Detecting Hull Fouling using Machine Learning Algorithms trained on Ship Propulsion Data to Improve Resource Management and Increase Environmental Benefits

Abstract. This study aims to develop a methodology to assess hull fouling based on ship propulsion data such as speed, draft and weather related data. Hull fouling is an unavoidable phenomenon in ships and results in higher fuel consumption and the maintenance frequency has be the optimal one. Despite the fact that until now this task has primarily relied on empirical rules, it turns out that it can be improved by employing machine learning techniques. Using data from clean-hull ships, we aim to isolate and consider only the weather in this study. Our goal is to replace empirical rules with machine learning, as the vast amount of data we possess can significantly aid us in this endeavor. It ends up to be a regression problem, and therefore, we experiment with several supervised algorithms using k-fold cross validation to finally select models based on ensemble methods or artificial neural networks. We propose the potential use of MLP Regressor, Random Forest Regressor and XGB Regressor since all of them yielded very good results in terms of some performance metrics. The timely detection of hull fouling can provide substantial benefits in terms of resource management and environmental sustainability.