Numerical simulations are used in a variety of computational problems. The error of the simulation is often of interest here. However, the error is usually not known and must be calculated using expensive methods such as adjoint error equations. We investigate the use of multiple Convulational Neural Networks (CNNs) architectures as efficient and reliable error estimators for muscle fiber simulations by opendihu.
To run this project you need to have opendihu successfully installed. In addition you need a python3 environment with numpy, pandas, scikit-learn and tensorflow.
Open the Main.ipynb in your favorite IPython-Notebook editor, e.g. JupyterLab and follow the steps described there.
- Robin Sasse
- Philipp Schmid
- Tobias Weinschenk
Last but not least we would like to thank our supervisor Felix Huber, for making this project possible and providing guidance and feedback.