Remaining Useful Life Prediction of Rolling Bearings Using Correlation Coefficient Based Long Short-Term Memory Neural Network

Abstract

Rolling bearings play a crucial component for machinery equipment, which affects the operation efficiency and safety of the machinery equipment. The remaining useful life (RUL) prediction of rolling bearings is helpful for intelligent maintenance of the mechanical equipment. Therefore, an improved correlation coefficient based long short-term memory (LSTM) neural network model is presented in this paper to realize the RUL prediction of rolling bearings. First, the time domain, frequency domain and time-frequency domain features are extracted from the rolling bearing vibration data. Meanwhile, the noise reduction and normalization of the extracted features are conducted. Then, some important features that represent the bearing degradation trend are selected as the training data set through the correlation coefficient method to construct the RUL prediction model of online rolling bearings. Finally, the superiority of the presented RUL prediction model is validated baesd on the XJTU-SY rolling bearing data. Results on the experimental data indicate that the presented RUL prediction model has high generalization ability and high accuracy.