Enhancing chemical identification efficiency by SAW sensor transients through a data enrichment and information fusion strategy--a simulation study

摘要

The paper proposes a new approach for improving the odor recognition efficiency of a surface acoustic wave (SAW) transient sensor system based on a single polymer coating. The vapor identity information is hidden in transient response shapes through dependences on specific vapor solvation and diffusion parameters in the polymer coating. The variations in the vapor exposure and purge durations and the sensor operating frequency have been used to create diversity in transient shapes via termination of the vapor-polymer equilibration process up to different stages. The transient signals were analyzed by the discrete wavelet transform using Daubechies-4 mother wavelet basis. The wavelet approximation coefficients were then processed by principal component analysis for creating feature space. The set of principal components define the vapor identity information. In an attempt to enhance vapor class separability we analyze two types of information fusion methods. In one, the sensor operation frequency is fixed and the sensing and purge durations are varied, and in the second, the sensing and purge durations are fixed and the sensor operating frequency is varied. The fusion is achieved by concatenation of discrete wavelet coefficients corresponding to various transients prior to the principal component analysis. The simulation experiments with polyisobutylene SAW sensor coating for operation frequencies over [55-160] MHz and sensing durations over [5-60] s were analyzed. The target vapors are seven volatile organics: chloroform, chlorobenzene, o-dichlorobenzene, n-heptane, toluene, n-hexane and n-octane whose concentrations were varied over [10-100] ppm. The simulation data were generated using a SAW sensor transient response model that incorporates the viscoelastic effects due to polymer coating and an additive noise source in the output. The analysis reveals that: (i) in single transient analysis the class separability increases with sensing duration for a given frequency of operation, and also with frequency for a given sensing duration, and (ii) the information fusion based on both the multiple sensing cycles and the multiple sensing frequencies enhances the class separability by nearly an order of magnitude.