Remaining useful life (RUL) prediction of aircraft auxiliary power units (APUs) is often limited by sparse terminal degradation samples, because on-wing removal events are much fewer than routine monitoring records. Although A13 reports provide operating and performance observations during APU service, exhaust gas temperature (EGT) trajectories are discrete, noisy, and vary with load conditions, which limits the degradation information available for model training. To address this problem, this paper proposes a physics-guided data augmentation method for sparse APU degradation samples. EGT is first transformed into a degradation residual by removing the healthy baseline value. A multi-distribution fusion strategy is then used to generate residual trajectories by combining stable degradation trends with multiple perturbation components. Boundary, terminal-trend, and local-fluctuation constraints are incorporated to reduce unreasonable generated patterns. The generated residuals are reconstructed into EGT degradation trajectories and combined with original A13 parameters as augmented inputs for RUL prediction. Experiments on APU data show that the proposed method improves generated data quality and enhances RUL prediction under sparse terminal degradation samples.

Aircraft auxiliary power units; Remaining useful life prediction; Physics-guided; data augmentation