Physics-informed Multi-fidelity Digital Twins for Prognostics and Health Management of Complex Engineering Systems

Abstract

Scalable and reliable prognostics and health management (PHM) has become increasingly important for safety-critical applications operating under dynamic environments. However, existing PHM digital twins often suffer from expensive high-fidelity modeling, limited scalability for large-scale systems, and lack of adaptability under evolving degradation behaviors. This research develops a unified physics-informed digital twin framework integrating multi-fidelity learning, physics-informed machine learning, and adaptive uncertainty-aware prognostics for complex multi-physics engineering systems. The proposed framework enables data-efficient diagnosis through multi-fidelity digital twins, mesh-free thermal-state monitoring through physics-informed spatial learning, and adaptive remaining useful life prediction through online state estimation and model adaptation. Preliminary studies on wet clutch systems, battery packs, and Hall thrusters demonstrate improved prediction accuracy, reduced dependence on high-fidelity data, and robust prognostics under dynamic operating environments.