Adaptive ensemble of surrogates for efficient fatigue reliability analysis of offshore wind turbines

Offshore wind turbines (OWTs) are crucial to the global transition towards renewable energy, but they operate under harsh marine conditions where fatigue damage from wind- and wave-induced loads is a critical design concern. Assessing fatigue reliability is further complicated by uncertainties in environmental and structural parameters, and conventional approaches such as Monte Carlo simulation (MCS) remain prohibitively expensive. To address this challenge, this study develops an adaptive ensemble of surrogates (AEOS) that integrates Kriging, Bayesian support vector regression, and polynomial chaos Kriging. A novel weighting strategy that balances local and global error measures, together with a reward-based learning function allocation scheme and a hybrid stopping criterion, enables efficient and accurate active learning. The proposed AEOS is validated on benchmark problems and a monopile-supported OWT case study, achieving failure probability estimates with less than 1.5% relative error while reducing computational cost by more than 95% compared to MCS. Sensitivity analysis further reveals that wind conditions and fatigue strength parameters dominate fatigue reliability outcomes. AEOS provides an efficient, accurate, and flexible framework for fatigue reliability assessment, supporting risk-informed maintenance, life extension, and sustainable operation of offshore wind infrastructure. • Novel active learning method for efficient fatigue reliability analysis of OWTs. • New weighting and learning function selection strategies to enhance adaptability and accuracy. • Achieves up to 95% reduction in computational cost while maintaining high accuracy. • Enables accurate long-term reliability assessment and risk-informed life extension decisions.