This conference paper was submitted for presentation at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

This paper presents a new hybrid method using simulation and data analytics for monitoring fatigue damage and estimating remaining useful life of marine vessels. This hybrid method takes reference from the smart functions introduced by the American Bureau of Shipping (ABS) to enhance approaches in structural health monitoring. The analysis of material fatigue is essential in ensuring the reliability of marine vessels such as ships and oil rigs, which undergo consistent cyclic loading from wind and sea waves. Fatigue lifecycles are currently determined based on a large variety of methods ranging from sensor experiments to high-fidelity simulations. The information and data are crucial in informing users to adopt pre-emptive measures to avoid premature structural failure during operation. While the current physics-based finite element (FE) simulation methods are accurate, they are too computationally intensive for continuous health monitoring across all loading combinations of marine vessels in operation. On the other hand, data-driven methods have gained traction due to their speed and versatility, but lack in capturing the required complexity of physical systems. This newly developed hybrid simulation and data-driven health monitoring method therefore aims to provide a continuous fatigue life prediction of the marine vessels in operation. High-fidelity FE simulations under potential sea states were used to generate training data for a surrogate Gaussian process (GP) model. This accelerates predictions of fatigue damage while preserving the underlying physics of the system. By imparting the physics as prior knowledge of the system, the GP model can also provide continuous predictions with appropriate confidence intervals. Based on the level of uncertainty, the GP model can be iteratively refined with additional training data to improve its robustness and effectiveness. In comparison to high-fidelity simulations, this method was able to predict fatigue damage with a maximum discrepancy of 10%, while achieving a speedup of up to 30%. The results successfully demonstrated the possibility and effectiveness of the method in continuous structural health monitoring.