Improving Residual Life Prediction of Rolling Bearings with CNN-LSTM Model
When using Convolutional Neural Networks(CNN)and Long Short-Term Memory(LSTM)models for predicting the remaining life of rolling bearings,the accuracy of the prediction results is influenced by experimental parameters.Therefore,a method was proposed to intervene in the model parameters using the Whale Optimization Algorithm,reducing the complexity of hyperparameter tuning.Firstly,three feature evaluation metrics of relevance,monotonicity,and robustness,as well as the similarity correlation coefficient,were used to rank features by weighted sorting,establishing a feature selection system.Secondly,the basic structure of CNN-LSTM was employed,and parameter optimization was carried out using the embedded Whale Optimization Algorithm.Finally,with the PHM2012 rolling bearing dataset,the prediction of bearing remaining life was achieved,validating the superior predictive performance of the improved model.
rolling bearingsCNN-LSTMresidual life predictionwhale algorithm
万方数据
维普
