A Hybrid Genetic Algorithm and Back-Propagation Classifier for Gearbox Fault Diagnosis

Abstract

An Artificial Neural Network (ANN) classifier trained by a hybrid GA-BP method for diagnosis of gear faults is presented here that can be incorporated in an online fault diagnostic system of vital gearboxes. The distinctive features obtained from vibration signals of a running gearbox; that was operated in normal and with faults induced conditions were used to feed the GA-BP hybrid classifier. Time domain vibration signals were divided in 40segments. From each segment features such as magnitude of peaks in time domain and spectrum along with statistical features such as central moments and standard deviations were extracted to feed the classifier. Based on the experimental results it was shown that the GA-BP hybrid classifier can successfully identify gear condition. It was also shown that the network trained by GA-BP hybrid method performs much better than ANN that is trained by standard BP or GA individually. Further, it was also shown that if prior to extraction of features; the vibration signals are pre-processed by Discrete Wavelet Transform (DWT) then efficacy of the GA-BP hybrid is significantly enhanced.