Sensors Research Laboratory
University of Warwick
Tel: 02476 574494
Novel Data Processing for Advanced Electronic Noses
Electronic noses are used to collect a wide variety of data. These data contain a large amount of information that is intended to be used to discriminate among the different odours detected by the electronic nose. However, the varying types and large quantity of information dilutes the discriminatory quality of the data ? noise, contamination and sensor variation information is contained within the data, obscuring the response information which is most desired.
In order to extract the useful information for electronic nose data, many established techniques have been used to process and classify response data. Various means are used to compensate for the excess of unwanted data ? including normalisation, feature selection and feature extraction. These compensation steps are aimed at removing interfering information such as noise and ensuring all responses are equally weighted, and pre-process the data so it is compatible with the primary processing methods. These pre-processing steps are chosen based on the nature of the data produced by the electronic nose, as well as the chosen application.
Process methods are used to classify and arrange the data in the most suitable and clear form. Electronic noses use a variety of established classification techniques, ranging from statistical linear methods (such as principle component analysis and discriminant function analysis) to non-linear neural net and clustering methods. These methods have been applied to a range of electronic nose applications with considerable success.
However, as knowledge of olfaction continues to improves, electronic noses also continue to improve. New generations of electronic noses generate new data containing new types of information. The aim of this research is to determine if novel data processing techniques not previously used in the realm of electronic nose processing can be used to extract this new information, and make more accurate classifications than when using currently established electronic nose processing methods.
Pre-processing and processing combinations are being explored used a new generation ?Tandem Electronic Nose?, developed at the University of Warwick. This Tandem Electronic Nose simulates the effect of nasal chromatography and generates spatio-temporal information not previously encountered in the field of electronic noses. Multiple related signals are generated, separated in space and time. While traditional electronic nose processing has been applied to this new electronic nose, and with success, the current focus of this research it to explore novel ways of combining the many signals in order to more effectively extract information, and use them in a variety of processing methods to determine if more accurate classifications can be made using this new information.
Principle component analysis plot of simple odours of ethanol and toluene, and a mixture of both
after signal combination processing based on Tandem E-Nose simulations