Discrete wavelet transform and principal component analysis based vapor discrimination by optimizing sense-and-purge cycle duration of SAW chemical sensor transients

摘要

Temporal evolution of response of chemical vapor sensors for step function like exposure and purge carries identity signatures of chemical analytes hidden in transient shapes. Representation of the transient response shapes by discrete wavelet transform (DWT) and principal component analysis (PCA) of wavelet approximation coefficients provides an efficient procedure for denoising and feature extraction. The present work is concerned with identification of volatile organic compounds (VOCs) by transient response analysis of polymer-coated surface acoustic wave (SAW) sensors. The sorption and diffusion kinetics of chemical molecules and polymer thickness determines the shapes of transient responses. The equilibration times for different vapor species are different for a given polymer coating. For a given exposure and purge cycle different vapor species with varied diffusion coefficients reach different stages of equilibration, hence loadings. Therefore, a pre-equilibrium termination of vapor exposure and purge durations can be expected to generate transient response shapes much richer in information compared to fully equilibrated condition. In this paper, we explore this aspect by carrying out a simulation based analysis of SAW sensor transients. The sense-and-purge cycle durations were varied for sensing of seven volatile organics by a polyisobutylene (PIB) coated SAW sensor. The transient signals were represented by DWT approximation coefficients based on Daubechies-2 mother wavelet. The vapor discrimination ability of an exposure and purge cycle was defined by the `class separability measure' calculated in the PCA generated feature space of DWT coefficients. The simulation experiments were carried out by varying exposure-and-purge durations and by adding different noise levels. It has been concluded that for obtaining best discrimination results in presence of noise the sense-and-purge duration must be optimized.