Combining Computational Fluid Dynamics and Gradient Boosting Regressor for Predicting Force Distribution on Horizontal Axis Wind Turbine

Round 1

Reviewer 1 Report

This paper focuses on the implementation of an AI/ML technique, specifically gradient boosting regression, as an option for expediting a traditional computational fluid dynamics analysis.  Improving understanding and prediction of forces and responses of wind turbine blades is incredibly important.

Paper focuses on the explanation of a technique that will hopefully be expanded on in future work.  I think it is an interesting application of AI/ML to expedite CFD work and the suggested future application of estimating fatigue life of blades could prove to be useful to the design process.

 

Items to address:

I don't agree with the change in force from 0.01m elements to 0.001 elements being negligible, it's a 7.5% change in force results.  You state that 0.005 would be best to chose, but you don't ever explicitly say which sizing is used.  I'm not sure if this is a 'lost in translation' issue or just needs additional information.

I don't see any value added from showing or discussing the Pearson correlation coefficients.  If the reader doesn't know that there is a correlation between velocity and drag/lift (and angle of attack) then they should be reading an introductory textbook and not a research paper. If you disagree, then the issue may be that there is not enough context/discussion on this topic to explain why it is important.

 

 

Other things worth considering:

You could use the 'blockage ratio' to help describe your domain sizing.  (frontal area of the blade divided by the total inlet area for your case).  It's a single number that is easily interpreted and quickly helps to show that your domain is an appropriate size.

 

The CFD plots could have unified coloring scales for easier comparison.

 

Your Position vs Force plots are labeled as percent values (%) on the x-axis, but only go from 0-1.

 

Grammar & word usage: There are some relatively minor issues with grammar and word usage. A few examples:  "One of the reason of increase",  "numerical based approach that solvers Euler", "It be observed"

 

Currently there are two sections labeled '3.2'.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The manuscript details the application of Gradient Boosting Regressor (GBR) to predict the wind turbine response to a combination of wind speed, angle of attack, and turbulence intensity when the air flows over the rotor blade. Overall, the manuscript is well structured and the relation between the used methods are clear. Discussion is detailed and interesting to the field. I would suggest, however, that the authors relate their method, results and findings with the thematic of vibrations, as it would widen the range of the paper and make it more suited for this journal.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The reviewer’s comments on a paper submitted to MDPI Vibration No. 1113400

“Coupling Computational Fluid Dynamics and Gradient Boosting Regressor for Predicting Force Distribution on Horizontal Axis Wind Turbine”

This study applies the Computational Fluid Dynamics (CFD) as well as Gradient Boosting Regressor (GBR) – as a machine learning method – to predict the response of a horizontal axis wind turbine (HAWT) to a combination of wind speed, angle of attack, and turbulence intensity when the air flows over the rotor blade. The theoretical contribution of the paper is limited and the paper cannot be accepted in its current form. The following comments are provided to authors to consider:

[1] Introduction is short and incomplete. It doesn’t clearly outline the research gaps in existing CFD modelling, the aim and objectives of the research paper, as well as the novelty (scientific contribution) of the research method.

[2] Only few papers from the literature (about predicting force distribution using CFD) were selected and reviewed. The theory section doesn’t explain the state-of-the-art of CFD and ML techniques adopted in predicting wind turbine loads.

[3] The reviewer wonders why the word ‘coupling’ has been used in the research. Figure 2 doesn’t justify this. This figure shows how two methods are combined. This figure requires further explanation.

[4] number of subsections in Section 3 is wrong. There are two 3.2 subsections. Sections 2, 3 and 4 represent all together “research methodology”. The reviewer suggests to merge them. Title of section 4 “Machine Learning Methodology” is too broad. The authors should specify what algorithm of ML is adopted.

[5]  Section 5 includes many figures with only little explanation. How was the data collected? What proportion of data was used for training purpose and what for testing. How was the performance of ML algorithm? Any comparison was made between this algorithm and other ML algorithms?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf