Skip to content

Latest commit

 

History

History
132 lines (75 loc) · 5.33 KB

README.md

File metadata and controls

132 lines (75 loc) · 5.33 KB
Raw
logo

conmech3d: simulations of contact mechanics in Python using JAX

What is contact mechanics and conmech3d?

Contact mechanics models the behavior of physical bodies that come into contact with each other. It examines phenomena such as collisions, normal compliance, and friction. Most contact problems cannot be solved analytically and require a numerical procedure, such as the classical finite element method (FEM).

Conmech3d is an implementation of FEM for soft-body mechanical contact problems. The project is almost entirely self-contained and mainly aimed at research and didactic applications. Conmech3d is written in Python and uses JAX, a library for high-performance numerical computing. Besides basic Python libraries, such as Numpy and Scipy, it also employs pygmsh for mesh construction and Numba along with Cython to increase the speed of initial setup. Various options for visualization of simulation results are included, such as Blender, Three.js and Matplotlib

Experimental implementations of model reduction techniques that include tetrahedral skinning used in computer graphics and a new approach using Graph Neural Network are included in this repository.

Features

  • ability to solve both two- and three-dimensional simulations,
  • static, quasistatic, and dynamic problems,
  • a selection of constitutive laws, e.g. linearized elasticity, lineraized viscoelasticity and hiperelasticity,
  • focus on general contact conditions describing friction and normal compliance,
  • inclusion of additional phenomena, such as temperature.

Because Conmech3d is based on JAX, it also offers

  • support for simulations on CPU, GPU, and TPU,
  • simple configuration of single (32-bit) and double (64-bit) modes of numerical precision,
  • support for automatic differentiation and applications in machine learning.

Sample results

3D with temperature (Blender) 3D (three.js) 2D with temperature (Matplotlib)

Installation

Main functionality

Create a virtual environment (Python 3.10 is recommended)

python -m venv .venv
source .venv/bin/activate

Install Gmsh used for mesh construction

apt-get install python3-gmsh

Install JAX following the guide at https://github.com/google/jax#installation. For example, to use Nvidia GPU with self-installed CUDA 12, run

pip install --upgrade "jax[cuda12_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

Install dependencies from "requirements.txt"

pip install -r requirements.txt

Additional dependencies required for tetrahedral skinning

Compile Cython files form folder cython_modules by running

cd cython_modules
python setup.py build_ext --inplace

Additional dependencies required for GNN model

Install Pytorch. Only CPU version is required, as operations on GPU are performed using JAX. Follow the guide at https://pytorch.org/get-started/locally/. For example, to install Pytorch on Linux with CPU support run

pip install torch --index-url https://download.pytorch.org/whl/cpu

Install dependencies from "requirements-deep.txt"

pip install -r requirements-deep.txt

Usage

To run sample simulations, start any file from the examples folder

PYTHONPATH=. python examples/examples_3d.py

To run three.js server (by default on port 3003) run (this will start screen "three" running in the background)

cd scripts
./three.sh