Multi-output Structure Combined with Separable Convolutional GRU Model for Aero-Engine RUL Prediction

Establishing an accurate and efficient mechanism to predict the Remaining Useful Life (RUL) is the core of aero-engine health management technology. Recently, techniques and frameworks related to deep learning have been shown to meet the needs of RUL prediction, and many models have been successfully applied to RUL prediction tasks. However, related research is still in its infancy, and there is still potential for enhancement and improvement in terms of prediction performance and model structure. In this paper, based on previous related efforts, we propose a RUL prediction model with multiple output structures incorporating separable convolutional gated recursive units. The model simplifies the number of required parameters and the computational process as much as possible to meet the needs of small mobile devices or embedded systems, while ensuring the prediction performance. Finally, the model is experimentally validated on the C-MAPSS dataset and compared with other models to demonstrate the feasibility of the model and achieve excellent prediction results.