Multi-Scale Patch-based Feature Fusion Model for Aero-engine Remaining Useful Life Prediction

Predicting the Remaining Useful Life (RUL) of aeroengines is of great significance for engine system maintenance and the safety of aviation. Existing multi-scale RUL prediction models typically use CNN, which lack the extraction of temporal features from local data and seldom consider the dynamic selection of global information. To address this issue, a patch-based feature fusion approach for aero-engine RUL prediction is proposed in this paper. Firstly, this method divides the original data into multiple patches along the temporal dimension based on the given different patch sizes. Subsequently, the LSTM model is employed to process the short-term sequential data within each patch, thereby extracting local temporal features. Following this, the self-attention mechanism is utilized to dynamically focus on and select global features among patches of different scales, effectively integrating global information. Moreover, one-dimensional convolutional operations are incorporated to extract the overall degradation features of the model, thereby enhancing its ability to perceive the degradation trends of the aero-engine. Ultimately, based on the fusion of these multi-dimensional features, the model is capable of accurately predicting the RUL of the aero-engine. The effectiveness of the proposed method is validated using the Commercial Modular Aero-Propulsion System Simulation (C- MAPSS) dataset provided by NASA. Comparative and ablation experiments on this dataset demonstrate that the method can adapt to more complex engine degradation processes and perform better in predicting the RUL of engines under relatively more complicated conditions.