Detecting information in gas sensor responses using analysis of variance

This article describes how Analysis of Variance may be used to select those regions of the curves generated by a gas sensor array which contain the most discriminant information for a particular application. The Analysis of Variance is performed on each point of the signals generated by the sensor array for a particular set of samples. The Group Variances and the Residual Variances are plotted as functions of their position in the signal. Regions of the signals that vary systematically from one group of samples to another will have high Group Variance values and low and randomly distributed Residual Variance values. This method has shown for a particular set of products for cats analysed with a thirty two polymer sensor array, that the most discriminant information is located at the end of the curves. It has also shown an absence of discrimination between standard and tainted pig fat samples with this sensor array. The advantage of this method is that it can be used on almost any sort of raw signal as a pre-analysis step in order to know whether it is worthwhile going on to more fastidious and time consuming signal analysis procedures.