Physics-Augmented Meta-ANN for Cross-Alloy Surface Roughness Prediction in Laser-Polished LPBF Metals: A Proof-of-Concept Study

Abstract

Surface roughness remains a critical challenge in Laser Powder Bed Fusion (LPBF), where laser polishing is often required to achieve the surface quality needed for functional applications. Conventional prediction approaches are frequently alloy-specific or empirical, which can limit their transferability across different materials. This study presents a compact physics-augmented Meta-ANN for predicting surface roughness after laser polishing of LPBF metals. The framework incorporates energy density (ED = P/v) and nonlinear terms as physically meaningful features to enhance the representation of laser–material interactions. The model is trained using a heterogeneous literature-derived dataset comprising 60 samples from four commonly used LPBF alloys: AlSi10Mg, Ti6Al4V, Inconel 718, and SS316L. The proposed model achieves encouraging predictive performance, with R² values above 0.96 across the evaluated datasets. Given the limited dataset size and heterogeneous data sources, the results are interpreted as preliminary proof-of-concept evidence rather than definitive validation of cross-alloy generalization. Nevertheless, the findings indicate that physics-augmented feature representation may improve data efficiency and provide a practical basis for developing more generalizable machine-learning approaches for laser-polished LPBF surfaces.