Establishing a common database of ice experiments and using machine learning to understand and predict ice behavior


Ice material models often limit the accuracy of ice related simulations. The reasons for this are manifold, e.g. complex ice properties. One issue is linking experimental data to ice material modeling, where the aim is to identify patterns in the data that can be used by the models. However, numerous parameters that influence ice behavior lead to large, high dimensional data sets which are often fragmented. Handling the data manually becomes impractical. Machine learning and statistical tools are applied to identify how parameters, such as temperature, influence peak stress and ice behavior. To enable the analysis, a common and small scale experimental database is established.