This repository contains an implementation of our paper "Learning the nonlinear elastodynamics of soft cellular mechanical metamaterials with graph networks" on International Journal of Mechanical Sciences. Here is an overview of the method:
There are three major steps: 1. Collect ground truth data by performing finite element simulation via FEniCS; 2. Train a graph network enabled by either a neural network or a Gaussian process regression model via JAX; 3. Deploy the trained graph network for dynamical simulation via JAX.
Here are a few examples.
Longitudinal wave propagation by direct numerical simulation using the finite element method:
dns.mp4
Longitudinal wave propagation by MGN:
P_wave_poreA.mp4
Shear wave propagation by MGN for a fully connected cross-spring system:
hierarchy_holow_0_shear_poreA.mp4
Shear wave propagation by MGN for a partially connected cross-spring system: