This study presents a structured approach for evaluating the impact of different control strategies (normal operation, wake steering and derating) on the structural integrity of offshore wind turbine support structures. The objective is to determine possible failure modes influenced by changes in control strategies and to prioritize these failure modes for further analysis.

The Failure Modes, Effects, and Criticality Analysis (FMECA) methodology is implemented using a semi-quantitative approach, wherein scores are assigned to the likelihood of occurrence, severity of consequences (including personnel safety, environmental impact, intervention complexity, spare part costs and operational losses) and the beta factor, which denotes the conditional probability of the failure end effect materializing. A Risk Priority Number (RPN) is determined by multiplying the likelihood, beta factor and sum of the consequence severity scores, and is finally categorized into low, medium and high risk.

Within this work, wake steering and derated WTG control scenarios have been compared to normal operation as a baseline. Only relevant failure modes and support structure components affected by the changed wind turbine operations are considered for the comparative risk assessment.

The effect of the control scenarios on the failure modes has been assessed within a case study focused on the gravity-based foundation of the Lillgrund offshore wind farm. A total of 41 failure modes have been identified and rated. As the severity of failure modes remains similar, only the likelihood and the beta factor differ between scenarios. All results are presented in a single overview, providing an effective means to compare the different scenarios.

The derating strategy did not result in any increased risk. In fact, 10 out of 18 failure modes classified as medium in the baseline were reduced to a risk priority level of low for the derating case due to reduced loads. The application of wake steering however resulted in a reduction of failure modes with a low risk level (from 23 in the baseline to 17 for wake steering), and an increase of failure modes with a medium (18 to 20) and high (0 to 4) risk level. Specifically fatigue-related failure modes of the tower and steel reinforcement bars were reclassified into a higher risk category because of expected increases in dynamic loads.

The applied FMECA methodology provides an effective framework for assessing the risks on offshore foundation integrity associated with wind farm control and can be applied to other components and operational scenarios. The insights from this study allow operators to better understand the correlations between control strategies and failure modes. The risk prioritization facilitates informed decision-making to ensure the asset integrity of offshore wind turbine foundations while optimizing the power output. For future studies, more quantitative indicators of the relevant failure modes could be integrated with minor adjustments of the framework.

In the presentation, the audience will learn about the specific failure modes identified in the study and foundation components affected. Furthermore, an extension of the analysis to other wind turbine components will be introduced.

Because of negative electricity prices or direct demand from the grid regulators, wind farms operators sometimes operate below optimal production. The frequency of these curtailment events is increasing and may now exceed the levels considered during design, which raise concerns about project lifetime. Past studies have shown that fatigue accumulation is reduced during curtailed operation in comparison to rated production but often ignored the transition between these states. Yet, these transitions are characterized by high-cycle stress fluctuations that can significantly reduce the lifetime of wind turbines if they occur frequently. The present study uses a coupled hydro-aero-elastic model of the offshore reference turbine NREL5MW to simulate curtailment. The damage accumulation at tower base and blade root is observed during de-rating operations across a range of environmental conditions. The pace and form of these transitions are analyzed in order to identify potential lifetime-conscious control practices. Additionally, simulations of similar turbines with onshore and bottom-fixed offshore foundations are compared to highlight the impact of waves on these curtailment-induced damages. The transition pace is identified as a key parameter for fatigue mitigation: de-rating operations faster than 10s can lead to substantially increased structural damage.

The increasing penetration of variable renewable energy sources—particularly wind and solar—has introduced significant challenges for grid operators, who must continuously balance supply and demand in real time. This has created a growing need for flexibility in the electrical grid, especially on short timescales, to accommodate fluctuations in generation and consumption. Offshore wind farms, with their large installed capacities and controllable output, are increasingly being called upon to provide such flexibility, notably through participation in intra-day electricity markets on a 15-min time interval basis. These markets enable dynamic adjustments in power delivery, often requiring turbines to undergo rapid curtailment or ramping events. While these interventions support grid stability, they also induce transient structural loads and complex control responses, potentially accelerating fatigue damage in critical components. Past studies focus on loads and damage in derated operation during curtailment events, once the turbine has been effectively downregulated. However, the effect of transients—the act of downregulating, not its resulting state—seems to have been mostly overlooked. Recently, structural health monitoring at the Belgian offshore wind farm NORTHER brought up the importance of transients during curtailment events and its potential for prematurely shortening the turbine tower’s lifetime, making it the principal topic of investigation in Horizon Europe project WILLOW.

In this context, this study aims to depict the relative significance of transient loads in curtailment events from a turbine control perspective. There, a degree of freedom is left in choosing the reference rotor speed the pitch controller should track, as well as its smoothing, and this appears to greatly affect changes in thrust force and hence fatigue damage. Turbine controllers are known to be subject to fierce intellectual property protection by turbine OEMs, and there is no reason to think downregulation is achieved in the same way across manufacturers and turbine models. For this reason, different control strategies are considered and first presented. Then, the corresponding fatigue damage on tower and blades is looked at, in both uniform and turbulent wind. The preponderance of transient loads (at start and end of curtailment) over stationary loads (during curtailment) is made clear, and the potential concern for the component’s lifetime is raised.

The increasing penetration of variable renewable energy sources—particularly wind and solar—has introduced significant challenges for grid operators, that must continuously balance supply and demand in real time. This has created a growing need for flexibility in the electrical grid, especially on short timescales, to accommodate fluctuations in generation and consumption.

Offshore wind farms, with their large installed capacities and controllable output, are increasingly being called upon to provide such flexibility, notably through participation in intra-day electricity markets on a 15-min time interval basis. These markets enable dynamic adjustments in power delivery, often requiring turbines to undergo rapid curtailment or ramping events. While these interventions support grid stability, they also induce transient structural loads and complex control responses, potentially accelerating fatigue damage in critical components. Past studies have focused on loads and damage in derated operation during curtailment events, once the turbine has been effectively downregulated. However, the effect of transients—the act of downregulating, and not its resulting state—seems to have been mostly overlooked. Recently, structural health monitoring at the Belgian offshore wind farm NORTHER brought up the importance of transients during curtailment events and its potential for prematurely shortening the turbine tower’s lifetime, making it the principal topic of investigation in the Horizon Europe project WILLOW.

In this context, this study aims to depict the relative significance of transient loads in curtailment events from a farm control perspective. Based on stationary Damage Equivalent Load matrices obtained within the project WILLOW—covering a wide range of wind speeds and turbulence intensities—this work extends the assessment of fatigue loading beyond stationary conditions. A hierarchical power dispatch approach is introduced, exploiting the degree of freedom available at the farm level to determine which turbines should adjust their production during curtailment. By distributing power setpoints across the farm, the controller aims to minimize the combined effect of stationary and transient fatigue damage. The achievable performance depends strongly on the plant’s control capabilities—from simple exclusion of turbines to turbine-specific setpoint adjustments—as well as on local wind conditions. In particular, correlations in wind speed between turbines play a key role in shaping optimal dispatch decisions, and are modeled using SINTEF’s FLAggTurb program. The controller's tuning and its effectiveness are assessed by replaying past curtailment events observed at the NORTHER offshore wind farm. The study highlights the potential trade-offs between flexibility provision and structural lifetime extension at the wind farm level.

Wind energy is a major source of clean, renewable power, but its efficiency depends strongly on wind speed, which is highly variable across spatial and temporal scales. Accurate short-term wind prediction is therefore essential for reliable power generation and grid integration. This study compares physics-based and machine learning approaches for short-term wind and power forecasting, focusing on accuracy and computational efficiency. Two methodologies were investigated: (i) physics-based modeling using an in-house mass-conservation code and the Weather Research and Forecasting (WRF) model, and (ii) data-driven modeling with a Long Short-Term Memory (LSTM) network trained on historical wind data. The test case involves an onshore wind turbine in Germany, a setting characterized by complex environmental interactions but readily generalizable to offshore conditions.
The in-house spatial resolution code failed to converge due to the limited availability of ground truth data from meteorological stations. Similarly, the WRF model proved difficult to optimize, as the large number of solvers and configuration options introduced instability and inconsistency in accuracy across spatial and temporal points. Moreover, the computational cost was significant, making WRF unsuitable for real-time wind prediction and active power estimation.
In contrast, the machine learning approach yielded more promising results. The LSTM model, known for its ability to capture both short- and long-term temporal dependencies, was trained, validated, and tested using over two years of turbine data with 10-minute averaged values. A hyperparameter optimization study was conducted to obtain the best-performing model, which aimed to predict the next wind speed value. Although the model achieved acceptable error metrics, it exhibited a temporal shift in predictions, phenomenon frequently reported in literature. This behavior, attributed to the high variability of wind and the model’s tendency toward persistence prediction, was mitigated through the application of Variational Mode Decomposition (VMD).
VMD was used to decompose the wind signal into distinct frequency modes (low-, medium-, and high-frequency components), facilitating more efficient learning. When applied offline and tested on an unseen dataset, the resulting model achieved excellent predictive accuracy, with R² = 0.997, RMSE = 0.11, and MAE = 0.085. However, while VMD has been studied in offline contexts, its online application remains underexplored. In real-time deployment, only incremental signal data are available, making full offline decomposition impractical. To address this limitation, we propose an online VMD strategy using a rolling window approach that performs decomposition in real time prior to prediction.
Our results demonstrate that machine learning, especially when combined with domain-specific corrections like online VMD, can deliver highly accurate forecasts at lower computational cost than traditional models. Physics-based approaches remain essential for understanding atmospheric processes and ensuring physically consistent predictions, but data-driven models offer a flexible, efficient framework for operational forecasting. We advocate a hybrid strategy that integrates physical interpretability with machine learning’s predictive power. Future work will explore ensemble methods, real-time data assimilation, and uncertainty quantification to enhance robustness and performance.