Remaining useful life prediction of a small sample of aero-engine based on an improved gray Markov model

• First hybrid data-physics architecture: fusion of feature engineering and adaptive gray Markov model, breaking through the bottleneck of aero-engine degradation modeling under the condition of small samples (n≤50). • Dynamic parameter optimization algorithm: Based on genetic algorithm to realize self-tuning of response parameters in time domain, and improve 12.7% accuracy of late failure prediction compared with traditional model. • Industrial-grade validation system: 2.04% MAPE error reduction on NASA C-MAPSS benchmarks, verifying the robustness of the method under real working conditions. • Migration engineering framework: establish a complete technology chain from algorithm innovation to maintenance decision-making, and provide zero-sample migration solutions for high-value equipment. Predictive maintenance of high-value aero-engines faces a serious challenge: the lack of sufficient failure samples for accurate remaining service life estimation, especially in the case of nonlinear degradation. To address this issue, this study presents a novel data-physics fusion framework using the NASA turbine engine public dataset as the operational data for the study. First, a two-stage feature engineering strategy is designed: (1) Spearman correlation analysis identifies degradation-sensitive physical parameters. (2) Principal component analysis reduces dimensionality while preserving degraded trajectory patterns. Second, a gray-Markov hybrid model is developed to capture nonlinear wear dynamics, where polynomial fitting simulates accelerated failure modes in late life. To enhance adaptability, a genetic algorithm dynamically optimizes the time response parameters of the grey model, overcoming the limitation of a fixed structure in traditional methods. The improved model achieves an average absolute percentage error reduction of 2.04% compared to the original gray Markov model. This research fills the gap between data scarcity and intelligent prediction, providing a viable informatics solution for high-cost devices.