A New Method of Bearing Remaining Useful Life Based on Life Evolution and SE-ConvLSTM Neural Network

The degradation process of bearing performance in the whole life cycle is a complex and nonlinear process. However, traditional neural network-based approaches usually consider as linear, which does not accord with actual situation degradation. To overcome this shortcoming, rolling bearing’s remaining useful prediction method based on Squeeze-and-Excitation-Convolutional long short-term memory (SE-ConvLSTM) network was proposed construction new health index evolution. considered change rule indicator during evolution bearings, then constructed by using SE-ConvLSTM network, effectively improving model accuracy training efficiency. Firstly, original data are filtered denoised Ensemble Empirical Mode Decomposition. Combined Principal Component Analysis (PCA) dimensionality reduction Local Outlier Factor (LOF) algorithm, interval divided. Then, model, bearings predicted particle filter double exponential model. compared other related methods PHM2012 dataset, results show that has higher predictions. Compared method, division lifespan practical significance.