A machine learning-based approach for active monitoring of blades pitch misalignment in wind turbines (Milani et al. 2024)

 Summary Comments:

Thank you for the opportunity to review this paper.

As discussed in this study, detecting misalignment of turbine blades is complex yet a key component to successful wind energy implementation, especially for offshore wind or in areas where maintenance is more difficult and costly. This study is an important contribution to the field, as it introduces a new method for detection of wind turbine misalignment, and more so, advances on any previous literature by the ability to detect which blades are misaligned. This paper is well written, and nicely walks the reader through their methodology; proving how this approach is more advanced than previously existing methods.

Overall, I think this paper has a lot to contribute to the field and demonstrates thorough analysis. My below comments are suggestions that I believe will help to improve the clarity of the paper.

Major Comments: 

Line 69: My one main comment/concern for this paper is the reproducibility. You mention that your method uses a "minimal set of sensors", but you never mention which sensors/measurements you use from the wind turbine. Without a clear discussion of what in-situ measurements are needed, this method will not be able to be reproduced or tested in the field.

Minor Comments:

Line 15-20: You may want to add a brief discussion about what typical misalignment ranges are and how frequently this occurs on a given turbine or within an array. Does turbine misalignment decrease energy production? Also, is there a significant threshold where turbine misalignment becomes more of an issue/concern?

Line 147: Would it be beneficial to show the power curve here?

Line 155: What ranges of wind speeds are you considering? Is there a difference in algorithm performance when the turbine is just above cut-in vs. at rated capacity?

Figures 2 and 3: I would recommend adding (a) and (b) to each figure. It will improve clarity when references the figures in the text.

Thank you for the opportunity to review this paper. Also, thank the authors for submitting this well written manuscript.



The paper presents a novel machine learning-based approach for automatic pitch misalignment detection and localization in wind turbines. As marked in the manuscript, the misalignment of blades is usually difficult to inspect and detect (especially for the offshore turbine systems). The reported approach can not only predict the exact entity of the misalignment, and localize the affected blades with high accuracy, but also only relies on features extracted from a limited set of sensors already integrated into modern wind turbine systems. This ensures the well-tuned model can be applied easily to modern turbine systems.

Overall, I consider this manuscript well written with clear overview of the addressed problem and provided contribution. The methodology used in the manuscript is well designed and presented. I only have one question and several minor comments below that I would like to get discussed and addressed by the authors before publication.

Questions:

I noticed that the manuscript focuses on the scenarios with unbalanced blade misalignments. These unbalanced cases ensure the Yawing moments differ from the healthy case. I was wondering if the Yawing moments will still be at the non-N Rev positions when the three turbine blades have the same amount of misalignments. In this case, can the proposed ML model make accurate predictions?

Minor Comments:

1. I recommend adding the explanation of “V21” (wind speed 7m/s) and “V07” (wind speed 21m/s) in the caption of the Figure 2 to increase the readability.
2. I recommend editing Figure 6 to show the overlapping between two data points. The current figure has the predicted result covering over the comparing data, which cannot provide a clear view of the matching status of the two sets of data.

Please find my comments on this paper in the attached document.