Data-Driven Vessel Operational Profile Based on t-SNE and Hierarchical Clustering

Measurements from the on-board monitoring systems of marine vessels are increasingly available for data analysis and are growing in importance as the ship industry enters a phase of digitalization. Sensor data can be applied to establish a real vessel operational profile that is often required by ship owners' and designers. Today, the vessel operational profile is created based on engineering and ship owner experience from similar vessel designs involving information regarding how the vessel has been commonly operated. Engineering knowledge applied to establish task definitions based on available signals gives an overview on what the vessel performed in reality. However, to fully utilize the data and receive a more detailed operational profile, we propose to apply data-driven techniques to find vessel operational modes. Data-driven techniques can be utilized for establishing finer resolution vessel tasks, for example, modes, and to find a more rigorous definition versus human definition. In this paper, we present a workflow to find vessel operational modes based on available signals involving machine learning techniques. We show that the data-driven techniques can be used to complement the engineering mode definitions.