An Optimized Machine Learning Technique for Fault Diagnosis of Roller Bearing in a Motor Drive System

Early fault detection and diagnosis (FDD) is an essential trait in reducing the maintenance cost and increasing the reliability of the manufacturing system. Considerable advancements have been achieved in recent years toward creating a reliable and automated condition monitoring approach that can detect the occurrence of a fault in advance, permitting the maintenance to be performed proactively to prevent trivial problems from building into catastrophic failures. In this study, we present a radial basis kernel based optimized Support Vector Machine (SVM) for classifying vibrational data obtained from the case western reserve university (CWRU) bearing data set into either good or faulty classes. Instead of using the raw data directly into the model, our method first involves extracting the relevant statistical characteristics from the sensor data. The hyper-parameters C and Gamma of the Support Vector Classifier (SVC) are optimized, resulting in an optimal value that maximizes the accuracy score on both the training and test data. A comprehensive assessment revealed that employing a basic SVM classifier could yield improved accuracy, provided that the features are appropriately extracted and the hyper-parameters are effectively tuned. A distinct demarcation was observed in the majority of the extracted features, resulting in a notable increase in the performance measures. Furthermore, the Fast Fourier Transformation (FFT) was conducted to calculate the Power Spectral Density (PSD) of the signal. This allows for the analysis of the attributes and trends displayed by the data in the frequency domain.