Review of physics-informed machine learning in laser metal additive manufacturing: a process-lifecycle perspective on prediction, monitoring, and post-process evaluation

Laser powder bed fusion (L-PBF) and laser directed energy deposition (L-DED) are widely applied across industries owing to their high geometric freedom and manufacturing precision. However, porosity and residual stress can arise from complex interactions among process parameters, degrading the mechanical performance and reliability of manufactured parts. This makes accurate prediction, monitoring, and quality assessment essential. Conventional analytical methods rely on simplified assumptions that limit their ability to capture complex, coupled physical phenomena, while data-driven approaches remain data-dependent and lack interpretability. Physics-informed machine learning (PIML), which integrates physical laws with data-driven learning, has recently attracted considerable interest as a means of overcoming these challenges. This review investigates PIML utilisation strategies across the process lifecycle of L-PBF and L-DED. First, the relationships between process parameters and the formation mechanisms of porosity and residual stress are analysed. PIML approaches applied in the pre-, in-, and post-process stages are then comparatively reviewed. Finally, current challenges and future research directions are discussed in terms of prediction accuracy, computational efficiency, generalisation capability, and industrial applicability. This provides an integrated perspective on PIML's application in next-generation intelligent laser metal additive manufacturing (LAM).