Bearing Remaining Useful Life Prediction Based on BERT Fine-tuning

Rotating machinery are the cornerstone of indus-trial, and the prediction of critical bearing life model has always been a research hotspot. However, the prediction accuracy of most bearing remaining useful life (RUL) prediction networks still needs to be improved, and the training time can be further shortened. This paper proposed a fine-tuning BERT model for predicting the RUL of bearings. Based on the output form of the BERT model word segmentation module and the NLP task, the health index from bearing vibration data was extracted, and combined with the BERT model to achieve bearing RUL prediction. The performance of the proposed method was verified on the XJTU-SY bearing dataset. The results showed that the BERT model achieved superior RUL prediction performance in training speed and prediction accuracy, indicating its great potential for the field.