Physics-Integrated and Uncertainty-Aware AI for Prognostics and Health Management
Session Organizers:
Junyu Qi, Karlsruhe Institute of Technology, Germany
Email: junyu.qi@kit.edu
Jie (Peter) Liu, Carleton University, Canada
Email: jie.liu@carleton.ca
Carsten Proppe, Karlsruhe Institute of Technology, Germany
Email: carsten.proppe@kit.edu
Gernot Schullerus, Reutlingen University, Germany
Email: gernot.schullerus@reutlingen-university.de
Zhuyun Chen, Guangdong University of Technology, China
Email: mezychen@gdut.edu.cn
Industrial systems, such as wind turbines, manufacturing equipment, robotic platforms, aerospace structures, and energy infrastructures, are fundamental to modern society and sustainable development. These systems typically operate under complex, dynamic, and harsh conditions, making them prone to degradation, faults, and unexpected failures. Conventional maintenance strategies based on reactive actions or fixed schedules are increasingly inadequate in terms of efficiency, adaptability, and cost-effectiveness.
Recent advances in artificial intelligence (AI), advanced signal processing, multi-sensor fusion, and digital twin technologies have led to a paradigm shift in Prognostics and Health Management (PHM). Data-driven and hybrid modeling approaches now enable real-time fault diagnosis, remaining useful life (RUL) prediction, and intelligent maintenance decision-making. However, despite impressive progress, purely data-driven approaches often lack physical consistency, robustness, interpretability, and reliable uncertainty quantification, which limits their adoption in safety-critical industrial applications. In parallel, physics-based modeling, supported by rigorous uncertainty quantification (UQ) and probabilistic methods, provide strong interpretability and extrapolation capabilities but often struggle with scalability, modeling complexity, and real-time deployment.
This special session aims to bridge physics-based modeling and modern AI, fostering the development of physics-integrated, uncertainty-aware, and explainable PHM frameworks. Particular emphasis is placed on digital twin–enabled PHM, physics-informed neural networks, hybrid modeling, and trustworthy AI, combining expertise from signal processing, probabilistic modeling, and deep learning.
Contributions are invited on, but not limited to, the following topics:
- Physics-based modeling and digital twins for PHM
- Robust and risk-aware decision-making under uncertainty
- Feature learning and heterogeneous data fusion for complex systems
- Physics-informed neural networks (PINN) for PHM
- Explainable AI (XAI) for interpretable representations of degradation mechanisms
- Industrial applications and case studies