Fusion-Based Dual-Task Architecture for Predicting the Remaining Useful Life of an Aeroengine

The prediction of the remaining useful life (RUL) of aeroengines is crucial for ensuring their safe operation and reducing maintenance costs. However, aeroengines are complex nonlinear systems that exhibit multiple degradation modes, making it challenging to extract predictive information from diverse feature fields. To address this issue, we propose a data-driven fusion-based dual-task architecture that explicitly models the degradation modes of aeroengines by leveraging degradation information to enhance RUL prediction. In our dual-task model, a residual network extracts feature information from observable values within a single flight, whereas a long short-term memory network captures feature information across multiple flights. These two models are fused into a unified RUL prediction model, which is subsequently fine-tuned using RUL labels reconstructed from degradation data. Evaluation of a data set containing comprehensive flight information demonstrates that the proposed method improves prediction performance by 13% compared to RUL prediction models that do not incorporate degradation information. Furthermore, comparisons with commonly used state-of-the-art methods confirm the superior performance and robustness of the proposed method.