RUL prediction of rolling bearings based on improved empirical wavelet transform and convolutional neural network

Accurate prediction the remaining useful life (RUL) of rolling bearings under complex environmental conditions is crucial for prognostics and health management (PHM). In this paper, A new method for rolling bearing RUL prediction based on improved empirical wavelet transform (IEWT) and one-dimensional convolutional neural network (1D-CNN) is proposed to overcome the interference of noise and other disturbance signals. Firstly, in view of the problem of too many spectrum divisions in the traditional empirical wavelet transform (EWT) process, the mutual information value is used to re-determine the frequency band demarcation point in the EWT. The IEWT method is introduced to adaptively divide the original vibration signal to obtain a series of empirical mode functions (EMFs). Secondly, the effective components after IEWT decomposition are extracted by mutual information and kurtosis criteria and used to extract multi-dimensional time-frequency domain features. Finally, the 1D-CNN is constructed with the percentage of remaining life as the tracking metric to predict the RUL of the bearings. Based on two publicly available rolling bearing datasets, the model proposed in this paper have high prediction accuracy, which is better than other prediction models. Compared to other methods, its mean absolute error (MAE) and root mean square error (RMSE) are reduced.

Automated summary

This paper proposes a new method for predicting the remaining useful life (RUL) of rolling bearings based on improved empirical wavelet transform (IEWT) and one-dimensional convolutional neural network (1D-CNN). The method overcomes the interference of noise and other disturbance signals. By re-determining the frequency band demarcation point in the empirical wavelet transform (EWT) using mutual information value, the IEWT adaptively divides the original vibration signal into a series of empirical mode functions (EMFs). The effective components after IEWT decomposition are extracted using mutual information and kurtosis criteria, and multi-dimensional time-frequency domain features are used for prediction. The proposed model shows high prediction accuracy and performs better than other prediction models, reducing the mean absolute error (MAE) and root mean square error (RMSE).