Research Experiences for Undergraduates (REU), NSF NHERI 2024: Unveiling Debris Flow Dynamics: Physics-Informed Machine Learning via MPM Simulation

Tsunamis, landslides, and storm-surges can mobilize debris, compounding the hazards faced by the built environments. Predicting the dynamics of entrained debris in these complex flow events is challenging due to the varying materials, phases, and the uncertainty associated with their respective properties. This study presents a novel approach to address this challenge using a combination of machine learning (ML) and high-performance numerical simulation. Our understanding of the dynamics of these hazards is underpinned by experimental facilities, which we aim to augment with a digital twin surrogate modeling workflow. In this manuscript, we introduce a prototype digital twin of the Hinsdale Wave Research Facility's Large Wave Flume at Oregon State University (OSU LWF) for studying wave-debris dynamics. The first-phase of our two-stage high-performance digital twin uses the Material Point Method (MPM), implemented within the Taichi programming language, to support classical numerical simulation of large-deformation, multi-material dynamics. For the creation of lightweight surrogate models in the second-phase, we leverage the highly flexible Graph Network Simulator (GNS), a graph neural network software package (GNN), to enable application of our workflow to a plethora of unexplored experimental facilities which may or may not center on wave-debris studies. By utilizing high-performance computing (HPC) and deep learning networks, this approach enables the efficient representation of complex physics and facilitates uncertainty quantification in debris hazard events. With this, the ongoing integration of these technologies within the NHERI SimCenter’s engineering workflow, specifically into HydroUQ, has great potential to provide a robust framework to run high-fidelity simulations. This work demonstrates the potential of ML-driven surrogate models to enhance the predictive capabilities and efficiency of simulation, paving the way for improved debris flow hazard mitigation strategies.