Key Technologies for Predictive Maintenance of General Aviation Aircraft Based on Digital Twin and Physics-Data Fusion

Abstract

Predictive maintenance is crucial for enhancing the reliability of aircraft and reducing operational costs. It involves analyzing various components of the aircraft and reviewing historical logs to identify previous failures. This study presents a digital twin maintenance system that integrates several physical models with real-time sensor analytics. Employ dynamic fusion to integrate the virtual model with real aircraft data to provide a robust flying machine and precise Remaining Useful Life (RUL) prediction in scenarios with limited data. Incorporate physics-informed neural networks and mechanistic failure principles to achieve interpretable and physics-consistent prognostics. This paper addresses the issue of insufficient real-world fault data by utilizing a synthetic dataset, which is generated through simulation technology and virtual fault injection. Additionally, it employs a framework incorporating Bayesian deep learning to enhance uncertainty and robustness. The system-level maintenance approach is enhanced using a genetic algorithm based on cost, risk, and fleet availability. Experimental validation on piston engine general aviation aircraft demonstrates superior defect detection accuracy, remaining useful life prediction accuracy, maintenance cost reduction, and operational availability relative to traditional and data-driven baselines. The results indicate that our proposed strategy is an excellent solution for predictive maintenance in the general aviation market.