Bearing Degradation Prediction by WPD and DPNN: Introducing a Novel Deep Learning Method

Abstract

In this article, a novel method of deep learning based on wavelet transform and deep perceptron neural networks (DPNNs) is proposed to predict the remaining useful life (RUL) of bearings. The proposed approach first extracts from the recorded signals the energy features by the wavelet packet transform. After training on these data, the DPNN model can be used to predict the RUL of a bearing. To verify the model, the proposed DPNN based on wavelet packet transform is compared with a least-squares support vector machine (LS-SVM) and long short-term memory (LSTM). The experimental results illustrate that DPNN can effectively predict the RUL of the bearing and is superior to the LS-SVM and LSTM in terms of prediction performance.