Role of Exposure and Recovery Transients in Classification of Gases/Odors With Thick Film Sensor Array

Abstract

In the present study, the effect of exposure (falling transients) and recovery (rising transients) responses of thick film gas sensor array on the classification and quantification accuracy of gas sensing system has been studied. The exposure and recovery data were extracted from the already published dynamic responses of a thick film gas sensor array. These datasets were individually fed to the multilayer feed-forward neural network with back propagation (BPNN) algorithm. The classification accuracy obtained for exposure responses was 81.2% while 84.3% for recovery responses. The same trend was obtained after applying principal component analysis, and our newly proposed average slope multiplication (ASM) feature techniques for data preprocessing. ASM transformed data showed 93.7% and 95.8% classification accuracy for exposure responses and recovery data, respectively. The simultaneous quantification results also followed the same trend, where ASM transformed data provided maximum 89.6% and 91.6% accuracy for exposure and recovery data, respectively, with relatively very less number of epochs required for network learning with simpler neural architecture. Thus, recovery or response readings alone can be used with the proposed ASM feature technique to get promising results in gas identification and/or quantification system. This can reduce data complexity, save computational time and, thus, can help in realizing real time gas sensing systems.