LIB E IZJTNIIVERSITY
MICHIGAN STAT
EAST LANSING, MICI-I 48824-1048

r-r

This is to certify that the
thesis entitled

DEVELOPMENT OF ELECTRONIC NOSE METHOD FOR
EVALUATION OF RESIDUAL SOLVENTS IN LOW DENSITY
POLYETHYLENE

presented by

ISINAY EBRU YUZAY

has been accepted towards fulﬁllment
of the requirements for the

MS. degree in School of Packaging

 

 

/" 17" 4

Major Professor’s Signature

10%. g, 270M

 

 

Date

MSU is an Affirmative Action/Equal Opportunity Institution

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

 

memo?

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DateOue.p65-p.15

 

DEVELOPMENT OF ELECTRONIC NOSE METHOD FOR EVALUATION OF
RESIDUAL SOLVENTS IN LOW DENSITY POLYETHYLENE

By

Isinay Ebru Yuzay

A THESIS

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

MASTER OF SCIENCE
School of Packaging

2004

ABSTRACT

DEVELOPMENT OF ELECTRONIC NOSE METHOD FOR EVALUATION OF
RESIDUAL SOLVENTS IN LOW DENSITY POLYETHYLENE

By
Isinay Ebru Yuzay

High levels of residual volatile organic compounds (VOCs) in packaging ﬁlms
can be a threat to the quality of food products. Currently, a wide range of
techniques are available for assessing the residual VOCs in packaging ﬁlms. An
electronic nose (e-nose) system appears to be a viable alternative to traditional
techniques, providing simplicity and objectivity. In this study, an objective method
has been developed for assessing the residual VOCs from low density
polyethylene ﬁlms. Three VOCs, ethyl acetate, ethyl alcohol, and toluene, were
chosen as model solvents. An Alpha Mos Fox 3000 e-nose system was used for
qualitative and quantitative analysis of residual solvents. Once the optimum e-
nose parameters were determined for all of the model solvent samples, the
measurements were made for single and binary solvent mixtures. The responses
obtained from the e-nose were processed with multivariate statistical analysis
methods (PCA, DFA, and PLS) and compared with gas chromatography, which
is currently used for determining the amounts of residual VOCs in packaging
ﬁlms. The results indicated that there was good agreement between the e-nose
responses and gas chromatography results. Both single and binary solvent

mixture concentrations were successfully predicted with the proposed method.

DEDICATION

This thesis is dedicated to my parents, Sevgi and Fuat Yuzay, and my brother
Atalay Yuzay, whose hearts and prayers were always with me during my
educaﬂon

iii

ACKNOWLEDGMENTS

Over the past two years, I have had lots of encouragement, words of wisdom,
and gracious contributions from many people that made this research possible.
But some of these people deserve special mentioning.

I would like to thank my major advisor, Dr. Susan Selke for her deep insight
and guidance. I would also like to thank my committee members, Dr Bruce Harte
and Dr. Janice Harte for their support and encouragement.

Center for Food and Pharmaceutical Packaging Research (CFPPR)
provided partial support funding for this project. This support is greatly
acknowledged.

A special note of recognition goes to Hsin-Yen Chung not only for being the
best lab-mate with whom I had stimulating scientiﬁc discussions, but also for
patiently helping me when an e-nose and computers were not friendly to me.

Finally, I sincerely express my appreciation to my parents and brother for
their love, support, and reassurance throughout this challenging endeavor. I do

not think words can be ever enough to express my gratitude to them. Thanks.

iv

TABLE OF CONTENTS

LIST OF TABLES
LIST OF FIGURES
1. INTRODUCTION

2. BACKGROUND AND MOTIVATION
2.1 Residual Solvents from Printing Inks and Adhesives
2.2 Residual Solvents from Film Manufacturing Process
2.3 Flexible Packaging Materials
2.4 Packaging Migration
2.5 Analysis of VOCs from Packaging Materials

3. MATERIALS AND METHODS

3.1 Materials

3.2 Statistical Analysis

3.3 Method Development
3.3.1 Representative VOC Selection
3.3.2 Sample Preparation
3.3.3 E-nose Parameter Optimization
3.3.4 Building a Model and Validation

4. RESULTS AND DISCUSSION
4.1 Identiﬁcation of the Single Solvents
4.2 Identiﬁcation of the Binary Solvents
4.3 Quantiﬁcation of the Single Solvents
4.4 Quantiﬁcation of the Binary Solvents

5. CONCLUSIONS and RECOMMENDATIONS

APPENDICES
APPENDIX A Properties of the Representative Solvents
APPENDIX B GC Analysis Conditions
APPENDIX C Outlier Detection
APPENDIX D Experimental and Predicted Area Response Values

BIBLIOGRAPHY

vi

viii

49

51
52
53

57

Table 2.1:

Table 3.1:

Table 3.2:
Table 3.3:

Table 3.4:

Table 4.1:

Table 4.2:

Table 4.3:
Table 4.4:
Table 4.5:
Table 4.6:
Table 4.7:
Table 4.8:

Table 4.9:

LIST OF TABLES

Evaporation rate for the solvents commonly used in printing inks
Effect of incubation temperature on discrimination indexes for
ethyl acetate, ethyl alcohol, and toluene

Effect of vial size on discrimination indices

Optimized data acquisition parameters

The combination of ethyl acetate, ethyl alcohol, and toluene at
Varying injection amounts

Validation data set “unknown” scores for single solvents

(A: Ethyl acetate, B: Ethyl alcohol, C: Toluene)

Identiﬁcation results of the DFA for single solvent validation data
set
Validation data set “unknown” scores for binary solvents
Identiﬁcation results of DFA for binary solvent validation data set
Predicted values for validation data set (for toluene samples)
Predicted values for the unknown toluene samples (5.0 ul)
Predicted values for the blind “unknown“ toluene samples (2.5 ul)
Correlation coefﬁcient values obtained from validation methods
Summary of the correlation coefﬁcient values from the PLS

calibration curves

Table 4.10: Summary of the correlation coefficient values from the PLS

calibration curves

vi

19

23
25

27

34

35

4O

41

43

44

45

46

47

48

Table A1: Typical properties of the representative solvents

Table D1: Experimental and predicted values for ethyl acetate samples

Table D2: Experimental and predicted values for ethyl alcohol samples

Table 0.3: Experimental and predicted values for “unknown” ethyl acetate
samples (5 pl)

Table 0.4: Experimental and predicted values for “unknown” ethyl alcohol
samples (5 pl)

Table 0.5: Experimental and predicted values for blind “unknown” ethyl
acetate samples (2.5 pl)

Table 0.6: Experimental and predicted values for blind “unknown” ethyl

alcohol samples (2.5 pl)

vii

51

54

54

55

55

56

56

LIST OF FIGURES

Figure 3.1: Alpha MOS Fox 3000 e-nose system

Figure 3.2: Typical e-nose sensor response curves for ethyl acetate samples
[A: ethyl acetate samples (1 pl), B: ethyl acetate samples (10pl)]

Figure 3.3: The steps of method development

Figure 3.4: Sample preparation and experimental steps

Figure 3.5: LDPE ﬁlm samples with three solvents at 50°C incubation
temperature in 10 ml vials

Figure 3.6: LDPE ﬁlm samples with three solvents at 60°C incubation
temperature in 10 ml vials

Figure 3.7: LDPE ﬁlm samples with three solvents at 75 °C incubation
temperature in 10 ml vials

Figure 3.8: LDPE ﬁlm samples with three solvents at 60°C incubation
temperature in 10 ml vials (Repeated procedure)

Figure 4.1: DFA plot of ethyl acetate with ﬁve different injection amounts

Figure 4.2: DFA plot of ethyl alcohol with ﬁve different injection amounts

Figure 4.3: DFA plot of toluene with ﬁve different injection amounts

Figure 4.4: DFA obtained from ethyl acetate (A), ethyl alcohol (B), and
toluene (C) training samples

Figure 4.5: The identiﬁcation of the single solvents validation data set

using DFA

viii

Figure 4.6: DFA plot of toluene and ethyl alcohol mixtures with different 37
combinations (30:70 , 50:50, 70:30, 100:0 v/v)
Figure 4.7: DFA plot of ethyl acetate and ethyl alcohol mixtures with different 37
combinations (30:70 , 50:50, 70:30, 100:0 v/v)
Figure 4.8: DFA plot of ethyl acetate and toluene mixtures with different 38
combinations (30:70 , 50:50, 70:30, 100:0 v/v)
Figure 4.9: DFA obtained from binary solvent training sets; ETAC/ETOH 38
(30:70, 50:50, 70:30), ETACfI'OL (30:70, 50:50, 70:30),
TOUETOH (30:70, 50:50, 70:30)
Figure 4.10: The identiﬁcation of the binary solvents - validation data sets 39
(“unknown” samples) using DFA
Figure 4.11: A typical PLS plot of the toluene samples (5 concentration - 5 42
replicates)
Figure 4.12: The PLS pot of toluene training data set validated by validation 43
data set
Figure 4.13: PLS plot of toluene samples constructed by leave-one-out 44
method (samples without 5.0 pl)
Figure 4.14: PLS model of toluene samples with blind ”unknown” samples 45
Figure 4.15: A typical PLS plot of TOLzETOH data sets at 1 pL injection 47

amount

ix

I. INTRODUCTION

Quality control of packaging ﬁlms can be considered as a critical issue in the
food industry due to the possibility of migration of substances from packaging
ﬁlms into food products (Robertson, 1993). Particularly, the volatile organic
compounds (VOCs) released from printing inks, adhesives, and polymers
present in ﬂexible food packaging may give off-odor or taste and signiﬁcantly
alter the quality of food products (Van Deventer and Mallikarjunan, 2002).
Therefore, it is highly desirable to have simple and effective quality control
methods for qualitatively assessing the volatile compounds as well as to
determine the level of residual VOCs in the packaging ﬁlms. Currently, there are
several methods available for analyzing VOCs from plastic packaging ﬁlm and
their printing inks. Gas chromatography (GC), gas chromatography-mass
spectroscopy (GCIMS), multiple headspace solid-phase microextraction (HS-
SPME) and sensory evaluation are extensively used for this purpose (Robertson,

1993)

The chromatographic techniques are used to provide quantitative and
qualitative information about the volatile compounds that are present in
packaging ﬁlms. These techniques allow accurate and objective analysis of
volatile compounds. However, they require high levels of expertise and time
(Pearce et al., 2003). In recent years, use of gas chromatography has become a
standard method of quantifying the level of residual VOCs in packaging ﬁlm.

ASTM F1884, standard test method for determining residual solvents in

packaging ﬁlms, is used to assess the VOCs from packaging ﬁlms in the facilities

manufacturing packaging materials.

Sensory evaluation can also be an effective tool in terms of qualitative
analysis of VOCs in packaging ﬁlms. Although it requires extensive time to train
sensory panelists, well-trained sensory panelists are consistent in their
evaluations. On the other hand, untrained (consumer) panelists can be subjective
in their evaluations. Hence, obtaining a consistent evaluation can be difﬁcult

(Pearce et al., 2003).

While all these techniques can help in the process of assessing and
controlling the amount of VOCs in food packaging materials, each technique has
its own limitations. Therefore, there is a need for an objective tool that can rapidly
provide results for quality control purposes. The electronic nose (e-nose)
systems appear to be a viable alternative to traditional techniques, providing
operational simplicity for assessing VOCs in packaging ﬁlms (Suman et al.,
2003). The e-nose is capable of detecting/recognizing and discriminating
volatiles emitted from liquid or solid samples, using sensor arrays. Moreover, it
can predict the amount of volatiles from a large range of materials (Alpha MOS
Fox 3000 manual, 2001). To the best of our knowledge, the majority of the
literature on the e-nose has been focused on identiﬁcation of VOCs. But, there
are limited amounts of information about quantiﬁcation of VOCs by using the e-
nose. Therefore, of particular interest to us is not only qualitative identiﬁcation of
the residual solvents, but also the prediction of the total amount of residual

solvents. Quantiﬁcation of residual solvents can be very desirable in real life, and

it is much more challenging to predict concentration levels of single solvents or
mixtures than qualiﬁcation of different residual solvents. As a result, in this
research, a considerable amount of effort was made to develop a standard
method for the quantitative determination of VOCs in low density polyethylene
ﬁlms, that can be extremely advantageous in quality control of packaging

materials.

The objectives of this study are:
1. Develop a standard method, by using the e-nose, to assess VOCs from low
density polyethylene ﬁlms.
2. To correlate e-nose results with gas chromatography, which is currently the
standard method for determining the amounts of VOCs in packaging ﬁlm.
It is assumed that the results of this study would enable better and faster
quality control of packaging materials, where quality is associated with total
amount of residual VOCs, since this new method offers operational simplicity

and reproducibility.

2. BACKGROUND AND MOTIVATION

It is well known that the use of plastic ﬁlms for food packaging may cause off-
odor and off-ﬂavor problems which alter the quality of the food product. In most
cases, the migration of residual volatile organic compounds (VOCs) from the ﬁlm
into the food is responsible for the off-odor and ﬂavor (Maneesin, 2001; Das,
2003). VOCs can be deﬁned as volatile compounds whose vapor pressure at
20°C (293.15°F) is 0.01 kPa or greater (Wypych, 2001).

The sources of VOCs in plastic ﬁlms can be classiﬁed as follows: solvents in
the printing inks; solvents in the lamination adhesives; additives in polymers;
possible residuals that form during manufacturing of the ﬁlms (Ezquerro et al.,
2003)

2.1 Residual Solvents from Printing Inks and Adhesives

Printing inks and adhesives composed of solvent-based systems are well
known for being a contributor to off-odor and off-ﬂavor. The residual solvent in
solvent-based printing inks and adhesives that are used in lamination of
packaging ﬁlms can be absorbed by the food and can impact the taste and
aroma of many food products, particularly chocolate, confections, snacks, coffee,
and tea (Lin, 1995). Therefore, the packaging ﬁlm materials are subject to strict
quality control requirements. The FDA has issued a guidance which sets limits on
the level of VOCs in food contact packaging materials. The European Union also
has published regulations on this issue (Robertson, 1993; Williams, 2001;

Begley, 1997). Since regulations become more restrictive for the residual

solvents, there is a growing tendency among converters to reduce and control
the amounts of VOCs in the printing inks and adhesives.

Most of the ﬂexible packaging material used for food products is printed.
During the printing process, the inks and adhesives can be diluted with solvents
which are removed by evaporation as the printed substrates are passed through
the dryers. Flexography and roto-gravure are the main printing techniques used
by converters. It is estimated that about 60% of printed packaging in the USA is
printed by the ﬂexographic process and less than 20% is printed by roto-gravure
(Savastotano, 2003). Flexography is a relief printing process which is widely
used in food packaging due to its high printing speed. In the ﬂexographic printing
process, ink is collected from an ink duct by a rubber roller and transferred on an
anilox roller, which is made of ceramic and is etched with small cells that collect
ink from the rubber roller and deliver it to the photopolymer plates (cliché). A
design on the photopolymer plates is put on the printing substrates. Gravure is
also used in food packaging. It is an intaglio printing process particularly suited to
very long print runs and where very high print quality is required. In the gravure
printing process, a design is mechanically engraved onto the surface of a copper
cylinder. This cylinder is submerged in ink and then the excess amount of ink on
the cylinder is removed by a doctor blade. The ink contained in the engraved
cells is transferred from the cylinder to the printing substrate due to the high
surface energy of printing materials (Robertson, 2003).

In many cases, solvent-based inks are used for both printing processes. A

typical solvent-based ink consists of four essential ingredients. The ﬁrst is

pigments which provide color. The second is binders whose function is to bind
the pigment to the printing surface. The third ingredient is additives which are
used for special purposes such as improving adhesion, controlling the viscosity,
etc. The forth ingredient is solvents. There are two basic functions of solvents: (1)
to dissolve a wide variety of binders such as nitrocellulose, cellulose ethers and
esters, polyamides, and acrylics and to carry the pigment and other components
to the surface of the substrate, and (2) to cause the drying of the printed ink ﬁlm
(Soroka, 1995). The great majority of solvent-based inks are based on solvent
mixtures, usually alcohol, ester, aliphatic and aromatic hydrocarbon and ether
mixtures. There seems to be an almost inﬁnite number of solvents possible for
formulating printing inks. However, there are several important factors to
consider in selecting a suitable combination from among these solvents; boiling
range, ﬂash point, evaporation rate, solvency (solvent power), chemical stability,
and toxicity (Apps, 1958). In this research, only the effect of evaporation rate was
covered.

The evaporation rate is one of the properties of solvents that is vital in printing
inks. A wide range of solvent-based printing inks dry by evaporation of solvents.
The evaporation rate is a function of the vapor pressure, heat of vaporization at
the liquid temperature, the speed of the air stream above the liquid, the thermal
conductivity of the air, the area of the surface, and the differences of temperature
of the evaporating liquid and the air. The rate of evaporation of the solvents is
often given as a numerical value relative to the some other solvent as a standard.

There are two solvents which are normally used as standards, diethyl ether and

n-butyl acetate (Apps, 1958: Thompson, 1998). It is interesting to note that for
unknown reasons, n-butyl acetate was chosen as the standard in the US, but
diethyl ether was used in Europe. Table 2.1 gives evaporation rates for some
solvents commonly used in ﬂexo and gravure printing inks.

Table 2.1 Evaporation rate for the solvents commonly used in printing inks

(Wypych. 2001)
Evaporation Rate

Solvent Name (n-butyl acetate:1.0)

 

Ethyl acetate 6.2
Butyl acetate 1.0
N-propyl acetate 2.76
Ethyl alcohol 3.2
lsopropyl alcohol 3.0
Acetone 10.0
Methyl ethyl ketone 5.7
Hexane 9.0
Benzene 6.1
N-butanol 0.45
Diethanolamine 0.001
Ethylene glycol 0.004
Water 1 .8

 

In addition, some resins (binders) and pigments used in printing inks are
solvent retentive, which can affect the evaporation rate of the solvents.
Generally, soft resins are apt to hold the solvents more than hard resins due to
their chemical structure (Apps, 1958; Thompson, 2001). A higher solid content in
hard binders reduces the amount of solvent that has to be evaporated and
therefore increases the drying speed. Conversely, the soft resins, having

relatively less solids content, tend to retain solvent. As a result, the printing

7

solvents that are retained on the packaging ﬁlm might migrate into food or the
headspace of the package and cause off-odor or ﬂavor. Hence, determining/
controlling the amount of residual solvents left in packaging ﬁlm is of primary
interest in the food packaging industry.
2.2 Residual Solvents from the Film Manufacturing Process

VOCs such as hydrocarbons, alcohols, aldehydes, ketones, and carboxylic
acids in packaging materials can also be formed by thermo-oxidative degradation
of polyoleﬁns during the extrusion coating process. Extrusion parameters such as
temperature and speed have an enormous impact on VOC formation (Ezquerro
et al., 2003). Therefore, it is necessary to develop an effective method to control
the process parameters. Further discussion of this subject is beyond the scope
of this research.
2.3 Flexible Packaging Materials

There are a considerable number of different single and multilayer polymer
ﬁlms used as food packaging materials. The polymer of interest in this study is
low density polyethylene (LDPE), because it is the most used polymer ﬁlm in the
food industry (Robertson, 1993). LDPE offers the advantages of inertness, good
barrier to water, heatsealability, and chemical resistance. However, due to its
polyoleﬁnic nature, LDPE tends to retain non-polar compounds (Lopez-Rubio,

2003). Therefore, it is worthwhile to explore the residual solvents in LDPEs.

2.4 Packaging Migration

Packaging migration is important in the food industry since it can change the
organoleptic properties of the food. As explained by Van Deventer and
Mallikarjunan (2001), the migration of compounds from polymer ﬁlm is controlled
by diffusion. In most cases the molecular transport obeys Fick’s laws of diffusion.
Fick’s ﬁrst law is F = - Dp * (de/dx). Further, if the diffusion coefﬁcient is
independent of concentration, Fick’s second law is de/dt = D,* (dZCp/dxz).

F is the rate of transport per unit area of the polymer, Dp is the diffusion
coefﬁcient of the migrant in the polymer, Cp is the migrant concentration in the
polymer, x is the thickness of the ﬁlm, and t is the elapsed time. There is much
information available about packaging migration, but it is beyond the scope of
this research.

2.5 Analysis of VOCs from Packaging Materials

There is a vast amount of scientiﬁc literature available regarding the use of
chromatographic methods (objective) and sensory panels (subjective) in
determining the residual solvents in packaging materials.

ASTM F 1884-98, standard test method for determining residual solvents in
packaging ﬁlms, and ASTM F151-86, test method for residual solvents in ﬂexible
barrier ﬁlms, have been developed to determine residual solvent levels. They are
used for evaluation of ﬂexible barrier ﬁlms in the facilities manufacturing
packaging materials. These methods are based on gas chromatography. The
known amount of specimen is enclosed in a container and heated to vaporize the

residual solvents into the headspace. Then, the headspace sample is analyzed

using gas chromatography. The studies of Kolb and Ettre (1997) and Leland et
al. (2001) represent up-to date reviews of different 60 techniques for
determination of residual solvents. For further discussion, these references are
recommended.

Another common method used for analyzing the residual solvents in
packaging materials is a sniff test (Huber, 2002). About 1000 cm2 of plastic
packaging material is placed into a clean 1 liter glass jar, which is then sealed.
After storage for a pre-determined time and temperature, evaluation of odor can
be made by sensory evaluation panels. This method is an effective tool in terms
of qualitative analysis of VOCs in packaging ﬁlms. Although it requires extensive
time and money to train sensory panelists, well-trained sensory panelists can
give precise and consistent results. On the other hand, untrained (consumer)
panelists can be subjective in their evaluations since they tend to fatigue. Hence,
obtaining a consistent evaluation can be difﬁcult (Pearce et al., 2003). In addition,
residual VOCs ,due to their toxicity effects, can pose a threat to sensory
panelists.

For these reasons, it is desirable to use an objective and consistent as well as
less labor intensive method. An electronic nose (e-nose) which uses special
sensors to mimic the human nose has been developed for this purpose. It
generates a characteristic ﬁngerprint of an odor which can be compared to data
from different samples, batches and mixtures (Alpha MOS Fox 3000 manual,
2001). Most publications deal with e-nose applications to food, cosmetics, and

environmental analysis (Schaller, 1998). However, there are a limited number of

10

publications referring to packaging applications. Benali et al. (1995) found that an
e-nose using multiple discriminate analysis could discriminate between cakes
wrapped in various qualities of ﬁlm, although a sensory panel did not ﬁnd
differences. Bohatier et al. (1995) used the e-nose to analyze off odors in
polypropylene samples. Polypropylene samples with off odor could easily be
distinguished from an undefective one. Van Deventer and Mallikarjunan (2002)
compared and optimized three electronic nose systems. The performance
analyses showed that based on discriminatory power and practical features, the
Fox 3000 and Cyronose 320 were superior. Suman et al. (2003) used an
electronic nose for qualifying the residual solvent from a wide range of single
layer and laminated substrates. But, the e-nose could not easily discriminate
between polyethylene (PE) samples with methoxyproponal and methoxypropyl

acetate and reference PE samples.

11

3. MATERIALS AND METHODS
3.1 Materials
The following items and equipment were used in this study;
A. Film Samples
The packaging material tested during the study was 1.5 mil thick low density
polyethylene (LDPE) supplied by Cello-Foil Products, Inc. (Battle Creek, MI).
LDPE was selected since it is one of the most often used ﬂexible packaging

materials.
B. Solvents

Three volatile organic compounds, ethyl acetate (esters), ethyl alcohol
(alcohols), and toluene (aromatic hydrocarbons), representing three different
solvent categories, were chosen as models for solvents of interest in ﬂexible food
packaging analysis. Ethyl acetate of 99.8% purity, ethyl alcohol of 95% purity,
and toluene of 100% purity were obtained from Aldrich Chemical Co. (Milwaukee,
WI). These solvents are commonly used in printing ink and adhesive
formulations. The typical physical and chemical properties of the representative
solvents are shown in Appendix A.

C. Equipment

. The Fox 3000 electronic nose (E-nose) system (Alpha M.O.S. SA, Toulouse,
France), HS100 Headspace auto sampler (Alpha M.O.S. SA, Toulouse,

France), Fox 3000 software (Alpha M.O.S. SA, Toulouse, France)

12

An Alpha MOS Fox 3000 E-nose system with two metal oxide sensor arrays
(consisting of 12 sensors) was used for qualitative and quantitative analysis of
residual solvents. Figure 3.1 shows the Alpha MOS Fox 3000 e-nose system. It
consists of three main components: an agitator, a headspace auto sampler, and
sensor arrays. The agitator is the place where the vials are heated up in order to
generate the headspace vapor from the samples. The auto sampler collects the
headspace vapor and injects it into the sensor array. The sensor array consists
of 12 metal oxide sensors. Each sensor has different sensitivity and selectivity to
various chemical compounds. Therefore, a combination of several sensors
provides a unique ﬁngerprint of the samples. This allows to track any variation in
the headspace of the samples. Figure 3.2 shows an example of sensor
responses as a function of time for one of the model solvents. In this ﬁgure the
sensor responses are represented by the relative resistance change, (R0 -R) / Re,
where R0 is the minimum of the resistance reading during the baseline purge
and R is the maximum resistance reading during the vapor exposure. The data
extracted from sensor responses can be used to detect, characterize, identify,
and eventually quantify the volatiles.

The optimized e-nose set up parameters are shown in Table 3.3. The detailed

optimization procedure is explained in section 3.2.3.

13

NT};

-.-1 .':' '
V‘ W 7“» '-;.‘_
W“. M¥IKQ~0',.I.‘. ..
o

 

Figure 3.1: Alpha MOS Fox 3000 e-nose system

iv-u.
a

GI QIII'B EEJE.Q *‘= 39$ Xaxisz 7mm Yaxisz magma/wt

 

   
   
 

- r/r0=0.055l
- r/r0=0.065]

  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 3.2: Typical e-nose sensor response curves for ethyl acetate samples
[A: ethyl acetate samples (1 pl), B: ethyl acetate samples (10pl)]

14

0 An HP 6890 gas chromatograph (Hewlett- Packard, Avondale, PA),
equipped with a gas ﬂame ionization detector and interfaced with
Empower-Waters software was used for quantitative analysis of the
residual solvents in LDPE ﬁlms. All the data were recorded as peak area
response.

The column used in this study was Supelcowax 10 fused silica capillary column
(60 m, 0.25 mm ID, 0.25 pm ﬁlm thickness) (Supelco Inc., Bellefonte, PA). A
detailed summary of the GC conditions employed in this study is presented in
Appendix B.
3.2 Statistical Analysis

The responses obtained from the sensor arrays were processed using
principle component analysis (PCA), discriminate factorial analysis (DFA), and
partial least squares regression analysis (PLS) in order to qualify and quantify the
residual solvents. The models obtained from DFA and PLS analysis were
validated by means of the partial cross-validation (leave-one-out) method. In this
validation method, the data sets obtained during measurements were randomly
divided into a training set, used for building up the model (learning process), and
a validation set, used for validation purposes. Finally, unknown samples were
used to check the performance of the new model.
3.3 Method Development

The method development for assessing the residual solvents from LDPE

using an electronic nose consists of four main steps which are representative

15

VOC selection, sample preparation, e-nose parameter optimization, and building

a model. Figure 3.3 illustrates the main steps of the method development.

 

 

Representative VOC Selection

I

Sample Preparation

E-nose Parameter Optimization
0 Head-space and sensor parameters

I

Building a Model and Validation
0 Analyzing the training and validation sample sets.

 

 

 

 

 

 

 

 

 

 

 

 

- Recognition of the validation and unknown samples (DFA)
0 Prediction of the validation and unknown samples (PLS)

0 Correlation between the e-nose and GC-FID

 

 

 

 

 

 

Figure 3.3: The steps of method development

3.3.1 Representative VOC Selection
In this study, ethyl acetate, ethyl alcohol, and toluene were chosen to
represent different solvent categories used in printing ink and adhesive

formulations.

16

3.3.2 Sample Preparation

LDPE ﬁlm samples were prepared by the method described by Suman et al.
(2003). The samples were cut into 2 x 25 cm strips and then placed into glass
jars. Filter paper inserted into each jar lid was injected with 1.0, 2.5, 5.0, 7.5, or
10 pl of the solvent. Then, the jars were tightly closed. The sealed jars were kept
at 50°C for three hours, and then at room temperature for one day in order to
equilibrate the solvent in the headspace of the jar. After this treatment, the ﬁlms
were used to prepare headspace vials for the e-nose and GC analysis. A

diagram summarizing the sample preparation and experimental steps is shown in

Figure 3.4.

 

2 x 25 cm Film

 

 

 

 

 

Placed into 1L Jar

 

 

 

 

 

 

"Spiked" with the solvent

 

 

 

V

 

Equilibrate

 

@ 50 °C, 3 hrs T room, 1 day

 

 

 

 

 

Put the ﬁlm into head-space vials

 

 

V

 

/—®_\

 

E-nose Analysis

 

 

 

 

 

DFA
Identiﬁcation

 

 

 

 

 

GC-FID Analysis

 

SJ

 

 

PLS
Quantiﬁcation

 

 

 

Figure 3.4: Sample preparation and experimental steps.

17

3.3.3 E-nose Parameter Optimization

The aim of method development is to select operating conditions for an e-
nose analysis which are consistent and reliable. There are several important
parameters that need to be optimized to obtain the maximum performance from
the e-nose (Alpha MOS Fox 3000 Manual, 2001). Therefore, before taking the
real measurements, a number of preliminary tests were made in order to ﬁnd
optimum analysis conditions which were acceptable for all of the solvent
samples.

There were thirteen variables that could be controlled directly in the system:
incubation time, incubation temperature, syringe temperature, syringe type, ﬁlling
speed, ﬂushing time, injection volume, injection speed, vial size, acquisition time,
acquisition period, agitation speed, and delay. To minimize unnecessary
complexity, all efforts were directed towards only the two most important
parameters, incubation temperature and vial size (Deventer and Mallikarjunan,
2002; Alpha MOS Fox 3000 Manual, 2001). The rest of the parameters were set
up as shown in Table 3.3.

Incubation Temperature:

The incubation temperature can be considered one of the salient parameters
that must be controlled since the generation of head-space volatiles varies with
the temperature. Any variation in the temperature may result in changes in
volatile concentrations in the head-space (Van Deventer and Mallikarjunan,

2002). Consequently, the changes in volatile concentrations would directly affect

18

the sensor responses. Table 3.1 shows how the temperature affects the e-nose

analysis results.

Table 3.1: Effect of incubation temperature on discrimination indexes for ethyl
acetate, ethyl alcohol, and toluene.

 

 

Incubation Discrimination Index
Trial Temperature, °C PCA
1 50 82
2 60 93
3 75 87

 

The Alpha MOS Fox 3000 Manual (2001) recommended that the incubation
temperature should be set between 40 and 50°C for highly volatile compounds
and between 80 and 100°C for less volatile compounds.

Van Deventer and Mallikarjunan (2002) compared the performance of three
commercial e-nose systems in terms of detection of volatiles from printed plastic
packaging ﬁlms. In this study, the volatile compounds of interest were not
speciﬁcally named due to a conﬁdentiality disclosure agreement with the
manufacturer; however, the setting parameters for the Fox 3000 e-nose were
reported. The optimum incubation temperature that was chosen for this study
was 75°C.

Suman et al. (2003) also analyzed six different solvents in order to identify
and control the retained solvents in printed ﬁlms. In this study, an incubation
temperature of 60°C was used.

Consequently, on the basis of previous studies, the incubation temperature

was studied within the range of 50 — 75°C. Figures 3.5, 3.6, and 3.7 illustrate the

19

plots of PCA for LDPE ﬁlm samples with three solvents at three different
temperatures. As can be seen, the ﬁlm samples were easily discriminated at 50,
60 and 75°C using principle component analysis (PCA). Since there was no
intersection among three sample groups, high discrimination indices were
obtained (82 for 50°C, 93 for 60°C, and 87 for 75°C). Generally, the higher the
discrimination index, the better is the discrimination between the groups (Alpha
MOS Fox 3000 Manual, 2001). In this study, increasing the incubation
temperature from 50°C to 60°C did give us a higher discrimination index,
whereas increasing the incubation temperature from 60°C to 75°C did not
provide us any signiﬁcant differences in discrimination index and sensor

responses. Therefore, the incubation temperature was optimized at 60°C.

 

 

 

 

 

 

 

 

 

 

 

 

C2: 0.03% Discrimination Index: 82
7
0.009- _
, 0.00% Toluene 11‘
i 0.004-
;g 0.002- MI“
1: ‘i\\ .
.. 0000- . . - Al _15
IN [1.114113
:0 \‘\
: 0.002" \I.»
50-9 BI 16
0004- B _8
0006- .
BI __12‘
”-003?“"—r'cwr"Mr":"iMW’rmn' * " I r" IITI r ““"i""'“i""'“i'“—"i
0.50 0.45 0.40 0.35 0.30 0.25 0.20 015 010 005 I100 0.05 0.10 0.15 0.20 0.25 0.30

 

C1: 99.96%

 

 

 

Figure 3.5: LDPE film samples with three solvents at 50°C incubation
temperature in 10 ml vials.

20

 

C2: 0.06%

 

 

Discrimination Index: 93

 

 

 

 

 

0035- 7 I , 41F: 10_ 3 I
0030- , l : ET AC .
0025- ' ; , L: jg

I

I

l

I

l

I l

L I l

Toluene , ; AIF _10l7

- I . I 41F01_10 0_1I
‘ l

l

.- -EL _. -“—- ~-.~--~_.-_._..__.\s.1

 

0.015-
0.010-
0.005-

circ 1033 I l , , ‘

0.000- = a z 1, f .
.0005— CIF- _;0_13 i l . . , . .
0.010- ' f I z I BIFC 10_e
{1015- I I . I I ;
.: '0'020' ~ % Q . I I ' I ' 'B‘IFr: ‘10_'_12""
0025- I I I I I I BIFC 104

I '0 D307 “’T‘T‘T’T'TT‘TTI 1 I '“"““’"i" I I I 1 1 - 1
1. 40 120 1.00 0.90 0.30 0.40 0.20 0.00 0.20 0.40 0.50 0.00

C12 99.91%

 

 

c2 Afr—1.06:

 

 

 

 

 

I
I
I
l
I

Figure 3.6: LDPE ﬁlm samples with three solvents at 60°C incubation
temperature in 10 ml vials.

 

 

 

02; 000% Discrimination Index: 86

 

 

 

 

 

 

 

 

 

0.002-

0001-

 

0.0004-4-4—7..-

 

 

CZ : 0.002

 

 

0001- ~- -

0.002--- _. ~ »

 

 

 

 

 

 

 

 

 

I

I

I
0.00:1—,, I i
0'50 0.50 0.40 0.30 m 0.0
C1: 100. 0%

 

 

 

 

Figure 3.7: LDPE film samples with three solvents at 75°C incubation
temperature in 10 ml vials.

21

Vial size:

Another important parameter that must be controlled is the vial size. The
literature review showed that there is a discrepancy in reported literature values
for vial size. The ASTM standard D1884-98 (Test Method for Determining
Residual Solvents in Packaging Materials) suggests using 20 ml head—space
vials for the gas chromatographic analysis. Suman et al. (2002) also preferred
using 20 ml vials in their research. However, Van Deventer et al. (2002) chose 10
ml vials for the e-nose analysis. Therefore, it was necessary to examine the
differences between the two vial sizes in terms of discrimination index. Using
PCA, discrimination of the three solvents from each other was excellent with the
10 ml vials, while discrimination failed with the 20 ml vials. It is well known that
the sensor responses depend on the incubation temperature. The solvent
concentration in the headspace of the vial also increases with the increase of
incubation temperature. Consequently, a high discrimination index can be
obtained. But, in the case of 20 ml vials, incubation temperature or time was
evidently not enough to get maximum sensor responses due to the large vial
volume. The effect of vial size on discrimination indices is shown in Table 2.2.

As a result, 10 ml headspace vials were found to be ideal for this study.

22

Table 3.2: Effect of vial size on discrimination indices.

 

 

Incubation Vial Size, ml Discrimination
Trial Temperature, °C Index PCA
1 60 10 93
2 60 20 O

 

On the basis of these preliminary tests, the incubation temperature of 60°C
and 10 ml vial size were found to be perfect for this study. These conditions gave
the highest discrimination index, which was also evidence of higher sensor
responses. After optimization of e-nose parameters, in order to control the
repeatability of the sample preparation and e-nose data acquisition parameters,
the LDPE ﬁlms were spiked with the solvents by the method described in section
3.3.2 and then analyzed by using PCA. We obtained exactly the same
discrimination index value as in previous measurements, which is evidence of the

repeatability of the method (Figure 3.8).

23

 

 

(:2; 0,27% Discrimination Index: 93
0030- , i

 

 

 

 

 

 

 

 

l .
l g
i L l l
0025-. . _.
I ’ g i I
l l l l

 

0010- l
0305-....-
-0.010-~- ' a
41015-... s . a 1
{1025- I

0030-. s ; i
{150 ~0.45 {MB {1.35 -D.

 

 

 

 

p..- ____._._._-_.._A..__. ..- V_ <..___,___--_ ‘_-z _- _A

     
   

 

Toluene

 

 

 

 

 

 

 

C2 0. 27"2

 

 

 

 

 

l

' I
-W\»_-">*-A —-‘ “w-gn'A-

4‘ ‘

. .

I .

) \

_.‘i,
l
i
i

A V .74 - ~ - - > \ 1‘ lax ¢*. A _z»
......' .. ...--. .-. ."' -........-z-.w.. ...
l i
; , 1
l I
l I .
a l l
. l I
l

l

 

 

 

 

 

 

 

 

30 0'25 0'20 0'15 010 0.05 0'00 0'05 010 0i5 0.20 0'25 0'30
C12 99.66%

 

 

 

 

Figure 3.8: LDPE ﬁlm samples with three solvents at 60°C incubation
temperature in 10ml vials (Repeated procedure)

24

Table 3.3 : Optimized data acquisition parameters.

 

 

PARAMETERS SETTINGS
Incubation Time, sec 300
Incubation Temperature, °C 60
Syringe Temperature, °C 65
Syringe Type, ml 5
Filling Speed, ul/sec 500
Flushing Time, sec 120
Injection Volume, ul 2000
Injection Speed, ul 2000
Vial Size, ml 10
Acquisition Time, sec 600
Acquisition Period, sec 0.5
Agitation Speed, rpm 500
Delay, sec 900

 

3.3.4. Building a Model and Validation
(Preparing and Analyzing the Training and Validation Sets)

Once the optimum parameters were set for the samples, measurements with
the e-nose using 12 metal oxide sensors were made for single (individual)
solvents, and binary mixtures of the three solvents. The single solvent groups
were composed of three solvents; ethyl acetate, ethyl alcohol, and toluene. The
data sets for the single solvents were prepared by using different injection
amounts (1.0, 2.5, 5.0, 7.5, and 10 ul). Five replicates were used for each single
solvent and their varying injection amounts. The data sets for the binary mixtures

were prepared for the three different solvent combinations using varying injection

25

amounts (1.0, 2.5, and 7.5 ul). Three replicates were used for the binary
mixtures. Table 3.4 shows the injection amounts of the single and binary solvents
at different concentrations.

The data matrix for one injection amount was constructed with 180 data
points, which were the responses from 12 sensors for three single solvent
samples with the ﬁve replicates (12 X 3 X 5 = 180). As can be seen, the data sets
for the e-nose are multivariate and difﬁcult to interpret. Therefore, it is necessary
to employ multivariate statistics in order to assess the classiﬁcation and
quantiﬁcation of the solvents in the LDPE ﬁlms. The most common multivariate
statistical techniques, principle component analysis (PCA), discriminate factorial
analysis (DFA), and partial least squares analysis (PLS), were used for
qualiﬁcation and quantiﬁcation of the samples.

PCA is an unsupervised learning technique which allows reduction of multi-
dimensional data to two dimensions while simplifying the interpretation of the
data (Delpha et al., 2001). For instance, the responses from 12 sensors (12
dimensions) can be processed and displayed in two dimensions. This allows us
to interpret the data easily. In addition, the samples can be classiﬁed without
prior information on the nature of the samples. Conversely, DFA and PLS require
prior knowledge about the samples. DFA is a supervised learning technique
which classiﬁes the samples by developing a model and then identiﬁes the
unknown samples in qualitative analysis. PLS is an algorithm based on linear
regression techniques. It is also used to build a model that is able to predict the

quantitative information for the samples (Alpha MOS Fox 3000 Manual, 2001).

26

Table 3.4: The combination of ethyl acetate, ethyl alcohol, and toluene at varying
injection amounts.

 

 

 

 

 

 

 

 

 

Percent Concentration Amounts of Injection
(leI (In)
ETAC ETOH TOL 1.0 2.5 5.0 7.5 1 0
(A) (B) (C)
3 100 0 0 J J J J J
:g 0 100 0 V V V V J
6
°‘ 0 o 100 J J J J J
In
30 70 0 V J J
50 50 0 V J J
70 30 0 V V V
g 30 0 70 V V V
35 50 o 50 J J J
D.
0
0‘ 70 o 30 J J J
0
0 30 70 V V V
0 50 50 V V V
0 70 30 V J J

 

 

 

 

 

 

In general, these algorithms can only be considered as valid when the data
(sensor response pattern) is well distributed in their domain. However, in many
cases the patterns from different classes can overlap. The overlapping groups
could lead us to believe that there is no difference between the samples
(Goodner, 2001). Thus, it is necessary to detect and remove outliers from the
data sets. The data sets were studied using PCA ,which is the best way to

detect rapidly the outliers (Alpha MOS Fox 3000 Manual, 2001). Among all the

27

measurements, only one sample was found to be an outlier and discarded from
the data set. A new sample was added to the data set. The outlier detection
method based on comparing the cluster shapes in PCA and analyzing the sensor
proﬁles is described in Appendix C.

After eliminating the outliers, the next step was a training phase. During this
phase, the algorithm is trained by the samples whose classiﬁcations are known,
and then this algorithm can be used to classify the unknown samples. It is clear
that the algorithm can only classify the unknown samples if they exhibit the same
behavior as the training samples (Carmel et al., 2003). As in all quantitative
methods, it is important to use training samples containing all expected ranges of
the future unknown samples (Vlasov and Legin, 1998). Therefore, in this study,
varying injection amounts for the single and binary solvents were used. The data
(the sensory response patterns) were divided into a training set and a validation
set. For the samples with ﬁve replicates, three replicates were randomly selected
for training the algorithm, and the remaining two replicates were used for the
validation. For the samples with three replicates, two replicates were selected for
the training set, and one replicate for the validation set. The validation data sets
were used to validate the algorithm.

When the identiﬁcation and validation were achieved on the DFA model,
estimation of the sample quantity was done by PLS analysis. The PLS analysis
was applied to the samples in order to assess the relationship between predicted
and experimental values for the concentration of the single and binary solvents.

For building the calibration curves for the single and binary solvent mixtures, pre-

28

determined GC area responses were used as a concentration value. Validation
of the PLS model was done by evaluating the correlation coefﬁcient and using a
modiﬁed leave-one out method where one randomly chosen sample was
removed from the dataset and considered as an “unknown“ sample.
In the traditional, leave-one out method (cross validation method), each data
point in turn is removed from the data set and tested as unknown using the
remaining data points. In this research, in order to minimize the unnecessary
complexity, a modiﬁed leave-one out method which omitting only one data point
was used.

Finally, in order to test the performance of the model, unknown test samples

were projected on the PLS model.

29

4. RESULTS AND DISCUSSION
4.1 Identiﬁcation of the Single Solvents

Figure 4.1, 4.2, and 4.3 illustrate the DFA plots of ethyl acetate, ethyl alcohol,
and toluene with different injection amounts, respectively. In these ﬁgures, the
data clusters which belong to different injection amounts were separated from
each other. It is interesting to note that the separation for all model solvents
followed similar patterns.

Figure 4.4 shows the DFA plots combining ethyl acetate, ethyl alcohol, and
toluene training sets at different injection amounts. The model developed by the
training sets was applied to the validation data set (“unknown samples”). Figure
4.5 shows the identiﬁcation of single solvent validation sets. As can be seen, the
validation data sets (so called “unknown” samples) were mostly placed into the
correct solvent category. The validation data set scores are listed in Table 4.1.
The ﬁrst column gives the codes of the validation data sets. The second column
indicates whether or not the validation data sets were recognized. The third
column also indicates the solvent categories (groups) and the last column shows
the percentage of recognition of the samples. Using this percentage, we can
understand how far the validation samples are from their recognized groups. If
the validation samples are in the group, the percentage is 100%. The overall
percentage correct identiﬁcation rates for the validation data sets are also
presented in Table 4.2. All ethyl alcohol and toluene samples were identiﬁed
correctly (100% correct recognition). Out of 10 ethyl acetate samples, only one

sample - ethyl acetate (1pl) - was misidentiﬁed as ethyl alcohol. However, the

30

overall percentage of correct identiﬁcation for ethyl acetate was found to be 90%
(Table 4.2). Hence, we can conclude that the e-nose is capable of clear

recognition of “unknown” single solvent samples.

 

CZ: 22.29%

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000— 1 . I 75 I
. . a g ! . u
l [“75 '3 f i
4000- 2 .051” ; ' ; g
i 5'“ . i
2.000- I ; , , 2 g ,
N t : € : j
R i i * l ,
H. 0.000- I z'l i , 1 . f
-. [95,) § ? 10m
3 -2.000- L.) g ., , w ,
2.5m E I g
0000- . é - ‘. 3 .i i .
-5.000-+ , ll“ 3
l l
0000-, i i ; g ; .
-1 5.00 0000 5.00 0.00 5.00 10.00 15.00
C1:70.79°/o

 

 

 

 

Figure 4.1: DFA plot of ethyl acetate with five different injectiorf amounts

31

 

 

 

 

5.000—
5.000»-
4.000-~-~-—
3.000-w
2.000---
1.000»—

-1.000-» ~

-2.000

-3.000-»~-

5000-, ,
05.00 0.00 -2.00 0.00 2.00
c1 : 68.912

 

 

 

 

 

 

 

 

 

 

C2 : 31 .082

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 4.2: DFA plot of ethyl alcohol with ﬁve different injection amounts

 

 

 

 

12.000—

 

2.5m 5m

 

 

 

10.0th-~-m .._.L--.. ~-——-1

 

 

 

 

a.000-~—--~-~--

 

 

 

 

 

“r

 

 

 

 

2.000“
0.000-
4300------ -~

8.000 ‘I I I I , . I I
24.00 20.00 -15.00 40.00 5500' 0.00 5.00 12.00
81 : 73.152

 

 

CZ : 23.882

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 4.3: DFA plot of toluene with ﬁve different injection amounts

32

 

 

 

 

 

02: 170% Percentage of recognition: 93

3.000— . I Toluene

2.000-

 

 

 

 

 

 

 

 

 

1.000»

0.000- '

 

-‘l .000-

   

CZ: 1.702

2.000-

D

 

 

-3.000-
ETAC

01>

41.000 -

 

W- 'h‘“"_*_ -r__...‘.._. _ - -_
N
UT

 

 

 

 

 

 

 

5000-, ,

. l .
-1 5.00 -1 0.00 5.00 10.00 15.00 2000
C1 : 98.30%

Figure 4.4: DFA obtained from ethyl acetate (A), ethyl alcohol (B), and toluene
(C) training samples.

 

 

 

 

 

 

 

 

 

 

 

 

 

. on Percenta e of reco nition: 93
C7 1 70 / 9 g (Toluene
3.000— ‘

 

 

 

 

 

 

2.000- ~-

1.000-

0.000-

-1 .000-

 

CZ: 1.702

-2. 000 -

 

3.000-

-4.000- - -.

 

 

 

 

5000—, , -, ,
-1 5. 00 -1 0. 00 -5.00 n m = M 10.00 15.00 20.00
C1 : 98.30%

Figure 4.5: The identification of the single solvent validation data set using
DFA.

 

 

 

 

 

 

 

 

33

Table 4.1: Validation data set “unknown” scores for single solvents
(A: Ethyl acetate, B: Ethyl alcohol, C: Toluene)

 

 

   

Sample Code Recognized Group % of Recagnition
A101_3 Yes A 100.0
A101_7 Yes A 100.0
A11__3 Yes B 82.8
A11_7 Yes A 100.0
A2.51_3 Yes A 100.0
A2.51_7 Yes A 100.0
A51_19 Yes A 100.0
A51_7 Yes A 70.8
A7.51_15 Yes A 100.0
A751 19 Yes A 100.0
B101_4 Yes B 100.0
B101_8 Yes B 83.7
B11_4 Yes B 100.0
B11_8 Yes B 100.0
32 51_4 Yes B 100.0
32 51_8 Yes B 100.0
B51_20 Yes B 100.0
B51_8 Yes B 100.0
B7 51_4 Yes B 100.0
37.51 8 Yes B 100.0
C101_5 Yes C 99.5
C101_9 Yes C 98.7
C11_5 Yes C 100.0
C11_9 Yes C 100.0
CZ.51_5 Yes C 97.9
CZ.51_9 Yes C 100.0
051_5 Yes C 100.0
C51_9 Yes C 100.0
C7.51_5 Yes C 99.2
C7 51 9 Yes C 100 0

34

Table 4.2: Identiﬁcation results of the DFA for single solvents validation data set.
(A: Ethyl acetate, B: Ethyl alcohol, C: Toluene)

 

% Correct

 

Code A B
Idenﬁﬁcaﬂon
A 9 1 0 90
B 0 1 0 0 100
C 0 0 1 0 100

 

4.2 Identiﬁcation of the Binary Solvents

Figure 4.6, 4.7, and 4.8 illustrate the DFA plots of toluenezethyl alcohol, ethyl
acetatezethyl alcohol, and ethyl acetateztoluene mixtures with different
combinations (30:70, 50:50, 70:30, 100:0 (vlv)), respectively. In these ﬁgures, the
data sets for different solvent combinations were clearly separated from each
other.

Figure 4.9 shows the DFA plots of binary solvent training sets at d'rfferent
injection amounts. The model developed by the training sets was applied to
validation data sets (“unknown samples”). Figure 4.10 shows the identiﬁcation of
binary solvent validation sets. The validation data set scores are listed in Table
4.3. The ﬁrst column gives the codes of the validation data sets. The second
column indicates whether or not the validation data sets were recognized. The
third column indicates the solvent categories (groups) and the last column shows
the percentage of recognition of the samples. Using this percentage, we can
understand how far the validation samples are from their recognized groups. If
the samples are in the group, the percentage is 100%. The overall percentage

correct identiﬁcation rates for the validation data sets are presented in Table 4.4.

35

ETAC/ETOH with all its combinations was identiﬁed correctly (100% correct
recognition). However, ETAC/TOL, TOUETOH and their combinations were
misidentiﬁed. Out of 9 ETAC/TOL samples, one sample was misidentiﬁed as
TOUETOH. Out of 9 TOUETOH samples, three samples were misidentiﬁed as
ETAC/TOL. The overall percentage of correct identiﬁcation for ETAC/1' OL and
TOUETOH were found to be 88% and 66%, respectively. As can be seen, the e-
nose was able to recognize ETAC/ETOH solvent mixtures whereas it can only
partially recognize ETAC/1' OL and TOUETOH samples. It is important to note
that when we considered the DFA analysis just within the same solventmixture
groups, a perfect separation was observed (Figures 4.6 - 4.8). But, when we
performed the DFA analysis for all binary solvent mixtures, some of the data
slightly overlapped (Figures 4.9 and 4.10). Consequently, this led to the
misclassiﬁcation on the DFA plots. Another reason for the misclassiﬁcation might
be the differences in solvents’ polarity. Ethyl acetate and ethyl alcohol are polar
solvents whereas toluene is a non-polar solvent. Therefore, non-polar LDPE ﬁlm
samples might be attracted to the non-polar toluene and tend to retain toluene
molecules relatively more than ethyl acetate and ethyl alcohol molecules. This

problem may be overcome by changing the e-nose setup parameters.

36

15.000—

 

125m4

10.000- . .. .

7.500-

5mg-.. _,

CZ : 0.272

1500—7

 

10.000—i ~-
450.00

I

. l
. I
I i
‘ I

10L amigo IMICLL 19 g
ImLfeImIi. 010000100 '

' l
1 z I I 1
.2 5110.. ...M._. .... _---. .. .---_._..... _.-...... -.......,.....,H .,. .
. 1 1
. L I

 

  
   
    
   

l
I
l
ToL_ ETOH 70 IMICL,_4

10L_ ETOH 0_ IMICLL 7
.101._510H 0104111710 .

I
I
I
I
I

I
I I
l I

I 1
I

10L_ ETOH 0730 110LL_ 22
1100100010 010.000.1000 5

125.00 000.00 -75'.00 .5000 25'00

TOL: ETOH (.30 .70)j

 

 

 

[:TOL EToH 100: 0)

T

 

 

 

 

 

 

 

 

TOL ETOH (50 50) :
TOL: ETOH (70: 30) _ 1

 

0.00

1 I I 3 I

25700 50.00 75.00 100.00 125.00 150.00 175.00 20000
01:99.72:

 

 

 

 

Figure 4.6: DFA plot of toluene and ethyl alcohol mixtures with different
combinations ( 30:70 , 50:50, 70:30, 100:0 vlv) - injection amount 1pl.

 

 

 

10.1130

anon-aw--. .2.-.“
BHII—
4.000- .
01110-
-1.lIIO--
-20[Il-— . , .
1M]-
4.0mm”-.- (r
-5.l]II~ -.

CZ : 0.282

 

6.1110-

5013010 00052 e w _

510015.10 3070,2510 _ ._

 

WAC.
I

J_,.g_15§ ,

3132;11:51fo .
: l

1 1 ‘.

t ' l

1 I I

. l .
”hwy. 77.-. . ._... .., _ .. 7.-.. _. -... H... . 7..

 
   
 

_ I
ETAC: ETOH (50:

ETAC_E105050_2511

. .
I l

 

ETAC: ETOH (30: 71B

5° ﬂ

I .
I !

 

I

 

 

 

 

1 1

‘ .....L-L1‘---]I- i '1
ETAC:ETOH (70:30) I L

~A25 _11. ' .

 

 

 

~1 00.00

A 5 -
I l l

05.00 50.00 25.00 0.00

150001

f , .
am dm mm IMm 15m

C1 : 99.722

 

Figure 4.7: DFA plot of ethyl acetate and ethyl alcohol mixtures with different

combinations ( 30:70 , 50:50, 70:30, 100:0 v/v) -

injection amount 1ul.

37

10.000-

910m---__---- I 1E:TAC TOL (70: 3mg---

 

I I
.. _ _-__ -._ _..T__....,.-....-....__ .1. _ _.~ _--- .. ...- .—.

I

I_

I . 1

-1. ..-_.... .
1ETAr: mime 75 24

ETHI: T015010 75721 _.
100030. In? ,.

0.000-w 1 1
70001 __ _.
woo-w ~ - ETAC TOL (50 50)

3000-..-.-0- W l 1 . :
0.000-—- ~-~~~ I '
.2000- :
-3.000~ ‘
4.000-~~
.5300- ..

ETAC TOL (100 DWI: C ”L (3° 70?; “000003?

 

 

   

 

 

I E
l l
1 .
i

(:2: 4.902

 

 

   

 

 

 

7.000»

'8-000 I 1 I I 1 1 I I I

40.00 35.0] 30.00 250] 20.00 45.00 -10.00 6.00 0.00 5.00 10.00 15.00 20.00
Ci : 95.082

 

 

 

 

 

 

 

 

 

 

 

 

 

09020.35 DFA ”blot Loft-et'hyl‘i‘acetate 'an'a ‘ toluene mi‘xthrés with” different
combinations ( 30:70 , 50:50, 70:30, 100:0 v/v) - in'Lection amount 1ul.

 

 

 

 

 

 

 

 

Percentage of recognition: 65 ET AC ITOL 3
5.000— 1 1

 

4300"” ETAC/ETOH

3000-....- 1......“ ..... - .

 

2000---..-

CZ: 1.242

0.000- ETAgE.

 

 

 

i

 

 

 

 

 

 

 

 

 

 

 

0000—,
43.00 0.00 .400 .200 0.00
CI : 90.75:

Figure 4.9: DFA obtained from binary solvent training sets; ETAC/ETOH (30:70,
50:50, 70:30), ETAC/1' OL (30:70, 50:50, 70:30), TOUETOH (30:70, 50:50,
70:30).

 

 

 

38

 

 

 

 

 

 

 

 

 

 

 

 

 

 
    

 

 

 

 

 

 

 

 

 

P .. :
ercentage of recognition 65 [ETAC/TOL
5000— l ,
l l i i
4.000' ‘ V Ii , .. --l- _, __
3.DUO-- ~ ~ I
2.000- : l 5103013070 v5_9,~3---_,, a i
a ‘ i l I i
T. "u_ ‘- ! Iml ﬂ ~".‘,'~, - _
E“: 0000- - - = ..-%L—9 EF:€JII«I313"'Sl'i-"wigs
' ”7'35 l ‘5’. ﬁt?“ =- “a 07520
i ' ED” TOL. '11! WWII"! nil-32' " '
-l.000- ~ ~ >5 - - l ______ V
l l g ' T0L_EIClH3 _7:i_7
2000 i . i g i .. l\.'_5’.’y
-. - f l l [TOL/ETOH? T0’L_E Hl5050_75_l9 ’
-3.000-, : ’ ; ' ;

 

 

-1 0.00 0.00 0.00 -4.00 2.00 0.00 2.00 4.00 0.00 0.00
1:1 :9070:

Figure 4.10: The identiﬁcation of the binary solvents - validation data sets
(“ unknown” samples) using DFA

39

Table 4.3: Validation data set “unknown” scores for binary solvents.

 

 

Sample Code Recognized Group % of Recgcmition
ETAC_ETOH3070_1 uL Yes ETAC / ETOH 100.0
ETAC_ETOH3070_2.5 pL Yes ETAC I ETOH 100.0
ETAC_ETOH3070_7.5 uL Yes ETAC / ETOH 100.0
ETAC_ETOH5050_1 pL Yes ETAC / ETOH 100.0
ETAC_ETOH5050_2.5 uL Yes ETAC / ETOH 100.0
ETAC_ETOH5050_7.5 pL Yes ETAC / ETOH 100.0
ETAC_ETOH7030_1 pL Yes ETAC I ETOH 1 00.0
ETAC_ETOH7030_2.5 uL Yes ETAC / ETOH 99.4

ETAC_ETOH7030_7.5 pL Yes ETAC / ETOH 99.3

  

    

ETAC_OTL3070__1 uL if ’ ’ es ETA TLo " 000. ”"— '
ETAC_TOL3070_2.5 pL Yes ETAC ITOL 100.0
ETAC_TOL3070_7.5 pL Yes TOL / ETOH 100.0
ETAC_TOL5050_1 pL Yes ETAC ITOL 100.0
ETAC_TOL5050_2.5 uL Yes ETAC ITOL 100.0
ETAC_TOL5050_7.5 pL Yes ETAC ITOL 100.0
ETAC_TOL7030_1 uL Yes ETAC ITOL 100.0
ETAC_TOL7030_2.5 pL Yes ETAC ITOL 100.0
ETAC_TOL7030_7.5 uL Yes ETAC ITOL 100.0
TOL_ETOH3070_1 uL Yes ETAC ITOL 100.0
TOL_ETOH3070_2.5 pL Yes ETAC ITOL 100.0
TOL_ETOH3070_7.5 uL Yes TOL / ETOH 100.0
TOL_ETOH5050_1 pL Yes ETAC ITOL 100.0
TOL_ETOH5050_2.5 pL Yes TOL / ETOH 100.0
TOL_ETOH5050_7.5 pL Yes TOL I ETOH 100.0
TOL__ETOH7030_1uL Yes TOL / ETOH 81.1
TOL_ETOH7030_2.5 uL Yes TOL / ETOH 100.0
TOL_ETOH7030_7.5 pL Yes TOL l ETOH 100.0

40

Table 4.4: Identiﬁcation results of the DFA for binary solvents validation data set.
( ETAC: Ethyl acetate, ETOH: Ethyl alcohol, TOL: Toluene)

 

 

Code ETAC:ETOH ETAC:TOL TOL:ETOH %Correct
ldenﬁﬁcaﬁon
ETAC:ETOH 9 0 0 100
ETAC:TOL 0 8 1 88
TOL : ETOH 0 3 6 66

 

4.3 Quantiﬁcation of the Single Solvents

Figure 4.11 shows a typical PLS calibration curve which was used to predict
the toluene concentrations. When constructing the PLS calibration curves, pre-
determined GC area response values were used as a concentration value. In this
typical calibration curve, the correlation coefﬁcient was found to be 0.9787. The
correlation coefﬁcient between experimental and expected values is the evidence
of how well the new model performed. In general, if the correlation coefﬁcient is
close to one, which indicates a perfect ﬁt, the new model created can correctly
predict the concentration. First, the PLS model was constructed by using training
data sets, then it was validated by using validation data dataset (Figure 4.12),
using the partially leave-one-out method (only one dataset, 5ul) (Figure 4.13),
and running blind samples (unknown samples, 2.5pl) (Figure 4.14). The detailed
experimental and predicted area response values for single solvents are
tabulated in Appendix D.

Table 4.8 summarizes the correlation coefﬁcient values between experimental
and predicted area responses obtained from validation methods I for single

solvents. As can be seen from the results, we obtained correlation coefﬁcient

41

values from 0.90 to 0.98 which shows there is a good ﬁt between the
concentration predicted by e-nose and the concentrations determined by GC.
Therefore, the e-nose can be considered to provide accurate information on the

concentration of residual single solvents in the LDPE ﬁlms.

 

 

 

Correlation: 0.9787

 

 

 

32000000000 , ; , _
31000000009 .j . 1 f . : . 01013
animun- ' z t 01000
2300000000- - V
20000000030- 2 ' - . . . . . .
27000000000- ~   . , . . .. 155;}? UOL.17,__

Toluene ; C791 5
260000000111- Data set ‘ * ' 0701.3
25000000000-

, 24000000000- ,

3'; 23000000000

3 22000000000-

?5 21000000000-

20000000000-
19000000000-
10000000000-
17000000000-
10000000.000-- .,
15000000000-
14000000010-
13000000000-

120000001110-I i I ~—w I ~ I I .I I I:
1000000000 1250000000 1500000000 1750000000 2001000000 22511100000 250000001!) 2750011111111 30000000..
Expected

ﬂ.”- A.-~..——-— #.

l
l
l
__..._ -_.."_-_-._ ~‘

 

 

 

 

 

 

 

 

 

Figure 4.11: A typical PLS plot of the toluene samples (5 concentration - 5
replicates)

42

 

 

Correlation: 0.9799
321111100000 j 1 1
MILE} _. ,__,._..._-.0I.-s.-,,-.-_ 2.-- .-_..__s--.s-..-_s-__.-- _...--“ _--- - .-_._..---.-..-.__.-_
mowm-sm~ --- --__- .-..-...--._- - .1. ”2.2.--- .-..--3_.---,.--. _.-- ........-. -._-.--_.- _-..-._.-.. _..“-..
moan- ._-_.--.-._...-_ .-.. -_ _ _.__:_.- .-..._... _ _ .. .. . . . _-.- .. ___.___.__-_..-_..
25000000000»
24mm00000-~ - .

Emmoomo-

 

 

 

 

 

 

 

 

 

    
 

 

 

 

10000000.000-- ~ « Validation data set
1 monogram .. ,I ,, _

. - - 1
12000000001. 1 . . . . l
. 1000000000 1250000000 1500000000 1750000000 2000000000 2250000000 2500000000 2750mm 300000001
1 Expected

i

 

 

1
. ... . .V...‘
1
1
1

 

 

 

 

 

 

 

Figure 4.12:3The PLS pot of toluene training data set validated by validation data
set ( . the ﬁlled circles represent training data set, 0 open circles represent,
validation data set)

Table 4.5: Predicted values for validation data set (for toluene samples)
Experimental Area Predicted Area

Sample Response (Average) Response

Code uV*sec, x10E+06 pV*sec, x10E+06

 

C1.0 13.0 13.4
01.0 13.0 13.9
C2.5 18.0 17.4
02.5 18.0 18.2
05.0 23 20.7
C50 23 20.3
C7.5 25 27.2
C7.5 25 24.0
C10 29 29.9
C10 29 27.7

43

 

 

 

 

Correlation: 0.9872

 

 

32000000000—
31000000000—
30000000000-
29000000000-
200000.000-
27000000000
20000000003-
25000000000-
24000000000—
”323000000000-
€22000000000-
0.221000000000-
2000000000- ,
13000000010- . . . 1 ,
10111001110111- 2
17000000000- 9.

16000000000- Unknown
15000000000-
14000000000- 7 -
130000000m- . 1' I I
12000000000- _----- 1 2”

1000000000 125000m00 1500000000 1750000000 2000001000 2250000000 2500000000 2750000000 30000000.
Expected

 

1'

  

  

Toluene
Data set

    

 

 

 

 

 

Figure 4.13: PLS plot of toluene samples constructed by leave-one-out method
(samples without 5.0 pl), then unknown samples (5.0 pl) were projected onto the
model.

Table 4.6: Predicted values for the unknown toluene samples (5.0 pl)

 

 

 

 

Experimental Area Predicted Area
Sample Response (Average) Response
Code pV*sec, pV*sec, x10E+06
x10E+06i1.1*
C5.0 23.0 19.0
05.0 23.0 19.0
C5.0 23.0 21.1
05.0 23.0 20.6
C50 23.0 19.7
AVERAGE 19.9
STDEV. : 0.9

 

 

 

*Mean value x 10E+06 i Standard deviation (three replicates)

44

 

 

 

Correlation: 0.9787

 

 

 

 

 

 

 

 

 

 

 

 

 

32000000000
3100000000
300000000001 ‘
23000000000- Toluene ;
29000000300, Data set
2700000000044 1» ..
20000000.000-- ~ '
25000000000- ~ .
24000000.000-~-----~--- ~ I
Emmwmm-«wmI j 1 ‘
19000000000W
'1 10000000.000-~--—-~-~ . I
15000000000— ‘
14000000000-
130000000001» 1 I I
’ 12000000000 , 1 1 1 1 1 I I I- I
1200000000 1400mm 1500000000 1000000000 2000000000 2200000000 2400mm 2500000000 2000000000 30000000..

Expected

Figure 4.14: PLS model of toluene samples with blind ”unknown” samples

 

O the ﬁlled circles represent toluene data set, Othe open circles represent
blind “unknown“ samples

Table 4.7: Predicted values for the blind “unknown “toluene samples (2.5 pl)

 

 

Experimental Area Predicted Area
Sample Response (Average) Response
Code pV*sec, pV*sec, x10E+06
x10E+06i1.3*
02.5 18.0 17.5
02.5 18.0 18.5
02.5 18.0 19.4
02.5 18.0 18.1
_QZé 1&9 12.2

 

 

AVERAGE 18.5

 

 

STDEV. : 0.7

 

 

*Mean value x 10E+06 i Standard deviation (three replicates)

45

Table 4.8: Correlation coefﬁcient values obtained from all validation methods for
single solvents.

 

Correlation Coefﬁcient
Using all data sets Using training dataset Using leave one data
set out (5pl)

 

ETAC 0.90 0.90 0.94
ETOH 0.94 0.95 0.95
TOL 0.97 0.97 0.98

 

4.4 Quantiﬁcation of the Binary Solvents

Figure 4.15 illustrates an example of a typical PLS plot for TOUETOH training
sets at 1 pL injection amount. The correlation coefﬁcients between experimental
and predicted values at varying injection amounts for binary solvent mixtures
were found to be between 0.84 and 0.99 (Table 4.9 and 4.10). The PLS analysis
results showed that the electronic nose can be used satisfactorily in quantitative
analysis of binary solvents.

These almost perfect correlation coefﬁcient values obtained from the model
for single and binary solvents can be used satisfactorily in quantitative analysis.
Moreover, if this new method is capable of predicting the concentrations, even
though there is a poor identiﬁcation/classiﬁcation between mixtures, it would be
possible to quantify the total residual solvents in the ﬁlms without
identifying/classifying them. This would bring a great advantage in quality control

of packaging materials.

46

 

 

14000000000
135000000004 ~ -

125000111000- ‘ --
12000001010“ - '-

11500010000+~~~~~~ -j-- ----— ---

105000m000- -- -- 4

1mm ., _

9500000000 -- -
3000000000-

Predicted

1

0500000000-
5000000000- , , .

SUJOOCIMmW . . I

 

Correlation: 0.9859

 

 

 

 

 

4000001010

400000000 000000000

0500000000- »
7000000000» ~

‘f“‘”"""“‘"” TOL:ETOH(30:70)

       
   
  

1

 

 

 

 

. ._.-.__.___._. .1
I .

l
1 .’ l

i 1

 

   

 

TOL:ETOH(70:30)

 

 

i
j 1
«a ,_---‘. _--

 

 

_.—.... y .__.._..-.. _.-- _.., _..... .__._ - ..

J TOL:ETOH(50:50)

 

 

 

1

e
1
_-... _--.. _-.. ...-..i.._-_-._-__- _-.- - .__...... ..'. ..._._-.._-... .. .

1 1
1
1

- _-.,._ ._.....s.. .1...,,.,-.. _... ,,..,.._.._.

.

.7
7

 

 

 

 

1000000000 1200000000 14000000,

Expected

800000000

 

 

Figure 4.15: A typical PLS plot of TOLzETOH data sets at 1 pL injection amount.

Table 4.9 : Summary of the correlation coefﬁcient values obtained from the PLS
calibration curves for binary solvents.

ETAC : ETOH
30:70, 50:50, 70:30, 100:0

ETAC : TOL
30:70, 50:50, 70:30, 100:0

TOL : ETOH
30:70, 50:50, 70:30, 100:0

 

Correlation Coefﬁcient

1.0 pL 2.5 pL 7.5 pL

0.91 0.92 0.95

0.89 0.96 0.84

0.98 0.96 0.96

 

47

Table 4.10 : Summary of the correlation coefﬁcient values obtained from the
PLS calibration curves for binary solvents.

 

Correlation Coefﬁcient

 

ETAC:ETOH ETAC:TOL TOL:ETOH

 

30:70

1.0, 2.5, 7.5 pl 0.99 0.98 0.99
50:50

1.0, 2.5, 7.5 pl 0.92 0.98 0.99
70:30

1.0, 2.5, 7.5 pl 0.95 0.98 0.98

 

48

5. CONCLUSIONS and RECOMMENDATIONS

A novel electronic nose method for assessing the residual solvents in low

density polyethylene has been established. The proposed method is very useful
for identifying and quantifying the residual solvents. The method presented is
also simple, quick and reliable in predicting the total amount of residual VOCs
present in low density polyethylene and thus useful for controlling the quality of
the process.
It has been found that the electronic nose with the proposed method is capable
of providing, for single and binary mixtures, quantitative analyses of residual
solvents and can also distinguish between similar types of solvents. This is
supported by high correlation coefﬁcient values obtained in PLS analysis.

Further investigation is required to assess the effectiveness of the proposed
method for identiﬁcation and quantiﬁcation of different binary and ternary
mixtures of other VOCs. Further research could also involve the adaptation of
this method into a quality control laboratory.

In recent years, with the introduction of new technologies such as
biodegradable polymers, water-based, UV, and EB curable printing inks, the
sources of off-odor started diversifying in the polymer and printing industry. The
methods and residual solvent levels based on solvent-based inks, coatings and
adhesives may not be applicable to these new technologies. Therefore, the
electronic nose could also be used to establish standards for low odor with these

new technologies.

49

APPENDICES

50

APPENDIX A

Table A1 : Properties of the representative solvents (Wypych, 2001)

 

 

 

 

 

 

 

 

 

ALCOHOLS ESTERS AROMATIC
HYDROCARBONS
Ethyl Alcohol Ethyl Acetate Toluene
Formula C2H5OH CH3COOC2H5 CaHsCH3
Density ( glcm 3) 0.789 0.901 0.867
Molecular Weight 46.1 88.1 92.1
_Lglmol )
Evaporation Rate 3.2 6.2 2.1
(BuAc=1 1
Boiling Point (° C) 78.3 77.2 110.8
Vapor Pressure 44 73 21
(mm Hg, 20 °C )
Polarity Index 5.2 4.3 2.3
Miscibility 14 19 23

Number

 

51

APPENDIX B

GC Analysis Conditions

Gas chromatographic analysis was performed using an HP 6890 (Hewlett-
Packard, Avondale, PA) gas chromatograph, equipped with ﬂame ionization
detector (FID) and interfaced with Empower-Waters software. Gas
chromatographic conditions used for single and binary solvent mixtures were as
follows:
Column : Supelcowax 10 ( Supelco lnc., Bellefonte, PA)

(60 m, 0.25 mm |.D., 0.25pm ﬁlm thickness)
Carrier gas : Helium at 30ml/min
Temperature : Injector temperature - 250 °C, Detector temperature - 250 °C

lnitial oven temperature- 40 °C isothermal

Initial time —10 min
First, 10 ml headspace vials were prepared using the method described in
section 3.3.2. Second, the vials were heated in the oven at 60 °C for 20 min in
order to equilibrate the solvent in the headspace of the vial with the sample.
Then, a 100 pl portion of headspace created in the vial was injected manually
into the gas chromatograph using a gas tight syringe. Since the solvent
concentration in the headspace is a function of the concentration in the ﬁlm, the
recorded area response values are also proportional to the solvent concentration
in the ﬁlm samples. Therefore, all recorded area response values were used to

build calibration curves for the e-nose analysis.

52

APPENDIX C

Outlier Detection

The outlier detection and rejection from the data set was one of the most vexing
issues in this study. But the time spent for this part lead us to build a concrete
model. First, it was necessary to have a valid reason for rejecting the point from
the data set. For this purpose, two methods were used to ﬁnd outliers.
1) Using PCA (Alpha MOS Fox 3000, Manual, 2001):

-Perform the group labeling in a way that repeats of each individual

sample are clustered together.

Build the PCA.

-Compare the cluster shape of all the groups.
The cluster shape must be quite similar, otherwise there is a good chance of
having one or several outliers in the dataset.
2) Using sensor responses
By analyzing the sensor proﬁles, it can be visually seen if there is a signiﬁcant

difference between the replicates of each sample for each sensor.

53

APPENDIX D

Experimental and Predicted Area Response Values for Solvents
The following tables show the experimental and predicted area response values
obtained from using three validation methods for ethyl acetate and ethyl alcohol.

Table D1: Experimental and predicted values for ethyl acetate samples.

 

Using Training Data Set

 

 

 

Experimental Area Predicted Area
Sample Response (Average) Response
Code pV*sec, x10E+06 pV*sec, x10E+06
A1.0 21.0 19.5
A1.0 21.0 21.6
A2.5 24.0 27.3
A2.5 24.0 27.4
A5.0 27.0 24.1
A5.0 27.0 26.0
A7.5 31.0 32.7
A7.5 31.0 33.0
A10.0 34.0 32.1
A10.0 34.0 33.0

 

Table 0.2: Experimental and predicted values for ethyl alcohol samples.

 

 

 

Using TraininLData Set
Experimental Area Predicted Area

Sample Response (Average) Response
Code pV*sec, x10E+06 pV*sec, x10E+06

B1.0 9.0 8.9

B1.0 9.0 9.8

B2.5 10.0 9.5

B2.5 10.0 10.3

B5.0 13.0 11.3

35.0 13.0 16.6

87.5 18.0 21.0

B7.5 18.0 16.9

B10.0 21.0 15.0

810.0 21.0 16.8

 

54

Table D3: Experimental and predicted values for “unknown” ethyl acetate
samples (5ul)

 

Using Leave One out Method

 

 

 

 

Experimental Area Predicted Area
Sample Response (Average) Response
Code pV*sec, pV*sec, x10E+06
x10E+06:t0.7*
A5.0 27.0 26.3
A5.0 27.0 26.2
A5.0 27.0 26.4
A5.0 27.0 25.3
A5.0 27.0 25.7
AVERAGE: 25.9
STDEV. : 0.5

 

 

 

*Mean value x 10E+06 1: Standard deviation (three replicates)

Table 0.4: Experimental and predicted values for “unknown” ethyl alcohol
samples (5pl)

 

Using Leave One out Method

 

 

 

 

Experimental Area Predicted Area
Sample Response (Average) Response
Code pV*sec, pV*sec, x10E+06
x10E+0610.9*
B5.0 13.0 12.0
B5.0 13.0 12.5
B5.0 13.0 13.2
B5.0 13.0 13.2
B5.0 13.0 10.3
AVERAGE: 12.2
STDEV. : 1 .1

 

 

 

 

*Mean value x 10E+06 i Standard deviation (three replicates)

55

Table 0.5: Experimental and predicted values for blind “unknown” ethyl acetate
samples (2.5pl)

 

RunnirLg Blind Samples

 

 

 

 

 

 

Experimental Area Predicted Area
Sample Response (Average) Response
Code pV*sec, pV*sec, x1 0E+06
x10E+06i0.9*

A2.5 24.0 24.1
A2.5 24.0 25.9
A2.5 24.0 24.8
A2.5 24.0 24.0
_AZé 2.4.9 J93
AVERAGE: 25.1
STDEV. : 1.2

 

 

 

*Mean value x 10E+06 : Standard deviation (three replicates)

Table 0.6: Experimental and predicted values for blind “unknown” ethyl alcohol
samples (2.5pl)

 

Runnigg Blind Samples

 

 

 

 

Experimental Area Predicted Area
Sample Response (Average) Response
Code pV*sec, pV*sec, x10E+06
x10E+06i0.5*
82.5 10.0 9.9
82.5 10.0 10.5
82.5 10.0 9.6
B2.5 10.0 10.9
32,5 10,9 1 1 .6
AVERAGE: 10.5
STDEV. : 0.8

 

 

 

*Mean value x 10E+06 i Standard deviation (three replicates)

56

BIBLIOGRAPHY

Alpha MOS Fox 3000 operating manual, Release Jan. 2001.
Apps, A.E. 1958. Printink ink technology. Leonard Hill Ltd. London, UK.

ASTM F151-86 Standard test method for residual solvents in ﬂexible barrier
ﬁlms. Annual book of ASTM Standards. Philadelpia, PA: 15.9, pp. 1070-1074.

ASTM F1884-98 Standard test method for determining residual solvents in
packaging ﬁlms. Annual book of ASTM Standards. Philadelpia, PA: 15.9, pp.
1465-1468.

Begley, TH. 1997. Methods and approaches used by FDA to evaluate the safety
of food packaging materials. Food Additive and Contaminants. 14:6—7.

Benali, F.: Henry, J.: and Delaunay, Y. 1995. Electronic nose potentially
utilization on packaged products. Olfaction and electronic nose, 2"d lntemational
Symposium, Toulouse.

Bohartier, C.; Vernat-Rossi, V.: Breysee, C.: and Theboult, J. 1995.
Caracterisation d’un emballage defectueux par analyse de sa fraction volatile.
Industries Alimenttaires at Agricoles. 122: 849-852.

Carmel, L.; Sever, N.; Lancet, D. and Harel, D. 2003. An enose algorithm for
identifying chemicals and determining their concentration. Sensor and Actuators
B 93:77-83.

Das, R. 2003. Development of electronic nose method for evaluation of HDPE
data, correlated with organoleptic testing. Master of Science Thesis. Michigan
State University, East Lansing, MI.

Delpha, C.; Siadat, M. and Lumbreras, M.2001. An electronic nose for
discrimination of forane 134a and carbon dioxide in humidity control atmosphere.
Sensor and Actuators B 78:49-56.

Ezquerro, O.; Pons, B. and Tena, MT. 2003. Headspace solid-phase
microextraction-gas chromatography-mas spectrometry applied to quality control
in multilayer-packaging manufacture. Journal of Chromatography A, 1008:123-
128.

Goodner, K.L.; Dreher, G.J. and Rouseff, R.L. 2001. The danger of creating false
classiﬁcations due to noise in electronic nose and similar multivariate analyses.
Sensor and Actuators B. 80:261-266.

57

Huber, M.; Ruiz, J. and Chastellain. 2002. Off-ﬂavour release from packaging
materials and its prevention: 3 food company’s approach. Food Additives and
Contaminants. 19:221-228.

Kolb, B. and Ettre, L. S. 1997. Static headspace-gas chromatography: theory and
practice. Wiley-VCH, USA.

Lau, O.W. and Wong, SK. 1995. Contamination in food from packaging material.
Packaging Technology and Science. 8:271.

Leland, J.; Schieberle, P.; Buettner, A. and Acree, T. 2001. Gas chromatography-
olfactometry. American Chemical Society, Washington, DC.

Lin, C. 1995, Permeability of organic vapors through a packaged confectionery
product with a cold seal closure: theoretical and practical consideration. Master
of Science Thesis. Michigan State University, East Lansing, MI.

Lopez-Rubio, A.; Cava, D.; Lagaron, J.M.; Catala, R. and Gavara R. 2003. Mass
transfer properties of organic compounds through EVOH copolymers as a
function of ethylene content and environmental humidity. 21st IAPRI Symposium,
Valencia proceedings. 92-95.

Maneesin, P. 2001. GC-MS and electronic nose analysis on off-ﬂavor
components in HDPE containers and correlation with sensory evaluation. Doctor
of Philosophy dissertation. Michigan State University, East Lansing, MI.

Pearce, T.C.; Schiffman,S.S.; Nagle, HT. and Gardner, J.W. 2003. Handbook of
Machine Olfaction. Wiley-VCH.

Robertson, G. 1993, Food packaging: Principles and practice. New York: Marcel
Dekker, Inc. 623-600.

Savastano, 0.2003. The packaging ink market. lnkworld Magazine.11(31): 45-51.

Schaller, E; Bosset, J.O., and Escher, F. 1998. ‘Electronic Noses” and their
application to food. Lebensmittel-Wissenschaft und technologie. 31:305-316.

Soroka, W. 1995. Fundamentals of packaging technology. Institute of Packaging
Professionals, Heindon, Virginia.

Suman, M.; Ricci, 0.; Dalcanale, E.; Bersellini, U. and Sensi, L. 2003. A08: from
study to quality control. ltalia Imballagio. 4: 195-206.

Thompson, B. 1998. Printing materials: science and technology. PIRA
International. Surrey, UK. 123-286.

58

Van Deventer, D. and Mallikarjunan, P. 2002. Comparative performance analysis
of three electronic nose systems using different sensor technologies in odor

analysis of retained solvents in printed packaging. Journal of Food Science.
67(8):3170-3183.

Vlasov, J. and Legnin, A. 1998. Non-selective chemical sensors in analytical
chemistry: from ‘electronic nose’ to electronic tonque. Journal of Analytical
Chemistry. 361:255-260.

Williams, C. 2001. Printing ink technology. Pira international Ltd. Surrey, UK.
120-230.

Wypych, G. 2001. Handbook of solvents. John Wiley and Sons Inc. New York.
1105-1450

59

 

31293 027324