A genetic algorithm search has been applied to the task of optimal array configuration and it is shown that multisensor array performance, in terms of the successful classification of wines, can be improved through the application of such techniques. This is achieved through the utilisation of an objective function, which accounts for sensor sensitivity and repeatability, in order to identify near-optimal subsets of the available parameters. When combined, these subsets of parameters provide discriminable and repeatable fingerprints of odorous materials in classification space which facilitate successful identification of wines to the same degree as much larger arrays. In particular, utilising a thermally cycled 7-element array, which generated 56 parameters, a subset of only 28 parameters was identified which gave an improvement above the classification rate derived from the 56-parameter set. Furthermore, several other optimised parameter sets showed similar results. The role of the objective function is also investigated and shows that further refinement is needed to improve the correlation with the final correct classification rate achieved by the array configuration. These results, along with previous studies, show that through the employment of optimisation techniques, it is often possible to reduce array dimensionality by around 50–70% while still achieving equal, or better, levels of system performance.

The use of multisensor systems is becoming widespread in a number of applications. Recent developments have led to the identification of multiple parameters that can be extracted from individual sensor elements. However, it is now accepted that the most promising employment of such technology will most probably be in application specific instruments, where a limited set of calibrands would be analysed, based upon a calibration database containing representative parameters of each calibrand. Given the large number of sensors, and the availability of multiple parameters, instrument designers must address the need to maximise the discriminatory information derived from a sensor array. This paper presents a structured approach to selecting the individual array elements (sensors or parameters). Ideally, given a database of the responses of n parameters to p species, it should be possible to identify a subset of n sensor responses that provide a specified level of performance when applied to the classification, or description, of a subset of the p species. This paper compares the utility of cost functions and search algorithms that can be applied to the task of optimal sensor array configuration. In particular, the novel use of genetic algorithms when applied to the task of array configuration for gas sensing is described.

Through the application of novel thermal modulation, pre-processing and feature extraction techniques the performance of an 8-element tin oxide gas sensor array has been significantly enhanced when applied to the task of classifying the aromas of three loose leaf teas. Array signatures were generated by thermally cycling the sensor array over the temperature range 250–500°C, whilst exposing the array to the odorous headspace of the three teas. The application of thermal modulation and an enhanced feature extraction algorithm, generating 208 parameters, proved to be highly successful giving a cross-validated classification rate of 90% for unseen samples of the three classes of tea. In comparison, a fixed temperature steady state metric, based upon the same array of sensors, yielded a cross-validated classification rate of only 69%. Furthermore, using a novel genetic algorithm optimisation technique to identify a near-optimal sensor parameter configuration for the task of tea classification, it was shown that a correct classification rate of 93% could be achieved with only 21 dynamic parameters.

Genetic algorithms have been applied to the task of optimal array configuration and it is shown that multisensor array performance. in terms of the successful classification of wines. cain be improved through the application of such techniques. This is;achieved by identifying;a subset of the available parameters which, when combined, provide discriminable and repeatable fingerprints of odorous materials in classification space. In particular. utilising a thermally cycled 8-element array, which generated 64 parameters, a subset of only 24 parameters was identified which gave an improvement of 5% above the classification rate derived from the 64-parameter set. These results.;along with previous studies. show that through the employment of optimisation techniques it is often possible to reduce array dimensionality by;around 50-70% whilst Will achieving equal. or better. levels of system performance.