| Abstract Scope |
Laser powder bed fusion (LPBF) is a widely used additive manufacturing process for producing complex metal components. However, localized thermal gradients during fabrication generate residual stresses that often lead to distortion, cracking, and build failure. While inherent strain simulations can estimate these effects, they are computationally expensive when applied to large-scale or complex design spaces. In this work, a data-driven surrogate model based on a sparse 3D transformer architecture is developed to provide a rapid approximation of such simulations. The model learns the mapping between voxelized geometries and the resulting mechanical fields, including stress, strain, and displacement. By leveraging a local attention mechanism over structured voxel neighborhoods, the model captures both short-range load transfer and long-range mechanical interactions driven by geometry and boundary conditions. The proposed framework significantly reduces computational cost compared to finite element simulations, providing a scalable tool for process-aware design and evaluation of complex LPBF geometries. |