Article No.

fs980376

Lebensm.-Wiss. u.-Technol., 31, 305316 (1998)

Review Article

Electronic Noses and Their Application to Food

Emmanuelle Schaller, Jacques O. Bosset and Felix Escher E. Schaller, J. O. Bosset: Federal Dairy Research Station, CH-3097 Liebefeld-Bern (Switzerland) F. Escher: Swiss Federal Institute of Technology (ETH) Institute of Food Science, CH-8092 Zurich (Switzerland) (Received August 30, 1997; accepted in revised form March 19, 1998)

Electronic nose systems involve various types of electronic chemical gas sensors with partial specicity, as well as suitable statistical methods enabling the recognition of complex odours. As commercial instruments have become available, a substantial increase in research into the application of electronic noses in the evaluation of volatile compounds in food, cosmetic and other items of everyday life is observed. At present, the commercial gas sensor technologies comprise metal oxide semiconductors, metal oxide semiconductor eld effect transistors, organic conducting polymers, and piezoelectric crystal sensors. Further sensors based on breoptic, electrochemical and bi-metal principles are still in the developmental stage. Statistical analysis techniques range from simple graphical evaluation to multivariate analysis such as articial neural network and radial basis function. The introduction of electronic noses into the area of food is envisaged for quality control, process monitoring, freshness evaluation, shelf-life investigation and authenticity assessment. Considerable work has already been carried out on meat, grains, coffee, mushrooms, cheese, sugar, sh, beer and other beverages, as well as on the odour quality evaluation of food packaging material.

1998 Academic Press Keywords: electronic nose; articial nose; gas sensor analysis; odour analysis

Introduction The perception of volatile compounds by the human nose is of great importance in evaluating the quality of foods, cosmetics and numerous other items of everyday life. Therefore, it is not surprising that repeated efforts have been made over the years to introduce instruments operating on a similar principle as the human nose. These systems would in most cases not replace but complete conventional analyses of volatile compounds by sensory methods and by traditional analytical techniques. The concept of an articial nose system was proposed in 1982 at the University of Warwick by Persaud and Dodd (1). In principle, such systems have to rely on gas sensors, which were rst developed more than 30 years ago (2, 3). At the beginning of the 1990s the term articial or electronic nose appeared, and several commercial instruments became available. More extended research began, and applications, especially in the food industry, could be tested. Gardner and Bartlett (3) dened the electronic nose as an instrument, which comprises an array of electronic chemical sensors with partial specicity and an appropriate pattern-recognition system, capable of recognising simple or complex odours. This seems very far from the human nose, and according to Mielle et al. (4, 5) this analytical system is obviously electronic but not nose.

In fact the only aspect in common with our odoursensing organ is its function. Like the mammalian nose, it detects gases by means of sensors which send signals to a recognition organ, that is to the brain or to a computer. The operating principle, the number of sensors as well as the sensitivity and selectivity are, however, very different (6). This is why some scientists prefer to call this instrument by other names, for example, avour sensor, aroma sensor (7), odoursensing system (3) or multi-sensor array technology (8). Electronic nose systems comprise a more or less sophisticated hardware, with sensors, electronics, pumps, air conditioner, ow controller, etc., and software for hardware monitoring, data pre-processing, statistical analysis, etc. As analytical instruments, these systems must be designed for long-term usage with high repeatability (the ability to obtain the same pattern for a sample on the same array over short intervals of time) and reproducibility (the ability of different sensor batches or different instruments to produce the same pattern for the same sample). As sensor technology plays a crucial role in reaching the desired analytical qualities (9), the following review concentrates on discussing the present status of sensor technology and statistical data analysis. Particular reference is then made to applications of electronic nose systems to food.
1998 Academic Press

0023-6438/98/040305 + 12 $30.00/0

305

lwt/vol. 31 (1998) No. 4

Sensors All types of sensors exhibit interactions with the gas to be measured so that a series of physical and/or chemical interactions occurs when volatile compounds ow over the sensor. A dynamic equilibrium develops as volatile compounds are constantly being adsorbed and desorbed at the sensor surface (8). The ideal sensors to be integrated in an electronic nose should full the following criteria (2, 1013): high sensitivity towards chemical compounds, that is, similar to that of the human nose (down to 1012 g/mL); low sensitivity towards humidity and temperature; medium selectivity, they must respond to different compounds present in the headspace of the sample; high stability; high reproducibility and reliability; short reaction and recovery time; robust and durable; easy calibration; easily processable data output; small dimensions. As the sensors are designed for industrial purposes, especially as on-line systems, a minimal working temperature with low power consumption, a high safety level, and low manufacturing costs present valuable advantages. Most manufacturers are looking for highly selective sensors. In the case of an electronic nose, every compound present in the gaseous phase should be detected by at least one sensor. If a new compound is added to a mixture, at least one sensor must detect this addition. The use of too many sensors leads to an over complex system with a large amount of unnecessary data. Various kinds of gas sensors are available, but only four technologies are currently used in commercialized electronic noses (Fig. 1): metal oxide semiconductors (MOS); metal oxide semiconductor eld effect transistors (MOSFET); conducting organic polymers (CP); piezoelectric crystals (bulk acoustic wave = BAW). Other such as breoptic (1418), electrochemical (13, 19) and bi-metal sensors (20) are still in the developmental stage and may be integrated in the next generation of electronic noses. Such sensors can be divided into two main classes: hot (MOS, MOSFET) and cold (CP, SAW, BAW). The former operate at high temperatures and are considered to be less sensitive to moisture with less carryover from one measurement to another. Therefore, they should offer the best ratio of drift and lifetime to sensitivity (8).

oxide, tin dioxide, titanium dioxide or iron (III) oxide) which responds to oxidising compounds, or of the p-type (mainly nickel oxide or cobalt oxide) which responds to reducing compounds (7, 22, 23). Thermal or photolytic excitation of an n-type (n = negative electron) or donor semiconductor results in an excess of electrons in its conduction band which increases the reactivity with oxidising molecules. An excited p-type (p = positive hole) or acceptor semiconductor shows an electron deciency in its valence band which promotes reactions with reducing compounds (22, 23). The lm deposition technique further divides each sensor type into thin (61000 nm) or thick (10300 m) lm MOS sensors. Film deposition includes physical or chemical vapour deposition, evaporation, or spraying for thin lms, and screen printing or painting for thick lms (10). The thin lm devices offer a faster response and signicantly higher sensitivities, but are much more difcult to manufacture in terms of reproducibility. Therefore, commercially available MOS sensors are often based on thick lm technologies. Due to the high operating temperature (typically 200650 C), the organic volatiles transferred to the
MOS Round Semiconductor Substrate Electrode Flat Semiconductor Substrate Electrode

CP Electrode Flat Polymer Substrate

BAW

Coating Crystal Electrode

SAW Coating Crystal

Electrodes

Metal oxide semiconductor sensors Metal oxide semiconductor gas sensors (MOS) were rst used commercially in the 1960s as household gas alarms in Japan under the names of Taguchi (the inventor) or Figaro (the companys name). These sensors, also called oxide or ceramic gas sensors, rely on changes of conductivity induced by the adsorption of gases and subsequent surface reactions (21). Commercially available MOS sensors consist of a ceramic substrate (round or at) heated by wire and coated by a metal oxide semiconducting lm. The metal oxide coating may be either of the n-type (mainly zinc

MOSFET

Electrodes Catalytic metal Insulator SiO2 Semiconductor Si

Fig. 1 Schematic diagrams of 5 different kinds of sensors. Caption: MOS = Metal oxide semiconductor; CP = Conducting polymer; BAW = Bulk acoustic wave; SAW = Surface acoustic wave; MOSFET = Metal oxide semiconductor eld effect transistor

306

lwt/vol. 31 (1998) No. 4

sensors are totally combusted to carbon dioxide and water on the surface of the metal oxide, leading to a change in the resistance (24). The two-step reaction mechanism is based on an oxygen exchange between the volatile molecules and the metal lm. First, oxygen from the carrier gas is adsorbed on the surface, and inside the metal coating, by lling the defects in the polycrystalline lattice structure of the metal. The electrons are attracted by the loaded oxygen, and the sensor conductivity decreases. Second, gas/surface interactions occur between the adsorbed volatile molecules and the metal lm following two mechanisms: (i) at relatively low temperatures, charges are transferred between the volatiles and the adsorbed oxygen; (ii) at higher temperatures, the lattice oxygen reacts with the adsorbed volatile compounds, changing the defects concentration in the crystal. The distinction between both mechanisms is not evident, and both may occur simultaneously depending on whether the electrophilic properties of chemisorbed oxygen (O2ads) and (Oads), or the nucleophilic properties of lattice oxygen O2 are involved (10, 25). To shift the selectivity of a metal oxide lm towards different chemical compounds the lm is doped with noble catalytic metals (e.g. platinum or paladium), or the working temperature is changed within the range of 50400 C (109). Although the selectivity is also greatly inuenced by the particle size of the polycrystalline semiconductor, the MOS sensors are usually less selective than other technologies such as CP, BAW, SAW or MOSFET (7, 24). MOS sensors are extremely sensitive to ethanol, which blinds them to any other volatile compound of interest. Furthermore, they may be poisoned by irreversible binding by compounds such as those of sulphur or weak acids. Finally, their high operating temperature means that they are inappropriate in environments containing large amounts of potentially ammable chemicals (7).

When a voltage is passed across the electrodes, a current passes through the conducting polymer. The addition of volatile compounds to the surface alters the electron ow in the system and therefore the resistance of the sensor (8). The volatiles may interact at least with (i) the polymer itself, (ii) the counterion, or (iii) the solvent (11). Therefore, good selectivity in the CP sensors may be achieved by altering one of these parameters or the electrical growth of the polymer coating. In general, these sensors show good sensitivities, especially for polar compounds. However, their low operating temperature ( < 50 C) makes them extremely sensitive to moisture (8). Although such sensors are resistant to poisoning (28), they have a lifetime of only about 918 months. This short life may be due to the oxidation of the polymer, or to exposure of the sensor to different chemicals that may develop contact resistances between the polymer and the electrodes. Unlike MOS sensors, the CP sensors are not yet widely marketed, and laboratory-scale manufacturing renders them expensive. The difculty of producing good batchto-batch reproducibility and a pronounced drift of the response are their main disadvantages (7).

Conducting organic polymer sensors Conducting organic polymer (CP) sensors have been under development for approximately 10 years (11) and, like MOS sensors, rely on changes of resistance by the adsorption of gas. However, their operating mechanism is more complex and not yet well understood. These sensors comprise a substrate (e.g. bre-glass or silicon), a pair of gold-plated electrodes, and a conducting organic polymer such as polypyrrole, polyaniline or polythiophene as a sensing element. The polymer lm is deposed by electrochemical deposition between both electrodes previously xed to the substrate (7, 26). As the conducting polymer is grown out of a solution, the deposited lm contains cation sites balanced by anions from the electrolyte and the solvent residue (11, 24, 27). The cation sites probably consist of polarons or bipolarons which are small regions of positive charge in the polymer chain providing mobile holes for electron transport.

Piezoelectric crystal sensors Piezoelectric sensors are based on a change of mass, which may be measured as a change in resonance frequency. These sensors are made of tiny discs, usually quartz, lithium niobate (LiNbO3) or lithium tantalate (LiTaO3), coated with materials such as chromatographic stationary phases, lipids or any non-volatile compounds that are chemically and thermally stable (2931). When an alternating electrical potential is applied at room temperature, the crystal vibrates at a very stable frequency, dened by its mechanical properties. Upon exposure to a vapour, the coating adsorbs certain molecules, which increases the mass of the sensing layer and hence decreases the resonance frequency of the crystal. This change may be monitored and related to the volatile present (11). The crystals may be made to vibrate in a bulk acoustic wave (BAW) or in a surface acoustic wave (SAW) mode by selecting the appropriate combination of crystal cut and type of electrode conguration (24). BAW and SAW sensors differ in their structure. BAW are 3-dimensional waves travelling through the crystal, while SAW are 2-dimensional waves (Rayleigh, Love and Bluestein-Gulyaev) that propagate along the surface of the crystal at a depth of approximately one wave length (29, 30). The principle of BAW sensors was introduced by King in 1964 with his Piezoelectric Sorption Detector (32). These devices are also called quartz crystal microbalance (QCM or QMB) because, similar to a balance, their responses change in proportion to the amount of mass adsorbed. BAW sensors vibrate with a frequency

307

lwt/vol. 31 (1998) No. 4

of 1030 MHz. Their thin coating (1 m10 nm) is deposited by spin coating, airbrushing or inkjet printing (7, 24, 30). More than 10 years later, in 1979, the rst gas sensor based on a SAW oscillator was introduced by Wohltjen and Dessy (3335). However, the use of such a sensor was rst reported by Martin et al. in the 1980s (3638). These devices operate at a higher frequency than BAW sensors, i.e. 100 MHz1 GHz. The manufacturing technique includes photolithography and airbrushing, and is fully compatible with planar integrated circuits fabrication, especially planar silicon technology. This enables SAW structures and conditioning circuits to be incorporated on the same silicon substrate, resulting in robust and inexpensive SAW sensors (39). Since piezoelectric sensors may be coated with an unlimited number of materials, they present the best selectivity (7, 11). However, the coating technology is not yet well controlled, which induces poor batchto-batch reproducibility (7, 24). SAW sensors, though limited by the noise caused by their high operating frequency, are more sensitive than BAW sensors. However, both sensors require a higher concentration of volatiles to elicit response levels comparable to other sensor types (7, 11). The difculty of integrating BAW and SAW sensors into an electronic nose resides in the more complex electronics and their high sensitivity to disturbances such as temperature and humidity uctuations (7, 24).

responds almost exclusively to molecules that dissociate hydrogen on the catalytic metal surface. It is implicitly assumed that the insulator is not exposed to the ambient molecules. The dissociated hydrogen atoms diffuse within microseconds through the metal causing a dipole layer at the metalinsulator interface, leading to a potential change in the transistor. Detection of molecules such as ammonia or carbon monoxide is not possible with such a layer since no hydrogen atoms are released (44). However, it transpired that the latter compounds respond well when the metal gate is thinned. The most probable mechanism is explained by the contribution of three voltage shifts: (i) at the surface of the metal (due to adsorption and chemical reactions); (ii) at the metal insulator interface (due to hydrogen diffusion through the metal); and (iii) at the insulator surface (due to reactions on the oxide metal surface of polar molecules or charges, which directly react with the oxide metal, or diffuse out on this surface) (41, 42, 44). The selectivity and sensitivity of MOSFET sensors may be inuenced by the operating temperature (50200 C), the composition of the metal gate, and the microstructure of the catalytic metal (41, 42). MOSFET sensors, like MOS sensors, have a relatively low sensitivity to moisture and are thought to be very robust. However, high levels of manufacturing expertise are necessary to achieve good quality and reproducibility.

Metal oxide semiconductor eld-effect transistor sensors The hydrogen sensitive PdMOS (palladium metal oxide semiconductor) device was developed in 1973 by a group of Swedish researchers and reported 2 years later by Lundstrom et al. (4042). The metal oxide semi conductor eld-effect transistor (MOSFET) sensors rely on a change of electrostatic potential. A MOSFET sensor comprises three layers, a silicon semiconductor, a silicon oxide insulator and a catalytic metal (usually palladium, platinum, iridium or rhodium), also called the gate. A normal transistor operates by means of three contacts, two allow the current in (source) and out (drain), and the third acts as the gate contact that regulates the current through the transistor. In the MOSFET transistor, the gate and drain contacts are shortcut, giving a diode mode transistor with convenient electronics for operation, characterized by an IV-curve. The applied voltage on the gate and drain contact creates an electric eld, which inuences the conductivity of the transistor. When polar compounds interact with this metal gate, the electric eld, and thus the current owing through the sensor, are modied. The recorded response corresponds to the change of voltage necessary to keep a constant preset drain current (41, 43). As in the coating of MOS sensors, the gate structure of a MOSFET sensor is either a thick, dense metal lm (100200 nm) or a thin, porous metal lm (620 nm). The thick, continuous metal gate

Statistical analysis techniques The raw responses generated by the sensors are analysed using various statistical methods (11, 30, 45) (Table 1). Commercially available techniques fall into three main categories: Graphical analyses: bar chart, prole, polar and offset polar plots. Multivariate analyses: principal components analysis (PCA), canonical discriminant analysis (CDA), feature weighting (FW) and cluster analysis (CA). Network analyses: articial neural network (ANN) and radial basis function (RBF). The choice of method depends on available data and the type of result that is required.

Table 1

Main statistical methods used in electronic noses PARC method Linear Supervised yes yes yes yes no no no no yes yes yes yes

Principal components analysis (PCA) Cluster analysis (CA) Canonical discriminant analysis (CDA) Feature Weighting (FW) Articial neural network (ANN) Radial basis function (RBF) PARC = Pattern recognition

308

lwt/vol. 31 (1998) No. 4

Graphical analysis Graphical analysis is the simplest form of data treatment that may be used with an electronic nose. This option is suitable when visually comparing samples to a single specied reference. However, when several references are used, analysis becomes more complicated and an alternative approach may be necessary (12).

Multivariate analysis Multivariate data analysis generally involves data reduction, it reduces high dimensionality in a multivariate problem where variables are partly correlated (e.g. sensors with overlapping sensitivities), allowing the information to be displayed in a smaller dimension (typically two or three) (45, 46). There are many multivariate techniques to choose from. A subset called pattern recognition (PARC) techniques are routinely used in electronic nose data analysis.

need more parameters, since some of them are used to describe the shape of the non-linearity (more input data than linear models). The main advantage of such a method is exibility, i.e. the ability to adjust to more complex data variations. However, caution is necessary when choosing model exibility; this can be achieved by selecting the number of parameters. If too many parameters are taken into account, the calculated model will be over-exible, tting to all relevant data variations and unwanted sensor noise. The best method to avoid an over-tted model is to use training data to build a non-linear model, and validation data to test this model (cross-validation) (30).

Supervised/unsupervised The PARC methods can be divided into unsupervised or untrained techniques, and supervised or trained techniques. Unsupervised learning methods are generally used in exploratory data analysis because they attempt to identify a gas mixture without prior information on the nature of the samples. These techniques, which include PCA, CA and multi-dimensional scaling, are useful when no example of different sample groups is available, or when hidden relationships between samples or variables are suspected (3, 45). Conversely, supervised learning techniques classify an odour by developing a mathematical model relating training data, i.e. samples with known properties, to a set of given descriptors. Test samples are then evaluated against a knowledge base and predicted class membership is deduced. These methods enable the system to reduce parameters other than volatile, for example temperature and humidity, and train a system to look only at particular combinations of sensors to measure a given odour. Generally, electronic nose data are best processed using trained techniques such as CDA, FW, ANN or RBF (3).

Articial neural network A neural network consists of a set of interconnected processing algorithms functioning in parallel. On a very simplied and abstract level, ANN is based on the cognitive process of the human brain. Mathematical functions, or neurones, link together to build a network which mimics the human nervous system (48). A weight is randomly assigned to each neurone and then adjusted by means of an iterative or learning process, for example, error back-propagation, until the desired outputs are obtained. The resultant set of weights and functions is then saved as a neural network. Like CDA, ANN is a supervised method and so needs a minimum of known data to correctly train the system. If the number of available data is not sufcient an erratic result will be obtained (12). Unlike other PARC methods, a neural network is a dynamic, self-adapting system that can modify its response to external forces using previous experience, offering a more exible and, due to the parallelism, faster method of analysis. In addition, it may more closely mimic mammalian neurone processing of odour stimuli (48, 49). A well trained ANN is very efcient in comparing unknown samples to a number of known references (11).

Applications to Food Electronic nose systems were designed to be used with numerous products, such as cars, food, packaging, cosmetics, etc. for an equally broad range of applications: quality control of raw and manufactured products; process, freshness and maturity monitoring; shelflife investigations; authenticity assessments of premium products; classication of scents and perfumes; microbial pathogen detection (50, 51); and environmental studies (7). At present, only a few basic studies have been published, because most topics have been examined either by manufacturers in preliminary feasibility studies or by industrial research units. Most publications deal with the application to foods (Table 2). Published studies are also available covering tobacco (49, 52), cosmetics (5356) and environment (48, 5763).

Linear/non-linear The above multivariate analyses are all linear PARC methods where a model is calculated using linear combinations of input data (30). Most sensors have a non-linear response versus concentration; however, these techniques work well if a low concentration of volatiles ensures an approximately linear response. In addition, the use of pre-processing algorithms, such as averaging, linearization or normalization, improve the performance of these analytical techniques (45, 47, 48). When high concentrations of volatile are measured, a non-linear PARC technique, such as ANN or RBF, would be more appropriate. Non-linear models usually

309

lwt/vol. 31 (1998) No. 4

Meat Within food, the majority of publications refer to meat products. Most of the studies were carried out by Berdague at the Station de Recherches sur la Viande at INRA Theix, Clermont-Ferrand (France). Berdagues studies used instruments based on MOS sensors, starting in 1993 with the Alabaster-UV, one of the rst commercially available instruments. This system consists of a stainless steel measurement chamber containing one semiconductor gas sensor, a UV-lamp, and air inputs and outputs connected to a fan. The signal obtained is shown as desorption curves in Alabaster units. The rst study was carried out in collaboration with Talou of the Ecole Nationale Superieure de Chimie de Toulouse (France). It was shown that this simple instrument could differentiate maxima in odour perception resulting from the maturation of dry non-

spiced sausages and rapidly detect sex-linked differences in meat product composition (64). This research group also initiated a study into the discrimination of micro-organisms using a multi-sensor MOS instrument (65). Analysing a headspace generated by bacteria grown on agar medium, the FOX 2000 correctly classied 90.5% of the analysed bacteria species (66). Bacterial strains commonly used for curing, or pathogenic strains sometimes found in meat products were also detected and correctly classied. Six types of French dry-sausages and 16 Iberian hams, both with and without off-avour were also correctly identied (6770). Another study, conducted over 3 days, demonstrated that non-controlled ambient air, which simulated an on-line quality control, could be used in the rapid discrimination of food products. Research was carried out on sausages and the data obtained subjected

Table 2 Meat

Application of different electronic nose systems to different foods Instrumental system Alabaster-UV FOX 2000 FOX 3000 NST 3210 AromaScan CPS NST 3210 FOX 3000 FOX 4000 FOX 3000 Laboratory-made Laboratory-made AromaScan Laboratory-made Laboratory-made AromaScan A20S ? FOX 2000 Laboratory-made AromaScan A20S QMB6 AromaScan A20S FOX 3000 FOX 2000 FOX 4000 FOX 3000 Laboratory-made NST 3210 Laboratory-made AromaScan FOX 2000 Laboratory-made FOX 2000 NST 3210 AromaScan A32S Sensor technology 1 MOS, 1 UV 6 MOS 12 MOS 10 MOSFET, 4 MOS, 1 IR ? CP 4 amperometric sensors 10 MOSFET, 4 MOS, 1 IR 12 MOS 18 MOS 12 MOS 12 MOS 6 MOS 32 CP 7 MOS 12 CP 20 CP ? CP 6 MOS 5 MOS 20 CP 6 BAW 20 CP 12 MOS 6 MOS 18 MOS 12 MOS 26 MOS 10 MOSFET, 4 MOS, 1 IR 2 MOS 32 CP 6 MOS 5 BAW 6 MOS 10 MOSFET, 4 MOS, 1 IR 32 CP Ref. 64 6572 73 74 75 76 77, 78 79 80 81 2, 8183 46 84 84 87 88 89 90, 91 92 92 93 28, 96 94 95 97 98 99 100 101 102 82 60 103, 104 105, 106 107

Application

Grains

Coffee

Beer

Mushrooms Cheese

Sugar Fish Blueberry Orange juice Cola Alcoholic beverages Packaging

MOS = metal oxide semiconductor; CP = conducting polymer; BAW = bulk acoustic wave; SAW = surface acoustic wave; MOSFET= metal oxide semiconductor eld effect transistor. Suppliers: AromaScan = AromaScan plc, UK-Crewe; FOX = Alpha MOS, F-Toulouse; NST 3210=Nordic Sensor Technologies, S-Linkping; QMB6 = HKR, D-Mnich.

310

lwt/vol. 31 (1998) No. 4

to modelling by the Gompertz sigmoidal function, and discriminant analysis with backward variable selection. In both cases satisfactory results were achieved, with 100% and 97% product recognition respectively (71, 72). Another study using a 12-MOS sensor FOX 3000 investigated boar taint and the contamination of meat products. It was concluded that the obtained results were promising; however optimization of the method, for example by adding extra acoustic wave sensors, was necessary (73). Winquist et al. (74) also studied the performance of an electronic nose with meat products. Using an NST 3210 Emission Analyser, samples of ground beef and pork that had been stored in a refrigerator between analysis were studied using ANN and an algorithm based on an abductory induction mechanism. Both types of pattern recognition software could predict the type of meat, but storage time was more problematic. Neural networks based on unsupervised training methods could be used for a storage time prediction of ground beef. Conversely, Shiers et al. (75) did not succeed in monitoring the spoilage of beef mince using a CP system. This failure was explained by a high sensitivity of CP sensors to moisture and a low sensitivity to the small quantity of volatile species generated during meat spoilage.

determined. In wheat, ANN predictions correlated well with measured ergosterol and with colony forming units of fungi and lactic acid bacteria. The Swedish Farmers Supply & Crop Marketing Association studied 80 wheat and barley samples using a 12-MOS-sensor FOX 3000 (79). The aim was to replace sensory evaluation for off-odours in grains with instrumental analysis. With the use of ANN samples offodours could be discriminated at more than 80%. Nevertheless, the conclusion was that the electronic nose needs further development and comparison with other methods before it may really be used as an alternative to sensory panel.

Grains The use of gas sensors in the classication of grains was rst reported by Stetter et al. 1993 (76). Using chemical parameter spectrometry (CPS), 83% of wheat samples could be correctly classied by the quality grades good, sour and insect infected. With CPS, a sample of dilute vapour is pyrolyzed on a heated catalyst of platinum or rhodium, then passed through four amperometric gas sensors of different selectivities. Each sample was measured with the four sensors at four different catalyst temperatures, generating 16 signals. The signals were treated with K-nearest-neighbour (KNN) analysis and ANN. Comparison of both pattern recognition systems revealed that after compensation for instrumental parameters, ANN was much more resistant to random error, and drift in sensor signal magnitudes. ANN therefore proved more adaptable to events occurring in real analytical situations, such as the failure of an instrument component. A group of Swedish researchers classied a total of 235 samples of wheat, barley and oats (77). The NST 3210 Emission Analyser, correctly identied 75% of the samples using a four-class system of descriptions by inspectors (mouldy/musty, acid/sour, burnt and normal), and 90% when using a two-class system (good and bad). These values exceeded corresponding percentages of agreement between two grain inspectors classifying the grain. Using the same combination of sensors to identify various off-odours in oats these results were conrmed to a high degree of accuracy (78). The percentage of mouldy barley or rye grains in mixtures with fresh grains could also be

Coffee Roasted coffee is considered to exhibit one of the most complicated food aroma compositions with more than 600 components; it has thus been chosen by the Alpha MOS company as a case study to validate the FOX 4000 instrument (80). Two Arabica coffees, one Robusta coffee, and mixtures of both were analysed over a period of approximately 6 months. Three instruments were used with pure products as a calibration material. Favourable results were obtained on all three systems over the entire 6-month period. Without calibration, the discrimination power decreased after one month. Gardners studies (2, 8183) used a laboratory-manufactured instrument based on 12 MOS sensors to classify three commercial coffees consisting of two different blends and two roasts. All three coffees could be identied, with the electronic nose being more sensitive to different roasts than to blends. The authors also concluded that improvements in experimental procedures, equipment and design, as well as advancement in the selectivity and stability of the sensors were necessary to obtain increased reproducibility and consistency of the system. In a static system, Brazilian and Colombian coffees were more accurately identied with the use of 12 CP sensors than with 12 MOS (2). Fukunaga et al. (46) conducted a study on different coffee blends using a 6-MOS-sensor laboratory-manufactured instrument. They also found a high correlation between roasting level and sensor outputs. By comparing two methods of analysis, it was illustrated that coffee with a medicine-like odour could be detected more easily with an electronic nose than with gas chromatography/mass-spectrometry (GC/MS) analysis. However, they concluded that other factors, additional to the medicine-like smell, may possibly contribute to the difference between tainted and non-tainted coffees. Delarue et al. (84) evaluated the ability of MOS and CP sensors to identify instant coffee powder coated with a avoured oil. The AromaScan A32/50S was not able to differentiate between coated and non-coated powder; this was attributed to the high sensitivity of the CP sensors to free water and carbon dioxide. A 6-MOSsensor laboratory-manufactured instrument was more successful, with discrimination between the samples being achieved. However, this system was found to be

311

lwt/vol. 31 (1998) No. 4

unsuitable for in situ food quality monitoring where substantial variation in oxygen content occurred during shelf-life. In addition to the classications of odours with electronic noses, studies were carried out in Japan discriminating between coffee tastes with an electronic tongue using lipid membranes taste sensors (85, 86).

groups. One mushroom species was clearly discriminated from the others. The CP could only differentiate between two mushrooms species. The study does not state whether this difference is due to the sensor technology or to the sampling method, i.e. static for the MOS system and dynamic with the CP system.

Beer The rst study on beer products was reported by Pearce et al. (87) in 1993. A laboratory-manufactured system, based on 12 CP was tested with three similar commercial beers, two lagers and one ale. The three brands were easily identied. An articially tainted lager could also be successfully differentiated from the same untainted beer. The intra-batch variation of one lager was examined, but no signicant difference could be detected. An English brewery studied the possibility of integrating the AromaScan A20S into their production process (88). The study investigated various beers (four lager and ve ale brands) and their raw materials (hop, leaf and malt). Signal variations produced by the introduction of diacetyl, phenolic and metallic off-taints to a lager beer were also included. The data obtained were treated with multi-dimensional analysis. A differentiation was shown between the two beer brands but with several overlaps within one brand. The diacetyl and metallic off-taints were found to lower the signal intensity, while the phenolic off-taint increased it. Measurements conducted on the raw materials indicated both clear differences and overlaps within one group of material. It was concluded that this analytical method could prove promising in the brewery industry, but further work is needed on sampling techniques, and on hardware and software. These results concur with those of Bailey et al. (89) who discriminated between beers adulterated with traces of diacetyl or dimethylsulde (200 g/kg) using CP sensors. Using a FOX 2000, Zimmermann and Leclercq (90, 91) studied 50 samples of two types of malt. They concentrated on the distinction between malt type, and the identication of ageing and improperly manufactured batches. From this feasibility study it was concluded that the electronic nose could be used as a tool to both improve product quality and develop new beer products.

Cheese Three investigations were conducted to discriminate between various cheese types. Two, using a QMB6 system and a FOX 3000 system, differentiated between six Swiss cheeses, Appenzeller, Emmental, Gruyere, ` Tete de Moine, Tomme Vaudoise and Tilsiter (93, 94). The third study, using an AromaScan A20S, was made with CP sensors on Gorgonzola and Cottage Cheese, and on Parmigiano-Reggiano, Gouda and Emmental, respectively (28). Within the two groups, the different cheese types could be easily distinguished. Some problems occurred with the repeatability of the results; this may be related to the product itself (different pieces of the same sample may have slightly different odours), the sampling technique (changes in temperature or duration may change the odour) or the air moisture content (the system used ambient air). Therefore, the analytical system was found to be useful as a research tool, but further development is necessary for use in routine tests. Using a GC and the FOX 2000 instrument on the American Swiss Cheese, Harper et al. (95) concluded that only ve lower molecular weight fatty acids were important to the aroma of this type of cheese, i.e. acetic, propionic, butyric, isovaleric and caproic. The English Cheddar cheese was investigated using an AromaScan A20S in mild, medium and mature states. A recognition rate of 80% was achieved, but only if a close control of the equilibration temperature was maintained (96).

Mushrooms In 1995, the Agro-industrial Chemistry Laboratory at the National Polytechnic Institute of Toulouse published preliminary results on the differentiation between nine wild species of freeze-dried mushrooms using three different instruments: a GC equipped with a snifng-port, a home-made system with 5-MOS sensors, and the AromaScan A20S (92). The GC/snifng analysis differentiated between four groups. Data from the MOS system, as analysed with PCA also classied the samples into the same four

Sugar A study on eight samples of beet and cane sugar with and without off-odour compared the FOX 4000 with the traditional GC/MS and sensory panel analyses (97). The electronic nose recognised six out of eight samples, which was better than the sensory panel. A static headspace with an autosampler showed an improved correlation with the GC/MS results over a manual dynamic headspace. Four commercial sugars of various qualities were classied using a sensory panel and then analysed with a 12-MOS-sensor FOX 3000 (98). It was concluded that the electronic nose could be useful to classify the quality of raw material and to complete technical specications. However, some improvement on the sampling method and on the equipment is necessary, such as a reduction in the dead volumes of the tubing and sensor chambers, and temperature control of the injection port and tubing.

312

lwt/vol. 31 (1998) No. 4

Fish Spoilage in three different sh species (haddock, cod and redsh) was followed using between two and six MOS sensors (99). The samples were kept at room temperature or in ice, and the results compared with sensory analyses. The results were promising enough to prompt further investigation into the use of electronic sensors in sh freshness assessment. A system based on MOSFET sensors was successfully trained to predict the age of cod let (100). The predictions were within the interval of 33.347.3 h.

totally between the ve classes of papers. The responses of nine different paperboards were correlated to the description of a sensory panel (106). Not all cases illustrated a good correlation with a selected number of sensors. Analysis using CP was successful in detecting offodours in canisters of pharmaceutical inhalant (107). Sensor responses suggested that specic elastomeric components were responsible for the off-odour.

Conclusion Blueberry A laboratory-manufactured instrument consisting of two MOS sensors detected damaged blueberries in packaging at a 5% level (101). It also successfully distinguished between four out of ve ripeness classes. Adding data such as pH, colour, berry rmness and titratable acidity, ten cultivars were differentiated. An electronic nose is a system originally created to mimic the function of an animal nose. However, this analytical instrument is more a multi-sensor array technology than a real nose. Whatever the sensor technology, it is still far from the sensitivity and selectivity of a mammalian nose. Therefore, its aim is not to totally replace either the human nose or other analytical methods. A sensory panel is necessary to dene the desired product quality which can then be used to train the system. Traditional analytical methods such as GC-analysis will always be needed to determine qualitatively or/and quantitatively why one food sample differs from others. The electronic nose can only perform quick yes or no tests in comparison to other products. It could occasionally replace sensory analysis and even perform better than a sensory panel in routine work, or in cases where non-odorous or irritant gases need to be detected. Therefore, an electronic nose can be regarded as an interesting tool for a quick quality test in various food applications. However, before it can be treated as a completely reliable, industrial instrument, much improvement is still needed, such as a reduction of the dead volume, development of calibration methods, etc. As current knowledge does not allow the replacement of the human nose, constructors tend to compensate by integrating several sensor technologies into one instrument. However, one single instrument to be used in every possible application would be over-complicated due to the large number of sensors and time consuming statistical analysis. The trend is to create a system for one specic application. This means that a compact and portable instrument would be desirable (108).

Beverages Orange juice. The AromaScan A32S was used to discriminate between freshly pressed orange juice at different ageing stages (102). A comparison of data obtained with the electronic nose, sensory panel and GC analysis revealed similar results. All three methods could differentiate between the three ageing groups, early, mid and late. However, the classication of the ve samples in the groups depended on the method used. Cola. Using a FOX 2000, two brands of cola and diet cola were compared. After the three samples were degassed, they could be discriminated using a cluster analysis (82). Alcoholic beverages. A bioelectronic nose consisting of 5-QCM sensors coated with bullfrog receptor proteins differentiated between sorghum liquors from different sources and brewing processes (60). Packaging. A FOX 2000 using multiple discriminant analysis could discriminate between cakes wrapped in various qualities of lm, although a sensory panel did not nd differences (103). Based on off-odours dened by a sensory panel, analyses obtained from GC, sensory panel and MOS electronic nose were correlated (104). A sample of polypropylene packaging, with off-odour could easily be distinguished from an undefective one. Five cardboard papers from commercial manufacturers were examined with the aid of an electronic nose comprising, ten MOSFET, four MOS and one infrared carbon dioxide sensor (105). Four of the papers originated from the same source, but from two different production lines, while the fth paper came from another manufacturer. The authors found that only four out of the 15 sensors were necessary to discriminate

References
1 PERSAUD, K. AND DODD, G. H. Analysis of discrimination mechanisms in the mammalian olfactory system using a model nose. Nature, 299, 352355 (1982) 2 BARTLETT, P. N., BLAIR, N. AND GARDNER, J. W. Electronic nose. Principles, applications and outlook. ASIC, 15e Colloque, Montpellier, pp. 478486 (1993) 3 GARDNER, J. W. AND BARTLETT, P. N. A Brief History of Electronic Noses. Sensors and Actuators B., 18, 211220 (1993) 4 MIELLE, P., HIVERT, B. AND MAUVAIS, G. Are gas sensors suitable for on-line monitoring and quantication of volatile compounds? In: ETIEVANT, P. AND SCHREIER, P.,

313

lwt/vol. 31 (1998) No. 4

5 6 7

10

11

12 13

14 15

16

17 18

19

20 21 22 23

(Eds,), Proceedings of Bioavour 95 Dijon: INRA editions, pp. 8184 (1995) MARQUIS, F. AND MIELLE, P. Le nez electronique en quete de credibilite. Odours & VOCs Journal, Special Issue, 4046 (1996) BARTLETT, P. N., ELLIOT, J. M. AND GARDNER, J. W. Electronic noses and their application in the food industry. Food Technology, 51, 4448 (1997) MIELLE, P. Electronic noses: Towards the objective instrumental characterization of food aroma. Trends in Food Science & Technology, Special Issue on Flavour Perception, 7, 432438 (1996) SHIERS, V. P. Electronic nose technology evaluations and developments for the food industry. In: MAARSSEN (Ed), Food Ingredients Europe: Conference Proceedings 1995. Frankfurt: Miller Freeman Technical, pp. 198200 (1995) GARDNER, J. W. AND BARTLETT, P. N. Performance denition and standardisation of electronic noses. Transducers 95 Eurosensors IX, The 8th International Conference on Solid-State Sensors and Actuators, and Eurosensors IX, Stockholm, Sweden, June 2529. 671686 (1995) DEMARNE, V. AND SANJINES, R. Thin lm semiconducting metal oxide gas sensors. In: SBERVEGLIERI, G. (Ed), Gas Sensors. Dordrecht: Kluwer Academic Publishers, pp. 89116 (1992) HODGINS, D. The Electronic Nose: Sensor Array-Based Instruments that Emulate the Human Nose. In: MARSILI, R. (Ed), Techniques for analyzing food aroma. New York: Marcel Dekker Inc., pp. 331371 (1997) HODGINS, D. A sixth sense. Dairy Industries International, 60, 3233 (1995) MARI, C. M. AND BARBI, G. B. Electrochemical gas sensors. In: SBERVEGLIERI, G. (Ed), Gas Sensors. Dordrecht: Kluwer Academic Publishers, pp. 329364 (1992) Anonymous. An articial nose. The Futurist, 30, 4747 (1996) DICKINSON, T. A., WHITE, J., KAUER, J. S. AND WALT, D. R. A chemical-detecting system based on a crossreactive optical sensor array. Nature, 382, 697700 (1996) NAKAGAWA, M., KAWABATA, S., NISHIYAMA, K., UTSUNOMIYA, K., YAMAMOTO, I., WADA, T., YAMASHITA, Y. AND YAMASHITA, N. Analysis of constituents in mixed odor vapors using an array of the -Al2O3 Chemiluminescence-based sensors. In: KUNZ, R. E (guest Ed), Proceedings of the 3rd European Conference on Optical Chemical Sensors and Biosensors: Europt(r)ode III, Zurich. Amsterdam, Lausanne: Elsevier, pp. 76 (1997) EGUCHI, K. Optical gas sensors. In: SBERVEGLIERI, G. (Ed), Gas Sensors. Dordrecht: Kluwer Academic Publishers, pp. 307328 (1992) SUTTER, J. M. AND JURS, P. C. Neural network classication and quantication of organic vapors based on uorescence data from a ber-optic sensor array. Analytical Chemistry, 69, 856862 (1997) BALTRUSCHAT, H., KAMPHAUSEN, I., OELGEKLAUS, R., ROSE, J. AND WAHLKAMP, M. Detection of volatile organic solvents using potentiodynamic gas sensors. Analytical Chemistry, 69, 743748 (1997) ANONYMOUS. Riechlandkarten im Gehirn. Kunstliche Nasen fur die Industrie. Neue Zurcher Zeitung, 63, (1997) KOHL, D. Oxidic semiconductor gas sensors. In: SBERVEGLIERI, G. (Ed), Gas Sensors. Dordrecht: Kluwer Academic Publishers, pp. 4388 (1992) HUHEEY, J. E. The solid state. In: HUHEEY, J. E. (Ed), Inorganic chemistry. New York: Harper & Row Publishers, pp. 201203 (1983) GREENWOOD, N. N. AND EARNSHAW, A. Silicium. In: GREENWOOD, N. N. AND EARNSHAW, A. (Eds), Chemie

24

25

26

27 28 29

30 31 32 33

34 35

36

37

38

39

40

41

der Elemente. Weinheim: VCH Verlag, pp. 424426 (1988) WUNSCHE, L. F., VUILLEUMIER, C., KELLER, U., BYFIELD, M. P., MAY, I. P. AND KEARNEY, M. J. Scent characterisation: from human perception to electronic nose. Proceedings of the 13th International Congress of Flavours, Fragrances and Essential Oils, 1519 October 1995, Istanbul, Turkey, 3, 295313 (1995) KOHL, D. Fundamentals and recent developments of homogeneous semiconducting sensors. In: GARDNER, J. W. AND BARTLETT, P. N. (Eds), Sensors and Sensory Systems for an Electronic Nose. Dordrecht: Kluwer Academic Publishers, pp. 5376 (1992) AMRANI, M. E. H., PERSAUD, K. C. AND PAYNE, P. A. High-frequency measurements of conducting polymers: development of a new technique for sensing volatile chemicals. Measurement Science and Technology, 6, 15001507 (1995) HODGINS, D. AND SIMMONDS, D. The electronic nose and its application to the manufacture of food products. Journal of Automatic Chemistry, 17, 179185 (1995) ZANNONI, M. Preliminary results of employ of an articial nose for the evaluation of cheese. Scienza e Tecnica Lattiero-Casearia, 46, 277289 (1995) NIEUWENHUIZEN, M. S. AND NEDERLOF, A. J. Silicon based surface acoustic wave gas sensors. In: GARDNER, J. W. AND BARTLETT, P. N. (Eds), Sensors and Sensory Systems for an Electronic Nose. Dordrecht: Kluwer Academic Publishers, pp. 131145 (1992) HOLMBERG, M. Data Evaluation for an Electronic Nose. Doctoral thesis at the Department of Physics and Measurement Technology, Linkoping University (1997) GUILBAULT, G. AND JORDAN, J. M. Analytical uses of piezoelectric crystal: a review. Critical Reviews in Analytical Chemistry, 19, 128 (1988) KING, W. H. JR. Piezoelectric Sorption Detector. Analytical Chemistry, 36, 17351739 (1964) WOHLTJEN, H. AND DESSY, R. Surface acoustic wave probe for chemical analysis. I. Introduction and instrument description. Analytical Chemistry, 51, 14581465 (1979) WOHLTJEN, H. AND DESSY, R. Surface acoustic wave probe for chemical analysis. II. Gas chromatography detector. Analytical Chemistry, 51, 14651470 (1979) WOHLTJEN, H. AND DESSY, R. Surface acoustic wave probe for chemical analysis. III. Thermomechanical polymer analyzer. Analytical Chemistry, 51, 14701478 (1979) MARTIN, S. J., SCHWARTZ, S. K., GUNSHOR, R. L. AND PIERRET, R. F. Surface acoustic wave resonators on a ZnO-on-Si layered medium. Journal of Applied Physics, 54, 561569 (1983) MARTIN, S. J., SCHWEIZER, K. S., SCHWARTZ, S. K. AND GUNSHOR, R. L. Vapor sensing by means of a ZnO-on-Si surface acoustic wave resonator. Proceeding IEEE Ultrasonics Symposium, Dallas. 207212 (1984) MARTIN, S. J., SCHWEIZER, K. S., RICCO, A. J. AND ZIPPERIAN, T. E. Gas sensing with surface acoustic wave devices. Proceeding of the 3rd International Conference on Sensors and Actuators, Philadelphia, 7173 (1985) CALIENDO, C. AND VERONA, E. Surface acoustic wave (SAW) gas sensor. In: SBERVEGLIERI, G. (Ed), Gas Sensors. Dordrecht: Kluwer Academic Publishers, pp. 281306 (1992) LUNDSTROM, I., SVENSSON, C., SPETZ, A., SUNDGREN, H. AND WINQUIST, F. From hydrogen sensors to olfactory images twenty years with catalytic eld-effect devices. Sensors and Actuators B1, 1314, 1623 (1993) LUNDSTROM, I., SPETZ, A., WINQUIST, F., ACKELID, U. AND SUNDGREN, H. Catalytic metals and eld-effect devices a useful combination. Sensors and Actuators, B1, 1520 (1990)

314

lwt/vol. 31 (1998) No. 4

42 LUNDSTROM, I., SHIVARAMAN, M. S., SVENSSON, C. AND LUNDKVIST, L. Applied Physics Letters, 26, 55 (1975) 43 LUNDSTROM, I., HEDBORG, E., SPETZ, A., SUNDGREN, H. AND WINQUIST, F. Electronic noses based on eld effect structures. In: GARDNER, J. W. AND BARTLETT, P. N. (Eds), Sensors and Sensory Systems for an Electronic Nose. Dordrecht: Kluwer Academic Publishers, pp. 303319 (1992) 44 SPETZ, A., WINQUIST, F., SUNDGREN, H. AND LUND STROM, I. Field effect gas sensors. In: SBERVEGLIERI, G. (Ed), Gas Sensors. Dordrecht: Kluwer Academic Publishers, pp. 219279 (1992) 45 GARDNER, J. W. AND BARTLETT, P. N. Pattern recognition in odour sensing. In: GARDNER, J. W. AND BARTLETT, P. N. (Eds), Sensors and Sensory Systems for an Electronic Nose. Dordrecht: Kluwer Academic Publishers, pp. 161179 (1992) 46 FUKUNAGA, T., MORI, S., NAKABAYASHI, Y., KANDA, M. AND EHARA, K. Application of avor sensor to coffee. ASIC, 16th International Scientic Collopuium on Coffee, 914 April, Kyoto, 478486 (1995) 47 GOPEL, W. Analysis of Complex Gas Mixtures by Pattern Recognition. Olfaction and Electronic Nose, 2nd International Symposium, Toulouse. (1995) 48 PERSAUD, K. C. AND PELOSI, P. Sensor arrays using conducting polymers for an articial nose. In: GARDNER, J. W. AND BARTLETT, P. N. (Eds), Sensors and Sensory Systems for an Electronic Nose. Dordrecht: Kluwer Academic Publishers, pp. 237256 (1992) 49 NEWMAN, A. R. Electronic noses. Analytical Chemistry, 63, 585A588A (1991) 50 GARDNER, J. W., CRAVEN, M., DOW, C. AND HINES, E. L. The prediction of bacteria type and culture growth phase by an electronic nose with a multi-layer perceptron network. Measurement Science and Technology, 9, 120127 (1998) 51 GIBSON, T. D., PROSSER, O., HULBERT, J. N., MARSHALL, R. W., CORCORAN, P., LOWERY, P., RUCKKEENE, E. A. AND HERON, S. Detection and simultaneous identication of microorganisms from headspace samples using an electronic nose. Sensors and Actuators B Chemical, 44, 413422 (1997) 52 SHURMER, H. V., GARDNER, J. W. AND CHAN, H. T. The application of discrimination techniques to alcohols and tobaccos using tin-oxide sensors. Sensors and Actuators, 18, 361371 (1989) 53 TULLETT, C. AromaScan: Tomorrows digital aroma technology today. Cosmetics and Toiletries Manufacture Worldwide, 205209 (1995) 54 BUET, D., BURGAUD, H. AND ROSSI, P. Electronic Nose: a real interface between sensory panels and ne analytical procedures in cosmetics. Olfaction and Electronic Nose, 3rd International Symposium, Toulouse (1996) 55 ORMANCEY, X., PUECH, S. AND COUTIERE, D. Electronic Nose: an efcient tool for the measurement and control of odors. Olfaction and Electronic Nose, 3rd International Symposium, Toulouse (1996) 56 TALOU, T., SANCHEZ, J. M. AND BOURROUNET, B. New trends in electronic noses for avour chemistry. In: TAYLOR, A. J. AND MOTTRAM, D. S. (Eds), Flavour Science: Recent Developments. Cambridge: The Royal Society of Chemistry, pp. 277282 (1996) 57 HODGINS, D. The development of an electronic nose for industrial and environmental applications. Sensors and Actuators B, 26-27, 255258 (1995) 58 HOBBS, P. J., MISSELBROOK, T. H. AND PAIN, B. F. Assessment of odours from Livestock wastes by a photoionization detector, an electronic nose, olfactometry and gas chromatography-mass spectrometry. Journal of Agricultural Engineering Research, 60, 137144 (1995) 59 LORANS, E. Identication des composes volatils du metro de Paris. Utilisation du nez electronique dans une

60 61

62

63 64 65

66

67

68

69

70

71

72

73 74

75

atmosphere diffuse et complexe. Olfaction and Electronic ` Nose, 2nd International Symposium, Toulouse (1995) WANG, H.-H. Assessment of solid state fermentation by a bioelectronic articial nose. Biotechnology advances, 11, 701710 (1993) MISSELBROOK, T. H., HOBBS, P. J. AND PERSAUD, K. C. Use of an Electronic Nose to measure odour concentration following application of cattle slurry to grassland. Journal of Agricultural Engineering Research, 66, 213220 (1997) BYUN, H. G., PERSAUD, K. C., KHAFFAF, S. M., HOBBS, P. J. AND MISSELBROOK, T. H. Application of unsupervised clustering methods to the assessment of malodour in agriculture using an array of conducting polymer odour sensors. Computers and Electronics in Agriculture, 17, 233247 (1997) KALMAN, E.-L., WINQUIST, F. AND LUNDSTROM, I. A new pollen detection method based on an electronic nose. Atmospheric Environment, 31, 17151719 (1997) BERDAGUE, J.-L. AND TALOU, T. Exemples dapplication aux produits carnes des senseurs de gaz a semi-con ` ducteurs. Sciences des Aliments, 13, 141148 (1993) BERDAGUE, J.-L. VERNAT, G. AND ROSSI, V. Revue. Caracterisation instrumentale de la qualite des matieres ` premieres et des aliments par analyse des composes ` volatils. Viandes et Produits Carn s, 14, 135138 (1993) e ROSSI, V., TALON, R. AND BERDAGUE, J.-L. Rapid discrimination of Micrococcaceae species using semiconductor gas sensors. Journal of Microbiological Methods, 24, 183190 (1995) ROSSI, V., GARCIA, C., TALON, R., DENOYER, C. AND BERDAGUE, J.-L. Discrimination rapide par senseurs de gaz a semi-conducteurs. Application aux produits carnes ` secs et aux souches bacteriennes. Viandes et Produits Carn s, 16, 2124 (1995) e VERNAT-ROSSI, V., GARCIA, C., TALON, R., DENOYER, C. AND BERDAGUE, J.-L. Rapid discrimination of meat products and bacterial strains using semiconductor gas sensors. Sensors and Actuators B, 37, 4348 (1996) ROSSI, V., GARCIA, C., TALON, R., DENOYER, C. AND BERDAGUE, J.-L. Rapid discrimination of meat products and bacterial strains using semiconductor gas sensors. In: ETIEVANT, P. AND SCHREIER, P. (Eds), Proceedings of Bioavour 95. Dijon: INRA editions, pp. 8589 (1995) MOY, L., VASIC, G., BERDAGUE, J.-L. AND ROSSI, V. Transient signal modeling for fast odour classication. In: ETIEVANT, P. AND SCHREIER, P. (Eds), Proceedings of Bioavour 95. Dijon: INRA editions, pp. 5558 (1995) VERNAT-ROSSI, V., VERNAT, G. AND BERDAGUE, J.-L. Discrimination de produits agroalimentaires par capteurs de gaz a semi-conducteurs fonctionnant avec lair ` de latmosphere ambiante du laboratoire, Differentes ` approaches de traitement du signal. Analusis, 24, 309315 (1996) VERNAT-ROSSI, V. Analyse de produits agro-alimentaires par capteurs de gaz a base de semiconducteurs. Tech` niques de collecte de la fraction volatile et traitement des donnees. These de doctorat de luniversite Blaise Pascal, ` Specialite Sciences des aliments, Clermont-Ferrand (1997) VAN DIJK, R. Quality control in the meat industry using gas sensor arrays. Olfaction and Electronic Nose, 3rd International Symposium, Toulouse (1996) WINQUIST, F., HORNSTEN, E. G., SUNDGREN, H. AND LUNDSTROM, I. Performance of an electronic nose for quality estimation of ground meat. Measurement Science Technology, 4, 14931500 (1993) SHIERS, V. P., FARNELL, P. J., DE BLACKBURN, W. C. AND KILCAST, D. Comparison of electronic nose, sensory panel and microbiological analysis for the detection of spoilage in modied-atmosphere-packed minced beef. Leatherhead report, 734, (1996)

315

lwt/vol. 31 (1998) No. 4

76 STETTER, J. R., FINDLAY, M. W., SCHROEDER, K. M. JR., YUE, C. AND PENROSE, W. R. Quality classication of grain using a sensor array and pattern recognition. Analytica Chimica Acta, 284, 111 (1993) 77 BORJESSON, T., EKLOV, T., JONSSON, A., SUNDGREN, H. AND SCHNURER, J. Electronic nose for odor classication of grains. Cereal Chemistry, 73, 457461 (1996) 78 JONSSON, A. WINQUIST, F. SCHNURER, J. SUNDGREN, H. AND LUNDSTROM, I. Electronic nose for microbial quality classication of grains. International Journal of Food Microbiology, 35, 187193 (1997) 79 BORJESSON, T. Detection of off-odours in grains using an electronic nose. Olfaction and Electronic Nose, 3rd International Symposium, Toulouse (1996) 80 TAN, T. T., LOUBET, F., LABRECHE, S. AND AMINE, H. The Electronic Nose: Discrimination, stability, calibration and transferability. Case Study: Coffee. Olfaction and Electronic Nose, 3rd International Symposium, Toulouse (1996) 81 GARDNER, J. W., SHURMER, H. V. AND TAN, T. T. Application of an electronic nose to the discrimination of coffees. Sensors and Actuators B, 6, 7175 (1992) 82 TAN, T., LUCAS, Q., MOY, L., GARDNER, J. AND BARTLETT, P. The Electronic Nose A New Instrument for Sensing Vapours. LC-GC International, 8, 218225(1995) 83 GARDNER, J. W. AND BARTLETT, P. N. Intelligent ChemSADs for Articial Odor-Sensing of Coffees and Lager Beers. In: KURIHARA, K., SUZUKI, N. AND OGAWA, H. (Eds), Olfaction and Taste XI: proceedings of the 11th Symposium on Olfaction and Taste and of the 27th Japanese Symposium on Taste and Smell. Tokyo: Springer-Verlag, pp. 690693 (1994) 84 DELARUE, J., GRETSCH, C., LIARDON, R. AND VISANI, P. Detection of Aroma above a Coffee Powder. Limits and Perspectives of Electronic Sensors. Olfaction and Electronic Nose, 3rd International Symposium, Toulouse (1996) 85 TOKO, K., MORI, S., NAKABAYASHI, Y., KANDA, M. AND FUKUNAGA, T. The taste sensor. ASIC, 16e Colloque, Kyoto, 292299 (1995) 86 KOMAI, H, NAITO, Y., SATO, K., IKEZAKI, H., TANIGUCHI, A. AND TOKO, K. Measurement of coffee taste using lipid membrane taste sensors. ASIC, 16e Colloque, Kyoto. 300308 (1995) 87 PEARCE, T. C., GARDNER, J. W., FRIEL, S., BARTLETT, P. N. AND BLAIR, N. An electronic nose for monitoring the avour of beer. Analyst, 118, 371377 (1993) 88 TOMLINSON, J. B., ORMROD, I. H. L. AND SHARPE, F. R. Electronic aroma detection in the brewery. Journal of American Society of Brewering Chemists, 53, 167173 (1995) 89 BAILEY, T. P., HAMMOND, R. V. AND PERSAUD, K. C. Application for an electronic aroma detector in the analysis of beer and raw materials. Journal of American Society of Brewering Chemists, 53, 3942 (1995) 90 ZIMMERMANN, D. AND LECLERCQ, C. Electronic nose for monitoring the avour of special malts. Olfaction and Electronic Nose, 2nd International Symposium, Toulouse (1995) 91 ZIMMERMANN, D. AND LECLERCQ, C. Le nez electronique: une technologie interessante pour la caracterisation des malts speciaux. Bios boissons conditionnement, 257, 2934 (1995) 92 BREHERET, S., TALOU, T., BOURROUNET, B. AND GASET, A. On-line differentiation of mushrooms aromas by combined Headspace/multi-odour gas sensors devices. In: ETIEVANT, P. AND SCHREIER, P. (Eds), Proceedings of Bioavour 95. Dijon: INRA editions, pp. 103107 (1995)

93 Anonymous. Elektronische Nasen. Die Ern hrungsina dustrie, 6, 5455 (1996) 94 MARIACA, R. AND BOSSET, J. O. Instrumental analysis of volatile (avour) compounds in milk and dairy products. Lait, 77, 1340 (1997) 95 HARPER, W. J., SOHN, S. AND DA JOU, K. The role of fatty Acids in the Aroma Proles of Swiss Cheese as determined by an Electronic Nose. Olfaction and Electronic Nose, 3rd International Symposium, Toulouse (1996) 96 PERSAUD, K. C. AND TRAVERS, P. J. Arrays of broad specicity lms for sensing volatile chemicals. In: KRESSROGERS, E. (Ed), Handbook of Biosensors and Electronic Noses. Frankfurt: CRC Press, pp. 563592 (1997) 97 KAIPAINEN, A., YLISUUTARI, S., MOY, L. AND LUCAS, Q. Comparison of two techniques for the analysis of headspace aromaproles of sugar: thermal desorption GC-MS and electronic nose. Olfaction and Electronic Nose, 3rd International Symposium, Toulouse (1996) 98 VIAUX, L., MAGNE, V. AND ROBILLARD, B. The electronic nose to help determination of the technical specications of some additives and technological aids used in the sparkling wines process. Olfaction and Electronic Nose, 3rd International Symposium, Toulouse (1996) 99 OLAFSSON, R., MARTINSDOTTIR, E., OLAFSDOTTIR, G., SIGFUSSON, P. I. AND GARDNER, J. W. Monitoring of sh freshness using tin oxide sensors. In: GARDNER, J. W. AND BARTLETT, P. N. (Eds), Sensors and Sensory Systems for an Electronic Nose. Dordrecht: Kluwer Academic Publishers, pp. 257272 (1992) 100 WINQUIST, F., SUNDGREN, H. AND LUNDSTROM, I. A practical use of electronic noses: quality estimation of cod llet bought over the counter. Transducers 95 Eurosensors IX, The 8th International Conference on Solid-State Sensors and Actuators, and Eurosensors IX, Stockholm, Sweden, (June 25-29), 1, 695697 (1995) 101 SIMON, J. E., HETZRONI, A., BORDELON, B., MILES, G. E. AND CHARLES, D. J. Electronic sensing of aromatic volatiles for quality sorting of blueberries. Journal of Food Science, 61, 967969 (1996) 102 BAZEMORE, R., ROUSEFF, R. AND SIMS, C. Comparison of Fresh Squeezed Orange Juice Analysis by Electronic Nose, Sensory Taste Panel and GC FID. Olfaction and Electronic Nose, 3rd International Symposium, Toulouse (1996) 103 BENALI, F., HENRY, J. AND DELAUNAY, Y. Electronic nose potentiality utilisation on packaged products. Olfaction and Electronic Nose, 2nd International Symposium, Toulouse (1995) 104 BOHATIER, C., VERNAT-ROSSI, V., BREYSSE, C., THEBAULT, J. AND BERDAGUE, J.-L. Caracterisation dun emballage defectueux par analyse de sa fraction volatile. Industries Alimentaires et Agricoles, 112, 849852 (1995) 105 HOLMBERG, M., WINQUIST, F., LUNDSTROM, I., GARDNER, J. W. AND HINES, E. L. Identication of paper quality using electronic nose. Sensors and Actuators B, 2627, 246249 (1995) 106 LJUNGBERG WILLING, B.-I. AND BRUNDIN, A. Sensory analysis versus Electronic Sensors- a comparison using multivariate analysis. Olfaction and Electronic Nose, 3rd International symposium, Toulouse (1996) 107 SCHIFFMAN, S. S., KERMANI, B. G. AND NAGLE, H. T. Analysis of Medication Off-odors using an Electronic Nose. Chemical Senses, 22, 119128 (1997) 108 NEAVES, P. I. AND HATFIELD, J. V. A new generation of integrated electronic noses. Sensors and Actuators B, 2627, 223231 (1995) 109 MIELLE, P. Une technique de pointe au service de la qualit aromatique. Le Technoscope de Biofutur, 174, 210 (1998)

316