Remaining Useful Life Prediction of Aero-engine Based on Multi-scale Spatio-temporal Attention Mechanism

Aero-engine remaining life prediction technology aims to mine engine degradation laws from massive sensor monitoring data, accurately predict remaining service life., and provide effective early warning before engine failure occurs, so as to avoid immeasurable losses caused by failures during flight missions. Existing methods tend to focus on sensor monitoring information at the final stage of engine operation and cannot effectively capture important features contained in long sequences. In this paper, a time series prediction model based on a multi-scale spatiotemporal attention mechanism is proposed to improve the situation. This model models the feature importance in the temporal and spatial dimensions of sensor data, so as to realize the amplification of important features and the suppression of interference factors, and improve the ability of the model to capture key sequence information. To test the effectiveness of this model, the simulation is carried out on four aero-engine datasets released by NASA. The primary contribution of this paper lies in its application of the multi-scale spatio-temporal attention mechanism to the field of aero-engine remaining life prediction, with a focus on numerical experiment. The results show that our model has higher life prediction accuracy than traditional data-driven methods and classic deep learning methods.