Ordinary neural-network-based bearing fault diagnosis methods do not strong enough in extracting features.For sce-narios with limited data samples or low robustness,these methods cannot obtain accurate diagnostic results.This paper devel-ops a fault diagnosis method for rolling bearings based on integration of CNN(Convolutional Neural Networks)with TL(Transfer Learning),i.e.TCNN(Transfer Convolutional Neural Network).The AlexNet network,one of the benchmark CNN models pretrained with the ImageNet dataset,is selected in the proposed transfer CNN.On this basis,TL is combined to freeze the underlying parameters in existing prediction models,and only train and update upper level parameters.When the number of iterations reaches a certain level,this approach can effectively improve the accuracy rate of fault diagnosis.The experimental results show that the TCNN can improve the noise resistance and generalization ability of the model with smaller dataset,and improve the accuracy of fault diagnosis.
rolling bearingfault diagnosisCNN(Convolutional Neural Networks)TL(Transfer Learning)
NETL
NSTL
万方数据
