Sensors and Actuators B 68 Ž2000. 53–57 www.elsevier.nlrlocatersensorb
Optimised sensor arrays with chromatographic preseparation: characterisation of alcoholic beverages I. Heberle a,) , A. Liebminger b, U. Weimar a , W. Gopel ¨ a a
Institute of Physical and Theoretical Chemistry, UniÕersity of Tubingen, Auf der Morgenstelle 8, 72076 Tubingen, Germany ¨ ¨ b Institute of Process Engineering, UniÕersity of Linz, Linz, Austria
Abstract High concentrations of alcohol increase the difficulties in identifying different alcoholic beverages with sensor arrays, because odour-relevant species usually occur at much lower concentrations. In this study, the characterisation of different beer brands was possible with a modular sensor system ŽMOSES II.. The latter was combined with a gas-chromatographic column thus enabling preseparation. The classification of beer brands was independent of the ethanol content. Fermentation by-products of the beer could also be detected and classified. In addition, different types of white wine and one red wine were investigated by the same method. q 2000 Elsevier Science S.A. All rights reserved. Keywords: Classification of beer and wine; Modular sensor system; Fermentation by-products; Chromatographic preseparation
1. Introduction As far as quality control is concerned, producers of beverages are interested in the question: Does our product reach the internal quality standards and is the taste the same? w1x. Analytical methods of measuring beer flavour, such as gas chromatography or human test panels are time-consuming, expensive and laborious and, in some cases, insensitive w2x. Therefore, the aim of this work was to develop an instrument enabling fast, cheap and reliable results. Over the past years, efforts have been made to use sensor arrays for the quality control of food and beverages w3,4x. In many applications, problems occurred due to the lack of sensitivity and selectivity of the sensors used. Poor selectivity results in the detection of components, which are present in high concentrations in the headspace but are of no relevance in relation to the food’s flavour w5x. Water and alcohol, for example, give a high sensor signal but their effect on flavour is limited. In our measurements, the Quartz Microbalance Sensors ŽQMB. in particular, showed to some extent cross-sensitivity to water. High concentrations of water and ethanol in a large variety of beverages have made classification difficult up to now. Therefore, in this work, we separated volatile sample components by
)
Corresponding author.
chromatographic preseparation and characterised the individual peaks by a subsequent sensor array test.
2. Experimental The modular sensor array set-up ŽMOSES II. w6x was extended by adding a preseparation unit between the headspace-sampler and the sensor array. The preseparation unit consists of a thermostatted gas chromatography column. Optimisation of the set-up included the choice of column material, temperature and pressure. Final measurements were carried out at a column temperature of 58C and a mass flow of 10 mlrmin ŽFig. 1.. New software was developed for the evaluation of the sensor data. First, the retention time was determined at the maximum of each individual peak. These individual sensor signal data served as input for the subsequent PCA analysis. In a first set of experiments, different beer samples were investigated. The samples were heated at 508C for 10 min in a headspace-sampler. The headspace gas was passed through the GC column to separate the major components. The individual compounds identified were ethanol, water and Aflavour-peaksB, and were detected with MOSES II. This sensor array consists of eight polymer-based QMB
0925-4005r00r$ - see front matter q 2000 Elsevier Science S.A. All rights reserved. PII: S 0 9 2 5 - 4 0 0 5 Ž 0 0 . 0 0 4 6 1 - 5
54
I. Heberle et al.r Sensors and Actuators B 68 (2000) 53–57
Fig. 1. Schematic experimental set-up.
and eight SnO 2-based Metal Oxide Sensors ŽMOX. in two individual measurement chambers. Typically, a PCA analysis was performed with all sensor signals and the results plotted in two dimensions as principal components PC1 against PC2 w7x. In a second set of measurements, the beer samples were doped with fermentation by-products, which influence the taste of the beers drastically and, in higher concentrations, lead to defective flavour and taste.
Fig. 3. Scoresplot of a variety of different mixtures Že.g., TrOrP means a mixture of toluene, n-octane and 1-propanol. measured with a regular MOSES II.
3. Results In testing the new system, the first aim was to separate a mixture of different organic volatiles Žsee Fig. 2. and to characterise these with MOSES II Žsee Fig. 3.. For this purpose, ethyl acetate, propan-1-ol, toluene and n-octane were mixed in concentrations of 50 ml each. All compounds could be separated from each other, which is a crucial requirement for their further analysis by MOSES II. In the second step, the four different chemicals were mixed in all possible combinations in order to carry out a PCA analysis. Therefore, the 15 samples obtained Žfour pure chemicals and 11 two- to four-component combinations. were measured. The scoresplot of the following PCA is shown ŽFig. 4.. The different samples are clearly separated by both methods. Compared to a normal MOSES II measurement,
Fig. 2. Time separation of a four-component mixture detected with the QMB module of the sensor-array MOSES II Žsignal responses of only three sensors Q1–Q3 are shown here for clarity..
the discrimination of volatiles is only slightly improved with the new system. However, the content of information is much higher Žto be seen in the quota of information in the PC 3.. To illustrate the increase in performance Žeach component provides an individual signal., a prediction of two different concentrations was made. Therefore, the samples were adjusted in concentrations of 2000 and 3000 ppm Žthree times each.. Two data measurement sets were used to generate a mathematical model in order to predict the concentrations of the third data set recorded ŽFig. 5.. The preseparated samples provide better results compared to the data obtained normally: The deviation of measured values is smaller and the error is minimised. For GCrMS measurements performed in parallel, the proportion of ethyl acetate was determined for up to 9.5% of the total headspace content. Thus, the preseparation is a suitable technique for the investigation of less volatile compounds. In the next step, eight different beer brands were analysed in the same arrangement.
Fig. 4. Scoresplot of a discrimination of a variety of different mixtures, measured with preseparation.
I. Heberle et al.r Sensors and Actuators B 68 (2000) 53–57
55
Fig. 7. Scoresplot of eight different beer brands; measured regularly with MOSES II.
Fig. 5. Prediction of the concentrations of the single-component ethyl acetate with and without a preseparation.
All beers, brewed by the same procedure Žby a Pilsen process., contain nearly the same amount of 5 vol.%
ethanol except of the brand AOettinger LeichtB with 2.8 vol.%. More than 100 beer samples have been measured in total. These measurements lead to a probability of greater than 80% in the correct classification of unknown samples. In Fig. 6 a typical result of a chromatographic separation is shown. The small peaks labelled A, B and C were used for the PCA analysis. In Fig. 7, the scoresplot of a PCA analysis is shown, which was obtained by using the original MOSES II system for beer measurements without preseparation. It is obvious that the influence of ethanol would be dominant in the sensor response. However, as the amount of ethanol in the different brands is similar, there is nearly no detectable separation. The scoresplot resulting from the PCA analysis of the Aflavour-peaksB A, B and C obtained after separation is shown in Fig. 8: The comparison of results with and without preseparation shows the drastic improvement in classification. The chemical identification of the flavour-peaks of this column is under current investigation with GCrMS.
Fig. 6. Chromatogram and magnified Aflavour-peaksB ŽA, B and C. of a typical beer sample Žthe signal responses of the eight MOX sensors are plotted as an example..
56
I. Heberle et al.r Sensors and Actuators B 68 (2000) 53–57
Fig. 8. Classification of eight different beer brands obtained by preseparation.
In a second series of measurements, regular beer samples were intentionally AdopedB with different fermentation by-products Žconcentrations in brackets indicate the values normally found in beer w8x.: Ø Ø Ø Ø
0.35 ppm dimethyl sulphide ŽDMS. Ž0.15 ppm., 5 ppm butan-2,3-dione Ž0.2–0.4 ppm., 5 ppm isopentyl-acetate Ž2–4 ppm., 50 ppm butan-1-ol Ž50 ppm..
The results with preseparation are shown in Fig. 9: All doped samples are clearly separated from each other and from the original sample. On the basis of the promising results obtained from measurements of different beer brands, we investigated different types of wine using the same method outlined above. Due to the fact that approximately more than 900 substances contribute to the flavour of wine w9x, our first aim was to discriminate four different types of vine Žwhite
Fig. 9. Scoresplot of a separation of beer samples differently AdopedB with fermentation by-products compared with an AundopedB beer.
Fig. 10. Scoresplot of four different white wines and one red wine ŽTrollinger..
wine. from several regions and one red wine. The following sorts of wine were available: Wurttemberger Kerner Ž1994., ¨ Ž1994., Gundelbacher Riesling ¨ Bourgogne Blanc Chardonnay Ž1996., Gruner ¨ Silvaner Ž1997., Haberschlachter Trollinger mit Lemberger Ž1996. Žred.. The installation of a capillary column between the transfer line of the headspace-sampler and the inlet of MOSES enabled partial separation of the Headspace components; as in the case of our beer measurements, the dominating ethanol peak could be separated from smaller ones; their identification still being under current investigation. The small Aflavour peaksB were used to perform the PCA in order to guarantee that the evaluation is independent of the amount of ethanol. The scoresplot of more than 70 samples is shown in Fig. 10. The small clusters demonstrate that the discrimination of different wines on the basis of Aflavour peaksB is
Fig. 11. Scoresplot of three different white wines without preseparation.
I. Heberle et al.r Sensors and Actuators B 68 (2000) 53–57
possible. For comparison, the scoresplot without preseparation is shown in Fig. 11: The percentage, given in brackets, represents the ethanol content of the individual wines. The reference only consists of distilled water and 10% ethanol. In theory, it is possible to distinguish between the different wines by taking the ethanol concentration into account, but it has to be mentioned that this discrimination is only caused by the larger range of different contents of alcohol. Compared with the beer samples where nearly all samples contain 5% ethanol, the scoresplot of wine forms a concentration cell from 10% to 12.5% ethanol, by volume.
4. Conclusions
Ø The combination of a gas-chromatographic system with a sensor-array shows improved separation of ethanol and water peaks from Aflavour-peaksB. Ø There is only a slight improvement in the detection of organic volatiles. Ø The preseparation method is a suitable technique to investigate less volatile compounds. Ø A drastic improvement in classification of beer and wine brands has been demonstrated. Ø The separation method is independent of ethanol concentration.
57
Ø Fermentation by-products of the beer are detectable and their classification is also possible
Acknowledgements We gratefully acknowledge Lennartz Electronic for providing us with the Modular Sensor System MOSES II.
References w1x H. Weyh, Zur Bestimmung der Geschmacksstabilitat ¨ von Bier, Brauwelt 8 Ž1986. 248–254. w2x J.W. Gardner et al., Electronic Noses, Principles and Applications, Oxford Univ. Press, 1999. w3x C. Di Natale et al., Electronic Nose and sensorial analysis: comparison of performances in selected cases, Sens. Actuators, B 50 Ž1998. 246–252. w4x T.C. Pearce et al., An electronic nose for the monitoring of the flavour of beer, Analyst 118 Ž1993. 371–377. w5x E. Landas, Off-flavours in beer, Brauwelt Int. Ž1991. 217–223. w6x J. Mitrovics et al., Modular sensor systems for gas sensing and odour monitoring: the MOSES concept, Acc. Chem. Res. 31 Ž5. Ž1998. 307–315, For details see www.lennartz-electronic.de. w7x W. Gopel, Chemical imaging: concepts and visions for electronic and ¨ bioelectronic noses, Sens. Actuators, B 52 Ž1998. 125–142. w8x L. Macher, Bier, Ullmann Vol. 81974, pp. 462–495, 4th edn. w9x G. Wurdig et al., Chemie des Weines, Verlag Eugen Ulmer, Stuttgart, ¨ 1989.