The Electronic Nose

The Electronic Nose

June 1995 -- QA/QC

By: Ray Marsili
Contributing Editor

  Analytical chemists and sensory scientists are two groups of food science researchers that often find it difficult to appreciate what the other contributes. Analytical chemists deplore the subjectivity, poor reproducibility and often inconclusive results that can occur with sensory tests (especially descriptive tests), while sensory scientists may grumble about the inability of long-awaited analytical data to correlate in any meaningful way to how a food actually tastes or smells.

  New technology, though, may have enabled the creation of an instrument that can make both types of researchers happy. The "Electronic Nose," as it is often called, excites sensory panelists because it works more like the human nose than the analytical chemists' esoteric instruments, yet it is high-tech and objective enough to make even the most stodgy analytical food chemist smile. In a sense, it allows flavor researchers to visualize food aroma and taste.

  The Electronic Nose is not superior to nor is it meant to replace sensory evaluation or gas chromatography/mass spectrometry analysis (GC/MS). Instead, it's a potentially valuable tool that finds application somewhere between the two disciplines. The development of the Electronic Nose is based on more than 10 years of research performed at Warwick and Southampton Universities in the U.K. and Toulouse University in France.

  Even though analytical chemists and sensory scientists sometimes act as rivals, they have learned that the best approach to studying food flavors is a combination of the two disciplines. The Electronic Nose could very well prove to be another tool that complements the analytical instrument analysis and sensory testing that food scientists have traditionally used for studying the flavor and aroma of foods.

  Since food flavor and aroma is such a challenging and complex field of study, the more tools the better. The main advantage of this instrument is that in a matter of seconds, it delivers objective, reproducible aroma discrimination with sensitivity comparable to the human nose for most applications.

How it works

  Except for salty, sweet and a few other taste sensations, it is generally accepted that the aroma and taste of most foods are due to the interaction of human sensory organs with the volatile and semi-volatile organic chemical constituents in food materials. Some foods may contain dozens or even hundreds of these volatile flavor-contributing chemicals, and analyzing for them with GC/MS or other sophisticated instrumental techniques in an attempt to resolve flavor-related problems can be a frustrating exercise.

  Unlike chromatography techniques, the Electronic Nose doesn't attempt to separate or resolve all individual volatile components. Instead, it uses an array of sensors that responds to each volatile chemical in a slightly different way -- much like the way the human nose functions.

  The human nose has approximately 10,000 odor sensors which are nonspecific but can be very sensitive to certain odors. It does not try to identify nor quantify the different constituents. Signals from human olfactory sensors are transmitted to the brain for processing. The brain then interprets what the sum of all these signals is describing in terms of odor. Today's Electronic Nose instruments attempt to do the same with many fewer sensors and a simulated brain consisting of a computer and sophisticated software.

  Designing a chemical sensor-based instrument to quantitate all the individual volatile chemicals in a complex mixture of food aroma chemicals is practical.

  "There are some misconceptions about chemical sensors," says Dr. David Ballantine, associate professor of chemistry at Northern Illinois University (DeKalb) and an expert on surface acoustic wave (SAW) sensors and other types of chemical sensors. "Some people think sensors are just mass sensitive devices where a vapor absorbs into the sensor coating and you get a response that's proportional to the mass of vapor that's absorbed. While that's true, it's not the only thing that's going on. The response you get isn't totally dependent on the mass."

  For example, Ballantine points out that polymer-based sensors frequently exhibit swelling when organic vapors are absorbed. This swelling causes changes in the viscoelastic properties of the polymer, which in turn affects the sensor's response. Another problem is that sensors suffer from cross-sensitivity -- i.e., they are not selective. If more than one type of vapor is present, a quantitative output from one sensor is likely to be erroneous.

  To get around these problems, instrument manufacturers employ sensor arrays of several sensors, each with a slightly different coating. So if you're using a sensor array of four sensors and are attempting to measure, for example, ethyl acetate in wine, sensor 1 might give a very high response to this chemical, sensor 2 might give a response of 75% of sensor 1, sensor 3 might give a response of 50% of sensor 1, and sensor 4 might not respond at all to the ethyl acetate vapors. After calibrating the sensors by spiking wine with various levels of ethyl acetate, pattern recognition software could then be used to determine from the response of the four sensors the actual concentration of ethyl acetate in unknown wine samples.

  The problem comes when you want to analyze vapors composed of many different chemicals. If you wanted to analyze wine for ethyl acetate and hexanal, you would need two sets of sensor arrays: one set coated with a material that would respond only to ethyl acetate, and another set coated with a material that would react only with hexanal. Analyzing five-component mixtures would require a minimum of five distinct coatings. Since most food aromas are complex multi-component mixtures (e.g., coffee has several hundred different volatiles), it is not necessarily a trivial exercise to quantitate each component unambiguously.

  "Now if all you want to do is check for a specific component that signals degradation of the product or signals product contamination and you are not trying to unambiguously identify and quantitate each component in a complex mixture, the problem becomes much simpler," explains Ballantine. In this case, all that's required is comparison of the sensor array output of a "control" and an "unknown" sample generated by the sensor array -- for example, in the form of a bar chart showing the response of each sensor after exposure to the food's vapor.

  Presently there are three major manufacturers of commercial Electronic Nose instruments: Alpha M.O.S. (DeMotte, IN), AromaScan (Hollis, NH), and Neotronics (Gainesville, GA). While the instruments may look different, they all have three primary functions that they perform in similar ways: data acquisition (detection of volatile flavor chemicals with specialized electronic sensors); data presentation (statistical graphic plots -- e.g., polar plots, offset polar plots, bar charts, difference plots); and data interpretation (software to assist the user in understanding the practical significance of the graphical outputs, usually accomplished through applications of artificial neural networks).


  As indicated in the accompanying table, each manufacturer uses its own proprietary sensor technology. "The type of sensors and how the sensors are made and used in the instrument is what really differentiates the instruments on the market today," says Ken Weber, technical sales representative for Neotronics. "Our sensors are electrochemically grown conducting polymers made with polypyrrole resins, although polyaniline resins are also commonly used."

  According to Weber, engineers electrochemically grow the resins at a defined potential across an electrode gap. By changing the solvent system or the counter ion associated with their manufacture, it is possible to modify the polymer chain and get subtle differences in the reactivity of each sensor.

  While Neotronics electrochemically grows each individual sensor, AromaScan uses an "inking" or "masking" process whereby it puts 32 sensors on one computer chip or board, says Weber. Neotronics uses an array of 12 different sensors which are basically soldered onto a sensor head. Alpha M.O.S., another instrument manufacturer, uses primarily tin oxide-based sensors in its array, but it also can configure the instrument with a limited number of conducting polymer sensors.

  Metal oxide sensors demonstrate good sensitivity to organic vapors (ppm or even ppb detection limits) for a very broad range of chemical compounds. Due to their poor selectivity -- i.e., all sensors can respond to a single volatile compound but in different magnitudes -- sensor arrays must be employed. Metal oxide sensors are made by depositing a thin layer (505 microns) of oxide film on a ceramic material. For proper functioning, metal oxide sensors are usually heated to between 175° and 425°C. The electrical resistance of the sensor decreases in the presence of an odor, with the magnitude of the response dependent on the nature of the detected molecule and the type of metal oxide used in preparing the sensor. Response time of metal oxide sensors is between 10 and 120 seconds.

  AromaScan and Neotronics instruments use sensors made of conducting polymer resins instead of metal oxides. Michele Greaney, science and technology manager for AromaScan, explains: "They have an inherent charge or base resistance, and as volatile components are absorbed onto the surface of the conductive polymers, they change their base resistance. You get what we call an FR value, and this FR is unique across the sensor array."

  By plotting the FR value for each sensor, a histogram or bar chart can be generated. The chart could for instance, represent a good-tasting coffee sample.

  According to Greaney, the conducting polymer-based instruments can detect organic compounds with a molecular weight range of between 30 and 300 -- essentially the same molecular weight range that the human nose can detect. Conducting polymer resins work best with polar compounds or molecules that have a charge associated with them.

  In general, discrimination of organic vapors is based on three primary modes of vapor/sensor interactions: molecular charge, molecular shape, and molecular size. The sensors are especially sensitive to organic molecules with sulfur and amine functional groups. High concentrations of acids, bases and amines can interfere with proper functioning of conducting polymer sensors.

  Sensor technology is a rapidly developing field, and new and improved sensors with different types of materials and entirely new technologies probably will be developed and applied in the future to ft analysis of aromas. One example is surface acoustic wave devices. It is likely that these new sensor technologies will be able to be chosen for different applications and will allow for the optimized performance of Electronic Nose instruments. However, that's no reason for postponing purchase of an instrument. It's relatively easy and painless -- and not too expensive -- to add or upgrade sensors in most existing commercial instruments.

Graphic outputs

  Sensor-array responses to organic vapors can be presented in numerous ways using a variety of statistical and pattern-recognition approaches. One form of output is a scaled polar plot. This way of visually displaying data is simple to interpret. Each vector on the polar plot represents the output from one sensor. As the relative response of each sensor changes when the sensor array is exposed to vapors from differing food samples, the overall shape and appearance of the polar plot vary.

  Another type of data presentation commonly used is the difference plot. For example, the scaled plot from one test is subtracted from the scaled plot from a different test. The resulting difference plot shows how the relative sensor output varies from sample to sample. If plots vary by a pre-defined value, then the food technologist could assume the sample differs significantly from the control sample. This type of analysis could be useful in quality control "go/no go" tests.

  One problem with this type of output, however, is that the technique is not suitable to plotting negative differences. Furthermore, the scaling data is lost. To overcome these problems, Neotronics uses a "difference ring" to display sensor output. For each sensor, a vector is taken from the radius. If the vector for a given sensor is outside the ring, the sample contains more of a particular type of aroma volatile than the control; if it is inside the ring, it contains less; and if it is on the ring, there is no difference between the sample tested and the control.

  Data-presentation techniques employing scaled polar plots, difference plots, difference rings, etc., allow for qualitative sample-to-sample differentiation. But can quantitative results be obtained? Not easily.

  The best way to use the Electronic Nose is as a fast screening tool to enhance the objectivity over the more subjective human sensory panel results. Fast analysis time, minimal sample preparation, and reproducible and reliable results are the major advantages of the Electronic Nose. Additional advantages include sensitivity, low cost of ownership, and ease of operation.

  Greaney says the instruments can be used to provide reasonable semi-quantitative results. For example, a group of samples can be prepared with an analyte concentration of 10 ppb, another group of samples with analyte levels of 50 ppb, and perhaps another group at 500 ppb. These calibration standards could be analyzed with the Electronic Nose and the sensor-array responses displayed in the form of a cluster analysis, a pattern recognition technique. Unknown samples could then be analyzed. By comparing where they fall in relation to the patterns of standard clusters, estimations of the concentration of the analyte in the unknown samples could be made. The most accurate and sensitive technique for quantitating specific aroma chemicals in food, however, is still gas chromatography.

Artificial neural networks

  The most powerful type of data processing technique being employed in Electronic Nose instruments is called Artificial Neural Network. ANNs are self learning; the more data presented, the more discriminating the instrument becomes. By running many standard samples and storing results in computer memory, the application of ANN enables the Electronic Nose to "understand" the significance of the sensor array outputs better and to use this information for future analysis.

  ANNs allow the Electronic Nose to function in the way a brain functions when it interprets responses from olfactory sensors in the human nose. The ANN's processing elements (or nodes) can be compared to the neurons in the brain. "Learning" is achieved by varying the emphasis, or weight, that is placed on the output of one sensor versus another. The learning process is based on the mathematical, or "Euclidean," distance between data sets. Large Euclidean distances represent significant differences in sample-to-sample aroma characteristics.

  "What we recommend," says AromaScan's Greaney, "is that you run a series of 23 replicates when developing the neural networks. Statistically, we've determined that that's the magic number when teaching the instrument. With these replicates you can incorporate sample variability, lot-to-lot variability, day-to-day variability -- whatever is important to your application."

  ANNs also can be trained to compensate for small response changes that occur when sensors degrade over time. Ideally, a sensor array would respond to a specific sample with the same precision over a long period of time. However, sensors can degrade with prolonged use and the output can vary. ANNs can correct for this problem.

Application examples

  Food companies can use the Electronic Nose as a quality control tool (e.g., to check raw materials, to check product deterioration during shelf life studies, to monitor product during transport to retailers, to ensure that packaging odors do not contaminate product) and as a tool for process control (e.g., to monitor food odors during critical stages of production to ensure that optimum processing conditions are being maintained).

  "We've run an application for the FDA grading the freshness of fish, where fresh fish has a grade of one and rancid fish has a grade of four," says Greaney. "Because the major components that contribute to off-flavors are amines, it is very easy for our instrument to classify fish according to freshness."

  Electronic Nose instruments have been used to verify the authenticity of Parmesan cheeses. The headspace of a Parmesan cheese has more than 160 volatiles. Some of the volatiles most important to the characteristic aroma of Parmesan cheese are esters and fatty acids which are most efficiently analyzed by using a "stripping" sample preparation technique. This technique is an alternative to the commonly used static equilibration method and involves dynamic sampling of the air above the cheese to remove volatiles and semi-volatiles. In addition to Parmesan cheese applications, the Electronic Nose has been used successfully to discriminate other types of cheeses at different stages of maturity.

  The coffee industry also can benefit from the Electronic Nose. It can use the instrument to precisely control the roasting and blending process and to objectively measure the shelf life of intermediates and final products. The Electronic Nose can detect green coffee beans damaged by insects or mold growth, as well as help coffee producers decide how to blend different bean types to obtain optimum taste attributes.

  Other applications that have been demonstrated include checking the freshness of fruit when it is harvested, during shipment, and at the point of sale; checking the freshness of snack products, such as potato chips, to optimize shelf life; characterizing vegetable oils and determining purity; performing quality assurance checks on expensive flavoring materials; and identifying different cuts of meat to minimize potential packaging problems.

  "We see a lot of interest from the tobacco industry," says Neotronics' Weber. "There are a lot of flavors associated with lacing tobacco for taste. For example, the tobacco industry is the No. 1 consumer of chocolate in the U.S."

  Sharp, the giant Japanese electronics company, is looking at incorporating the technology in its microwave ovens to allow the ovens to shut off automatically when they begin to detect chemicals associated with overcooked food.

  Whitbread Breweries, outside London, uses an Electronic Nose instrument to screen incoming ingredients it uses to make beer. The company can minimize the expense associated with discarding improperly brewed beers by catching bad raw materials before they are used to make the beer. According to John Tomlinson, a research scientist at Whitbread Breweries, although the instrument performs well, one problem is that it is so sensitive it can detect small amounts of harmless odors in the water used in beer instead of malodors contributed by the ingredients.

  When asked for his evaluation of the Electronic Nose in classifying wine varieties, Robert Joseph, publishing editor of Wine magazine, commented in a recent Wall Journal Article "I was impressed."

  Joseph compared an Electronic Nose instrument to his sensory evaluation of wines in a blind tasting. The Electronic Nose failed to match one of the white wines in the tasting but fared well with Joseph on most of the rest. "I have seen the future," commented Joseph, "and I have the horrible feeling that (the Electronic Nose) works."

Analytical Chemists Study Food Aromas and Flavors

  In the past the analytical chemist's primary tool for characterizing the aroma and flavor of foods was gas chromatography with flame ionization detection. The recent development of relatively low-cost, highly sensitive bench-top mass spectrometry detectors for gas chromatographs has proven very useful to the study of food flavors and aromas because they permit qualitative identification of chromatographic peaks.

  In addition, by incorporating olfactory detectors and special calibration techniques, analytical chemists have gone one step further in making their work more relevant to sensory interpretation. The technique -- called gas chromatography-olfactometry (GCO) -- involves sniffing the effluents from GC columns.

  Using the combination of GCO with mass spectrometry detection, chemists first inject a representative sample of the aroma vapors associated with a particular food material. This is usually done by employing a device that purges the volatiles from the food matrix then captures them on an absorbent trap (usually Tenax), and this is followed by thermal desorption into a GC capillary column. Numerous other techniques for introducing a representative sample of food vapors into the GC can be employed, depending on the specific application, the sample matrix, the type of aroma chemicals being studied, and other factors.

  As the slug of volatile organic chemicals from the food matrix passes through the heated capillary column, the chemical components, which have varying affinities and reactivities for the coating on the column, pass through the column at different rates. As a result, some emerge from the column quickly, while others take longer. In this way, separation of the various organic chemicals is accomplished.

  Chemicals eluting from the exit end of the capillary column are then split into two separate streams. One stream may be directed to a mass spectrometry detector and the other stream to the "olfactory detector."

  Chemists or trained sensory scientists can then sniff the effluent stream from the olfactory detector (also called a sniff port) and assign characteristic aroma descriptors for each chromatographic peak, or at least write down odor descriptors at specific retention times. It is not uncommon for some of the largest chromatographic peaks to have no detectable odors and for the very small peaks or, indeed, even undetectable peaks to have significant odor intensities.

  With this approach, the food chemist can use the mass spec results to identify the chemical responsible for each peak and the sniff port results to determine the aroma contribution of each component peak. Furthermore, after calibration with standards, chemists also can use results from the mass spectrometry detector to quantitate the levels of specific chemical components by using a computer integrator to determine the area under the peak. With this approach, modern food chemists can identify specific chemicals responsible for important odors in foods and calculate how much of each odor-contributing chemical is present, with sensitivities in the ppb or, in some cases, in the ppt range. Results from this type of experiment are often expressed in a graph called an aromagram, a chromatogram with aroma descriptors written above appropriate peaks.

  While this type of instrumental study provides extremely useful information for the evaluation of food flavors and aromas, it does have limitations. Cost for instrumentation -- including a purge-and-trap device, a capillary gas chromatograph, a mass spectrometry detector and an olfactometry detector -- can run $100,000 or more, and operation of the instrumentation requires a highly trained chemist. Analysis time is significant -- from one to two hours for a single sample. In other words, this sophisticated instrumental technique is more of a research tool than a tool for routine quality control applications.

  One excellent experimental strategy for studying food flavors and aromas is to employ GCO/mass spectrometry detection as a research tool and then correlate findings with the Electronic Nose so it could then be used as a rapid, simple-to-use QC tool.

Sensor Technologies



Type of Sensors

No. of Sensors in the Array

Approximate Cost


Alpha M.O.S.

Fox 2000

Metal Oxides or conducting polymers; expect to have surface acoustic wave (SAW) sensors by 1996

6, 12 or 18


Uses ANN* technology for calibration/data processing



Conducting Polymers



Uses ANN* technology for calibration/data processing


The Nose

Conducting Polymers



Doesn’t currently employ ANN, but should by 1996
*Artificial Neural Networks
Back to top

comments powered by Disqus