PIML Improves PHM: A Systematic Review of Physics-Informed ML.

Prognostics and Health Management (PHM) is vital for maintaining the reliability and efficiency of complex industrial systems. While Machine Learning (ML) has driven significant advancements in diagnostics and prognostics, purely data-driven models often suffer from limitations such as poor generalization, lack of causal inference, and limited interpretability. Physics-Informed Machine Learning (PIML) addresses these issues by integrating prior physical knowledge directly into the ML pipeline, a strategy gaining considerable interest in PHM applications. A comprehensive systematic literature review of 212 studies investigates the application of PIML in PHM. The review categorizes approaches into observational bias, inductive bias, learning bias, and hybrid methods, and further classifies them by specific PHM tasks. Across all categories, PIML consistently demonstrates superior predictive performance compared to conventional data-driven baselines. Despite these clear benefits, the literature shows a heavy focus on specific assets like lithium-ion batteries and bearings, and a prevalence of problem-specific solutions. While PIML offers tangible performance improvements, claims regarding enhanced interpretability or causal inference often lack sufficient supporting evidence. Future research should prioritize developing transferable design patterns, establishing benchmarks for integration strategies, and creating lightweight, uncertainty-aware models robust enough for online deployment in real-world industrial settings.