Artificial neural networks for strength and behavior prediction of ice samples
Current ice-related simulations are limited by available material models, which don't reflect the complexity of ice. Various phenomena are documented and well understood. Yet no material model can include all effects. Prior to modeling, decisions on which and how to include phenomena and parameters in the model must be made. A possible remedy is the correlation of parameters with outcome in experiments. For instance, how does a higher temperature influence the compressive strength of ice? An importance ranking of parameters is sought to support decision making during modeling.
Machine learning and statistical tools can be used to establish such a ranking. In this thesis, an existing data base of experiments (~3000 data points) is to be extended and analyzed with artificial neural networks. Once the model is able to predict the outcome of experiments, the SHAP (SHapley Additive exPlanations) method should be used to understand the black box neural network, see Figure and the SHAP package on Github
- Python with Jupyter notebooks or Spyder/Anaconda. Easy to use and beginner-friendly programming language with a development environment of your choice.
- MATLAB, for additional analyses and post-processing.
Leon Kellner, firstname.lastname@example.org, Office C 4.004
Saša Milaković, email@example.com, Office C 4.005
Merten Stender, firstname.lastname@example.org