Sensor Faults Detection and Classification using SVM with Diverse Features

Sensors in industrial systems fault frequently leading to serious consequences regarding cost and safety. The authors propose support vector machine-based classifier with diverse time- and frequency-domain feature models to detect and classify these faults. Three different kernels, i.e., linear, polynomial, and radial-basis function, are employed separately to examine classifier's performance in each case. Furthermore, the respective kernel scales, delta and p of radial-basis function kernel and polynomial kernel, are varied manually to obtain the optimal values. Leave-one-out cross validation is adopted to overcome the overfitting problem. The dataset was acquired from a temperature-to-voltage converter through Matlab and Arduino Uno microcontroller. The efficiency in terms of percent accuracy of proposed time- and frequency-domain feature models can be seen in experimental results.