Performance Portable Additive Manufacturing Simulations with MPM

Sam Reeve, Kwitae Chong, Austin Isner, Stuart Slattery

Keywords: additive manufacturing, Kokkos, Cabana

TL;DR: Performance Portable Additive Manufacturing Simulation with MPM

Abstract:

PicassoMPM is a new material point method application with a focus on high-fidelity simulation of additive manufacturing (AM). We first describe the software design as part of the Exascale Computing Project (ECP), where PicassoMPM is built on three libraries: Kokkos for performance portability across hardware architectures, Cabana for general particle algorithms and communication, and Picasso for PIC and MPM motifs. This strategy enables sharing development across the widest set of applications possible, e.g. entirely different particle methods or MPM for a different application domain. This design also enables simulation across HPC architectures, including ORNL Summit and Frontier from which we show performance and discuss continuing optimizations. Finally, we discuss the calibration, comparison, and predictive capability of PicassoMPM against experimental data for laser powder bed fusion in various alloys. These comparisons include the shape and size the AM melt pool, the fluid flow within the liquid, and the transition across regimes of AM processing.

Supported by ECP (17-SC-20-SC).