| || Development of Surface Acoustic Wave Electronic Nose using Pattern Recognition System
Author : Jha, S.K.;Yadava, R.D.S.
Source : Defence Science Journal ; Vol:60(4) ; 2010 ; pp 364-376
Subject : 53 Applied Physics;681.586 Sensors;Defence Science Journal
Keywords : SAW sensor array;electronic nose;TNT vapour detection;SVD denoising;pattern recognition;pattern recognition system;singular valve decompostion based denoising
Abstract : The paper proposes an effective method to design and develop surface acoustic wave (SAW) sensor array-based electronic nose systems for specific target applications. The paper suggests that before undertaking full hardware development empirically through hit and trial for sensor selection, it is prudent to develop accurate sensor array simulator for generating synthetic data and optimising sensor array design and pattern recognition system. The latter aspects are most time-consuming and cost-intensive parts in the development of an electronic nose system. This is because most of the electronic sensor platforms, circuit components, and electromechanical parts are available commercially-off-the-shelve (COTS), whereas knowledge about specific polymers and data analysis software are often guarded due to commercial or strategic interests. In this study, an 11-element SAW sensor array is modelled to detect and identify trinitrotoluene (TNT) and dinitrotoluene (DNT) explosive vapours in the presence of toluene, benzene, dimethylmethylphosphonate (DMMP) and humidity as interferents. Additive noise sources and outliers were included in the model for data generation. The pattern recognition system consists of: (i) a preprocessor based on logarithmic data scaling, dimensional autoscaling, and singular value decomposition-based denoising, (ii) principal component analysis (PCA)-based feature extractor, and (iii) an artificial neural network (ANN) classifier. The efficacy of this approach is illustrated by presenting detailed PCA analysis and classification results under varied conditions of noise and outlier, and by analysing comparative performance of four classifiers (neural network, k-nearest neighbour, naïve Bayes, and support vector machine).