A wiener-prior-guided hierarchical network for aero-engine RUL prediction with uncertainty quantification

Abstract Accurate aero-engine remaining useful life (RUL) prediction is essential for ensuring flight safety and enabling reliable maintenance strategies. Owing to the strong coupling among engine components and the highly nonlinear characteristics of degradation processes, existing methods often produce non-monotonic RUL trajectories inconsistent with physical degradation mechanisms, whilst failing to explicitly account for end-of-life risks and prediction uncertainties. To address these challenges, this paper proposes a domain knowledge integrated prediction-correction framework, named CTWR, for aero-engine RUL prediction. First, stochastic degradation modeling is performed via a generalized Wiener process to extract a robust health indicator and construct a degradation parameter library as structured priors. Second, a hybrid network employs multi-scale feature fusion to capture global–local dependencies, achieving joint preliminary estimation of RUL and degradation parameters from multivariate sensor sequences. Third, a knowledge-regularized adaptive correction mechanism refines the predictions by embedding risk-aware and monotonicity constraints, while Monte Carlo dropout quantifies uncertainty to generate confidence intervals. Experimental results on C-MAPSS and N-CMAPSS demonstrate that CTWR outperforms multiple state-of-the-art methods in terms of root mean square error and NASA-score. The generated RUL trajectories exhibit greater smoothness and better alignment with degradation patterns, while providing valuable uncertainty information, thereby validating the proposed framework’s effectiveness.