A Novel GA-BP Neural Network for Wireless Diagnosis of Rolling Bearing

Abstract

Rolling bearings are pivotal components in industrial rotating equipment and the issue of fault will occur inevitably due to long-term abrasion. This study proposes a novel GA-BP neural network (GA-BPNN) algorithm to improve the accuracy of fault diagnosis of industrial rolling bearings. The genetic algorithm (GA) is employed to optimize the structure, initial weight and threshold of BP neural network, which can improve the ability of diagnosis and reduce the time of network training. At first, the structure of network is determined so that the optimal parameters of GA can be given, then the population of GA will be encoded. At second, the individual fitness function is calculated based on the test error norm of the BP neural network, as a criterion for distinguishing the individual from individual. The optimal weight and threshold are obtained by means of the corresponding selection, cross and variation, etc. Finally, the simulation experiment is carried out in Matlab and massive vibration experimental data of industrial rolling bearing are utilized. To verify the ability of the proposed novel GA-BPNN, compared with BP neural network algorithm (BPNN), the convergence speed and accuracy of GA-BPNN are better. The results of experiment illustrate that the optimized GA-BPNN method can identify the fault-type quicker, and has higher feasibility, which can be used to assist diagnosis of industrial bearing and improve efficiency.