Next-generation process monitoring and control for metal additive manufacturing: toward a physics-informed digital twin framework

Metal additive manufacturing (MAM) enables fabrication of geometrically complex components but faces process-induced defects that undermine quality consistency. The digital twin (DT) paradigm, a synchronised virtual counterpart of the physical process, offers a powerful framework for intelligent closed-loop quality assurance. However, MAM-DT’s full potential is limited by the divide between its two foundational modelling approaches: physics-based models offer accuracy but are computationally prohibitive for real-time use, while data-driven models are fast yet require extensive data and lack physical interpretability. Physics-informed machine learning (PIML) emerges as the pivotal technology for their fusion. This review first elaborates a four-layer MAM-DT system architecture. It then critically examines the virtual twin’s two pillars: accelerated physics-based and surrogate modelling, and data-driven in-situ monitoring and control, highlighting their limitations. The core contribution is a systematic taxonomy and analysis of PIML methodologies, categorised as Physics-Augmented Data, Physics-Constrained Learning, and Physics-Guided Architecture, which serve as the core fusion engine for MAM-DT. Their principles, applications in MAM, and roles in enhancing generalizability, data efficiency, and physical plausibility are detailed. Finally, the paper concludes with a critical discussion of current challenges and future research directions, outlining a path toward autonomous, self-optimizing, trustworthy intelligent manufacturing systems.