From correlation to causation: data-mechanism dual-driven analysis of defects induced by melt pool dynamics, vapour recoil, and spatter in PBF-LB and mitigation

Laser beam powder bed fusion (PBF-LB) is reshaping high-end manufacturing. However, the non-equilibrium metallurgical process involves transient phenomena governed by multiphysics coupling, and their correlation with build quality remains unclear. Consequently, defect control still relies on empirical trial-and-error. This paper analyzes the formation mechanisms of critical physical phenomena during PBF-LB. It assesses the principles, capabilities, and limitations of in situ monitoring techniques, including synchrotron X-ray, high-speed optical, schlieren, and acoustic emission. Mechanistic modelling approaches, spanning macroscale thermomechanical modelling, mesoscale melt pool dynamics, and microstructural evolution, together with their applications in defect prediction, are critically analyzed. Furthermore, data-mechanism dual-driven methods, integrating multimodal data and physics-informed machine learning, link melt flow, keyhole instability, vapour recoil, and spatter to porosity, lack of fusion, cracking, and segregation. Finally, recent advances in defect-mitigation strategies are analyzed, including process-window optimisation, spatiotemporal beam shaping, and multiphysical-field assistance, to support defect-free PBF-LB manufacturing.