A hybrid method of support vector machine and Dempster-Shafer theory for automated bearing fault diagnosis

Abstract

The rapid growth of many critical industries in the past decades, such as power generation and oil and gas, has increased the demand for more reliable machines and mechanical parts. One of the most critical parts of a machine is the bearing, of which a failure can lead to total machine malfunction. Therefore, an effective bearing fault diagnosis is essential in ensuring the integrity of the machine. In recent years, the popular approach for bearing fault diagnosis isby analyzing the bearing signal using advanced processing algorithms such as wavelet analysis, empirical mode decomposition, Hilbert-Huang transform, etc. The success of these methods, however, is highly dependent on the experience and knowledge of the individual personnel. As such, the automated bearing fault diagnosis provides an alternative solution to this pitfall. This paper studies the effectiveness of a hybrid SVM-DSas compared to SVM models for automated bearing fault diagnosis. Results show that the proposed SVM-DS method increased the accuracy of the diagnosis of SVM from 82% to 89% by further refining and eliminating the conflicting results of SVM. Therefore, the hybrid SVM-DS model was found to be more superior and effective than the sole SVM approach for automated bearing fault diagnosis.