A genetic algorithm(GA)optimization probabilistic neural network(PNN)diagnosis method(GA-PNN)is proposed to address the fault diagnosis problem of rolling bearing.Firstly,GA was used to optimize the diffusion constants in PNN net-works.Secondly,the vibration signals of rolling bearing under inner ring fault and normal condition were collected from the labo-ratory.Considering the defects of the acquisition system,the least square method(LMS)and exponential smoothing method were used to eliminate the drift and weak noise in the vibration signals.Thirdly,several time-domain feature parameters were extract-ed,and six different models were established according to the input variables involved in the establishment of the diagnosis model.Finally,six models were diagnosed and comprehensively analyzed by GA-PNN and PNN diagnosis model.The results show that GA-PNN can achieve more than 95％for the diagnosis of 6 models.However,PNN has a large difference in diagnosis results due to the setting of the diffusion constants.Besides,the diffusion constants and input variables will affect the diagnostic results of PNN.Therefore,from the perspectives of convergence error and test set diagnostic accuracy,GA-PNN is more suitable for rolling bearing fault diagnosis than PNN.
关键词
滚动轴承/故障诊断/概率神经网络/遗传算法/最小二乘法/指数平滑法
Key words
Rolling Bearing/Fault Diagnosis/Probabilistic Neural Network/Genetic Algorithm/Least Squares/Ex-ponential Smoothing
维普
万方数据