Australian Journal of Forensic Sciences

ISSN: 0045-0618 (Print) 1834-562X (Online) Journal homepage: http://www.tandfonline.com/loi/tajf20

Analysis of writing/printing paper via

Thermogravimetric Analysis: application in
forensic science

Raj Kumar, Vishal Sharma, Neha Verma, Pawan Kumar Diwan, Vinay Kumar
& Vijay Kumar

To cite this article: Raj Kumar, Vishal Sharma, Neha Verma, Pawan Kumar Diwan, Vinay Kumar &
Vijay Kumar (2017): Analysis of writing/printing paper via Thermogravimetric Analysis: application in
forensic science, Australian Journal of Forensic Sciences

To link to this article: http://dx.doi.org/10.1080/00450618.2017.1310921

Published online: 04 Apr 2017.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=tajf20

Download by: [University of Exeter] Date: 04 April 2017, At: 14:50

Australian Journal of Forensic Sciences, 2017
http://dx.doi.org/10.1080/00450618.2017.1310921

Analysis of writing/printing paper via Thermogravimetric

Analysis: application in forensic science
Raj Kumara, Vishal Sharmaa  , Neha Vermaa, Pawan Kumar Diwanb, Vinay Kumarc 
and Vijay Kumard 
a
Institute of Forensic Science and Criminology, Panjab University, Chandigarh, India; bDepartment of Applied
Science, University Institute of Engineering & Technology (UIET), Kurukshetra University, Kurukshetra, India;
c
Department of Physics, Shri Mata Vaishno Devi University, Katra, India; dDepartment of Physics, Chandigarh
University, Mohali, India

ABSTRACT ARTICLE HISTORY

In this research, we characterized and discriminated 24 different Received 14 December 2016
kinds of writing/printing papers that may or may not have similar Accepted 16 March 2017
compositions by using Thermogravimetric Analysis (TGA). All
KEYWORDS
paper samples were obtained from the local market in and around Thermogravimetric
Chandigarh (Northwest India). Large differences were observed in the analysis; activation energy;
estimation of the kinetic parameters of cellulose and hemicellulose. multivariate analysis;
The calculated percentage weight loss and activation energy for writing/printing papers;
Reaction I and II was further used to determine the thermal stability forensic science
of each sample. Although few differences were observed on visual
examination of the samples, significant differences were obtained on
the basis of their kinetic parameters. Multivariate analysis was further
applied to the results and considerable enhancement of the results
was observed. The current study provides 99.28% discrimination
among all the possible pairs of paper samples. The technique is cost-
effective, involves no sample preparation, consumes less time and
provides reproducible results.

Introduction
In the perspective of archeological and historical artifact studies, thermal analytical methods
have been successfully used to solve identification problems. Although samples such as
foods, organic/inorganic compounds, pharmaceuticals, polymers, electronic materials, soil/
cement and biological samples are analyzed effectively by using thermal techniques, thermal
methods have not gained much importance in the field of forensic science. Rather, modern
analytical methods are the most favored tools for forensic examination of exhibits recovered
from crime scenes. The fundamental advantage of thermal techniques is the ability to meas-
ure changes in the property of a substance with the function of temperature, which helps
in understanding the reaction mechanism, activation energy and its thermal stability1. These
properties further reveal the composition of the exhibits, which can be highly useful for the
identification and classification of the sample under examination.

CONTACT  Vishal Sharma  vsharma@pu.ac.in

© 2017 Australian Academy of Forensic Sciences
2   R. KUMAR ET AL.

The analysis of concrete2, cement3,4, human hair5, individualization of polymers6,7, explo-

sive substances8, analysis of soil samples from shoes and clothes, which ultimately helps in
relating the culprit to the crime scene9, analysis of latex gloves10, pipes11, medicine12, sexual
assault cases13, the effect of heating on human bone14,15, paints16,17 etc., have been studied
using thermal methods within the field of forensic science. However, thermal methods have
not been applied to the forensic examination of paper to date.
In many legal cases, paper becomes prominent evidence to ascertain the connection
between forged and genuine documents. In establishing an opinion of the case, analyses of
substrate (paper) and writing instrument (pen/pencil) are equally significant to determine the
authenticity of the document received. In cases where multiple pages of a document are to
be examined (such as wills, examination sheets, passports, bank passbooks, loan agreements
etc.), each piece of paper must be analyzed individually in order to determine if they belong
to the same or different sources. Finding the integrity and authenticity of the paper sample
requires an in-depth examination of the nature of the paper’s constituents, in order to link the
source of suspected paper samples, an in-depth examination of paper constituents are required.
In ancient times, paper was made from rags, i.e. long chains of cellulose fibers with sizing
(animal glue) as an additional compound. This form of paper possessed superior properties,
such as high durability and long life, compared to modern paper. Manufacturing of modern
paper is solely dependent on wood pulp as a raw material in the form of short-chain cellulose,
hemicellulose and lignin along with various non-fibrous coloring agents and coating/fillers.
Animal glue, which was earlier used as a sizing component, is replaced by alum and rosin in
present manufacturing processes. The use of wood pulp in modern paper reduces its quality
and hence shortens its shelf life in comparison to ancient papers.
The degradation process of wood samples has been successfully analyzed by studying
Thermogravimetric Analysis (TGA) curves in order to classify the samples18−23. Again, the
percentage of the wood constituent varies, depending on the species24−27. Therefore, if the
expert dealing with the samples has a basic knowledge of TGA and its output, the results
obtained can prove to be highly useful for the identification of various species of wood.
The literature reveals work on pyrolysis of paper (cellulose) carried out by many experts.
These studies are mainly focused on the thermal degradation of pure cellulose28−30 and
various types of paper, i.e. coated/uncoated printing and writing31,32, newspaper33,34 and
tissue paper35, in order to convert paper debris into marketable fuels.
However, the possibility of using differential thermogravimetric analysis (DTG) data on
writing/printing papers in their statistical discrimination has not been reported to date. The
DTG curve is a plot of the rate of change of mass (m) with respect to time (t) or temperature
(T) against time or temperature. Moreover, the thermal stability of two essential components
of paper, i.e. cellulose and hemicellulose, is studied by TGA, mostly using two decomposition
processes – one with the decomposition at a lower temperature, which is linked to hemi-
cellulose pyrolysis, and the other with decomposition at a higher temperature, which is
associated with cellulose. The thermogravimetric method is used to analyze the percentage
weight loss with respect to the temperature of paper samples, which is further used in the
calculation of activation energy. The de-convolution of the TGA curve helps to determine
the temperature at which maximum weight loss of different organic components occurs.
Thus, the current study helps in the characterization of organic contents present in writ-
ing/printing papers for their differentiation. It was also found that the quality of paper can
also be determined by its activation energy and percentage weight loss. The various
 AUSTRALIAN JOURNAL OF FORENSIC SCIENCES   3

advantages linked to this process include minimal sample size (5–10 mg), cost-effectiveness,
reliable results and a user-friendly nature. The only limitation of thermal analysis is that the
method is destructive in nature.
Other destructive methods such as Pyrolysis Gas Chromatography - Mass Spectrometry
(Py GC-MS) and Atomic Absorption Spectroscopy (AAS) can also be used for analysis of
paper samples. However, the cost-effectiveness of the TGA instrument allows it to be used
more often than Pyrolysis GC-MS and AAS, and hence it can be easily procured in forensic
science laboratories. While the TGA method involves no sample preparation, consumes less
time and provides reproducible results similar to those of other methods like Fourier
Transform Infrared Spectroscopy (FTIR) and Pyrolysis GC-MS, it also helps in the determina-
tion of the thermal characteristics of the paper sample. Such characteristics can play a sig-
nificant role in quality control assessment, for instance the thermal stability of security
documents (passport, currency etc.) at varying temperatures. TGA could be used in combi-
nation with non-destructive FTIR and destructive Mass Spectroscopy (MS) for forensic anal-
ysis of other samples such as soil, cement etc.

Materials and methods

Writing/printing paper samples
The paper samples selected for the study were A4 size white paper, commonly used for office
work, writing, printing and photocopying purposes. Twenty-four paper samples from a var-
ious manufacturers with the same or different brands were used for the analysis and assigned
letters from A to X. All the samples were collected by random sampling without repetition
from different stationery shops in and around Chandigarh (Northwest India). The number
has no effect on the analysis and its results, but it is necessary for the study to have a sufficient
number of samples to achieve significant results. The use of white office paper in the present
study is based on the fact that this type of paper is most commonly encountered in forgery
cases, as seen in various questioned document laboratories in India. However, the same
study can be extended to the analysis of other paper samples, including colored, legal,
photographic, craft paper etc. A comparison is performed only on those samples that appear
similar to the naked eye and thus require some analytical method to investigate the differ-
ences in their chemical make-up, coatings and color (if any).
The paper samples were kept in a poly bag and stored in a dark place. All sets of paper
were kept in the same environmental conditions to avoid any differences caused by environ-
mental parameters. This method also reduces the chances of differences due to the aging of
paper. Latex gloves, sterilized scissors and forceps were used for the handling of paper
throughout the study in order to prevent any deposition of fingerprints or other contamina-
tion. Five to ten mg of each paper sample was cut to study its thermal nature by using TGA.
The intra-variability of samples was checked by analyzing the single sheet of paper from
three different positions, i.e. top (t), bottom (b) and middle (m). However, the TGA curves of
t, b and m overlap exactly, as shown in Figure 1, and hence show a similar type of thermal
behavior (i.e. percentage weight loss with respect to temperature). Similar results were
obtained when two different sheets, X and Y, from a single ream (single brand) were examined
for weight loss (Figure 2). Such behavior confirms the negligible intra-sample variability. This
may be due to the fact that the single sheet/ream of paper comprises a uniform coating and
homogenous chemical composition.
4   R. KUMAR ET AL.

Figure 1. TGA curve of single paper sample from three different positions.

Figure 2. TGA curve of two randomly selected paper sheets from same paper ream.

Thermogravimetric Analysis (TGA)

The analysis was carried out using a Simultaneous Thermal Analyzer (STA) STA7200 (Hitachi).
This gives simultaneous thermogravimetric, Differential Scanning Calorimetry (DSC) and
Differential Thermal Analysis (DTA) curves, which provide useful information about the ther-
mal properties of the materials. A heating rate of 10 °C min−1 was programmed in the tem-
perature range of 40 °C to 500 °C with a Nitrogen (N2) gas flow rate of 100 ml min−1.
Isothermal and constant heating rate TGA was employed to obtain kinetic information
with the ‘constant heating rate method’ developed by Flynn and Wall36. The advantage of
this method is that less time is required to obtain results. The selected heating rate helps in
obtaining a proper balance between the time of the test and resolution. These parameters
provide an effective analysis of paper samples. All documents were analyzed under similar
conditions.
The paper sample was kept in an alumina crucible and weighed using the automated
balancing system available with the TGA instrument. Experimental factors such as heating
rate, type of atmosphere, the amount of sample etc. affect TGA measurements extensively.
Thus, all factors were kept constant to achieve better results for characterization and
discrimination.
 AUSTRALIAN JOURNAL OF FORENSIC SCIENCES   5

Multivariate analysis
Principal Component Analysis (PCA)
Principal Component Analysis (PCA) is a statistical method for highlighting differences and
similarities in a data-set. PCA was used on the DTG data range 250–400 °C with an interval
of 1 °C. This range was selected because the major weight loss usually occurs in this range
only. A matrix (151 rows × 24 columns) was entered into the software for multivariate
analysis.
The data were normalized before conducting the multivariate analysis. The normalization
was done by dividing the individual data point by the sum of all the data points multiplied
by 100. The variation caused by the number of paper samples was minimized by the nor-
malization process. Therefore, the number of samples has no issue with the differentiation
of papers. Varimax rotation was used to extract the maximum amount of information from
the data-set as it changes the coordinates used in PCA and maximizes the sum of the vari-
ance. All the assumptions are followed for statistical analysis. These statistical approaches
are used to obtain more significant results in a short time. Manual comparison is not possible
for a large number of samples as it may create errors in the interpretation of results.
For a detailed study of the mathematical basis of PCA and cluster analysis calculations,
references37,38 may be consulted.

Results and discussion

Characterization of writing/printing papers by TGA
Writing/printing paper is a complex material mainly consisting of cellulose with a fraction
of hemicellulose, CaSO4, CaCO3, additives, binders, resins, coating materials and some impu-
rities. TGA of all the paper samples was performed under the controlled atmosphere of N2
gas, and their thermal as well as kinetic behavior was studied in detail.
The TGA curves along with derivative of mass loss with respect to the temperature curve
(i.e. DTG curves) related to the thermal decomposition of the writing/printing paper samples
are given in Figure 3. It was observed that the completion of pyrolysis takes place in two
main reactions because two significant and distinct weight loss phenomena occurred in the
temperature range 250–400 °C, as shown by the TGA curves.

Figure 3. TGA and DTG curves of Class I and Class II paper samples.
6   R. KUMAR ET AL.

The overall reaction comprises the degradation reactions of cellulose, hemicellulose and
other additives of writing/printing papers. The first pyrolysis reaction takes place in the
temperature range 250–320 °C. The outcomes of the reaction were volatile gasses and inter-
mediates. The second reaction was completed between the temperature regions 320 °C and
400 °C. This reaction gives volatiles and solid residues as products.
The parabolic peak, between 40 °C and 80 °C, shown by all the paper samples is due to
the presence of moisture content attributed to humid environmental conditions. The amount
of moisture absorbed by the paper samples depends on the storage conditions (humidity)
in which the paper is kept. In the current research, all samples were stored in the same
environmental conditions and hence may have acquired the same moisture content, which
is evident from a common peak (40–80 °C) in all paper samples. Hence, this temperature
range has not been included in the PCA analysis.
Again, as hemicellulose requires a lower temperature than cellulose for degradation, this
type of thermal behavior is explained by detailed examination of their chemical structure
and formulation, as detailed below.
Hemicellulose is made up of low molecular weight polysaccharide units, including
D-mannose, D-xylose, D-glucose, D-galactose, L-arabinose etc., which are linked together
with β-1,4 glycosidic bonds and sometimes with β-1,3 glycosidic bonds. Because of their
amorphous nature, short chains and the low molecular weight of sugar molecules, a lower
temperature is required for their thermal degradation, as summarized in Figure 4. Degradation
of hemicellulose starts at 250 °C and is completed at 320 °C. Its maximum weight loss occurs
at 319 °C at the rate of 2.03 wt. % °C−1. Hence, the appearance of the shoulder in the convo-
luted peak confirms the presence of hemicellulose.
However, cellulose is a linear polymer that is composed of D-glucose units linked in con-
junction with β-1,4 glycosidic bonds. These glucose units are linked together by hydrogen
bonds and van der Waals forces to form long chain macromolecules. Therefore, cellulose has
higher thermal stability than hemicellulose, and its pyrolysis is completed in the temperature
range of 321–400 °C. It shows maximum weight loss at 348 °C at the rate of 4.23 wt. % °C−1.
The temperature at which maximum weight loss occurs for cellulose and hemicellulose is

Figure 4. TGA and DTG curves of Class V paper samples along with the degradation steps observed during
thermal analysis.
 AUSTRALIAN JOURNAL OF FORENSIC SCIENCES   7

Figure 5. The temperature at which cellulosic compounds lost their maximum weight.

Table 1. Class of paper samples with their characteristic differences.

Class Sample ID Characteristic differences Organic constituents
I A, K Parabolic peak at 40–80 °C and 400–500 Cellulose, Hemicellulose
°C, main peak with a shoulder
II B, C, G, I, J, M, N, O, Q, R, S, U, W Parabolic peak at 40–80 °C, main peak Cellulose, Hemicellulose
with a shoulder
III D, T, X Parabolic peak at 40–80 °C, two main Cellulose, Hemicellulose
peaks
IV E, L, P, V Parabolic peak at 40–80 °C, convoluted Cellulose, Hemicellulose
main peak
V F and H Parabolic peak at 40–80 °C, main peak Cellulose
without a shoulder

determined by the de-convolution of the DTG curve in the range of 240–400 °C, and the
results are shown in Figure 5.
Besides the organic degradation, some samples, such as A, D, K, T and X, along with
moisture and organic curves, showed some extra peaks in the temperature range 400–500
°C. This type of behavior may be explained by the presence of some impurities or of thermally
less stable inorganic ingredients during the coating of paper in the manufacturing process.
In other paper samples, this region belongs to the presence of only solid residues.
Based upon the thermal nature of organic and inorganic compounds, we classified the
paper samples into five different classes. The classification is based on the characteristic
differences observed between the thermal curves obtained after thermal degradation. Each
paper sample gains a different number of organic and inorganic compounds during the
manufacturing process, and hence each class shows distinct characteristic thermal curves.
The variation in their concentration is often made by the manufacturer to provide differences
in the quality of the paper. These variations in concentration may result in discrimination
within the same class of paper. Table 1 shows the paper samples with their respective class
and characteristic differences among the class.
The grouping of papers has been done by examining the TGA and DTG curves as shown
in Table 1 and Figure 3, studying the characteristic differences in the TGA and DTG curves
8   R. KUMAR ET AL.

such as parabolic peaks at 40–80OC and 400–500OC, main peaks with and without a shoulder,
two main peaks and convoluted main peaks.
The segment of the curve where sharp changes occur in the weight loss of samples should
be examined carefully. In the present case, the weight of paper samples reduces throughout
the TGA, but a sharp decrease in weight occurs in the temperature range 250–400 °C and
around 500 °C. Therefore, a cautious examination of thermal curves helps in explaining the
intra- as well as inter-variations present in the class of paper samples. From Table 1, it can
be concluded that all samples except samples F and H have cellulose and hemicellulose as
their main organic constituents. These two samples carry only cellulose as the organic mate-
rial in their formulation, and hence both are included in a separate class, i.e. Class V.
Class I consists of two samples, i.e. samples A and K. This class shows characteristics of
two parabolic peaks and one main peak with a shoulder as distinguishing features from
Class III. Class III shows two prominent peaks and a parabolic peak, which are apparent only
in this group. Class II contains the maximum number of paper samples, i.e. 13 samples. This
class shows the most common thermal behavior of paper samples and hence is discriminated
from other classes. Similarly, Class IV samples show convoluted peaks, which were not shown
by any other class.
The above discussion concludes that intra-variations are present in each class of paper
sample. To study the presence of inter-variation within the class, some other parameters
such as kinetic behavior and percentage weight loss are studied in detail, as follows.

Kinetic study of writing/printing paper

The kinetic study is an essential in-depth examination of pyrolysis reaction mechanisms.
Numerous kinetic models have been suggested for the pyrolysis of wood species. There are
two reaction models31,32 that were proposed to show the pyrolysis of coated/uncoated print-
ing and writing papers:

However, from a forensic point of view, the basic degradation process with a simplified
reaction mechanism is adequate to explain the kinetic behavior of materials. Therefore, the
minimum volatile segments with conjugated one-step reaction parameters were used to
describe maximum kinetic information for all 24 paper samples.

Calculation of rate constant

The rate constant is dependent on the order of reaction (n). The order of reaction is generally
taken as 139 for most unknown reactions. Selecting a different value for order of reaction
mathematically indicates that the initial mass has an impact on the rate of weight loss. As
the TGA data are normalized in order to overcome the problem related to differences in the
weight of samples, all the kinetic parameters, i.e. activation energy and rate constant, are
calculated with the order of reaction = l. The reactions considered below will, therefore, be
written with n = 1. For first-order reaction, the rate constant is calculated as follows;
[A]t = [A]0 e−kt
 AUSTRALIAN JOURNAL OF FORENSIC SCIENCES   9

where,
[A]0 = Paperweight recorded initially at time = 0 sec,
[A]t = Paperweight at time = t sec,
k = is the rate constant.

Calculation of activation energy (Ea)

Activation energy is the energy required by the reactant molecules to form products. Without
the threshold Ea, the reaction between the reactant molecules cannot be accomplished.
According to the Arrhenius equation40, the rate constant of the reaction is related to the
temperature as follows:

k = A × e−E a ∕RT
where,
A = a constant or pre-exponential factor,
k = rate coefficient or rate constant,
R = the gas constant (8.314 JK−1 mol−1),
T = kelvin temperature.
By taking the natural log of the equation, the rearranged equation follows a straight line
layout:
Ea [ 1 ]
ln k = − + ln A
R T

y = mx + b
Therefore, the activation energy is calculated by plotting the graph between ln k vs. 1/T. The
resulting slope of the graph is equal to –Ea/R.

Eight temperature data points from the temperature range, i.e. 250–400 °C, are selected
for calculation of the activation energy of hemicellulose and cellulose. Four data points
belong to Reaction I and four belong to Reaction II. Figure 6 represents such a plot between
ln k vs. 1/T for both reactions, i.e. Reaction I for sample A and Reaction II for sample B.
The slope of ln k vs. 1/T helps in the determination of activation energy. Those reactions
that require less activation energy take place more easily than those requiring high activation
energy. Each sample shows positive activation energy. However, in the temperature range
400–500 °C, some of the samples show negative activation energy. This might occur because
of the sudden change in the reaction mechanisms. However, in the current study we are not
focusing on this, as the main aim here is to characterize and differentiate the paper samples
for forensic application.
10   R. KUMAR ET AL.

Figure 6. Plot between ln k vs. 1/T for determination of activation energy of the reactions.

Table 2. Calculated activation energy and R square value for all paper samples.
Reaction I Reaction II
Sample name Activation energy (kJ/mol) R square Activation energy (kJ/mol) R square
A 40.98 0.89 2.15 0.82
B 44.09 0.91 62.25 0.90
C 40.04 0.91 59.86 0.88
D 39.92 0.85 54.49 0.87
E 55.00 0.98 45.71 0.84
F NA NA 76.67 0.89
G 31.00 0.88 79.26 0.93
H NA NA 73.99 0.92
I 46.99 0.91 58.40 0.87
J 52.10 0.93 37.03 0.75
K 43.46 0.89 53.10 0.84
L 52.96 0.99 48.72 0.88
M 43.42 0.89 59.26 0.88
N 44.75 0.90 60.83 0.90
O 38.78 0.90 80.10 0.94
P 52.45 0.99 40.10 0.84
Q 40.33 0.91 57.36 0.86
R 43.17 0.88 56.68 0.88
S 44.95 0.89 59.80 0.90
T 37.91 0.92 62.51 0.90
U 37.61 0.90 79.03 0.93
V 53.62 0.99 45.03 0.86
W 42.69 0.91 64.20 0.90
X 42.09 0.91 15.64 0.53

Assuming the Arrhenius equation, the calculated activation energies are presented in
Table 2. This shows activation energy in the temperature ranges of 250–320 °C and 320–400
°C, which corresponds to the two principal reactions occurring during the pyrolysis of writing
paper. The calculated Pearson coefficient shows very high correlation values in Reactions I
and II. Moreover, the activation energy is high in Reaction II as compared to Reaction I, except
for samples A, E, J, L, P, V and X. These samples show specific characteristics in the classifica-
tion of paper, as already shown in Table 1. Water evaporation requires lower activation energy,
which is not shown in the table. Further, the activation energy helps in the differentiation
of paper samples as all samples have a different energy. The differences in the energy may
be caused by the quality as well as the quantity of organic and inorganic materials used
during the paper manufacturing process.
 AUSTRALIAN JOURNAL OF FORENSIC SCIENCES   11

Figure 7. Paper samples with a varied heating rate.

Table 3. Comparison between heating rate and obtained activation energy.

Reaction I Reaction II
Heating rate Activation energy (kJ/mol) R square Activation energy (kJ/mol) R square
25 °C Heating Rate 44.63 0.91 79.48 0.97
10 °C Heating Rate 39.92 0.85 54.49 0.87

Effects of heating rate on activation energy

The rate of heating significantly affects the activation energy of the paper samples. This is
due to the presence of inorganic materials and additives, which may affect the pyrolysis
mechanisms and hence the activation energy. Therefore, to study the effects of heating rate,
a paper sample is taken randomly and analyzed using the TGA instrument. The DTG curves
of paper samples with heating rates of 10 °C and 25 °C are shown in Figure 7.
A comparison is made between the activation energy of organic materials with a heating
rate of 10 °C vs. 25 °C, and it is concluded that by accelerating the rate of heating from 10 °C
to 25 °C, there is an increase in the activation energy of the paper sample. Moreover, the
value of R square also increased significantly, as shown in Table 3.

Discrimination of paper samples by using kinetic and thermal analytical data

To differentiate the paper samples, the most useful features of TGA and DTA curves, i.e. the
percentage weight loss and the kinetic parameter (activation energy), are considered. Five
to ten mg of paper samples were taken for this study and percentage weight loss was cal-
culated. The weight loss prior to 80 °C corresponds to the weight loss due to moisture con-
tent. Even though the environmental conditions for all the paper samples were kept the
same, the amount of moisture was subtracted from the initial weight of paper samples to
get the weight loss only as a result of organic and inorganic components. The obtained
results are presented in Figure 8. The minimum weight loss occurs for sample J, and hence
this sample shows maximum thermal stability compared to the rest of the samples. However,
sample A shows maximum weight loss and hence has the lowest stability among all the
samples. Some samples also show close resemblance in weight loss, i.e. samples B and D,
samples F, P and T etc.
12   R. KUMAR ET AL.

Figure 8. Percentage weight loss for all paper samples.

Similarly, discrimination based on activation energy could not be significantly achieved

as there are small differences in the activation energy of samples like T–U, N–S etc., as shown
in Table 2. Moreover, the task is very time-consuming and there are chances of error in manual
comparison. Therefore, chemometrics is used to obtain significant results in paper discrim-
ination. The chemometric method is very fast, reliable, and gives repeatable and significant
results, which can be used in a court of law as evidence. A detailed discussion follows.

Discrimination of paper samples using chemometrics

PCA is a statistical multivariate method used to highlight the variation present in a data-set,
which otherwise is difficult to measure. It is helpful in exploring a complex data-set and
visualizes it graphically. The DTG matrix, i.e. 24 × 151 (samples × observations), is subjected
to PCA analysis. Before going into the detailed analysis, the sample adequacy is checked by
KMO/Bartlett’s test, and the significant values permit the expert to further analyze the data
for discrimination purposes. The detailed results and discussion are as follows.
The primary purpose of using PCA was to identify and compute all the data-sets obtained
from DTG analysis. The first four factors explained 83.66%, 11.15%, 3.02% and 1.15% of the
variance respectively in initial eigenvalues. However, only the first two factors match the
Kaiser Criteria41 of eigenvalue ≥ 1. Therefore, 94.81% variance is explained by the first two
PCs in the initial eigenvalue solution, which is significant. However, when we apply the
Varimax rotation, the first three PCs match the Kaiser Criteria and the total variance explained
was increased, i.e. 97.74% (PC1 = 45.54%, PC2 = 44.78%, PC3 = 7.42%), as shown in Table 4.
Thus, Varimax rotation helps in extracting the maximum amount of variation of the data-set
without the loss of information from parent data.
Depending on the loading values of respective PCs, all three principal factors contain a
different number of samples. The results of an orthogonal rotation of the solution are shown
in Table 5. When we use the maximum loading value of the PCs, as shown in bold in the
table, the analysis yields a three-factor solution with a simple structure. Factor 1 contains 11
samples, i.e. B, F, G, H, M, N, O, S, T, U and W. Factor 2 contains 12 samples, i.e. A, C, D, E, I, J,
 AUSTRALIAN JOURNAL OF FORENSIC SCIENCES   13

Table 4. Total variance explained by respective PCs.

Initial eigenvalues Rotation sums of rescaled loadings
Cumulative Cumulative
Component Total % of Variance % Total % of Variance %
1 17.64 83.66 83.66 10.93 45.54 45.54
2 2.35 11.15 94.81 10.75 44.78 90.32
3 0.64 3.02 97.83 1.78 7.42 97.74
4 0.24 1.15 98.98

Table 5. Orthogonal rotated component loading for 24 paper samples.

Samples PC1 PC2 PC3
A .083 .706 .614
B .786 .611 .055
C .661 .744 .059
D .616 .770 .154
E .359 .907 .143
F .787 .592 −.104
G .953 .245 −.069
H .914 .399 .013
I .650 .752 .081
J .195 .824 .514
K .392 .877 .245
L .681 .722 .026
M .698 .695 .151
N .814 .564 .124
O .956 .112 .028
P .342 .900 .157
Q .566 .815 .106
R .677 .715 .165
S .736 .665 .089
T .818 .567 .082
U .943 .319 .005
V .484 .857 .069
W .787 .612 .057
X −.082 .162 .939
Eigenvalue 10.93 10.75 1.78
Percentage of Total Variance 45.54 44.78 7.42
Number of Test Measures 11 12 1

K, L, P, Q, R and V. Factor 3 contains only one sample, i.e. X. To observe the differences among
the samples within each principal factor, a scatter plot is drawn between variables PC2 and
PC3. The variable selection is based on the correlation value42. As these two variables are
not correlated with each other, they are selected for the scatter plot.
The scatter plot between PC2 and PC3 gives a clear visual picture of the relationship
between the variables. It is clear from the plot that there is no correlation between the PC2
and PC3 samples. All sample pairs show significant differences except sample pairs B–W, C–I,
E–P and M–R, as shown in the scatter plot (Figure 9). The four sample pairs show a close
resemblance to each other. The sample paired t-test43 is used to observe whether any differ-
ence is present in their mean value.
The paired sample t-test compares the mean of two related groups of the values to zero.
The output of the t-test is presented in Table 6.
It is clear from Table 6 that out of four sample pairs, two pairs (i.e. C–I and M–R) show
significant p-values and hence the null hypothesis is rejected for these two pairs. Therefore,
14   R. KUMAR ET AL.

Figure 9. Scatter plot of PC2 vs. PC3 loading values.

Table 6. Paired sample t-test for samples showing close resemblance in the scatter plot.
Mean Std. deviation Std. error mean T Df Sig. (2-tailed)
Pair 1 B–W −0.005 0.038 0.003 −1.706 150 .090
Pair 2 C–I −0.066 0.058 0.005 −13.855 150 .000
Pair 3 E–P −0.014 0.139 0.011 −1.236 150 .218
Pair 4 M–R −0.011 0.034 0.003 −3.899 150 .000

Table 7. Reported studies in other forensic domains using multivariate statistical approaches.
References Types of sample Multivariate statistics Concluded remarks
[44] Soil CA, PCA and DA 85% correct classification of soil
samples
[45] Body fluids PCA, Partial Least Square (PLS) Postmortem interval determina-
tion with an error of ±1.92 h
[46] Pharmaceutical tablets Support Vector Machine (SVM) Identify 25 tablets’ product families
model without an error
[47] Paint PCA and Hierarchical Cluster Characterized and discriminated
Analysis (HCA) paint samples
[48] Hair PCA, Soft Independent Modeling of Spectra of differently treated single
Class Analogies (SIMCA), Fuzzy hair can be discriminated
Clustering (FC)
[49] Gasoline ANOVA, CA and PCA Differentiated the gasoline and
light gas condensate fractions
[50] Fire debris PCA Identified plastic materials from
fire debris
[51] Ballpoint pen ink CA and PCA Achieved 99.46% discrimination
power

these two sample pairs are statistically different from each other. However, the two sample
pairs (i.e. B–W and E–P) show non-significant p-values. In this case, the null hypothesis is
accepted, and it is concluded that these two sample pairs are statistically similar. Moreover,
these two sample pairs show similar activation energy, and therefore these pairs are differ-
entiated neither visually nor statistically.
 AUSTRALIAN JOURNAL OF FORENSIC SCIENCES   15

It is clear from the above discussion that only two sample pairs among the 276 sample
pairs are not differentiated. Therefore, a total discriminating power of 99.28% is achieved
together with chemometrics and t-test, which is highly significant.
Table 7 summarizes reported multivariate statistical approaches coupled with analytical
methods in other forensic domains such as soil discrimination, postmortem interval deter-
mination by body fluids, identification of unknown tablets, paint characterization and dis-
crimination, hair discrimination, differentiation of gasoline, identification of plastic materials
from fire debris etc. Moreover; in recent times, chemometrics has been used extensively in
the forensic questioned document for discrimination of ink and paper samples. These sta-
tistical approaches do not suffer from any limitation, except the fact that prior knowledge
of statistics and an adequate number of samples is required.
The probability of different storage conditions and other environmental factors of various
locations in the same geographical region is low, which allows the current study to be conducted
on real case works as well. This study may also help in other fields of forensic science for the
analysis of other exhibits such as gunshot residue, toxicological analysis, wood trafficking deter-
mination, anthropological analysis etc., where destructive techniques are generally practiced.

Conclusion
The current approach for classification, characterization and differentiation of 24 different
types of writing/printing paper has been investigated by TGA and multivariate analysis. The
pyrolysis parameters of cellulose and hemicellulose were explored using TGA. Hemicellulose
degraded at a lower temperature in comparison to cellulose. The percentage weight loss
and activation energy of both reactions were calculated, and are used to determine the
thermal stability of each paper sample. Principal component analysis is used to provide
significant discrimination among the samples. The current study provides 99.28% discrim-
ination among all the possible pairs of paper samples, as confirmed by the significant value
of the paired sample t-test.
Unlike basic methodologies that are currently practiced in many laboratories (visual exam-
ination by altering light/UV radiations, thickness, pH and ash content measurements), the
TGA method is cost-effective, with no sample preparation, and provides reliable results that
can serve as leading testimony in a court of law. In addition, chemometrics for such forensic
discrimination of paper samples enables reduction of the possible classification errors and
better interpretation of data in less time. The present study allows this technique to be used
in the analysis of different kinds of samples related to other forensic disciplines.

Acknowledgment
The first author is very grateful to the UGC, New Delhi for awarding an SRF fellowship. A fellowship
from DFSS (MHA, Govt. of India) is gratefully acknowledged by the second author. The corresponding
author is grateful to DST (Govt. of India) for providing funds through the PURSE grant.

Disclosure statement
No potential conflict of interest was reported by the authors.
16   R. KUMAR ET AL.

ORCID
Vishal Sharma   http://orcid.org/0000-0002-5130-1626
Vinay Kumar   http://orcid.org/0000-0002-6366-079X
Vijay Kumar   http://orcid.org/0000-0002-7438-8496

References
 1. Cheng SZD, Li CY, Calhoun BH, Zhu L, Zhou WW. Thermal analysis: the next two decades.
Thermochimica Acta. 2000;355:59–68.
  2. Alqassim MA, Jones MR, Berlouis LEA, NicDaeid N. A thermoanalytical, X–ray diffraction and
petrographic approach to the forensic assessment of fire affected concrete in the United Arab
Emirates. Forensic Sci Int. 2016;264:82–88.
  3. Alonso C, Fernandez L. Dehydration and rehydration processes of cement paste exposed to high
temperature environments. J Mater Sci. 2004;39:3015–3024.
  4. Gabrovsek R, Vuk T, Kaucic V. Evaluation of the hydration of Portland cement containing various
carbonates by means of thermal analysis. Acta Chim Slov. 2006;53:159–165.
  5. Ionashiro EY, Hewer TSR, Fertonani FL, de Almeida ET, Ionashiro M. Application of differential
scanning calorimetry in hair samples as a possible tool in forensic science. Ecletica. 2004;29:53–56.
  6. Ihms EC, Brinkman DW. Thermo gravimetric analysis as a polymer identification technique in
forensic applications. J. Forensic Sci. 2004;49:505–510.
  7. Tsukame T, Kutszaw M, Selane H, Saitoh H, Shibasaki Y. Identification of polyethylene by DSC.
Application to forensic science. J Therm Anal Calorim. 1999;57:847–851.
  8. Nazarian A, Presser C. Forensic analysis methodology for thermal and chemical characterization
of homemade explosives. Thermochim Acta. 2014;576:60–70.
  9. Lee CS, Sung TM, Kim HS, Jeon CH. Classification of forensic soil evidences by application of THM–
PyGC/MS and multivariate analysis. J Anal Appl Pyrolysis. 2012;96:33–42.
10.  Causin V, Marega C, Marigo A, Carresi P, Guardia VD, Schiavone S. A method based on
thermogravimetry/differential scanning calorimetry for the forensic differentiation of latex gloves.
Forensic Sci Int. 2009;188:57–63.
11. Sajwan M, Aggarwal S, Singh RB. Forensic characterization of HDPE pipes by DSC. Forensic Sci Int.
2008;175:130–133.
12. Pyramides G, Robinson JW, Zito SW. The combined use of DSC and TGA for the thermal analysis of
atenolol tablets. J Pharm Biomed Anal. 1995;13:103–110.
13. Maynard P, Allwell K, Roux C, Dawson M, Royds D. A protocol for the forensic analysis of condom
and personal lubricants found in sexual assault cases. Forensic Sci Int. 2001;124:140–156.
14. Ellingham STD, Thompson TJU, Islam M. Thermogravimetric analysis of property changes and
weight loss in incinerated bone. Palaeogeogr Palaeoclimatol Palaeoecol. 2015;438:239–244.
15. Lozano LF, Peña-Rico MA, Heredia A, Ocotlán-Flores J, Gómez-Cortés A, Velázquez R, Belío IA, Bucio
L. Thermal analysis study of human bone. J Mater Sci. 2003;38:4777–4782.
16. Burmester A. Investigation of Paint Media by Differential Scanning Calorimetry (DSC). Stud Conserv.
1992;37:73–81.
17. Odlyha M. Investigation of binding media of paintings by thermoanalytical and spectroscopic
techniques. Thermochim Acta. 1995;270:705–727.
18. Francisco-Fernandez M, Tarrıo-Saavedra J, Mallik A, Naya S. A comprehensive classification of wood
from thermogravimetric curves. Chemom Intell Lab Syst. 2012;118:159–172.
19. Burhenne L, Messmer J, Aicher T, Laborie MP. The effect of the biomass components lignin, cellulose
and hemicellulose on TGA and fixed bed pyrolysis. J Anal Appl Pyrolysis. 2013;101:177–184.
20. Ding Y, Wang C, Chaos M, Chen R, Lu S. Estimation of beech pyrolysis kinetic parameters by shuffled
complex evolution. Bioresour Technol. 2016;200:658–665.
21. Chen Z, Zhu Q, Wang X, Xiao B, Liu S. Pyrolysis behaviors and kinetic studies on Eucalyptus residues
using thermogravimetric analysis. Energy Convers Manage. 2015;105:251–259.
 AUSTRALIAN JOURNAL OF FORENSIC SCIENCES   17

22. Tarrıo-Saavedra J, Naya S, Francisco-Fernandez M, Lopez-Beceiro J, Artiaga R. Functional

nonparametric classification of wood species from thermal data. J Therm Anal Calorim. 2011;104:87–
100.
23. Sebio-Punal T, Naya S, Lopez-Beceiro J, Tarrio-Saavedra J, Artiaga R. Thermogravimetric analysis of
wood, holocellulose, and lignin from five wood species. J Therm Anal Calorim. 2012;109:1163–1167.
24. Muller-Hagedorn M, Bockhorn H, Krebs L, Muller U. A comparative kinetic study on the pyrolysis
of three different wood species. J Anal Appl Pyrolysis. 2003;68:231–249.
25. Wang S, Wang K, Liu Q, Gu Y, Luo Z, Cen K, Fransson T. Comparison of the pyrolysis behavior of
lignins from different tree species. Biotechnol Adv. 2009;27:562–567.
26. Kilzer FJ, Broido A. Speculations on the nature of cellulose pyrolysis. Pyrodynamics. 1965:151–159.
27. Branca C, Blasi CD. A unified mechanism of the combustion reactions of lignocellulosic fuels.
Thermochim Acta. 2013;565:58–64.
28. Bradbury AG, Sakai Y, Shafizadeh F. A kinetic model for pyrolysis of cellulose. J Appl Polym Sci.
1979;23:3271–3280.
29. Agrawal RK. Kinetics of reactions involved in pyrolysis of cellulose. I. The three reaction model. Can
J Chem Eng. 1988;66:403–412.
30. Yang J, Chen H, Zhao W, Zhou J. TG–FTIR–MS study of pyrolysis products evolving from peat. J Anal
Appl Pyrolysis. 2016;117:296–309.
31. Wu CH, Chang CY, Lin JP, Hwang JY. Thermal treatment of coated printing and writing paper is
MSW: pyrolysis kinetics. Fuel. 1997;76:1151–1157.
32. Chang CY, Wu CH, Hwang JY, Lin JP, Yang WF, Shih SM, Chen LW, Chang FW. Pyrolysis kinetics of
uncoated printing and writing paper of MSW. J Environ Eng ASCE. 1996;122:299–305.
33. Agrawal RK, McCluskey RJ. The low pressure pyrolysis of newsprint. J Appl Polym Sci. 1983;28:367–
382.
34. Nguyen T, Zavarin E, Barrall II EM. Thermal analysis of lignocellulosic materials. Part I. Unmodified
materials. J Macromol Sci Part C. 1981;20:1–65.
35. Wu CH, Chang CY, Hwang JY, Shih SM, Chen LW, FW, Yang WF. Pyrolysis kinetics of tissue paper of
MSW. J Chinese Inst Eng. 1994;17:659–669.
36. Flynn JH, Wall LA. A quick, direct method for the determination of activation energy from
thermogravimetric data. Polymer Lett. 1966;4:323–328.
37. Brereton RG. Chemometrics: data analysis for the laboratory and chemical plant. Chichester: John,
Wiley and Sons Ltd; 2003.
38. Gemperline P. Principal Component Analysis. In: Practical guide to chemometrics. 2nd ed. USA:
CRC Press; 2006.
39. Font R, Martin–Gullon I, Esperenza M, Fullana A. Kinetic law for solids decomposition. Application
to thermal degradation of heterogeneous materials. J Anal Appl Pyrolysis. 2001;58:703–731.
40. Chesworth W. Encyclopedia of Soil Science. Netherlands: Springer; 2008.
41. Kaiser HF. The application of electronic computers to factor analysis. EducPsycholMeas.
1960;20:141–151.
42. Yong AG, Pearce S. A beginner’s guide to factor analysis: focusing on exploratory factor analysis.
Tutor Quant Methods Psychol. 2013;9:79–94.
43. Kim TK. T test as a parametric statistic. Korean J Anesthesiol. 2015;68:540–546.
44. Thanasoulias NC, Piliouris ET, Kotti MSE, Evmiridis NP. Application of multivariate chemometrics in
forensic soil discrimination based on the UV–Vis spectrum of the acid fraction of humus. Forensic
Sci Int. 2002;130:73–82.
45. Zhang J, Li B, Wang Q, Li C, Zhang Y, Lin H, Wang Z. Characterization of postmortem biochemical
changes in rabbit plasma using ATR–FTIR combined with chemometrics: a preliminary study.
Spectrochim. Acta Part A. 2017;173:733–739.
46. Roggoa Y, Degardina K, Margotb P. Identification of pharmaceutical tablets by Raman spectroscopy
and chemometrics. Talanta. 2010;81:988–995.
47. Muehlethaler C, Massonnet G, Esseiva P. The application of chemometrics on Infrared and Raman
spectra as a tool for the forensic analysis of paints. Forensic Sci Int. 2011;209:173–182.
48. Panayiotou H, Kokot S. Matching and discrimination of single human–scalp hairs by FT–IR micro–
spectroscopy and chemometrics. Anal Chim Acta. 1999;392:223–235.
18   R. KUMAR ET AL.

49. Rudneva VA, Boichenkob AP, Karnozhytskiy PV. Classification of gasoline by octane number and
light gas condensate fractions by origin with using dielectric or gas–chromatographic data and
chemometrics tools. Talanta. 2011;84:963–970.
50. Rodrígueza JG, Sissonsa N, Robinsonb S. Fire debris analysis by Raman spectroscopy and
chemometrics. J Anal Appl Pyrolysis. 2011;91:210–218.
51. Kumar R, Sharma V. A novel combined approach of diffuse reflectance UV–Vis–NIR spectroscopy
and multivariate analysis for non–destructive examination of blue ballpoint pen inks in forensic
application. Spectrochim Acta Part A. 2017;175:67–75.