Identification of Dominant Perturbation Segments for Voltammetric E-Tongue Sensor Applications

The perturbation voltage signal of a voltammetric electrochemical sensor system is an essential aspect that brings out characteristic information about an analyte. The shape and sequence of the input voltage signal should vary with the required application and analyte. One such perturbation input, known as large amplitude pulse voltammetry (LAPV) signal, is used extensively for electronic tongue (E-Tongue) applications. The study suggests that the response obtained from a voltammetric E-Tongue system contains significant redundancy that requires elimination using different feature extraction techniques. This work investigates a novel approach to estimate an information-rich, customized input sequence of pulses in place of the entire LAPV signal for a given analyte sample, thereby enabling feature reduction. The dominant perturbation segments of the response have been identified by using a binary-coded genetic algorithm (GA) optimization to maximize the separation among response clusters. Multivariate data analysis techniques, such as principal component analysis (PCA) coupled with separability index (SI) values, are often used to explore E-Tongue response clusters for different samples in a visual and quantitative manner. Therefore, the SI value of response clusters has been adopted as the fitness value for optimization. In this work, experiments were performed with different standard samples at various concentration levels and a complex sample. The analysis was carried out using a voltammetric sensor system. It has been found that the responses obtained from specifically modified perturbation signals result in more distinct and well-defined clusters compared to that obtained using an entire set of data from a full-length LAPV signal.