One may think an electronic nose would tell you everything you ever wanted to know about a sample, but, in fact, an electronic nose functions like a black box—it knows nothing until it is taught. It is simply a series of sensors that respond to volatile components of the headspace above a sample.
The experimenter optimizes operating conditions and teaches the electronic nose what to recognize through teaching sets and data libraries. The software of the electronic nose uses a series of algorithms and chemometric methods to provide meaningful data from the sensor responses. Because this is a correlation technique, one must remember that the electronic nose will only be, at best, as accurate as the analytical or sensory data it is correlated against.
Several companies manufacture electronic noses, using a variety of detectors. The most commonly used detector is composed of a series of sensors, which can be metal oxides, conducting polymers, or quartz microbalance sensors. These sensor technologies, with regard to the food industry, have been described in detail by Bartlett et al. (1997). Recently, mass selective detectors have been added in place of sensors. Unlike other electronic nose technologies, these instruments function similarly to a mass spectrometer to characterize sample aroma.
The electronic nose can be used for a number of purposes:
• Analysis. Although the electronic nose can be used for analytical purposes, some consideration is needed to determine whether it could be a valuable tool for an analytical laboratory, since this technology is fairly expensive and requires a well-defined teaching set of samples. Because the electronic nose is not a primary analytical technique, teaching sets of samples are necessary. The data for the teaching sets can be obtained in a number of ways: from different types of sensory panels, from chromatography or wet chemistry analyses, or from samples that have been carefully selected with legitimate information provided by suppliers.
Product variability can be rapidly measured within a sample set or series of product lots. The electronic nose can compare samples to each other or to a reference standard. The benefits of this technique are rapidity and reproducibility. Once an application has been established, the system can be run by anyone, with minimal training. Because this technology is relatively new, the longevity and stability of the sensors have not been determined over time.
--- PAGE BREAK ---
• Quality Control. If the electronic nose is calibrated to recognize acceptable and unacceptable samples, the number of sensory panels to evaluate raw materials and finished products can be minimized. The electronic nose is not a replacement for sensory evaluations, however, because sensory data are required to set the limits of what is acceptable and unacceptable. Furthermore, the electronic nose only analyzes the volatile component present in the sample; it does not sense components that would be associated with tasting the sample.
• Product Matching. The electronic nose can also assist competitor matching by showing changes in a complex flavor system caused by the addition of flavors and other ingredients to a base formula. When matching a competitor’s sample, the electronic nose can help formulators selectively use cost-effective flavor components as part of the process.
The electronic nose system in our laboratory is the Fox (Alpha-M.O.S. America, Hillsborough, N.J.). This system consists of 12 metal oxide sensors, an agitator/heated headspace controller, an air conditioner/humidity controller, and an autosampler. The detector responses are interpreted by a Windows-based computer and the Fox 3000™ software package provided by the manufacturer.
Before any samples are analyzed by this technique, a number of operating conditions must be established. These include setting up the sample incubation temperature, sample size, injection rate, injection quantity, and use of added solvent. These conditions are optimized, so that differences between samples are enhanced.
In the experiments discribed below, 1.00±0.02 g of sample was weighed directly into a 10-mL headspace vial, which was then capped and placed into the autosampler tray. The autosampler moved the sample vial into the heating compartment, where the sample was agitated and warmed to a prescribed temperature. The autosampler withdrew an aliquot from the headspace of the vial and injected the volatilized sample for sensor analyses. The normal acquisition time was 2 min, and the maximum sensor responses were recorded. Because this is a technique based on statistical analysis, all samples and standards were prepared and run in triplicate.
The results were processed by chemometrics, which uses mathematical, statistical, or other methods of formal logic to design or optimize experiments and to provide maximum relevant chemical information from analysis of the chemical data. The electronic nose uses multivariate statistics as its chemometric technique to evaluate and classify samples that have been analyzed. The examples presented below use two different multivariate statistical approaches to evaluate the data.
Principal component analysis (PCA) examines linear combinations of the original data for maximum variance. It allows one to reduce the number of variables, or sensor responses, to a smaller number without compromising or losing data and information. The first component accounts for the greatest portion of the variance and is typically plotted on the x-axis. The second largest component is plotted orthogonally to the first axis (i.e., on the y-axis), and so forth.
Discriminant factor analysis (DFA) is a technique aimed at determining which sets of variables best discriminate one group from another. In a DFA experiment, the objective is to determine where an unknown can be grouped based on values assigned to samples in different groups.
--- PAGE BREAK ---
• Spice Origin Analysis. The traditional techniques that have been used for determining the origin of spices are sensory and chemical analyses. Regardless of what technique was used, it was important to obtain a fairly large number of samples whose origins were known. The samples would be used to develop reference libraries and identify specific compounds or attributes that were common to the origin.
The same approach can also be used when considering the electronic nose to identify spice origins. Sensory and chemical data support the findings. For example, if we look at Jamaican and Guatemalan allspice origins, the sensory responses from trained panelists show a larger difference in the minimum and maximum values related to each attribute of Jamaican allspice (Fig.1), compared to those of Guatemalan allspice (Fig. 2). Hence, there are sensory differences between the two allspice origins.
The Jamaican and Guatemalan allspice samples were also analyzed chemically by gas chromatography. The gas chromatograms of the volatile oil fraction from each sample show that the chromatographic peak profiles are different (Fig. 3). Therefore, one can conclude that the Jamaican and Guatemalan allspice samples are chemically different.
The electronic nose sensor responses appear similar when comparing the change in sensor response vs time (Fig. 4). However, PCA shows significant differences. The PCA plot demonstrates separate regions for each allspice origin (Fig. 5). Each geometric shape represents the data boundary of replicate analyses performed by the electronic nose. Since there is no overlap, both origins are easily identified. Allspice is an easy example, since there are only two origins to identify.
In another example, we considered five origins of cinnamon. The DFA plot of cinnamon shows origins from Sri Lanka, China, Korintji, Madagascar, and Vietnam (Fig. 6). The DFA plot of the cinnamon samples shows no overlap. Since each region is separate, the origin of an unknown sample could easily be identified.
• Formula Matching. A salad dressing base formula was prepared, then compared to a competitor control sample. A PCA plot showed two separate regions relative to each formula. Next, two different quantities of tarragon and onion flavor were added to the base formula in an effort to match the flavor of a competitor control sample. The PCA plot shows movement in the data toward the region of the control sample (Fig. 7).
A flavorist also recommended the addition of ethanol and a proprietary flavor component, each in two concentrations. When various amounts of the proprietary flavor component were added to the base formula, the PCA plot demonstrated a close match to the competitor sample (Fig. 8).
When the proposed salad dressing formula and the competitor formula were evaluated by trained panelists, sensory responses supported the electronic nose data.
Training Set Is Most Important
In any electronic nose experiment or application, it is extremely important to remember that the data derived from the electronic nose will only be as good as the samples that are used in the calibration and teaching sets. Therefore, selection of the training set is the most important part in developing this technology.
Based on a paper presented at the IFT Annual Meeting of the Institute of Food Technologists, Chicago, Ill., July 24–28, 1999.
The authors acknowledge Alan Harmon, Kris Spence, and Rebecca Norwat at the McCormick Technical Innovation Center for their sample analyses and advice.
by MICHAEL G. MADSEN AND ROMAN D. GRYPA
Author Madsen, a Professional Member of IFT, is Analytical Chemist, and author Grypa is Senior Scientist, McCormick & Co., Inc., 204 Wight Ave., Hunt Valley, MD 21031. Send reprint requests to author Madsen.
by Edited by Neil H. Mermelstein,
Bartlett, P.N., Elliott, J.M., and Gardner, J.W. 1997. Electronic noses and their application in the food industry. Food Technol. 51(12): 44-48.