Detecting and recognizing chemical targets in untrained backgrounds with temperature programmed sensors

Applications for artificial olfaction typically require analytical performance in the context of diverse backgrounds. Therefore, to deal with practical challenges posed by chemical species recognition in the presence of pre-trained and untrained backgrounds, a desirable feature is the ability to rapidly detect fresh analyte introductions (foreground odor) and segment their contributions from the foreground-background response cocktail. Here, we present a simple approach for this purpose based on the moving-window pair-wise correlation between sensor responses measured at multiple temperatures. We show that pairwise-correlation across isotherm segments can be used as a robust measure to rapidly detect chemical events (onset and offset), as well as to track and compensate for sensor baseline changes due to background variations. We demonstrate this approach for the problem of identifying three toxic industrial chemicals-ammonia, hydrogen cyanide, and chlorine-in several untrained backgrounds. Additionally, we show that the proposed scheme could be used to reduce baseline differences in response signatures between sensors of equivalent manufacture and thereby allow training and testing using different but comparable sensors.