Bearing Fault Diagnosis Based on Cyclic Correntropy and One-dimensional Shallow Convolutional Neural Network
The traditional two-dimensional convolutional neural network(2D CNN)not only has high computational complexity and is easier to over fitting,but also is difficult to deal with low signal-to-noise ratio(SNR)signal effectively.In order to overcome the shortcomings of 2D CNN,a new fault diagnosis method based on cyclic correntropy(CCe)and one-dimensional shallow convolutional neural network(1D SCNN)is proposed.This new method(CCe-1D SCNN)makes fully use of the advantages of 1D SCNN and CCe,in which the 1D SCNN has simple structure and low computational complexity.Firstly,the cyclic correntropy function,the cyclic correntropy spectral density(CCSD)function and generalized degree of cyclostationary(DCS)of bearing fault vibration signal are calculated.Secondly,the one-dimensional normalized generalized degree of cyclostationary is used as the input layer of one-dimensional shallow convolutional neural network.The fault feature extraction and pattern classification are automatically realized by one-dimensional shallow convolutional neural network.Finally,the CCe-1D SCNN method is applied to fault feature extraction and pattern classification of motor bearing fault.The experimental results show that the CCe-1D SCNN technique can still maintain a high accuracy of pattern recognition in the case of very low signal-to-noise ratio,which is an effective method for automatic fault feature extraction and pattern recognition.
国家科技期刊平台
NSTL
万方数据
维普
