llamm-de / thermalCANN

Model of thermo-hyperelasticity using Constitutive Artificial Neural Networks

Thermoelastic Constitutive Artificial Neural Network (thermalCANN)

This is the repository for a model of thermo-hyperelasticity using a Constitutive Neural Network (CANN) architecture build with TensorFlow and Keras.

The architecture is based on the publications of K. Linka et al. (2021): Constitutive artificial neural networks: A fast and general approach to predictive data-driven constitutive modeling by deep learning and K. Linka & E. Kuhl (2023): A new family of Constitutive Artificial Neural Networks towards automated model discovery.

Getting started

This model was developed under Python 3.8.10. The required packages for executing the example scripts can be found in the requirements file. To install them into a virtual environment please proceed as follows:

virtualenv .venv
source .venv/bin/activate
pip install -r requirements.txt

Running the code

Pretrained model

TBC

Training the model

TBC

Running unittests

To run the unit tests provided by this implementation, please use the following

python -m unittest

from the root directory of this repository. Please be aware that there is not a 100% code coverage. Coverage reports can be generated using the coverage module as

coverage run -m unittest

Datasets

The repository contains various datasets. Datasets can be loaded using the functionality provided within the src/data.py package.

Treloar data for hyperelasticity

The data from the famous paper of Treloar (1944): Stress-Strain data for vulcanised rubber under various types of deformation is located in data/Treloar. The original dataset is a copy of the version published in P. Steinmann, M. Hossain, G. Possart (2012): Hyperelastic models for rubber-like materials: consistent tangent operators and suitability for Treloar’s data.

The original data is availabel as both numpy arrays stored in .pkl files as well as in .csv format. To open the data from .pkl files use the load_treloar() function from src/data.py. To load a fitted verison of the original data using the Arruda & Boyce model, please use load_treloar(arruda_boyce_fit=True). This will give you more datapoints to train your model.

Artificially generated data for thermo-hyperelasticity

You can findartificially generated data from a thermo-elastic, incompressible Arruda & Boyce model in data/artificially_generated. It can be loaded using the load_artificial_Arruda_Boyce() function from src/data.py. The dataset contains three numpy arrays:

1. Stretch in loading direction
2. Temperature delta (T - T_ref) for each sample
3. First Piola-Kirchhoff stress values in loading direction for each temperature delta sample

Get involved

For feature requests, bug reports etc. please open an issue on github or get in contact with us directly.

License

This model is licensed under the MIT License. More information on that is given in the LICENSE.md file.