Odour & Flavour tests: Human panel and electronic … report compares results obtained with an electronic tongue, electronic nose and an odour and flavour panel for non-concentrated

Odour & Flavour tests: Human panel and electronic testing compared Deliverable 3.4.6.

Techneau NOVEMBER 2007

© 2006 TECHNEAU TECHNEAU is an Integrated Project Funded by the European Commission under the Sixth Framework Programme, Sustainable Development, Global Change and Ecosystems Thematic Priority Area (contractnumber 018320). All rights reserved. No part of this book may be reproduced, stored in a database or retrieval system, or published, in any form or in any way, electronically, mechanically, by print, photoprint, microfilm or any other means without prior written permission from the publisher

Odour & Flavour tests: Human panel and electronic testing compared Deliverable 3.4.6.

Techneau, 07.346 NOVEMBER 2007

This report is: PU = Public Co-financed by the American Water Works Association Research Foundation

Colophon

Title

Odour & Flavour tests: Human panel and electronic testing compared Author(s)

Bram van der Gaag, Kiwa Water Research Sandrine Isz, Alpha-MOS René Braakman, Atas Benelux Hans van der Jagt, Kiwa Water Research Quality Assurance By Dr. A. P. van Wezel, Kiwa Water Research Deliverable number D 3.4.6

Odour & Flavour tests: Human panel and electronic testing compared

© TECHNEAU - 1 - November 30, 2007

Summary

Materials used for water distribution have to be tested on odour and flavour. The NEN-EN 1622:2006 and NEN-EN 1420-1 methods are used to determine if polymers or rubbers release chemicals that influence the odour and/or flavour of drinking water. In these methods an odour and flavour panel tests drinking water after long-term contact with distribution material. This report compares the results of an electronic nose and/or electronic tongue with the obtained results by an odour & flavour panel. The aim is to investigate the ability of the electronic systems to reduce the number of panel tests. The results support the decision of use of online electronic noses and –tongues for water quality control at different places in the chain from source to tap.

Electronic noses and tongues contain an array of sensors with different chemical interfaces. Each sensor respond in a different way to the presence of chemicals. Based on the signals produced the electronic instruments are able to distinguish between one or the other sample tested. An odour and flavour panel lacks these discriminating abilities. The results obtained show a moderate to good correlation between the panel results and electronic methods. It is foreseen that in the future electronic nose and tongue systems can reduce the number of panel tests, and even be implemented at vendors factories to perform quality control during production. The authors do not foresee that the instruments will fully substitute the odour and flavour method.



Contents

1 Introduction 5

2 Materials and Methods 6

2.1 Sample preparation 6

2.2 Protocol for odour - and flavour threshold assessments 6 2.2.1 Model validation for odour and flavour assessments by a panel 7

2.3 Electronic Nose 7 2.3.1 Electronic nose measurement protocol 9

2.4 Electronic Tongue 10 2.4.1 Electronic tongue measurement protocol 11

2.5 Data treatment methods 11

2.6 General remarks 13

3 Results 14

3.1 Odour & flavour panel 14

3.2 Electronic Nose 14 3.2.1 Discrimination capability of the electronic nose system 14 3.2.2 Quality control tool capability of the electronic nose system 16 3.2.3 Panel correlation capability of the electronic nose system 17

3.3 Electronic Tongue 20 3.3.1 Discrimination capability of the electronic tongue system 20 3.3.2 Quality control tool capability of the electronic tongue system 22 3.3.3 Panel correlation capability of the electronic tongue system 22

3.4 Combination of Electronic Nose & Electronic Tongue 25 3.4.1 Discrimination capability of the combined system 26 3.4.2 Quality control tool capability of the combined system 27 3.4.3 Panel correlation capability of the combined system 28

4 Discussion 32


4.2 Statistics 32

4.3 Electronic nose system 34

4.4 Electronic tongue system 34

4.5 Combination of Electronic Nose & Electronic Tongue 35

5 Conclusions 36

6 References 38


6.2 Electronic nose & Electronic tongue 38



6.3 Statistics 39

6.4 Technical notes of Alpha MOS products 39

7 Abbreviations 40





1 Introduction

Today, rubber materials (ancilleries) are widely used in water distribution systems. A potential problem with rubber materials is that the possibility exists that odour and taste-causing organic compounds from the rubber material may leach into drinking water and may result in odour and/or taste problems (4). To avoid odour and flavour problems all rubbers used in distribution systems have to be certified before it may sold. The certification is based on the NEN-norms NEN-EN 1420-1 and NEN-EN 1622:2006 (1, 2). An odour and flavour panel tests a defined dilution of drinking water that has been in contact with rubber. Based on the panel results a rubber is accepted or rejected. In practice the end product is tested and certified before the first market introduction. During production no batch to batch quality control on odour and flavour is performed due to high costs. In place the products are controlled annually. It is in favour to have a more frequently odour and flavour quality control on rubber products produced. Electronic noses and tongues are devices that can measure respectively volatile or non-volatile compounds. These devices contain an array of sensors each coated with a different interface layer. The (non)-volatile compounds from the sample enter into non-covalent, non-specific interactions with the interface layers resulting in a response per sensor. From the responses of all the sensors, patterns are distinguished using neural networks. On the basis thereof, changes in composition of a sample stream or the presence of particular substances can be monitored.

Various applications , mainly for the food industry, are developed for electronic noses and tongues. There are also methods available for environmental monitoring (water analysis and contaminated soil analysis) and medical applications (5, 6, 8, 10).

The aim of the research performed was to identify if the performances of electronic noses and tongues are comparable to that of an odour and flavour panel. If so, these electronic devices can be of help to increase the frequency of quality control for rubbers. This report compares results obtained with an electronic tongue, electronic nose and an odour and flavour panel for non-concentrated aqueous samples. Samples were prepared by bringing rubbers in contact with drinking water which are subsequently tested by an odour and flavour panel, an electronic nose and electronic tongue.



2 Materials and Methods

2.1 Sample preparation

Odour & flavour panellists are offered a diluted sample of the migration water. In the NEN-test setup the concentration of migrated molecules into the drinking water is too high for the human nose and tongue and even irritate them. The concentrations obtained are also not representative for operational use of the rubbers. The NEN-norms stated that the dilution factor is set to 150 for rubbers (3). Sample preparation was performed according to NEN-EN 1420-1 by immersing Ethylene-Propylene-Diene-Monomer (EPDM) and Acrylonitrile butadiene rubber (NBR) materials into drinking water during 3 x 72 hr. at a temperature of 23ºC. The water extracts (first migration) were collected and immediately diluted to 150-fold and tested by the odour and flavour panel. Fresh drinking water was added to the same test specimens and incubated for a second 72 hours. These water extracts were collected (second migration) and immediately diluted to 150-fold and assessed by the panel. The test specimens are brought into contact for the third time with drinking water and incubated for an other 72 hours. The water extracts were collected again and diluted to 150-fold and assessed by the panel. For the electronic-nose and –tongue experiments the samples were first collected, frozen at 20°C and stored until use. Before testing by electronic nose and tongue the samples were defrosted and subsequently tested undiluted. The samples for this research were obtained from regular performed odour and flavour testing. All samples were coded (to protect suppliers privacy) and blind tested by the odour and flavour panel, electronic nose and tongue. No pre-information on sample identity were communicated upfront. The NBR and EPDM rubbers are provided by different suppliers each having their own specific production process. The processes slightly differ from each other resulting in product deviations. The production process are undisclosed and not relevant for this research. All samples are correlated to a blank drinking water sample (blanco).

2.2 Protocol for odour - and flavour threshold assessments

The odour and flavour experiments were performed with 8 panellists according to the forced-choice paired test (EN 1622-1). Odour and flavour testings were performed immediately after migration as described in 2.1 Water dilutions from the migration water were prepared in flavour-free bottled drinking water. Water samples were brought to expected temperature of 23º for odour and flavour threshold concentration before being submitted to panellists (30 minutes in a water bath or incubator). Odour and flavour threshold concentrations were calculated in two ways, geometric means and a graphical procedure (logistic regression by inverse curve). For the graphical



approach, threshold concentrations were calculated for 50% of right panellist answers, in addition 95% confidence intervals were calculated.

2.2.1 Model validation for odour and flavour assessments by a panel

Panellists were selected on their ability of selectivity to descriptors, preliminary training and experience with the procedure. The odour and flavour results of new panellists need to be normal distributed. The panel should not be more sensitive than the normal population. The test procedure was carefully explained to the panellists. They choose the bottle in a pair or triangle with the stronger odours and flavours, regardless of how confidence they are in their judging. Next, the panellists participate in an initial sensitivity trial under local circumstances. The objective of these experiments was to validate the results provided by the kinetic model under typical drinking water conditions. For that purpose, laboratory scale experiments were performed in drinking water with added concentration of methyl tertiary butyl ether (MTBE) descriptor. MTBE is considered primarily a taste and odour concern and a toxicity issue at the level of drinking water. Thus, setting up drinking water standards with a MTBE descriptor has become an issue to selecting and training of a panel. The Threshold Odour Number (TON) and Threshold Flavour Number (TFN) is not allowed to be lower than 150 for rubbers. Therefore the migration water of tested rubbers was diluted 150 times and offered to the odour and flavour panel. The panellists are offered two cups of water (A and B) for each sample tested. In one cup the diluted migration water was added while the second cup contained untreated drinking water. The panellists had to judge the paired samples, by taking into their mouth whatever volume of water is comfortable, and to hold in the mouth for several seconds before discharging it without swallowing. The panellists selects the sample one over the other on the basis of stimulus values in degree of intensity or specific taste. The panellists are instructed to guess when in doubt. When 6 or more panellists out of 8 assessed the sample as the deviating one, the sample was rejected. For this research we counted the number of panellists that identified the deviating sample. For the comparison between the odour and flavour panel and the electronic nose and tongue we used these figures as results for the odour and flavour panel.

2.3 Electronic Nose

Electronic olfaction, here called “electronic nose” technology (7, 9, 11, 15, 16), is designed to measure an odour in a way similar to humans. It does not analyze parts of an odour and even tolerate some ambiguity, just as humans do. It provides an estimate of human acceptance.



If electronic olfaction technology can predict human response successfully, it can add confidence to the supplier/manufacturer relationship. The system reacts to most odours. The technology combines gas sensor technology .with artificial intelligence. Artificial intelligence shows the difference in the sensors reaction to various samples thus creating a recognizable pattern or fingerprint. From this fingerprint and from previous panel test results the system predicts the most likely human response to the new pattern. The response is put in simple terms like “recognized”, “good”, or “bad” (5, 8, 17, 18). Figure 1 illustrates the general architecture of the instrument and the basic principles of operation :

Figure 1: General architecture and basic principles of operation of the αFox electronic nose of Alpha-MOS, France.

HS100 Static Headspace Auto sampler

ααααFox

Gas Ss Sensor Array System

Acquisition

Statistical data processing

DATA ACQUISITION

The values obtained from the Gas Sensors are summarized in a data bank or library

DATA PROCESSING

Chemometrics associate sensor response pattern with analytical

and/or sensory panel differences (good, bad, origin, contamination,

weak, strong..) to display the sample pattern or identify groups.



The auto sampler uses the static headspace principle : The sample (liquid, solid or gas) is put into a sealed glass vial (10 or 20ml). The filled vial is then placed into the auto sampler's incubator in order to generate a headspace containing the volatile compounds. The headspace is finally taken from the vial using a gas syringe (from 1 to 5ml) and injected into the instrument.

The sensor array is based on gas sensors which produces a fingerprint from an odour. This array can be made up to 18 different sensors. The choice of sensors is highly linked to the application objectives. A first selection of sensors is based on robustness and then further tests allow to select the most discriminating ones in order to tune a specific system of the target applications. The detection is based on conductivity measurements. In the presence of a combustible gas, the oxygen species are adsorbed on the metal oxide, reacts and are removed from the surface. As a result, the conductivity of the metal oxide film changes.

Multivariate Analysis is used to discriminate between the various samples. (The MVA’s used are described in paragraph 2.5)

2.3.1 Electronic nose measurement protocol

The system used for this study was a αFox Electronic Nose system and AlphaSoft V11 software both developed and produced by Alpha-MOS. All the tests have been conducted using the same instrumental set-up. Twelve of the shelf sensors from Alpha-MOS are used for this study with product codes: LY2/LG, LY2/G, LY2/AA, LY2/GH, LY2/gCTL, LY2/gCT (Chamber CL), T30/1, P10/1, P10/2, P40/1, T70/2 and PA2 (Chamber A). As carrier gas a synthetic air was used with a composition of 21% oxygen, 78% nitrogen, 1% argon, < 0.1 ppm carbon and < 1 ppm carbon dioxide. The flow speed was 150 ml/min To generate a consistent headspace 5 ml of migration sample was added to a 10 ml glass vial. The vial was then automatically placed in an incubator at a temperature of 60°C for 20 minutes. During incubation the vial was agitated with a speed of 500 rpm according to a repeated sequence of 5 seconds agitation and 2 seconds of rest. After incubation a volume of 2500 µl headspace was collected with a thermo stated gas syringe of 2500 µl at 70°C. The fill speed of the syringe was 500 µl/sec. The sample was sequential injected into the carrier stream of the system with an injection speed of 1250 µl/sec. The sensors in chamber CL and A operated at an ambient temperature of 60°C and 65°C respectively. The sensor signals were collected during a period of 120 sec after injection of the sample and saved to disk. To achieve stabilisation of the sensor in the



system a delay of 1080 sec between sequential injection was introduced. During this delay the syringe was washed for 30 sec to avoid cross contamination of the samples.

2.4 Electronic Tongue

The electronic tongue is developed to analyse taste by dissolved organic and inorganic chemicals and is complementary to the electronic nose. The electronic tongue system is designed to supplement conventional techniques as sensory panels or liquid chromatography.

For this research the αASTREE of Alpha-MOS is used as electronic tongue and consists of the following parts: An auto-sampler (A) and an array of sensors (B) and a unit for data acquisition (C) and a chemometric software package (D).

Figure 2: General architecture αAstree electronic tongue of Alpha-MOS,

France. The electronic tongue needs to be trained with selected samples to ensure good recognition and reproducibility. Sample vials are filled with the test samples and sequential injected to the sensor array. A washing cycle ensures that there is no carry over from one sample to the other. Total analysis cycle (i.e. sample measurement and sensor cleaning times) takes 3 minutes. Typically, up to 5 replicates measurements are made for each sample in order to check measurement repeatability (Residual Standard Deviation RSD < 5 %). Sensors are made from silicon transistors with different organic coatings. The signals produced are the potentiometric difference between each individually coated sensor and a Ag/AgCl reference electrode. The electronic tongue is equipped with an array of seven different sensors. The choice of sensors is highly linked to the application objectives. A first selection of sensors is based on robustness and then further tests allows to select the most discriminating ones in order to tune a specific system of the target applications.

D

C B

A

A

C



Multivariate Analysis is used to discriminate between the various samples. (The MVA’s used are described in paragraph 2.5)

2.4.1 Electronic tongue measurement protocol

The system used for this study was a α-Astree electronic tongue system and AlphaSoft V11 software both developed and produced by Alpha-MOS. All the tests are conducted using the same instrumental set-up. Seven of the shelf sensors from Alpha-MOS are used for this study with product codes: ZZ, BA, BB, CA, GA, HA and JB. Migration samples are automatically tested by the sensor system without any sample pre-treatment. The samples are poured into vials on a carrousel with a volume of 25 ml. The sensor head is moving from one sample to the other. The sensor signals were collected during a period of 120 sec after dipping of the sensor head into the analysed vial and saved to the disk. A delay time of 60 sec was used before the next sample due to the carrousel movement (up/down of the sensor head, rotation of the carrousel for analysis the next sample, cleaning of the sensor head)

2.5 Data treatment methods

The electronic nose and tongue systems provided patterns of data. It is unfeasible to perform the data interpretation by hand so, in order to compare obtained patterns, multivariate statistical analysis has been performed (12, 13, 14). With multivariate statistical analysis more then one statistical variable can be observed and analysed. The fingerprints produced by an electronic nose and tongue contains more then one variable, 12 and 7 respectively. Multivariate statistic algorithms identify which of the variables are of most importance to describe the differences or uniformity of tested samples. The following multivariate statistical techniques are used in this study: � Principal Component Analysis (PCA) is used for dimensionality reduction in a data set, and to assess discrimination performances (transforming the data to a new coordinate system such that the greatest variance by any projection of the data comes to lie on the first coordinate (first principal component), the second greatest variance on the second coordinate, and so on). PCA is the first data analysis step performed prior to DFA, SIMCA or PLS. The PCA results are not presented in this report but are available on request.

� Discriminant Factorial Analysis (DFA) is used to identify unknown samples into one of the training groups DFA algorithm gives access to qualitative application for sample group identification application. Based on the known characteristics of the materials tested a DFA algorithm has been developed which separate and predict the sample group partition.



This method provides a two dimension map and a validation score allowing to judge the performance of the discrimination and compare it among different techniques (in this report: electronic nose, electronic tongue and their combination). DFA models can also be used to classify unknown samples by projecting them on the map. For each unknown sample, the distance to the centroïd of each group is computed. The sample is then identified as part of the group associated to the minimum distance.

� Soft Independent Modelling by Class Analogy (SIMCA) is used to compare an unknown sample to a reference; The interest of the soft independent modelling of class analogy (SIMCA) method comes from the ability to build a model by taking into account only one group considered as the "reference". In this research the data of the blanks are used to build up the model. The model is going to identify unknown samples as belonging or not to the one and the only one group previously defined. The standard deviation of the blanks used as references define the limits for judging which sample is comparable to the blank and which is not. When the standard deviation is small a narrow area is defined. The advantage of the SIMCA model is that it is not necessary to collect samples from different groups but it is only necessary to collect representative samples of the group of interest.

� Partial Least Squares regression (PLS) is used to find the correlation coefficient between the odour & flavour panel results and the electronic nose and tongue results. In statistics, the method of partial least squares regression (PLS) bears some relation to principal component analysis; instead of finding the hyperplanes of maximum variance, it finds a linear model describing some predicted variables in terms of other observable variables. This method is used to correlate the electronic nose and tongue measurements with quantitative sample characteristics. There are mainly two types of quantitative applications:

• Determination of concentration of specific compounds within the product (basic taste substance, off-taste substance, bitterness unit…)

• Sensory panel score The objective is to build a model that is able to predict this quantitative information for each sample to be analysed. This method provides a correlation coefficient between to two sets of data allowing to judge the performance and the obtained sample marks for comparing values.

In summary, the following multivariate statistic results are presented in this report: - DFA map will allow to check the discrimination capability of the electronic nose or electronic tongue system, analysing the same sample set,

- SIMCA model will allow to check the capability of each system to differentiate samples from the blank (reference),



- PLS model will allow to check the capability of each system to correlate the obtained data with the sensory panel scores.

2.6 General remarks

To avoid that the operators of the electronic nose and tongue could influence the results, no pre-information on sample origin and odour & flavour panel results were provided to them. All samples were tested under the same instrumental conditions and analysis protocols. In the graphical representation of the results the codes M1, M2 and M3 are used representing the results of migration 1, migration 2 and migration 3 respectively. Each migration area is the result of 5 measurements of the same sample. Drinking water composition change over time and the experimental setup reckon with this fact. For each migration experiment a new blank drinking water is used. When to compare a sample with the blank, one should compare to the blank of a specific migration and not compare it to a blank with an other migration number.



3 Results

The human nose and tongue are more sensitive for chemicals in solution than the electronic nose and tongue. In this research we investigated if electronic measurements produces comparable results as an odour & flavour panel on which an acceptation or rejection of rubbers can be based. In this chapter we demonstrate the discriminating properties of electronic nose and tongue. Secondly, we show that the acceptation or rejection of rubbers based on electronic measurements are comparable to the judgement of an odour & flavour panel.

3.1 Odour & flavour panel

Table 1 is summarising the odour & flavour panel results. The results represent the number of persons out of eight recognising the sample. Table 1: Panel results for NBR and EPDM rubbers.

1st migration 2nd migration 3rd migration

Sample Type Odour Flavour Odour Flavour Odour Flavour

Blank -

0235 NBR 3 5 5 3 6 3

0396 EPDM nd nd nd nd 5 3

0715 EPDM 3 1 4 3 5 4

1568 NBR 5 6 4 3 4 3

2159 NBR nd nd nd nd 3 4

2435 EPDM 1 5 5 3 4 4

nd: These products were extensively tested in the past. For batch testing only the third migration has to be performed by the odour and flavour panel. In those cases no results were obtained for the 1st and 2nd migration (samples 0396 and 2159).

3.2 Electronic Nose

3.2.1 Discrimination capability of the electronic nose system

The discrimination capability of the system can be checked by performing a Discriminant Factorial Analysis (DFA). Figure 3 presents the DFA map of all NBR samples with the three different migrations. Not all sensor signals contributes in the same way to the discrimination of samples. Therefore a sensor selection allows to present the best possible discrimination capability of the system. In this case 4 sensors out of 12 have been selected: T30/1, P40/1, T70/2, PA/2.



Blank 2159 1568 235

M1

M1

M1

M1

M3

M2

M2

M2

M2

M3

M3

M3

Figure 3: DFA map of E-nose results of all NBR samples with the three

different migrations. DFA only can discriminate sample 2159 from the blank sample. No discrimination can be obtained between samples 235 & 1568 and the blank. Based on the DFA calculation of the electronic nose results the last two samples are rather similar to the blank. Figure 4 presents the DFA map of all EPDM samples with the three different migrations. In this case 12 sensors out of 12 have been selected

blank 2435

715

396

M1

M1

M1

M1

M2

M2

M2

M2

M3 M3

M3

M3

Figure 4: DFA map of E-nose results of all EPDM samples with the three

different migrations. Based on the DFA analysis rubber 396 is the most similar to the blank. Rubbers 715 & 2435 are the most different to blank. The electronic nose is able to discriminate between the different samples per migration experiment. However, the differences in signal are small resulting in overlapping areas of the different samples when all the migration experiment results are plotted.



3.2.2 Quality control tool capability of the electronic nose system

Taking into account the blank measurements of each rubber migration as the reference , it is possible to build a Soft Independent Modelling of Class Analogy model (SIMCA). This SIMCA model identifies if the samples are equal or not to the reference (blank in this case). If the samples are equal to the blank they passes the test as accepted. In all other cases the samples will be rejected. As an example, figure 5 shows the graphical representation of the SIMCA model of the first migration of the NBR rubbers.

235

1568

2159

Figure 5: SIMCA model of E-nose results of the first migration of the NBR

rubbers Based on the blank sample results a conform territory is defined. (in blue on the map based on the standard deviation of the measured signals of the blank). The samples are then projected over this model. If the samples are plotted into the blue area the rubbers are accepted. Plotted outside the blue area a rubber is rejected. Sample 2159 is projected outside the conform territory, this rubber is rejected. Rubber 1568 is a borderline and has to be further checked before accepted or rejected. Rubber 235 is projected inside the conform territory, this rubber is accepted. The SIMCA calculations are performed for all six samples for all three migrations. Table 2 summarises the SIMCA results of the electronic nose measurements, by expressing if a sample is reject, accepted or a borderline based on the SIMCA calculations.



Table 2: SIMCA results of electronic nose measurements obtained for 3 NBR and 3 EPDM samples for all 3 migrations.

Rubber type Rubber process

Migration 1 Migration 2 Migration 3

235 A A A

1568 B A A NBR

2159 R A R

396 A A A

715 A A A EPDM

2435 A A A

A means accepted B means borderline R means rejected Based on the SIMCA results of the electronic nose measurements all samples are accepted except sample 2159. When acceptation or rejection was only based on the results of the first migration sample 1568 is a borderline and should be closer investigated.

3.2.3 Panel correlation capability of the electronic nose system

The correlation capability of the system can be checked by performing a partial least squares (PLS) model. Correlations are given for both the odour and flavour results of the odour & flavour panel. This is done because when a human is tasting a sample (taking water into the mouth) both flavour and odour are judged. So the flavour results are build up by the experience of flavour and odour. For the PLS calculations all migration results of the NBR or EPDM samples are used. The PLS could be calculated based on the results of the individual migrations but the statistical relevance of these calculations should be too low. The electronic nose results are correlated with the odour and flavour results of the odour & flavour panel. The figures 6 and 7 show the PLS calculations for the NBR samples correlating the electronic nose results to the odour and the flavour results of the odour & flavour panel.



Figure 6: Graphic representation of the PLS calculations of the NBR-samples

(235, 1568 and 2159). Electronic nose and the odour results of the odour & flavour panel are compared. (Reference = panel score, Measured = electronic nose score)

Figure 7: Graphic representation of the PLS calculations of the NBR-samples

(235, 1568 and 2159). Electronic nose and the flavour results of the odour & flavour panel are compared. (Reference = panel score, Measured = electronic nose score)

The figures 8 and 9 show the PLS calculations for the EPDM samples correlating the electronic nose results to the odour and the flavour results of the odour & flavour panel.



Figure 8: Graphic representation of the PLS calculations of the EPDM-

samples (235, 1568 and 2159). Electronic nose and the odour results of the odour & flavour panel are compared. (Reference = panel score, Measured = electronic nose score)


samples (235, 1568 and 2159). Electronic nose and the flavour results of the odour & flavour panel are compared. (Reference = panel score, Measured = electronic nose score)

Table 3 summaries the obtained correlation coefficient for the different sample types in the comparison of electronic nose results to the odour & flavour results of the panel.



Table 3: Obtained correlation coefficient for the different sample types in the comparison of electronic nose results to the odour and flavour results of the odour & flavour panel.

Rubber type Sample ID Panel odour results

Panel flavour results

235

1568 NBR

2159

0.67 0.79

396

715 EPDM

2435

0.77 0.79

Table 4 summaries the average correlation precision for the different sample types in the comparison of electronic nose results to the odour and flavour results of the odour & flavour panel. Table 4: Obtained average correlation precision for the different sample types

in the comparison of electronic nose results to the odour and flavour results of the odour & flavour panel.

Rubber type Sample ID Panel odour results (%)

Panel flavour results (%)

235

1568 NBR

2159

16 13

396

715 EPDM

2435

13 8

3.3 Electronic Tongue

3.3.1 Discrimination capability of the electronic tongue system

The discrimination capability of the system can be checked by performing a Discriminant Factorial Analysis (DFA). Figure 10 presents the DFA map of all NBR samples with the three different migrations. Not all sensor signals contributes in the same way to the discrimination of samples. Therefore a sensor selection allows to present the best possible discrimination capability of the system. In this case 6 sensors out of 7 have been selected here : ZZ, BB, CA, GA, HA & JB.



blank

235

2159

1568 M1

M1

M1

M1

M2

M2

M2

M2

M3

M3

M3

M3

Figure 10: DFA map of E-tongue results of all NBR samples with the three

different migrations. A good discrimination is obtained between the different NBR-samples. All samples differ from the blank although sample 235 is almost similar to the blank. The performance of the DFA analysis is better than the one of the electronic nose (See chapter 3.2.1). The electronic tongue is able to see differences and to rank the different rubbers based on the signals produced. The panel is not able to discriminate between one or the other rubber. Globally, the panellists notice off taste or off odour of rubbers compared to the blank. Figure 11 presents the DFA map of all EPDM samples with the three different migrations. All sensors contributed to the discrimination between the samples. Therefore all sensors have been used incorporated in this DFA analysis.

blank

396

715 2435

M1

M1

M1

M1

M2

M2

M2

M2

M3

M3

M3

M3

Figure 11: DFA map of E-tongue results of all EPDM samples with the three

different migrations.



Based on the DFA analysis the highest discrimination is obtained for the samples 396 and 715. Sample 2435 is almost similar to the blank. The performance of the DFA analysis is better than the one of the electronic nose (See chapter 3.2.2). The electronic tongue is able to see differences and to rank the different rubbers based on the signals produced. The panel is not able to discriminate between one or the other rubber. Globally, the panellists notice off taste or off odour of rubbers compared to the blank.

3.3.2 Quality control tool capability of the electronic tongue system

Taking into account the blank of each rubber migration as a “reference” , it is possible to build a Soft Independent Modelling of Class Analogy model (SIMCA). This SIMCA model identifies if the samples are equal or not to the reference (blank in this case). If the samples are equal to the blank they passes the test as accepted. In all other cases the samples will be rejected. The SIMCA calculations are performed for all six samples for all three migrations. Table 5 summarises the SIMCA results of the electronic tongue measurements, by expressing if a sample is reject, accepted or a borderline based on the SIMCA calculations. Table 5: SIMCA results of electronic tongue measurements obtained for 3

NBR and 3 EPDM samples for all 3 migrations.



235 A A A

1568 A A A NBR

2159 A A A

396 A A A

715 A A A EPDM

2435 A A A

A means accepted Based on the SIMCA results all samples are accepted.

3.3.3 Panel correlation capability of the electronic tongue system

The correlation capability of the system can be checked by performing a partial least squares (PLS) model. Correlations are given for both the odour and flavour results of the odour & flavour panel. This is done because when a human is tasting a sample (taking water into the mouth) both flavour and odour are judged. So the flavour results are build up by the experience of flavour and odour. For the PLS calculations all migration results of the NBR or EPDM samples are used. The PLS could be calculated based on the results of the individual



migrations but the statistical relevance of these calculations should be too low. The electronic tongue results are correlated with the odour and flavour results of the odour & flavour panel. The figures 12 and 13 show the PLS calculations for the NBR samples correlating the electronic tongue results to the odour and the flavour results of the odour & flavour panel.

Figure 12: Graphic representation of the PLS calculations of the NBR-

samples (235, 1568 and 2159). Electronic tongue and the odour results of the odour & flavour panel are compared. (Reference = panel score, Measured = electronic tongue score)


samples (235, 1568 and 2159). Electronic tongue and the flavour results of the odour & flavour panel are compared. (Reference = panel score, Measured = electronic tongue score)

The figures 14 and 15 show the PLS calculations for the EPDM samples correlating the electronic tongue results to the odour and the flavour results of the odour & flavour panel.




samples (396, 715 and 2435). Electronic tongue and the odour results of the odour & flavour panel are compared. (Reference = panel score, Measured = electronic tongue score)


samples (396, 715 and 2435). Electronic tongue and the flavour results of the odour & flavour panel are compared. (Reference = panel score, Measured = electronic tongue score)

Table 6 summaries the obtained average correlation coefficient for the different sample types in the comparison of electronic tongue results to the odour & flavour results of the panel.



Table 6: Obtained average correlation coefficient for the different sample types in the comparison of electronic tongue results to the odour and flavour results of the odour & flavour panel.



235

1568 NBR

2159

0.83 0.73

396

715 EPDM

2435

0.74 0.78

Table 7 summaries the obtained average correlation precision for the different sample types in the comparison of electronic tongue results to the odour & flavour results of the panel. Table 7: Obtained average correlation precision for the different sample types

in the comparison of electronic tongue results to the odour and flavour results of the odour & flavour panel.



235

1568 NBR

2159

9 16

396

715 EPDM

2435

13 10

3.4 Combination of Electronic Nose & Electronic Tongue

The odour panel and the electronic nose system only characterise the volatile part of each sample. Depending on the operational conditions, the volatile part of the sample can differ. This is why the correlation between both techniques might differ; the difference is only provided by the relative dispersion of each techniques. The taste panel and the electronic tongue system characterise both the volatile and the dissolved parts of each sample. This should be the most complete characterisation of a sample. However, the electronic tongue has a better discrimination capability to analyse dissolved compounds then volatile ones. This is why it is highly important to analyse samples with both systems (EN & ET) and if needed to improve the results taken by performing a combination. In this paragraph the results of the electronic nose and tongue are combined to investigate if the combined results gave an increase of correlation with the odour & flavour panel.



3.4.1 Discrimination capability of the combined system

The discrimination capability of the system can be checked by performing a Discriminant Factorial Analysis (DFA). Figure 16 presents the DFA map of all NBR samples with the three different migrations. Not all sensor signals contributes in the same way to the discrimination of samples. Therefore a sensor selection allows to present the best possible discrimination capability of the system. In this case 18 out of 19 sensors are selected. (6 out of 7 for the electronic tongue and 12 out of 12 for the electronic nose)

blank 2159

1568

235

M1

M2

M3

M1

M2

M2

M2

M1

M3

M3

M3

M1

Figure 16: DFA map of E-nose + E-tongue results of all NBR samples with

the three different migrations. Based on the DFA analysis of the combined sensor results a better discrimination performance is obtained compared to electronic nose or electronic tongue measurements alone (See paragraph 3.2.1 and 3.3.1). Figure 17 presents the DFA map of all EPDM samples with the three different migrations. All In this case 19 out of 19 sensors are selected..



blank 715

2435

396

M1

M1

M1

M1

M2

M2

M2

M2

M3 M3

M3

M3

Figure 17: DFA map of E-nose + E-tongue results of all EPDM samples with

the three different migrations. The discrimination performance increases also in the case of the EPDM rubber. Sample 2435 is still close related to the blank but in comparison with the electronic nose results it has no overlap anymore with the blank results (See paragraph 3.2.2 and 3.3.2).

3.4.2 Quality control tool capability of the combined system

Taking into account the blank of each rubber migration as a “reference” , it is possible to build a Soft Independent Modelling of Class Analogy model (SIMCA). This SIMCA model identifies if the samples are equal or not to the reference (blank in this case). If the samples are equal to the blank they passes the test as accepted. In all other cases the samples will be rejected. Table 8 summarises the SIMCA results obtained for all samples and migration experiments. Table 8: SIMCA results obtained with E-nose + E-tongue for 3 NBR and 3

EPDM samples for all 3 migrations.



235 R A A

1568 R A A NBR

2159 R A R

396 A R R

715 A A R EPDM

2435 A A A

A means accepted R means rejected



The SIMCA results of the combined systems show more rejections then the individual systems. The SIMCA model based on the combined results is able to identify smaller differences between blank sample and the migration samples. The question raises if these differences measured are a good basis for the determination of acceptation or rejection of a sample. In this case the SIMCA model is too sensitive and not representative for making decisions.

3.4.3 Panel correlation capability of the combined system

The correlation capability of the system can be checked by performing a partial least squares (PLS) model. Correlations are given for both the odour and flavour results of the odour & flavour panel. This is done because when a human is tasting a sample (taking water into the mouth) both flavour and odour are judged. So the flavour results are build up by the experience of flavour and odour. For the PLS calculations all migration results of the NBR or EPDM samples are used. The PLS could be calculated based on the results of the individual migrations but the statistical relevance of these calculations should be too low. The combination of electronic nose & tongue results are correlated with the odour and flavour results of the odour & flavour panel. The figures 18 and 19 show the PLS calculations for the NBR samples correlating the combined electronic nose & tongue results to the odour and the flavour results of the odour & flavour panel.


samples (235, 1568 and 2159). Electronic nose & tongue results and the odour results of the odour & flavour panel are compared. (Reference = panel score, Measured = combined electronic score)




samples (235, 1568 and 2159). Electronic nose & tongue results and the flavour results of the odour & flavour panel are compared. (Reference = panel score, Measured = combined electronic score)

The figures 20 and 21 show the PLS calculations for the EPDM samples correlating the combined electronic nose & tongue results to the odour and the flavour results of the odour & flavour panel.


samples (396, 715 and 2435). Electronic nose & tongue results and the odour results of the odour & flavour panel are compared. (Reference = panel score, Measured = combined electronic score)




samples (396, 715 and 2435). Electronic nose & tongue results and the flavour results of the odour & flavour panel are compared. (Reference = panel score, Measured = combined electronic score)

Table 9 summaries the obtained average correlation coefficient for the different sample types in the comparison of the combined electronic nose and tongue results to the odour & flavour results of the panel. Table 9: Obtained average correlation coefficient for the different sample

types in the comparison of the combined electronic nose and tongue results to the odour & flavour results of the panel.



235

1568 NBR

2159

0.84 0.82

396

715 EPDM

2435

0.93 0.81

Table 10 summaries the obtained average correlation precision for the different sample types in the comparison of the combined electronic nose and tongue results to the odour & flavour results of the panel. Table 10: Obtained average correlation precision for the different sample

types in the comparison of the combined electronic nose and tongue results to the odour & flavour results of the panel.



235

1568 NBR

2159

9 9

396

715 EPDM

2435

4 9





4 Discussion


The test water used for the odour & flavour panel test is obtained from Tull & Waal. The panel has determined that this water has no smell or taste. The blanks for this research are test water samples which are treated in the same away as the tested materials are except that the blank was not exposed to any test material. Blank water, after exposure, was used to dilute the water coming from the material exposure. The odour & taste panel method is based on the forced-choice paired test. In this method the panellist make a choice between the test sample and a blank sample. The score of the panel indicates if there is a weak or strong deviation between the blank and the test sample. For statistical use the panel scores of the blank samples are defined as “0”. The authors indicate that this assumption could have influenced the performance of the statistical calculations. Future research has to investigate the relevance of this assumption.

4.2 Statistics

Within this research a limited number of samples is tested. This has a direct influence on the performance of the statistical calculations. To secure the statistical relevance of the results we performed the calculation on the combined results of three NBR or EPDM samples. The results included respectively all data from all three migration experiments. As result average correlation coefficient and correlation precision are obtained for the comparison with the odour & flavour panel results. To increase the reliability of the results, increasing numbers of samples have to be tested. Typically over 300 samples have to be tested per sample type to achieve this reliability. Based on experiences from other areas the discriminating property of the DFA analysis is judged as follows: When the validation score is between 60 and 80 the discriminating property is defined as “sufficient”. Above a validation score of 80 the discriminating property is defined as “good”. The best is being the closest as possible to 100. A comparable judgement can be made for the PLS analysis resulting in the following limits: A correlation coefficient between 0.60 and 0.70 is judged as “moderate”, between 0.70 and 0.80 as “sufficient” and above 0.80 as “good”. The best is being the closest as possible to 1. In this research the panel score of the blank samples are set to “0” as discussed in paragraph 4.1. As said before this is under discussion. But when the blank samples are removed from the data set, the PLS calculation of the other samples have to be recalculated. Before recalculation the average signals of the blank samples have to be subtracted first from the signals of the individual signals of all samples measured. This recalculation is not



performed by the authors. In general the correlation coefficients will change but no estimation can be made if these will increase or not. The odour & flavour panel judge if a sample is accepted or rejected. All samples tested in this research were accepted by the odour & flavour panel. This knowledge is used to judge if the SIMCA model is appropriate to determine if a sample has to be accepted or rejected. Table 11 provides qualifications for the statistical calculations performed on the results obtained in this research. Qualifications are used as defined above. Table 11: Qualifications for the statistical calculations for electronic nose,

electronic tongue or the combination

Electronic nose

Electronic tongue

Combination

NBR rubber

DFA Sufficient Good Good

SIMCA Appropriate Appropriate Not

appropriate

PLS (Correlating odour panel)

Moderate Good Good

PLS (Correlating flavour panel)

Sufficient Sufficient Good

EPDM rubber

DFA Sufficient Good Good

SIMCA Appropriate Appropriate Not

appropriate

PLS (Correlating odour panel)


PLS (Correlating flavour panel)


The combination of both electronic systems provides better statistical performances for the DFA and PLS methods then the individual electronic nose and tongue systems. The opposite is the case for the SIMCA method which is too sensitive for the judgement of acceptation or rejection of the samples. The individual electronic nose or tongue systems are better equipped for this purpose. Depending on the purpose of use one can judge based on financial consideration if the combination is worthwhile or if one of the two system is sufficient.



4.3 Electronic nose system

In general an electronic nose is able to analyse aqueous migration samples from different rubber types. Based on the DFA results the electronic nose is able to distinguish between one or the other rubber sample per migration experiment. However, the differences in signal are small resulting in overlapping areas of the different samples when all the migration experiment results are plotted. The electronic nose results are sufficient enough to use as input for building up a SIMCA model that is able to judge if a sample has to be accepted or rejected. The PLS results show a moderate to sufficient comparison between the electronic nose and the odour & flavour panel. The average error obtained is below 15 %, which can be judged sufficient for at least a first screening of samples. The poor correlation coefficient between the electronic nose and odour & flavour panel might be explained by the heating step in the protocol of the electronic nose. To speed up the release of volatile compounds from the drinking water sample, the sample is heated up to 60°C for 20 minutes. It is conceivable that at this temperature volatile compounds reacts with either water, molecules dissolved in water or gas molecules. If so, the composition of the headspace tested by electronic nose differs from the headspace tested by the odour & flavour panel resulting in a poor correlation between the electronic system and the panel.

4.4 Electronic tongue system

The electronic tongue is also able to analyse aqueous migration samples from different rubber types. A better discrimination between the rubber samples is obtained with the electronic tongue system according to the electronic nose system. The DFA calculations show a better performance on which it is able to distinguish between one or the other rubber sample. The odour & flavour panel lacks this possibility of discrimination. Also the electronic tongue results are sufficient enough to use as input for building up a SIMCA model that is able to judge if a sample has to be accepted or rejected. The PLS results show a sufficient to good comparison between the electronic tongue and the odour & flavour panel. This is better then the PLS results of the electronic nose. The average error obtained is below 15 %, which can be judged sufficient for at least a first screening of samples.



4.5 Combination of Electronic Nose & Electronic Tongue

The combination of electronic nose and electronic tongue is also able to analyse aqueous migration samples from different rubber types. The highest discrimination between the different rubber samples is obtained with the DFA results of the combination of electronic nose and tongue. The odour & flavour panel lacks the possibility of discriminating between one or the other rubber. The SIMCA model based on the combined results is able to identify smaller differences between blank sample and the migration samples. In this case the SIMCA model is too sensitive and results are showing that compared to the odour & flavour panel results a sample can not be rejected or accepted based on the combined electronic results. The PLS results show an increase correlation coefficient and correlation precision for the different samples in the comparison of the combined electronic nose and tongue results to the odour & flavour results of the panel. The average error obtained is below 10 %, which is significant lower then the average errors of the individual electronic nose and tongue systems. The use of the combined systems is for sure sufficient for at least a first screening of samples.



5 Conclusions

To investigate the value of use for an electronic nose and/or electronic tongue in the production of materials in use for distribution of drinking water a comparison was made with an odour & flavour panel. Materials that are used for the transport of water have to be tested according to the NEN-EN 1622:2006 and NEN-EN 1420-1 methods. These validated methods are in use to determine whether polymers or rubbers are releasing chemical compounds which influences the odour and/or flavour of drinking water. The acceptation or rejection of a material is based on the results of the odour & flavour panel testing end-products or in some cases raw product. The results obtained from the same samples offered to the odour & flavour panel, electronic nose and electronic tongue are showing moderate to good correlation between the compared methods. The combination of results from the electronic nose & -tongue demonstrate the best correlation with the odour & flavour panel results. The combination of instrumental analysis and statistical analysis of the data is able to discriminate between different production processes of the same material (NBR or EPDM). The best discrimination is obtained with the combination of results of the electronic nose & tongue systems. The odour and flavour panel does not have these discriminating abilities. To increase the certainty of the discriminating ability a higher number of samples have to be tested. The result of this future testing can be used to implement electronic nose and –tongue systems at producers factories to determine the quality of the product in an earlier stage of production. Based on the results of this report the question raises if the odour and flavour panel test can be substituted by electronic nose and/or –tongue methods. Although the results are promising the authors suggests a consolidating approach for the (near) future. The introduction of instrumentation at producers factories will open the opportunity to test the raw materials before production will start. This might reduce the number of rejections of end-products by odour and flavour tests. In the meanwhile odour and flavour tests can be used to validate the use of instrumental methods. At some point in the future, based on gained experiences, it can be decided to adapt nowadays practises in testing, resulting in a reduction of number of odour and flavour tests. At that time it is preferred to have validated instruments and according methods. It is not foreseen that the instruments will fully substitute the odour and flavour method. The procedure of an electronic nose differs from the procedure of an electronic tongue. With an adaptation of the instrumentation it is possible for the electronic tongue to perform continuous measurements. An electronic nose only can perform discontinuous measurements of water. This is due to



the fact that an electronic nose is measuring the headspace of a water sample. A typical measurement run for an electronic nose is 25 – 30 minutes. It has to be investigated if online electronic nose measurements can be performed without lost of robustness of the measurement. Up-to-now no investigations have been performed to determine the online application of an electronic nose or –tongue for water quality management. The results of this project supports the need to investigate the possibilities for online use of electronic noses and –tongues for water quality control. The preferred instrument to start with is the electronic tongue. The adaptation of the instrument and according method is less drastic then an adaptation for an electronic nose measurement. Online applications of interest in the chain from source to tap are:

• Intake protection

• Purification monitoring o Membrane o Slow sand bed o Ozonation o UV irradiation

• Distribution control By the choice of the next application it has to be taken into account that the strength of electronic nose and –tongue measurements is the measurement of differences. The instrumentation has to be trained to determine if a change in signal indicates a desirable or undesirable water quality change. The choice for the next application can be made in collaboration with the people from WA2, WA4 and WA5 based on the investigation they have performed.



6 References


1. EN 1420-1 Influence of materials on materials on water intended for human consumption. Organic materials. Pipes, fittings and their coatings used in piping systems. Odour and flavour assessment of waters. Part 1: test method.

2. EN 1622 Water Quality - Determination of the threshold odour number (TON) and threshold flavour number (TFN)

3. Dits, J.S., Jagt, H.van der, Smeenk, J.G.M.M. (1992). The odour/taste dilution factor an objective measurement for determining water quality.

4. Jagt, H. van der (1999). Evaluation of public health related quality aspects of materials, coming into contact with drinking water – odour and flavour tests in particular, Wat. Sci. Tech., Vol 40, No. 6, pp 239-244.

6.2 Electronic nose & Electronic tongue

5. Li, Zhenfeng; Wang, Ning; Vigneault, Clément (2006), “Electronic nose and electronic tongue in food production and processing“, Stewart Postharvest Review, Volume 2, Number 4, August 2006 , pp. 1-5(5), Publisher: Stewart Postharvest Solutions

6. Ak Deisingh, DC Stone, M Thompson (2004), “Applications of electronic noses and tongues in food analysis”, International Journal of Food Science and technology, 2004, 39, 587-604

7. Handbook of Machine Olfaction: Electronic Nose Technology, Hardcover - Wiley VCH (January. 2003) - 624 Pages , ISBN 3527303588 - ISBN-13 9783527303588

8. Roy, G. and McDevitt, J. T., "In vitro taste sensors: Technology and applications," Chemistry of Taste: Mechanisms, Behaviors, and Mimics, 2002, 825, 262-275 (Acs Symposium Series), An American Chemical Society Publication (août 2002) - 376 Pages, ISBN 0841237344 - ISBN-13 9780841237346

9. Instrumentation and Sensors for the Food Industry, Woodhead Publishing (sept. 2001) - 818 Pages, ISBN 1855735601 - ISBN-13 9781855735606

10. Electronic Noses and Olfaction 2000: Proceedings of the Seventh International Symposium on Olfaction and Electronic Noses, Brighton, Uk, July 2000, Institute of Physics Publishing (janv. 2001) - 310 Pages, ISBN 0750307641 - ISBN-13 9780750307642

11. Ronald T. Wojcik, Electronic Nose and Sensor Array Based Systems, Technomic Publishing Co (déc. 1999) - 348 Pages, ISBN 1566767601 - ISBN-13 9781566767606



6.3 Statistics

12. Jane Miller, James Miller, Statistics and Chemometrics for Analytical Chemistry, Prentice-Hall (mai 2005) - 268 Pages, ISBN 0131291920 - ISBN-13 9780131291928

13. Brian Everitt, Graham Dunn, Applied Multivariate Data Analysis, Hodder Arnold (mars 2001) - 352 Pages, ISBN 0340741228 - ISBN-13 9780340741221

6.4 Technical notes of Alpha MOS products

14. N-DT-01 Statistical data treatments 15. N-P-01 Electronic Nose technology overview 16. N-SAS-01 Metal oxide sensors 17. T-P-01 Electronic Tongue overview 18. T-SAS-01 Electrochemical sensors



7 Abbreviations

DFA Discriminant Factorial Analysis EN Electronic Nose EPDM Ethylene-Propylene-Diene-Monomer ET Electronic Tongue PCA Pricipal Component Analysis PLS Partial Least Squares NBR Acrylonitrile butadiene rubber SIMCA Soft Independent Modelling by Class Analogy

Documents

Odour & Flavour tests: Human panel and electronic … report compares results obtained with an electronic tongue, electronic nose and an odour and flavour panel for non-concentrated