An efficient active learning Kriging approach for expected fatigue damage assessment applied to wind turbine structures

Abstract

Estimating fatigue damage in wind turbine structures over their lifetime is a time-demanding task, necessitating simulations encompassing all the wind and wave conditions. Typically, this involves expending several thousand hours in simulation efforts. Given formidable computational demand, this study proposes an efficient active learning Kriging approach named ALK-EFDA for estimating expected fatigue damage in wind turbines. The ALK-EFDA framework leverages a Kriging surrogate model to predict fatigue damage levels across various wind-wave scenarios, and a learning function is designed to reduce Kriging prediction error in the high probability of occurrences of wind-wave cases. An active learning approach is proposed to estimate the expected fatigue damage efficiently. To validate the effectiveness of this approach, we applied it to a 15 MW Semi-submersible floating wind turbine model. The proposed approach estimates the expected short-term fatigue damage of the wind turbine tower and mooring lines. Our findings demonstrate that the ALK-EFDA method efficiently and accurately estimates fatigue damage in wind turbine structures. Compared to simulation methods, the proposed ALK-EFDA approach significantly reduces computational expenses by over 30 times. Furthermore, the absolute error, when compared to simulated results, remains at approximately 1%. This method will be a useful tool for offshore wind turbine designers.