A highly efficient computational approach for fast scan-resolved microstructure predictions in metal additive manufacturing on the scale of real parts

Abstract

In metal additive manufacturing (AM), fast and efficient simulation approaches are essential to explore the full potential of these promising processes, particularly in generating components with tailored microstructures via laser powder bed fusion (LPBF). Due to the inherent multiscale nature of LPBF, existing approaches often need to resort to strong simplifications, such as layer-wise heating models, to make part-scale simulations feasible. In contrast, the present article proposes a scan-resolved approach, which consistently resolves the laser scan path in a coupled thermo-microstructural model of LPBF. Building on a high-performance computing model for the thermal problem, we propose a highly efficient implementation of a recently developed microstructure model for Ti-6Al-4V with three main constituents: stable αs-phase, martensitic αm-phase and β-phase. The implementation is tailored to modern hardware features using vectorization and fast approximations of transcendental functions. A performance model and selected numerical examples of LPBF manufacturing of parts on the centimeter scale are studied to verify the high degree of optimization. Depending on the specific example, results were obtained with moderate computational resources in a few hours to days. We demonstrate how the proposed scan-resolved model allows us to predict the correlation between scan strategy and resulting microstructure composition, an aspect that layer-wise heating models cannot capture. The numerical examples include scan-resolved thermo-microstructure simulations of the full NIST AM Benchmark cantilever specimen. It is shown that varying the build plate temperature by only 100K can significantly change the microstructure composition from αm- to αs-dominated.