Bearing Remaining Useful Life Prediction based on TCN-Transformer Model

Bearing remaining useful life (RUL) prediction is the process of estimating the RUL of bearings according to the current condition and historical data, which helps to determine the best time for equipment repair, replacement and maintenance in advance. The Long Short Term Memory (LSTM) network is widely used in the field of bearing RUL prediction, but suffering from the problem of weak long-term prediction ability and low prediction accuracy, this paper proposes Temporal Convolutional Network—Transformer (TCN-Transformer) model solving these problems. Firstly, in feature extraction, the vibration signal is decomposed using empirical modal decomposition (EMD) technique to extract the sensitive feature, and the kernel principal component analysis (KPCA) technique used to obtain the integration of sensitive features. Then, the time domain convolutional network (TCN) and residual structure are introduced into Transformer network to further improve the model's ability to extract features and predict RUL. Experiments on bearing RUL prediction are using the TCN-Transformer model and the PHM2012 bearing public dataset. The experimental results show that the bearing RUL prediction accuracy of TCN-Transformer model is significantly improved.