Nominal features-based class specific learning model for fault diagnosis in industrial applications

Abstract Fault Detection and Isolation (FDI) is an initial stage of the real-time fault diagnosis in industrial applications. The increase in a number of faults will increase the size of the features that limit the accuracy. This paper proposes the suitable technique to extract the relevant features and classify the faults accordingly. Initially, the preprocessing removes the unfilled entries in the fault dataset (after the data collected from the sensor). Then, the Minimal Relevant Feature extraction predicts the features that correspond to the six types of fault classes. The minimum and maximum ranges of voltage, current, vibrations and speed due to the above classes regarded as the features. The modified objective function for class-specific support vector machine (CS-SVM) classifies the fault classes which highly contribute to the early diagnosis. The relevant feature prior to the Classification increases the accuracy effectively. The variations of voltage, current, and motor speed according to the injection of vibration faults from the motor bearings determine the impacts effectively. The comparison between the proposed Minimal Relevant Features-based Classification (MRFC) with existing SVM regarding the accuracy, precision, recall, sensitivity, specificity and coefficient (Jaccard, Dice and Kappa) confirms the effectiveness of proposed MRFC in earlier fault diagnosis in industrial applications.