Machine learning techniques for ship performance predictions in open water and ice

The primary purpose of the work is to explore the practicality of using Artificial Intelligence (AI); specifically, Machine Learning (ML) and Deep Learning (DL), to predict ship performance characteristics based on time-averaged and time-dependent data. Three application cases are studied. The first modelling case is a time-averaged ship propulsor performance dataset, the second and third modelling cases are a time-averaged and time series prediction of forces on a dynamic positioning ship operating in a broken ice-field. An ML-based model was developed to predict various propulsor coefficients of a podded propulsor, given the advance coefficient, cavitation condition, hub geometric variations, pod configurations and the azimuthing angle. The second modelling case involved developing an ML algorithm to predict time-averaged ice forces on DP-controlled ships at the given ranges of ice concentration, floe size, ice thickness, strength, density, drift speeds and direction. The third modelling case involved predicting the time-dependent forces on a DP-controlled ship at specific operating conditions and ice-field parameters. The AI-ML-based predictive models showed reasonable accuracy compared to the corresponding measurements and performed better than conventional regression-based models.
31st International Ocean and Polar Engineering Conference, June 20-25, 2021, Rhodes, Greece