Diffusion-Based Generative Artificial Intelligence for Optical Microscopy Images of Additively Manufactured Microstructures in Laser Powder Bed Fusion

This study synthesizes multi-scale, two-dimensional (2D) microstructural images of powder-bed fusion laser (PBF-L) additive-manufactured (AM) metals using denoising diffusion probabilistic models (DDPMs). By leveraging experimentally derived data from light optical microscopy (LOM), 16,811 image patches of size 256×256 pixels were used to train the model. These patches represent microstructural features such as porosities and melt pools across scales ranging from sub-micron to several millimeters. The DDPM-generated images achieved a Fréchet inception distance (FID) score of 34.6, indicating satisfactory quality synthesis. This approach addresses the challenges of limited experimental data by enabling the generation of nuanced, high-fidelity representations of AM microstructures, advancing the understanding of their multi-scale interactions.