Olfactory Imaging: An Electronic Nose Using Tiered Artifical Neural Networks and Quartz Piezoelectric Gas Sensors

Abstract

Our group is interested in the potential use of gas sensors for detection of key odorants in industry and as alternative sensory mechanisms for guidance control in robots. It has been recently shown that chemically modified, resonating quartz piezoelectric (QPZ) crystals exposed to different odorants, generate characteristic response patterns (termed "kinetic signatures"). To demonstrate their utility in an electronic olfactory system, artificial neural networks (ANNs) were trained using the back propagation method, to associate the kinetic signature responses of 6 differently treated sensors to 18 trialed odorants. Arranging each of the separate networks corresponding to each sensor in a layer-like fashion (hereafter referred to as tiers to avoid confusion with network layers), the weight states of the output processing elements (PEs) of the amalgamated ANN combine to produce weighted, two dimensional 'olfactory response maps' that uniquely identify each of the odorants. Using simple image processing techniques, we discuss how these response maps can give an automated system a degree of feedback as to its physical state, allowing it to detect and potentially rectify problems encountered during normal operation.