关键词: 轴承故障诊断, 经验模态分解, 最小均方算法, 离散小波变换, GoogLeNet模型

Abstract: Aiming at the problems that the traditional mechanical bearing fault diagnosis model is easy to be disturbed by system noise and low efficiency of feature recognition,a bearing fault diagnosis method based on deep modeling analysis of signal inherent mode was proposed.The collected bearing vibration signals were subjected to Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to obtain local characteristic signals of different time scales.Correlation coefficients were used to identify and remove false intrinsic mode function.The remaining IMF components were denoised and reconstructed by Variable Step-Size Least Mean Square algorithm (VSSLMS).Then,the vibration signal after noise reduction was obtained by Discrete Wavelet Transform (DWT),and the feature was enhanced by morphological operation.The improved GoogLeNet network model was used to train the feature map,and the feature classification was completed by Softmax classifier,so as to realize the bearing fault diagnosis.The proposed fault diagnosis method was applied to the bearing fault data set under different working conditions,and the test results showed that the diagnosis accuracy was higher under noise interference.

Key words: bearing fault diagnosis, empirical mode decomposition, least mean square algorithm, discrete wavelet transform, GoogLeNet model