MPM as a Multi-Physics Engine for Experimental Digital Twins, Surrogate Models with Uncertainty Quantification, and Regional Natural Hazard Recovery Simulations.
Keywords: Digital Twins, Uncertainty Quantification, Surrogate Modeling, Material Point Method, Taichi, Multi-Scale Physics, Scientific Workflows, Hurricane Simulations, Tsunami Simulations, Landslide Simulations, Exa-Scale, High-Performance Computation, Coastal Engineering, Debris
TL;DR: MPM is integrated into open-source computational workflows used in the creation of digital twins of experimental facilities, surrogate models for uncertainty quantification, and regional recovery studies for natural disasters.
Abstract: Due to its "jack-of-all-trades" multi-physics and excellent properties for near real-time simulations, the Material Point Method (MPM) is integrated into the NHERI SimCenter's open-source computational workflow. It serves as a flexible multi-physics engine for the creation of experimental facility digital twins (e.g., wave flumes at Oregon State University), surrogate models for uncertainty quantification (UQ), and regional natural hazard recovery studies. The process of loosely-coupling two massively-parallel, high-performance implementations of MPM, specifically in Taichi Lang (Python-esque scripts) and ClaymoreUW (C++ CUDA configured by JSON), into both local and remote workflows is detailed. Capabilities unlocked by connecting popular UQ engines (e.g., Dakota, SimCenterUQ), robust structural analysis software (e.g., OpenSees), generative AI (e.g., BRAILS, Point-e), surrogate modeling techniques (e.g. GNS, Gaussian Processes, Probabilistic Learning on Manifolds), and cutting-edge community resources for computation and storage (e.g. TACC, DesignSafe DataDepot) with the high-performance multi-physics of MPM are demonstrated. Emphasis is placed on intended exa-scale applications for the social sciences and regional policy-making. All workflows are reproducible in the NHERI SimCenter's open-source, graphical desktop applications for Mac, Windows, and Linux (e.g., HydroUQ).
Submission Number: 30
Loading