Remaining Useful Life Prediction for Train Bearing Based on an BiLSTM-KAN

As one of the key components of the train bogie, accurate bearing remaining useful life (RUL) prediction and timely maintenance play a vital role in the safe and reliable operation of the train. The environment of rail transit trains is complex, and the vibration signal of train bearings shows the characteristics of non-linearity and non-smoothness. Meanwhile, the safety requirements of rail transportation system are comparatively demanding, and the time series RUL prediction of bearings should consider the long-term and multi-data problems. For the complex degradation process of rail transit train bearings, a hybrid bidirectional long and short-term memory (BiLSTM) networks and Kolmogorov-Arnold Networks (KAN) RUL prediction method is proposed. Based on the BiLSTM network, KAN is used to replace the fully connected layer, which improves the parameter utilization and enhances the ability to obtain the nonlinear pattern information in the hidden state of BiLSTM. Compared with the traditional time-series prediction method, the method has better prediction accuracy, stronger interpretability, and is more suitable for the prediction of train bogie bearing RUL in high safety requirement scenarios.

Keywords: Train bearing, RUL, BiLSTM, KAN.