Detection of ice using ship propulsion and navigation measurements

Abstract

The presence of sea-ice influences the operation of ships in polar waters. Ice navigation leads to increased rates of damage accumulation in propulsion machinery. Methods and algorithms used to monitor the condition of the propulsion machinery can benefit from indicators that detect either operation in ice-infested waters or direct detection of propeller–ice interaction. This paper explores existing methods that inform the novel application of automatically detecting sea-ice based on high frequency shaft response measurements, low frequency propulsion control measurements, and navigation data. These methods include rule-based approaches, machine learning based classification, and statistical change detection. Methods are evaluated based on misclassification errors during cross-validation. It is shown that detection using the selected data is possible and able to give satisfactory prediction accuracy based on a case-study vessel. The results also show that machine learning approaches provide a solution that balances prediction accuracy and the rate of false negative predictions. The recommended method is to use support vector machine classifiers with a radial basis function kernel.