Predicting icebreaker resistance using machine learning and scale model testing

Icebreakers are essential assets to enable safe Arctic and subarctic operations. These specialized ships have strengthened hulls and robust propulsion systems to enable them to manage ice and open shipping lanes through sea ice. Accurately predicting icebreaker performance is critical for designing vessels that are fit for purpose. A key factor in icebreaker performance prediction is understanding ice resistance, which determines an icebreaker’s capability to operate effectively in icy conditions. In recent years, Machine Learning (ML) methods have been increasingly utilized to predict ship efficiency, typically using parameters such as length, beam, draft, and speed. This study expands on this approach by integrating both fundamental vessel parameters and environmental factors, including ice thickness, along with detailed hull geometry data into ML models. The objective is to assess how these factors enhance the accuracy of ice resistance predictions. The dataset includes ten different icebreakers, with model tests conducted by the National Research Council of Canada’s Ocean, Coastal, and River Engineering Research Centre (NRC-OCRE). We trained boosting models to predict total ice resistance. This study demonstrates how data-driven approaches can result in novel multivariate regressions of ice resistance and highlights the improvements in prediction accuracy achieved by incorporating hull geometrical characteristics.