Transformer-LSTM Based Health State Prediction Method for Rotating Machinery

Abstract: As a key rotating component in mechanical equipment, the health of the bearing directly affects the operational stability and safety of the equipment. Deep learning technology has achieved remarkable results in bearing fault diagnosis and health assessment in recent years. This paper proposes a two-branch parallel network model based on a Transformer and LSTM for the health assessment of large rotating equipment. The Transformer sub-module extracts global features of the bearing vibration signal, fully capturing the long-range dependence of the equipment operating state in the signal. At the same time, the LSTM sub-module is used to model local temporal features, which enhances the model's ability to capture non-linear time series data. This hybrid model fully combines the advantages of the Transformer's long sequence feature recognition with LSTM's local sequence modeling capabilities, improving the accuracy and robustness of rotating equipment health assessment. Experimental results show that the proposed model can effectively identify the health status of the bearing under different fault modes and has better diagnostic performance than traditional methods. This method can support bearing health management and fault prediction and has excellent practical application value.