Fault Detection Method for Wind Turbine Generators Based on Attention-Based Modeling

Round 1

Reviewer 1 Report

Can you explain the experimentation procedure used to obtain two types of wind turbine gearbox datasets, along with images of the experimental setup?

What is the speed of the wind turbine gearbox, and what is the sampling frequency used to collect vibration signals?

The research is not reported the influence of speed of the wind turbine gear box on the characteristics of vibration signal

Table 1 and Table 2 are not presenting any vibrational signals

EMD can be sensitive to noise in the data, especially when the noise level is comparable to the amplitude of the modes. What steps have been taken care of during the data acquisition?

The details of the signal characteristics of the gearbox faults are not presented.

Explain the method used to remove unstable data from the signal of the faulty gearbox to maximize the accuracy of the diagnosis

Why sample data was decomposed into only 7 IMF components

Why did the author not use regression methods, considering that the architecture is simple for the study, given that there is only one neuron in the output layer?

Present convergence results

 

Address the novelty of the work presented in the paper

Improve the quality of the enlish

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The article is well written and presents interesting topic. Please include some pseudocode or set of steps in which the respective equation number and/or section number are mentioned so that the readers can follow the work in a better way. Moreover, in Fig. 6, please prefer using the different line styles with lesser thickness so that the trends can be easily differentiated. Some of the figures are blur, please improve them.

Thanks

 

 

nil

 

Author Response

Point : The article is well written and presents interesting topic. Please include some pseudocode or set of steps in which the respective equation number and/or section number are mentioned so that the readers can follow the work in a better way. Moreover, in Fig. 6, please prefer using the different line styles with lesser thickness so that the trends can be easily differentiated. Some of the figures are blur, please improve them.

 

Response : Thank you for the questions asked by the reviewer and I am very sorry that the paper is deficient. The paper has been revised according to your comments. In addition, Figure 6 has been made more legible by increasing the clarity and enlarging the font size of the legend.

This part has been made changes and marked yellow in the revised manuscript.

Author Response File: Author Response.pdf

Reviewer 3 Report

1. Improvement is needed for the Abstract section. The background of research is missing and need to be included. The quantitative results should be reported in the Abstract too.

2. Authors are advised to summarize those existing works explained in Lines 39 to 80 in table form. Please explain the difference between the proposed work and these existing works too.

3. The problem statements and research challenges encountered in the existing works are not clearly explained and need further elaboration. It is also advised to present this in separate paragraphs since the Paragraph 3 (Lines 39 to 85) is already quite lengthy.

4. Please elaborate the technical contributions and novelty of current work to justify the research significance.

5. Some inconsistency is found in the equation of alpha_i in Eq. (5). Why the index in summation symbol is presented as j instead of i?

6. Please check the contents presented in Lines 198 to 203. The parameter X has been mentioned several time and it represents different meanings in different sentence.

7. The descriptions of implementation details need further improvement. It is quite confusing to see numbering such as (1), (2), (3), (4) as well as Stage 1, Stage 2 and etc. together in the same paragraph. It is advised to explain each essential stage in separate paragraph for better clarity.

8. The details used for model training are not provided. What are the loss functions used?

9. Table caption should be presented on the top of table instead of bottom.

10. Fig. 4 needs to be enlarged. 

11. Fig. 6 - The font sizes used for descriptions of x-axis, y-axis and legend are too small. Please increase the front sizes used.

12. Authors are encouraged to share their source code publicly to benefit other researchers.

13. Please discuss the limitations of current works and the potential future works can be derived. 

Some typos and grammatical errors are found. Please proofread the paper again before resubmission.

Author Response

请参阅附件。

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors tried to address my comments. The paper can be accepted.

Reviewer 3 Report

All comments have been addressed by the authors. No further comments.