the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 10 Dec 2025
Enhancing yaw control resilience in wind turbines with CFD-informed digital twins
Abstract. Wind turbines are pivotal in the transition towards renewable energy. The operational conditions of these machines are continuously monitored through sensors that measure key indicators of efficiency and performance, including yaw angles, rotational speed, and vibrations. However, sensors are subjected to wear, degradation and consequent reduction in data reliability over time, which provides scope for developing a consistent and effective method to detect misinterpretation of turbine operating conditions caused by faulty measurements.
This research presents a novel method that integrates Computational Fluid Dynamics (CFD) simulations into a Digital Twin (DT) model to detect and correct yaw misalignment caused by faulty wind direction readings. Yaw error is estimated by interpolation across CFD-based performance data using live sensor measurements. The novel DT-based method was validated through experimental testing on a small-scale horizontal-axis wind turbine.
The results provide scope for a significant improvement in the resilience of wind turbines under conditions of sensor malfunctions, without the need for human intervention or supervision.
The proposed method is intended to be adaptable, enabling analysis of diverse failure modes under varying operational conditions. This work also advances condition monitoring and sustainable asset management, offering potential for a larger adoption across different turbomachinery applications.
- Preprint
(1639 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on wes-2025-216', Anonymous Referee #1, 22 Jan 2026
-
AC1: 'Reply on RC1', Lorenzo Carrattieri, 18 Feb 2026
We thank the reviewer for the comprehensive assessment and constructive remarks. The manuscript has been substantially revised to address the concerns raised.
First, the description of the Digital Twin has been clarified. The proposed method is now explicitly defined as a CFD-informed DT based on inverse interpolation, and its architecture and control logic are described more clearly, including how rotor operating conditions were prescribed in the simulations.
Second, the limitations of the steady RANS approach under real atmospheric inflow have been explicitly acknowledged. The CFD simulations are intentionally idealised (uniform inflow, turbulence modelled rather than resolved) to generate a structured aerodynamic response space. The link between the steady CFD framework and the outdoor experiments is now clarified through the numerical–experimental power-matching procedure.
Third, the issue of hub-height measurement and shear has been addressed, and the proof-of-concept nature of the small-scale experimental campaign has been emphasised.
Fourth, all 3D plots have been replaced with 2D line and contour plots to improve clarity, and a dedicated section has been introduced to explain the yaw inference mechanism. Quantitative validation metrics (MAE, RMSE, bias, cross-validation) have been added to demonstrate the performance of the CF correction and yaw detection accuracy.
Fifth, the term “novel” has been removed, and claims have been moderated to position the work as a physics-informed proof-of-concept rather than a universally superior alternative to existing yaw control strategies.
Finally, the Introduction and Conclusions have been revised to improve structure, reduce overstatement, and ensure a more conventional technical presentation.
We believe these revisions significantly strengthen the clarity, rigour, and positioning of the work. Thank you for your comments.
Detailed answers to the comments are given in the attached document.
-
AC1: 'Reply on RC1', Lorenzo Carrattieri, 18 Feb 2026
-
RC2: 'Comment on wes-2025-216', Anonymous Referee #2, 23 Jan 2026
General Comments
- The authors present an interesting experimental campaign of a small wind turbine rejecting yaw misalignments using a surrogate model created from CFD simulations. However, few details about the detection algorithm are provided and the results are sometimes redundant and difficult to understand. Please guide the reader through your process and experiment in your own words.
- The abstract would benefit from including more quantitative summaries of the results and methods.
- The manuscript would be strengthened by a clearer and more detailed explanation of the correction factor and misalignment detection process.
- Overall, the structure of the paper makes it difficult to follow. Most results are deferred to Chapter 3, while the reader is expected to accept claims in Section 2 without supporting evidence. A more integrated presentation, which introduces results alongside the methods would improve clarity and readability.
Methodology
- There is extensive discussion of implementation details (e.g., programming language, database structure, interpolation methods), but insufficient methodological detail for another researcher to reproduce the detection algorithm that is the primary contribution of this work. Consider refocusing this section on the scientific procedure and assumptions rather than software choices.
- It is not clear whether the mapping \gamma = f(C_p, TSR) yields a unique solution. Please address the potential for non-uniqueness and how it is handled.
Figures and Results
- Figure 5 would be clearer as a heatmap rather than a 3D surface plot. The current format makes it difficult to interpret gradients and trends.
- Consider combining figures where possible, as several convey largely overlapping information. For example, Figures 4, 5, and 6 appear to present similar content with limited incremental value.
- Figure 8 raises questions: Why does the CFD-derived yaw angle remain near zero for long periods and then exhibit sudden jumps? This behavior should be explained or justified.
- Please clarify your conventions for yaw angle, wind direction, and yaw misalignment.
- What are their respective bounds?
- Are any of these defined as inverses of one another?
- How are sign conventions handled consistently throughout the paper?
Citation: https://doi.org/10.5194/wes-2025-216-RC2 -
AC2: 'Reply on RC2', Lorenzo Carrattieri, 18 Feb 2026
General Comments
The authors present an interesting experimental campaign of a small wind turbine rejecting yaw misalignments using a surrogate model created from CFD simulations. However, few details about the detection algorithm are provided and the results are sometimes redundant and difficult to understand. Please guide the reader through your process and experiment in your own words.- We thank the reviewer for the positive assessment and constructive feedback. The organisation of the manuscript has been revised to improve clarity and guide the reader more clearly through the detection algorithm, experimental campaign, and validation process. Redundant elements have been streamlined, additional explanations have been added where needed, and the overall structure has been adjusted to improve the delivery of the methodology and results. To do so, a dedicated section 2.4 has been included in the manuscript to better differentiate between the measured power coefficient and the one derived through CFD.
The abstract would benefit from including more quantitative summaries of the results and methods.
- We thank the reviewer for the suggestion. The abstract has been revised to include quantitative details on the CFD campaign, validation procedure, and yaw correction performance, thereby providing a clearer summary of the methodology and key results.
The manuscript would be strengthened by a clearer and more detailed explanation of the correction factor and misalignment detection process.
- We thank the reviewer for the comment. The manuscript has been revised to provide a clearer and more detailed explanation of both the correction factor (CF) formulation and the yaw misalignment detection process. The steps used to construct the CFD-based interpolator, compute the pointwise CF residuals, fit the correction surface, and apply the corrected power coefficient within the DT inference loop are now explicitly described, including the relevant equations. These additions are intended to guide the reader more transparently through the algorithmic workflow.
Overall, the structure of the paper makes it difficult to follow. Most results are deferred to Chapter 3, while the reader is expected to accept claims in Section 2 without supporting evidence. A more integrated presentation, which introduces results alongside the methods would improve clarity and readability.
- We thank the reviewer for this important observation. The manuscript structure has been revised to improve logical flow and readability. Key methodological elements are now more clearly motivated, and explanatory transitions have been added to avoid unsupported claims. Additionally, selected results and clarifications have been repositioned to reduce forward referencing and improve coherence between methods and validation sections.
Methodology
There is extensive discussion of implementation details (e.g., programming language, database structure, interpolation methods), but insufficient methodological detail for another researcher to reproduce the detection algorithm that is the primary contribution of this work. Consider refocusing this section on the scientific procedure and assumptions rather than software choices.- We thank the reviewer for this valuable suggestion. The section has been substantially revised to refocus on the scientific procedure and underlying assumptions rather than implementation details. The DT yaw detection algorithm is now presented as a structured four-step methodology (data conditioning, performance correction, yaw inference, and fault decision logic), with explicit definitions of the correction factor, decision thresholds, and sign inference strategy. These additions clarify the reproducible detection logic and isolate the methodological contribution from software-specific aspects.
It is not clear whether the mapping \gamma = f(C_p, TSR) yields a unique solution. Please address the potential for non-uniqueness and how it is handled.
- We thank the reviewer for the comment. The manuscript has been revised to clarify that the inversion \gamma=f(C_p,TSR,U_\infty) is performed within CFD-derived response surfaces that exhibit smooth, monotonic behaviour. Under these conditions, the inversion is well-posed and yields a unique solution for \gamma. This assumption and its validity within the investigated operating domain are now explicitly stated in Sect. 2.5.3.
Figures and Results
Figure 5 would be clearer as a heatmap rather than a 3D surface plot. The current format makes it difficult to interpret gradients and trends.- We thank the reviewer for the suggestion. The original 3D surface plot has been removed and replaced with 2D line plots (Fig. 5), which allow clearer interpretation of performance trends and quantitative comparison across yaw angles and inflow regimes. In addition, the yaw inference process is now explained in a dedicated section (Sect. 3.7) with an accompanying 2D contour representation (Fig. 7), further improving clarity and readability.
Consider combining figures where possible, as several convey largely overlapping information. For example, Figures 4, 5, and 6 appear to present similar content with limited incremental value.
- We thank the reviewer for the feeback. Figures have been consolidated to remove redundancy and improve clarity. In particular, the former Fig. 6 has been absorbed into the revised Fig. 5, resulting in a more compact and coherent presentation of the CFD performance data.
Figure 8 raises questions: Why does the CFD-derived yaw angle remain near zero for long periods and then exhibit sudden jumps? This behavior should be explained or justified.
- We thank the reviewer for the comment. The periods during which the CFD-derived yaw angle remains close to zero correspond to intervals where the turbine aerodynamic performance is consistent with the CFD reference surface. The sudden jumps occur during rapid wind direction changes (up to −18°) under gusty inflow conditions, which induce transient variations in C_{p,new}^{meas} and TSR. These short-term aerodynamic fluctuations momentarily project the operating point onto regions of the CFD performance space associated with non-zero yaw angles.
However, since the misalignment criteria were not persistently satisfied across consecutive intervals, no fault state was triggered and the yaw control remained governed by the wind vane measurements. This explanation has now been clarified in the manuscript.
Please clarify your conventions for yaw angle, wind direction, and yaw misalignment.
-What are their respective bounds?
-Are any of these defined as inverses of one another?
-How are sign conventions handled consistently throughout the paper?- We thank the reviewer for this important request. A dedicated subsection 2.3.1 (Definition of angular quantities and sign conventions) has been added to explicitly clarify the conventions used throughout the manuscript. The wind direction \theta_w, nacelle yaw \theta_y, and yaw misalignment angle \gamma are now formally defined within a shared ground-based reference frame, including their bounds and sign conventions.
Citation: https://doi.org/10.5194/wes-2025-216-AC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 324 | 156 | 26 | 506 | 16 | 30 |
- HTML: 324
- PDF: 156
- XML: 26
- Total: 506
- BibTeX: 16
- EndNote: 30
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The paper investigates whether parametric sweeps of steady-state CFD simulations can be used to detect and correct real-time wind turbine yaw misalignment. The authors test their method on a small-scale outdoor experimental rotor setup.
The authors strongly emphasize that they developed a digital twin of the rotor, however it seems that they are running steady-state CFD simulations and use linear interpolation to create a simple surrogate of the tsr, wind speed, yaw misalignment, Cp surface for the particular turbine they tested. They do not explain clearly how the turbine is controlled in the simulations. Simulations are performed for uniform inflow and without modelling turbulence, even though the turbine was tested close to the ground, which usually exhibits strong shear and variability. The wind speed reference is taken just above ground height, far below the hub height. Mostly 3D plots are used, which are difficult to interpret and it is difficult to follow how the surrogate model was built and how steady-state simulations should be adequate to correct yaw misalignment in real-time. It is not shown whether the yaw misalignment from their “digital twin” is actually correct. In the plots they seem quite off from the actual yaw misalignment. Without comparing their control method to other established forms of yaw control or more recent proposals, they claim that their “novel” method is superior, which is a strong claim.
The structure needs to be improved as well as the introduction that only focuses on very recent work. The conclusion seems to be written by an AI and is very general.
More detailed comments are given in the attached document.