Deep Belief Network and Dempster-Shafer Evidence Theory for Bearing Fault Diagnosis

Abstract

Bearing takes an important part in rotary machines. In industrial manufacturing systems, bearing fault diagnosis is a critical task which helps to reduce the cost for maintaining. This paper proposes a novel bearing fault diagnosis algorithm for rotary machine in which multiple sensors are installed using Deep Belief Network and Dempster-Shafer evidence theory. First, for signals from each sensor, a Deep Belief Network is used to extracted features. Each feature set generated by the corresponding Deep Belief Network is classifier by one softmax classifier. Finally, prediction results of all softmax classifiers are fused by DS evidence theory to generate the final prediction of bearing fault. Experiments are carried out with bearing data from Case Western Reserve University Data Central.