An Automatic Approach to Diagnose Bearing Defects Using Time-Domain Analysis of Vibration Signal

Bearing defects are the most frequent occurring faults in any electrical machine. In this perspective, this paper presents a novel time-domain methods incorporating feature reduction method and back propagation feedforward neural network (BPNN) to identify bearing defects. For this, thirty-six standard vibration datasets related to healthy, inner raceway, and ball defects were derived from the Case Western Reserve University (CWRU) website. Four single point defects levels as 7, 14, 21, and 28 mils of inner raceway and ball defects were investigated for effective diagnosis of bearing defects. Initially, nine time-domain features were extracted from each vibration datasets, and then these features were ranked using Fisher’s ranking method to selected top four most discriminating features for effective classification of bearing conditions using BPNN algorithm. The effectiveness of proposed scheme to diagnose bearing defects was corroborated using performance parameters as accuracy (ACC), sensitivity (SE), and specificity (SP). The proposed algorithm has achieved maximum fault classification ACC as 94.87%.