Interannual variability in fatigue damage estimation from short-term strain monitoring of offshore wind turbines

Abstract. Extending the service life of offshore wind turbines demands fatigue assessments that reflect site-specific loading rather than conservative design assumptions. Structural health monitoring can provide strain-based damage estimates, but such monitoring campaigns are typically short. Consequently, long-term damage is often estimated from short-term strain data conditioned on available environmental and operational condition (EOC) records. Despite widespread use of such estimation approaches, the representativeness of a one-year strain window remains insufficiently quantified. In particular, it is unclear whether estimation errors are primarily driven by within-year statistical uncertainty or by interannual variability in the conditional damage–EOC mapping. This distinction is difficult to assess in practice due to the scarcity of long-term paired strain–EOC datasets. In this work, an eight-year strain-EOC dataset from an in-service offshore wind turbine is used to quantify damage estimation error by systematically shifting a one-year strain monitoring window across years. A hierarchy of damage-mapping strategies is evaluated, from unconditional extrapolation to EOC-conditioned estimators based on binned damage models. Unconditional extrapolation yields substantial window-dependent errors, with deviations up to 30 % in the estimated long-term mean 10-minute damage, and exhibits pronounced interannual variability. Conditioning on informative EOCs generally reduces errors to around 10 % and decreases sensitivity to the considered monitored year. However, these improvements are not monotonic with the dimensionality of the EOC-conditioned model. Bootstrap-based estimates of within-year statistical uncertainty are consistently small (<1 %), indicating that estimation error is dominated by interannual variability. Longer strain monitoring periods reduce window-to-window variability, but do not eliminate damage estimation error under non-stationary conditions, where the EOC–damage mapping changes over time. These results show that long-horizon damage estimates remain sensitive to both the timing and duration of the strain-monitoring window. This sensitivity highlights the need to verify that the considered EOCs are representative of long-term conditions and that the damage–EOC mapping remains stable over time.