Fatigue crack propagation and life prediction in offshore steel jackets using multi-fidelity modeling and machine learning

Offshore steel jacket structures are exposed to harsh marine environments and cyclic loading, leading to fatigue damage that undermines their structural integrity. Among the most vulnerable areas are the welded joints, which are susceptible to crack initiation and propagation, highlighting the need for reliable fatigue assessment during design and operation to ensure durability and safety. To address this challenge, this study introduces a machine learning (ML)-based surrogate modeling framework for efficient and reliable fatigue assessment. The framework comprises three integral components: (1) a multi-fidelity finite element (FE) modeling to substantially minimize computational demand; (2) a surrogate model for predicting stress intensity factor (SIF); and (3) a crack propagation and fatigue life prediction module. The surrogate model was trained on a dataset from simulations in Abaqus and Franc3D. The multi-fidelity models reproduced the first five vibration periods with mean errors below 3.3 %, and mode shapes showed strong agreement with a high-fidelity reference. Among eight ML models assessed for SIF prediction, a deep neural network (DNN) achieved the highest accuracy (MAE ≈ 3 %), whereas XGBoost attained a balanced trade-off between accuracy and computational efficiency (MAE ≈ 11 %). Beyond model-level assessment, two additional full-case verifications at the weld toe (crown and saddle) matched FE-computed SIF and fatigue life within ± 5 % and 1.6 %, respectively, while delivering results in 3 s compared with about 5 h for the FE simulations. These findings demonstrate that the proposed framework provides a reliable, efficient alternative to fracture mechanics-based FE simulations for fatigue assessment of complex structures under realistic loading.