Härtill 6 uppsatser

Diss. (sammanfattning) Göteborg : Chalmers tekniska högskola, 2023

Due to digitalization in the maritime industry, a huge amount of ship operation-related data has been collected. The main objective of this thesis is to exploit machine learning/big data analytics to build data-driven ship performance models, focusing on speed-power relationship modeling, and fatigue accumulation assessment during a ship’s operation at sea. The speed-power performance models are established in three different ways: 1) semi-empirical white-box models, 2) machine learning black-box methods, and 3) physics-informed grey-box models. The white-box models include improved semi-empirical formulas for ship added resistance due to head waves, and further developed formulas in arbitrary wave headings. Validation studies using three case study ships show good agreement between the speed predictions by the white-box models and the long-term averages of full-scale measurements. Different supervised machine learning methods’ capabilities have been compared for black-box modeling. The XGBoost algorithm is found to have the most reliable predictive ability, with the highest efficiency suitable for onboard devices. The novel grey-box models are proposed by considering the physical principles in model tests and big data information from real sailing. It has been demonstrated that the proposed grey-box models can improve prediction accuracy by approximately 30% for ship speed estimation and provides 50% less cumulative error of sailing time than the black-box methods. The impact of voyage optimization-aided operations on the encountered wave conditions and ship fatigue damage is investigated in this thesis.