Physics-guided metamodel for vertical bending-induced fatigue damage monitoring in container vessels

• Published in: Ocean Engineering
• Main Authors: Lang, Xiao, Zhang, Mingyang, Zhang, Chi, Ringsberg, Jonas, Mao, Wengang
• Language: unknown
• Published: 2024
• Subjects: Materials Engineering; Mathematics; Vehicle Engineering; metamodel; machine learning; physics-guided; full-scale measurements; container vessel; Fatigue damage; North Atlantic
• Online Access: https://doi.org/10.1016/j.oceaneng.2024.119223; https://research.chalmers.se/en/publication/542980

2800TEU container vessel navigating the North Atlantic, based on data from the vessel’s hull monitoring system. The metamodel combines two XGBoost-based base learners: a black-box model utilizing ship heave and pitch motion measurements, and a gray-box model using spectral moments from numerical analysis. Predictions from both models are refined through a meta-learner Gaussian process regression to enhance accuracy. The metamodel was evaluated against black-box and gray-box models across various training data volumes. The proposed model adapts to varying data volumes, from months to over 2 years, effectively integrating the strengths of both base learners to provide reliable predictions in both seen and unseen scenarios. The model consistently demonstrated superior performance, enhancing fatigue damage accumulation accuracy by up to 35% over traditional machine learning methods. This advancement can aid the maritime industry in effectively monitoring ship fatigue and implementing predictive maintenance strategies, marking a significant step forward in applying data-driven techniques in shipping.