A Deep Learning Approach for Predicting Surface Topography in Laser Powder Bed Fusion (PBF-LB) Additive Manufacturing

Abstract Metal additive manufacturing (AM) processes like Powder Bed Fusion - Laser Beam (PBF-LB) produce complex structures that directly impact part functionality. In this study, we propose a novel deep learning-based conditional denoising diffusion probabilistic model for generating synthetic yet physically realistic PBF-LB surface topographies. We fabricate metal AM specimens across controlled parameter spaces, then capture high-resolution 3D morphology data using confocal microscopy. The point cloud data are processed into 2D height maps and paired with critical process parameters. Our model, trained on datasets with multiple types of morphology pattern, demonstrates the ability to generate synthetic surfaces through parameter interpolation while maintaining key stochastic features. This approach offers a significant advancement in the ability to anticipate and optimize surface quality in metal additive manufacturing, providing valuable insights for both academic research and industrial applications.