Session: 07-05-01 Ice Model Tests

Paper Number: 87434

87434 - Analyzing Flexural Strength Data of Ice: How Useful Is Explainable Machine Learning? 

As consequence of climate catastrophe huge amounts of ice melt in polar regions, making vessel routes traversable for longer durations and therefore increases maritime traffic and ice-structure interactions in arctic environment. In this context, the flexural strength is probably the most important material property of ice. In addition to the technological aspect, the flexural strength is also an important factor in the process of ice ridge formation. Due to the complexity of natural materials, whose properties are often determined by a large number of influencing variables, the development of a suitable material model for ice remains a major challenge. Many investigations have been carried out in the past, for example by (Butkovich, 1959; Ervik, 2013; Frankenstein, 1959; Frederking & Timco, 1983; Ji et al., 2011; Kujala et al., 1990; Weeks & Anderson, 1958), in which the flexural strength of ice was determined by experimental methods. These experimental results can often be used to draw conclusions about the material properties and its multi-variable influences. Thus, for example, porosity is named as the most important influencing variable with respect to the flexural strength of sea ice (G.W. Timco & S. O.'Brien, 1994; Timco & Weeks, 2010). For fresh water ice this might be the ice volume (Williams et al., 1994). However, a major drawback is that these experiments can differ significantly in e.g. test method, ice conditions, measured variables etc., which makes a comparison as well as the derivation of general laws for material properties of ice challenging. For this reason, we present a machine learning approach to investigate data from various bending tests and use XAI to identify relevant features regarding the flexural strength of ice. In (Kellner et al., 2019; Kellner et al., 2020) it has already been shown how machine learning methods can be used to determine the relevance of influencing variables regarding the failure mechanisms and strength of ice under tensile and compressive loading. Similarly, in this work we establish a flexural strength database based on a literature review. Furthermore, the data will be used to create machine learning models, which subsequently will be interpreted with a suitable XAI method. The goal is to show a new approach to investigate the flexural strength of ice and to get a better understanding of how suitable the use of XAI for this problem is.

Presenting Author: Patrik Buil University Duisburg-Essen

Authors:

Patrik Buil University Duisburg-Essen
Leon Kellner Hamburg University of Technology
Sören Ehlers Hamburg University of Technology
Franz V. Bock U. Polach Hamburg University of Technology