Research on Bearing Remaining Useful Life Prediction Method Based on TCN-LSTM Hybrid Model
作者
Wenjie Hu,Pengfei Zhan
标识
DOI:10.1145/3804601.3804777
摘要
Accurately predicting the Remaining Useful Life (RUL) of bearings is of great significance for equipment health management. This paper proposes a hybrid model integrating Temporal Convolutional Network (TCN), Bidirectional Long Short-Term Memory (BiLSTM), and Light Gradient Boosting Machine (LightGBM) for RUL prediction of bearing vibration signals. Through a three-level collaborative architecture of “TCN multi-scale feature extraction — BiLSTM temporal dependency modeling — LightGBM feature selection regression”, combined with a multi-dimensional feature fusion strategy of time domain and frequency domain, the model achieves refined modeling of the bearing degradation process. Experiments on the IEEE PHM 2012 dataset show that the model's Relative Percentage Error (RPE) of fault time prediction reaches 8.3% and Exponential Transformation Accuracy (ETA) is 0.91, both superior to traditional models such as CNN-LSTM and single TCN. Verified by sliding window data augmentation and noise robustness tests, the model balances accuracy, efficiency, and stability, providing an efficient solution for bearing predictive maintenance.