An online incremental support vector machine for fault diagnosis using vibration signature analysis

A major challenge facing modern industries today is the appropriate handling of faults. A machine fault that goes untreated results into undesirable downtime. The information about machine's health can be obtained by monitoring its vibration signals. In this paper, we apply Support Vector Machine (SVM) to vibration signals in order to detect and classify faults in rotating machines. As the environment changes with time, the machines undergo the phenomena of concept drift and the classifier loses its accuracy. To tackle this issue, we reformulate the classical batch-trained SVM and solve it by an incremental approach, referred as Incremental Support Vector Machine (ISVM). Moreover, an online method is introduced to learn over newer samples of data and make the SVM model adapt to the changing concept trend. The simulation are carried out using Intelligent Maintenance System (IMS) bearing vibration data-set. The results obtained show that the Online Incremental SVM algorithm outperforms the static classical SVM.