Physics-informed machine learning for reliability and systems safety applications: State of the art and challenges
Citation
Xu, Y., Kohtz, S., Boakye, J., Gardoni, P., Wang, P. (2023). Physics-informed machine learning for reliability and systems safety applications: State of the art and challenges. Reliability Engineering & System Safety, 230: 108900. Link to paper
Abstract
Physics-informed machine learning (PIML) is a set of hybrid modeling approaches that incorporate physics knowledge and machine learning (ML) models for real-world engineering applications. PIML has become an emerging frontier for scientific computing, and has shown tremendous success in solving differential equations and physical modeling. However, limited attention has been devoted to utilizing PIML methods to enhance the reliability and safety of complex engineering systems. In this paper, we provide a comprehensive and critical review of the state-of-the-art PIML approaches for reliability and systems safety applications. We first provide a systematic review of both data-driven ML and physics-based methods in reliability, followed by a comprehensive summary of the PIML approaches. A new taxonomy is proposed to classify existing literature into three groups based on the ways physics is integrated with ML: physics-guided, physics-encoded, and physics-integrated ML. The physics information is encoded in different formats, including governing equations, constraints, and fundamental laws. For each category, representative methods are analyzed in detail, with the benefits and challenges discussed. Finally, we identify the existing gaps, limitations, and open challenges of PIML for reliability and safety applications, and propose several promising future directions.