PGI-RUL: Prior-Guided Generative Imputation and Physics-Informed Prediction for Aero-Engine RUL under Sparse and Noisy Conditions

Predicting the Remaining Useful Life (RUL) of aero-engines is a critical safety task that depends on highdimensional sensor data. In Industrial Internet of Things (IIoT) environments, data integrity is often compromised by sparse missing values and noise, leading to performance degradation in existing prognostic models. To solve these challenges, this paper presents a novel framework, Prior-Guided Generative Imputation with Physics-Informed RUL Prediction (PGI-RUL). First, we introduce a Bayesian Augmented Tensor Factorization (BATF) module to capture the global low-rank structure of the corrupted sensor data, providing a robust structural prior. This prior then guides a Conditional Diffusion Model to iteratively refine local temporal details, ensuring high-precision imputation, even with high missing rates. Lastly, a Physics-Informed PatchTST (PI-PatchTST) predictor is developed, incorporating monotonic degradation constraints into the loss function to guarantee physically consistent prognostic results. Extensive experiments on the C-MAPSS and N-CMAPSS datasets show that PGI-RUL outperforms state-of-the-art baselines in both imputation accuracy and RUL prediction, particularly under severe conditions with 50% missingness and high noise.