Data-Driven Fault Classification Using Support Vector Machines

Detecting faulty condition of rolling-element bearings is significant in improving system reliability and preventing machine failure in industrial operations. In this paper, a machine learning pipeline is developed using filtered data through time domain features to train support vector machines with radial basis function, polynomial and linear kernels for multi-level fault diagnosis and classification. Overall accuracy rate and F-score values were used as figures of merit to evaluate and validate the performance of the machine learning model. SVM classifier showed significantly high overall accuracy rate of 91% to 99% and F-score of 0.81 to 0.99 with time domain statistical features due to the capability of this method in elimination of irrelevant features as well as reducing the dimensionality of the data. In addition, the high accuracy rate of SVMs in class-specific detection of localized bearing faults show the significant potential of data-driven classification modeling for fault detection in industrial and manufacturing operations.