This study proposes a random-forest-based machine learning (ML) model for fatigue life prediction of offshore wind turbine monopile foundations compared to traditional approaches, using long-term strain data from turbines in the Belgian North Sea. This study shows that the ML model predicts the fatigue life of monopile foundations more reliably when only short-term measurements are available, whereas for longer monitoring periods of greater than 12 months, simpler binning methods perform equally well.

We show that simple vibration sensors on wind turbines can reveal how each machine is operating without relying on control system data. By learning patterns from short acceleration segments, our model identifies turbine behavior, detects changes in operation, and tracks events over time. These patterns also support estimating fatigue, providing a new way to understand turbine performance using only vibration measurements.

Analysis of long-term fatigue damage of four offshore wind turbines using 3 years of measurement data was performed  for the first time to gain insight into the low-frequency fatigue damage (LFFD) impact on overall consumed life. The LFFD factor depends on the (linear) stress–life (SN) curve slope, heading, site, signal, and turbine type. Up to ∼ 65 % of the total damage can be related to LFFDs. Therefore, in this case study, the LFFD effect has a significant impact on the final damage.

The estimation of fatigue in offshore wind turbine is highly relevant, as it can help extend the lifetime of these assets. This article attempts to answer this issue by developing a methodology based on artificial intelligence and data collected by sensors installed in real-world turbines. Good results are obtained, and this methodology is further used to learn the value of eight different sensor setups and employed in a real-world wind farm with 48 wind turbines.