_N0 0:0 <_~0

Um~0
mm~0
<m~0
030
030
<v~0
UMNO
020
<m~0
UNNO
m-0
<-0
U_ N0
0 _ N0
< _ N0

110

Table A.12 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed by

Method E.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ Noood oNood

moood oooo._ _ mood

oNood good oooo._

5 .od oo_od Nomod oooo._ moood _mood

ooood ooood o. mod moood oooo._ o_oo.o

o _ ood moood oomod _moo.o o_oo.o oooo._

.Nnod ooood ooood oonod _ _oo.o moood oooo._ mmood ovood

mmood omood mnood oowod nnvod ownod mmood oooo._ mmood

ooood mmood ooood oovod oo _ od momod ovood mNood oooofl

mmvod o_mo.o oovod ~o~od _fl.od oovod oomod flood momod oooo._ ooood ooood

nmwod nonod omwod _omod oSod ooVod _oNod momod oomod ooood oooo._ Noood

ovnod o_ood vonod _mvo.o vovod Nmmod ommod oovod oovod ooood Noood oooo._

movod _oood voood ooood ooood moood mmood ooood vmood oZod oiod Nomod oooo._ good vmood
ooNod oflod moood omood good Noood _oood _ooo.o _owod oNood owood oo_od good oooo._ omood
ovmod oomod oomod oomod monod ooood ooood _oood oovod oo_o.o voood fllod vmood oNood 88..
020 030 <20 va0 mv~0 <30 UMNO mm~0 <mm0 UN~0 mmm0 <NNO 0:0 030 <_N0

030
020
<m~0
03.0
mvm0
<v~0
Um~0
090
<m~0
UNNO
0~N0
<-0
0:0
0 _ N0
< _ N0

111

Table A.13 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed by

Method F.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ ooood oNood

ooood oooo._ omood

oNood omood oooo._

oomod flood vomod oooo._ flood ooood

ovmod oomod onvod flood oooo._ ooood

oomod ogod moNod ooood ooood oooo._

ooood voood o_ood good good mmood oooo._ omood moood

o_ood ovood oomod o_oo.o ooood moood omood oooo._ onood

moood ooood o—ood ooood ovood voood moood omood oooo._

oflod «Sod oflod Nomod momod oo_od omood _~oo.o ovood oooo._ mmood Nvood

mo_od oflod nflod oomod vwood mo_od omood o_ood ovood mmood oooo._ omood

vo_o.o oo_o.o oo_o.o ooood ooood ocmod o3od good good mvood omood oooofl

oomod womod vflaod ooood .mood oovod mnood Nnood _ooo.o vflod moflod onmod oooo._ mmood ooood
nmood ovood mnmod Nmood omood voood ooood NNood flood oNood .Nood ooood mmood oooo._ flood
oomod nNood vmvod :ood moood oovod .Nood :ood ooood oo.od oo_o.o mmood ooood flood oooog
UmN0 mm~0 <20 03.0 030 <30 030 03.0 <m~0 020 080 <N~0 U_ N0 9 N0 < _ N0

Um~0
mm~0
<m~0
Uv~0
030
<v~0
030
980
<m~0
U~N0
0N~0
<NNO
0:0
0:0
<_N0

112

Table A.14 PPMC Coefficients for the Alkane EIP of Diesel Samples Analyzed by

1que.

Techn

njectlon

Method D Using a Pulsed Automated I

 

 

 

 

 

 

 

 

 

 

 

 

oooofl .oood enood

_ooo.o oooo._ ooood

«mood ooood oooo._

omvod oveod oomod oooo._ moood voood

Novod Smod mmvod moood oooo._ «mood

oomod oovod oovod voood NNood oooo._

ooood .mood ooood oowod ommod oovod oooo._ ooood ooood

woood o .ood o _ ood Novod o _ nod vovod ooood oooofl voood

moood o _ ood good mood Nomod Nnmod ooood voood oooo._

voood vomod omood flood omood omood flood ommod ovnod oooo.. ooood ooood

ooood Noood moood flood omood _oood nomod momod Nomod ooood oooo._ good

omood :ood ovood mmood _moo.o moood ommod ommod oVwod ooood vmood oooo._

omood ooood ovood ooood woood vmood voood Vmood :ood ommod movod _omo.o oooo._ voood flood
ooood ooood _oood ooood ooood ooood emood flood omood Novod vovod Evod voood oooo._ moood
ooood Noood flood mmood omood mmood moood wood wood o_vo.o vfluod omvod flood noood oooo._
Um~0 030 <mm0 Uvm0 03.0 <v~0 Um~0 0mN0 <m~0 UNNO 0N~0 <NNO U _ N0 0 _ N0 < _ N0

Omm0
mmN0
<m~0
U¢N0
9&0
<v~0
Um~0
mm~0
<m~0
030
020
<NNO
U _ N0
0 _ N0
< _ N0

113

Table A. 1 5 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed by

Method A.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ moood mNood

moood oooo._ _.nood

mmood .mood oooo._

oNood ooood Noood oooo._ .mood ooood

ovood _oood voood .mood 88.. m mood

vmood ooood ooood ooood m mood oooo._

flood voood omood flood omood o _ ood oooo._ ovood ooood

ooood omood flood flood omood good ovood oooo._ omood

o_ood ooood _ _ood nNood mmood oNood ooood omood oooo._

flmod omvod oovod _ono.o oomod .omod _mood ooood omood oooo._ voood onood

:nod oovod .ovod ooood oNood Nowod onood ooood vmood voood oooo._ ooood

oonod vovod _omod flood ooood omood voood :ood Nmood omood ooood oooo._

_Nmod good ommod ooood flood vmood ooood omood ooood oflod oflod _o_o.o oooofl flood moood
Nvmod onod _omo.o wmood ovood omood Noood omood ooood Izod _o_o.o oo_od flood oooo._ ooood
Nomod onnod vomod good good vmood moood ooood ooood Slod vflod mo_od moood ooood oooofl
020 030 <m~0 UVNO 0E0 <30 UMNO 0mm0 <m~0 020 020 <~m0 050 0:0 <_N0

020
020
<m~0
Uvm0
0v~0
<30
Umm0
0m~0
<m~0
UN~0
0~N0
<N~0
0_ m0
0_ N0
< _ N0

114

Table A. 16 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed by

Method B.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ voood mvood

voood oooo. _ ovood

flood ovood oooo._

_oood _oood ooood oooofl omood _oood

flood _oood flood omood oooo._ flood

Noood moood ooood _oood wmood oooo._

o _ ood ooood _ _oo.o ooood ooood ooood oooofl ovood ooood

ooood mmood ooood good flood ovood ovood oooofl emood

mmood flood moood n mood ooood omood ooood enood oooofl

oomod flood onod fléod oovod Novod fléod onod _ovod oooofl flood good

momod oomod oomod movod oomod _ovod flmod Nomod ommod flood oooo._ ovood

oovod Evod oomod flnod Noood ommod Nvmod oNood _onod «mood ovood oooo._

ommod Nonod oonod mmood ooood omood mvood Noood mmood mmood omood _oood oooo._ onood ovood
_omod Nomod venod :ood flood :ood ovood ooood owood ooood ooood ooood omood oooo._ ooood
momod Nomod onod flood ooood ooood omood moood moood ooood omood ooood ovood ooood oooo._
020 020 <mm0 030 030 <v~0 UMNO 0mm0 <m~0 UNNO 0NNO <-0 U _ N0 0. m0 < _ N0

030
0m~0
<30
030
0v~0
<v~0
Um~0
0m~0
<m~0
UN~0
0NNO
<-0
n: N0
0 _ N0
<_N0

115

Table A. 1 7 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed by

Method C.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ «mood voood

vmood oooo._ onood

voood omood oooofl

flood _omod omood oooofl flood ovood

mmood ooood mmood flood oooo._ ooood

Nmood ooood omood ovood ooood oooo._

ovood ooood ovood v _ ood ovood flood oooo._ flood ooood

mmood ooood ooood omood oNood omood flood oooo._ flood

moood mmood moood ooood ooood ooood ooood flood oooofl

momod ommod oovod ommod ommod momod momod flood flood oooofl good omood

ozod NNmod vovod onnod onod monod vonod mmood ovood good oooo._ omood

oomod oomod oovod oomod onod oonod nomod mmood _vood omood omood oooo._

ommod oomod flood ooood oNood NNood ooood mnood _Nood ooood ooood omood oooo._ mmood mmood
ommod vovod mmmod flood wmood _nood ooood ooood ovood ooood ooood ooood mmood oooofl ooood
venod NNmod ommod ooood omood vvood moood ooood omood ooood ovood _mood mmood ooood oooofl
Um~0 0m~0 <m~0 030 0v~0 <v~0 UMNO 0mm0 <m~0 U~N0 0NNO <-0 U _ N0 0 _ N0 < _ N0

020
03.0
<n~0
030
0vm0
<v~0
Um~0
0mm0
<m~0
Um~0
0NNO
<~N0
U. N0
0 _ N0
<_N0

116

Table A. 18 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed by

Method D.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ mmood omood

mmood oooofl oNood

omood omood oooo._

ooood ooood ooood oooo._ ooood ooood

_oood noood moood ooood oooo._ ooood

ooood ooood _ _oo.o ooood ooood oooofl

_oood _oood _ _oo.o v _ ood ovood _oood oooo._ ooood Noood

voood Noood onood flood _ _oo.o ooood ooood oooofl ooood

omood ooood omood Nmood Vmood «wood Noood ooood oooo._

oNvod moood _mvod mvmod oomod oomod momod emmod ommod oooofl ooood ooood

oovod oowod novod oonod _oood ooood moood nNood vmood ooood oooo._ ooood

flood omnod flmod omood floood ooood _oood moood voood ooood ooood oooo._

Nonod mnmod Noood moood flood mmood _mood Noood Noood ooood ooood flood oooo._ oNood moood
ooood momod noood ooood omood omood moood oNood _oood ooood omood Noood omood oooo._ omood
ooood monod _Nood flood flood Vnood voood omood ooood NNood omood ooood moood ooood oooo._
Um~0 0w~0 <m~0 0:0 030 <30 Umm0 0mm0 <m~0 UNNO 0NNO <NNO U_N0 0:0 <_N0

Unm0
0m~0
<m~0
Uvm0
0:0
<v~0
Umm0
0mm0
<m~0
UNNO
0NNO
<NNO
U _ N0
0 _ m0
<_N0

117

Table A.19 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed by

Method E.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ nmood flood

nmood oooo._ ooood

flood ooood oooo._

momod Noood voood oooo._ ooood ooood

oomod oNood mmood ooood oooo._ ooood

omood Noood _Nood ooood ooood oooo._

ooood v~ood omood nmood ooood _oood oooo._ _oood ooood

ooood omood Vmood _Nood ooood vnood _oood oooofl ooood

ooood emood mmood ooood ooood vmood ooood ooood coco.—

_moo.o oflod mmood mmmod _omod flmod mmmod moflod Nvmod oooo._ Nmood mvood

_o_od ommod moNod momod omvod ooNod ommod Nomod omvod Nmood oooo._ ooood

oo_od ovmod o_~od flood ogod flood oomod oomod Novod flood ooood oooo._

ommod monod oomod flood voood mmood ooood ooood ooood mmood _mood mmood oooofl ooood ooood
emmod mmmod oomod omood voood omood moood omood ovood ommod ooood flood ooood oooofl voood
Zwod flmod omnod ovood moood ooood omood :ood mmood nmood voood «mood ooood voood oooo._
Umm0 0m~0 <m~0 0:0 0v~0 <30 UMNO 0mm0 <mm0 UNNO 0N~0 <-0 0. N0 0 _ N0 < _ N0

020
020
<m~0
Uvm0
0vm0
<30
Um~0
0mm0
<m~0
U-0
0NNO
<~N0
U _ N0
0 _ N0
< _ N0

118

Table A.20 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed by

Method F.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ moood woood

moood oooo._ Noood

moood Noood oooo._

ooood ooood ovood oooo._ ooood voood

mmood Noood _oood ooood oooo._ ooood

omood ovood flood voood voood oooofl

mmood vmood omood ooood Nnood _mood oooo._ oNood ooood

ooood WNood _oood ooood voood ooood oNood oooo._ ooood

ovood emood vvood omood ooood noood ooood ooood oooofl

momod omvod onmod ooood ooood omood oonod noood ooood oooofl omood _oood

oovod vovod oovod noood ooood _oood _Nood ooood Noood omood oooo._ mmood

onvod Novod omvod voood :ood ooood _oood moood _oood _oood mmood oooofl

:ood oNood _omod voood moood Noood ooood nmood _oood noood ooood oflod oooo._ moood voood
flood omood _vood moood mmood ooood flood _oood emood ooood ooood moood moood oooofl mmood
vmmod oomod flnod ooood ooood _oood ooood ooood ovood ooood oflod Slod voood mmood oooo._
020 030 <m~0 Uv~0 0vm0 <v~0 Um~0 0m~0 <m~0 020 030 <NNO U _ N0 0_ N0 <_ N0

Um N0
020
<30
Uv~0
0E0
<v~0
Um~0
0m~0
<m~0
UNNO
0N~0
<~N0
U _ N0
0 _ m0
< _ N0

119

Table A.21 PPMC Coefficients for the Aromatic EIP of Diesel Samples Analyzed by

Method D Using a Pulsed Automated Injection Technique.

 

 

 

 

 

 

 

 

 

 

 

 

oooo._ mmood _mood

mnood oooofl flood

_mood flood oooo._

ooood vvood _oood oooo._ _oood moood

_oood m mood _nood _oood oooofl ovood

Noood Nmood ooood moood ovood oooo._

mmood _oood ooood omood ooood o _ ood oooofl moood ooood

omood ooood ooood oNood ooood flood moood oooo._ ooood

voood flood v _ ood vmood voood NNood ooood ooood oooo._

omnod oovod vovod ooood voood voood ooood _oood ooood oooo._ _oood moood

ommod ommod oonod ooood _oood flood _vood o_ood flood _oood oooo._ moood

omwod oomod movod ooood moood flood flood moood ooood moood moood oooofl

ooood _omod ommod owood ovood ooood moood voood ooood oflod oflod oflod oooofl voood ooood
o_oo.o oomod _onod ovood omood nmood moood voood moood oflod vflod mflod voood oooo._ _oood
_Nood momod vomod flood omood ovood ooood moood _oood _o_od oo_od mflod ooood _oood oooo._
Um~0 0mm0 <mN0 030 0v~0 <vm0 Umm0 0mm0 <m~0 030 020 <NNO U _ N0 0 _ N0 <_ N0

020
030
<mm0
Uv~0
0v~0
<v~0
Um~0
0mN0
<MNO
UNNO
0NNO
<NNO
U _ N0
0_ N0
< _ N0

120

APPENDIX B

Total Ion Chromatograms of Neat Ignitable Liquids

121

1-5EO6 -
B.1

toluene /C2-alkylbenzenes
C -alkylbenzenes
l—"_I / 3

l—_I

 

 

 

 

 

 

 

 

 

 

 

    

 

 

 

 

 

 

 

 

 

 

 

 

2.21306
3.2 C
14
C13 C
C12 15
C16
C17
C11
C18
aromatics
I I
(1 H A Retention Time (min) 31

Figure 8.1 Total ion chromatogram of gasoline with major components labeled.
Figure 8.2 Total ion chromatogram of diesel with major components labeled.

122

 

 

3.5EO6 * 1

 

 

 

 

 

 

 

 

 

 

B.3 C 13 1
C14

C12

1

C15 1

9n - ‘

6.01306 T
3.4 E
/C2-alkylbenzenes i

/ C3-a1kylbenzenes

1—.

. 1111

0 Retention Time (min) 31

Figure B.3 Total ion chromatogram of lamp oil with major components labeled.
Figure B.4 Total ion chromatogram of adhesive remover with major components labeled.

123

 

3 .2E06

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

3.5 C13
C12
C
11 C14
3-5 Branched C7-C12
I I

I
I I

1' . .

0 Retention Time (min) 31

Figure B.5 Total ion chromatogram of torch fuel with major components labeled.
Figure 3.6 Total ion chromatogram of paint thinner with major components labeled.

124