Introduction

Accurate simulation of granular flow dynamics is crucial for assessing geotechnical risks, including landslides and debris flows. Granular flows involve dynamic particle rearrangement involving complex solid-fluid-like behaviors. Traditional numerical methods are limited by their computational cost in simulating large-scale systems. Statistical or machine learning-based models offer an alternative.  Still, they are largely empirical, based on a limited parameter set. Due to their permutation-dependent learning, traditional machine learning-based models require huge training data to generalize. 

To resolve these problems, we use a graph neural network (GNN), a state-of-the-art machine learning architecture. Graphs represent the state of dynamically changing granular flows, and GNN learns the interaction law which governs the granular flow behavior, such as energy and momentum exchange between grains. GNN-based simulator (GNS) takes the granular flow’s current state and predicts the next state using Euler explicit integration by learning the local interaction laws. We implement a multi-GPU GNS capable of handling different material types. 

GNS accurately predicts flow dynamics for column collapses with different aspect ratios and interaction with barriers for the configurations unseen during training. GNS is up to a few thousand times faster than high-fidelity numerical simulators.

Demonstration

We train GNS on a limited set of granular flow trajectories. We then assess the performance of GNS by predicting granular column collapse. GNS accurately predicts flow dynamics for column collapses with different aspect ratios and friction angles unseen during training.  

GNS also can accurately learn the granular flow dynamics including interaction with barriers.  The GNS is trained on granular flow trajectories with a limited set of initial conditions (in terms of granular mass, geometry, and barrier configuration). Our training data only includes the cube-shaped granular mass geometry interacting with one or two barriers. 

We showcase its generalizability performance by upscaling the geometry size and evaluating its performance on different geometry and barrier configurations (watch the video below). The GNS successfully replicates the ground truth behavior simulated by MPM. 

Computation efficiency

We compare the computation time between high-fidelity numerical simulator (MPM) and GNS. As shown in the following table, GNS shows up to a few thousand times of speed up.

Publication

• Choi, Y., & Kumar, K. (2023) Graph Neural Network-based surrogate model for granular flows. Computers and Geotechnics, 166, 106015. https://doi.org/10.1016/j.compgeo.2023.106015

• Open-source code and data at https://github.com/geoelements/gns