Regression Approach for Concentration Estimation of Volatile Chemical Compounds by E-nose Response Analysis

Abstract

This study presents regression methods for response analysis of electronic nose (E-nose) in concentration estimation of volatile organic chemicals (VOCs). Particularly study focuses on nonlinear support vector regression (SVR) methods and its performance comparison with linear regression analysis (LRA) method. E-nose system is realized with combination of polymers and carbon molecular sieves based odor filtering and 8 metal oxide semiconductor (MOX) sensors based sensing system. MOX sensor resistance Ra, due to chemical vapor adsorption is measured for five target VOCs including acetone, benzene, ethanol, pentanal, and propenoic acid and used in analysis. VOCs are exposed on sensor array at distinct concentration in between 3‒500 parts per million (ppm). Scatter plot and Spearman's rank correlation coefficient (ρ) is used to determine the dependence of sensor resistance on VOCs concentration prior to the regression analysis. Coefficient of determination R (R-squared) is computed to compare the performance of build regression models and to search optimal filtering material for VOCs quantification. 6 sensor in array containing pure polyvinyl chloride (PVC) as filtering material results maximum value of R = 0.9825 for ethanol vapors with radial basis kernel function based SVR method. Though the 5 sensor in array results minimum value of R = 0.0864 for pentanal vapors using linear kernel function based SVR method.