Feature extraction of metal oxide gas sensors using dynamic moments

Phase-space methods combined with metal oxide gas sensors are introduced for the first time. The usefulness of the method is assessed by analysing the transient response of metal oxide gas sensors either to a step change in gas concentration or to thermal modulation. The first application considered consists in sorting samples from the headspace of crisps according to their rancidity state. It is shown that using the new dynamic features results in an improvement of rancidity classification: 85.4% instead of the 68.7% reached when standard features are used (ΔG). In the second step, the method is applied to analyse the response of temperature-modulated micro-hotplate sensors. In this case, phase-space methods give the opportunity to use novel features that describe sensor trajectories while adsorption and reaction kinetics are altered by the temperature modulation. Tungsten oxide-based sensors are employed to detect ammonia, nitrogen dioxide and their binary mixtures at different concentrations. Dynamic features are used to build and validate PLS-DA classifiers and good results are obtained, not only ameliorating the gas identification rate (95.6%), but also dramatically reducing the time needed to perform measurements (less than 2% of the response transient is needed to perform the analysis).