Fault Diagnosis of Rolling Bearing Acoustic Signal Under Strong Noise Based on WAA-FMD and LGAF-Swin Transformer

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. The quantitative results should be added to the abstract.
2. The type of bearing and fault should be reflected in the abstract. Revise the abstract.
3. At the end of section 1, various approaches available can be presented as a Table with their merits and demerits. That could be easy to understand the research gap.
4. An image of the experimental setup needs to be added.
5. For noise reduction, convergence of the optimization algorithm needs to be verified.
6. What are the changes that have been incorporated into the proposed algorithm (E)?
7. Figure 12 needs to be enhanced for better visibility.
8. The authors considered only Inner and outer race failures. Why?
9. Consistency should be maintained for LGAF-Swin Transformer or LGAF-ST.
10. Kindly ensure that no overfitting for the proposed method.
11. The proposed LGAF-ST yields enhanced performance. Why? What are modifications present in the model to improve the accuracy?
12. In Figure 16, the sensor used was a Microphone, not an AE sensor, Kindly revise it.
13. While capturing the sound signal using a Microphone, how the noise signals are filtered or canceled? Since it can capture the sound in all 360 degrees.
14. Figure 17 is not clear to see the fault conditions.
15. Classification reports with TPR, FNR, F1 score, Kappa statistics, and ROC need to be added.
16. A separate discussion section is required to brief the research findings.

Author Response

Dear Reviewer,

It is a great honor and pleasure to receive your valuable review and insightful comments on our manuscript entitled "Fault Diagnosis of Rolling Bearing Acoustic Signal under Strong Noise Based on WAA-FMD and LGAF-Swin Trans-former" Your constructive suggestions have significantly contributed to improving the overall quality and depth of our work, and have also provided us with new directions and objectives for our future research. We are sincerely grateful for your thoughtful guidance.

 

Throughout the revision process, we have carefully studied and addressed each of your comments with the utmost diligence and respect. We fully recognize that every excellent academic paper benefits greatly from the rigorous scrutiny and insightful feedback of experienced experts like yourself. Your suggestions have helped us optimize the model design, refine the experimental validation, and enhance the clarity and scientific rigor of the paper.

 

If you have any further comments or suggestions, we would be truly grateful to receive them. We remain committed to addressing any issues to the best of our abilities, in pursuit of the highest academic standards. Under your guidance, we sincerely hope this work will evolve into a more robust and valuable contribution to the field.

 

Thank you once again for your time, support, and professional insight. We genuinely appreciate the opportunity to revise our work under your careful review.

 

With deep respect and sincere gratitude.

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript introduces a novel combination of the Weighted Average Algorithm (WAA) to optimize Feature Mode Decomposition (FMD) and a Swin Transformer–based architecture for fault diagnosis. The authors were requested to address the following comments.

Clarify how each new module (Bilateral Augmentation, Depthwise Separable Convolution, Local-Global Attention Mechanism, and Adaptive Feature Selection Module) interacts with the others.

Specify how the time-domain signals are preprocessed prior to applying the Continuous Wavelet Transform (CWT). 

Provide more details about the hyperparameter settings for training the transformer models, such as batch size, learning rate (beyond “0.001”), optimization strategy (e.g., rationale for choosing SGD), and the number of training epochs for each model variant.

Consider including statistical tests or confidence intervals for the reported improvements (especially given that multiple independent tests were conducted) to demonstrate the significance of the performance gains.

Provide the discussion of potential limitations or scenarios where the proposed method might be less effective (e.g., under different noise conditions, with alternative bearing types, or on much larger datasets).

Please include Suggestions for future work to address these limitations, such as extensions to other types of fault signals or integration with multi-modal sensing data.

Some equations appear to be embedded in run-on text. Setting them apart in proper mathematical environments and ensuring they are consistently numbered and referenced can improve clarity.

Comments on the Quality of English Language

Proofreading is required

All acronyms (e.g., FMD, WAA, BA, DSC, LGAM, AFSM, HAL) should be defined clearly on their first occurrence. After their definition, ensure that their usage remains consistent throughout the manuscript.

Author Response

Dear Reviewer,

It is a great honor and pleasure to receive your valuable review and insightful comments on our manuscript entitled "Fault Diagnosis of Rolling Bearing Acoustic Signal under Strong Noise Based on WAA-FMD and LGAF-Swin Trans-former" Your constructive suggestions have significantly contributed to improving the overall quality and depth of our work, and have also provided us with new directions and objectives for our future research. We are sincerely grateful for your thoughtful guidance.

 

Throughout the revision process, we have carefully studied and addressed each of your comments with the utmost diligence and respect. We fully recognize that every excellent academic paper benefits greatly from the rigorous scrutiny and insightful feedback of experienced experts like yourself. Your suggestions have helped us optimize the model design, refine the experimental validation, and enhance the clarity and scientific rigor of the paper.

 

If you have any further comments or suggestions, we would be truly grateful to receive them. We remain committed to addressing any issues to the best of our abilities, in pursuit of the highest academic standards. Under your guidance, we sincerely hope this work will evolve into a more robust and valuable contribution to the field.

 

Thank you once again for your time, support, and professional insight. We genuinely appreciate the opportunity to revise our work under your careful review.

 

With deep respect and sincere gratitude.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

• Why Swin Transformer is choosed?
• What dynamic convolution kernel adjustment?
• mechanical fault which feature is required to choose decomposition method?
• What is manual feature selection?
• Why time-frequency maps inputs for the deep learning model?
• Use of An Integrated Signal Quality Index (ISQI)? Waht is the range ?
• Why author choose Dragonfly Algorithm (DA) and Grey Wolf Optimizer (GWO)  as benchmark any siginificance?
• When the number of iterations P is satisfied?
• In Figure 3. Model structure diagram of Swin Transformer what is first block and second block represent?
• What is the impact of Parameters of LGAF-Swin Transformer model?
• Why Global Pool require kernel size 4x4?
• Why we need to perform both Depth wise Convolution and Point wise Convolution?
• Flow of the paper need to check before resubmission.

Author Response

Dear Reviewer,

It is a great honor and pleasure to receive your valuable review and insightful comments on our manuscript entitled "Fault Diagnosis of Rolling Bearing Acoustic Signal under Strong Noise Based on WAA-FMD and LGAF-Swin Trans-former" Your constructive suggestions have significantly contributed to improving the overall quality and depth of our work, and have also provided us with new directions and objectives for our future research. We are sincerely grateful for your thoughtful guidance.

 

Throughout the revision process, we have carefully studied and addressed each of your comments with the utmost diligence and respect. We fully recognize that every excellent academic paper benefits greatly from the rigorous scrutiny and insightful feedback of experienced experts like yourself. Your suggestions have helped us optimize the model design, refine the experimental validation, and enhance the clarity and scientific rigor of the paper.

 

If you have any further comments or suggestions, we would be truly grateful to receive them. We remain committed to addressing any issues to the best of our abilities, in pursuit of the highest academic standards. Under your guidance, we sincerely hope this work will evolve into a more robust and valuable contribution to the field.

 

Thank you once again for your time, support, and professional insight. We genuinely appreciate the opportunity to revise our work under your careful review.

 

With deep respect and sincere gratitude.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

This paper proposes an innovative hybrid fault diagnosis method for rolling bearings by combining an optimized Feature Mode Decomposition with an optimized Swin Transformer. The paper presents experimental results that demonstrate the proposed method's superior performance compared to traditional state-of-the-art methods in both denoising and fault diagnosis.

1. The abbreviations used in the title (WAA-FMD, and LGAF) were not expanded in the abstract, the full name of the methods presented in the paper would be more helpful, for example, the word Weighted Average Algorithm (WAA) was not mentioned until the third page (line 112).
2. While the paper discusses the prior denoising methods (EMD, VMD, FMD…) as well as fault diagnosis methods, an explanation of the reason why existing Transformer-based approaches fail in noisy acoustic scenarios would strengthen motivation.
3. In equation (1), the author does not explain why the test functions F1, F2, F3, F4 are the standard. More details about the function variables used in the equation should be given.
4. In section 2.1 in Methodologies, the paper does not discuss how some critical parameters (e.g. w1 = 0.3, w2 = 0.7, in ISQI) were selected, the paper cites another study, however a brief explanation of how and why these weights’ values were selected would be helpful. Were these values selected experimentally or derived theoretically?
5. In the methodologies section, algorithm tables should be presented. Flowcharts are good but lack details.
6. In figure 3 and figure 8, no explanation of the variables shown was provided (e.g. H, W, Z^l, Z^l+1…).
7. In the open-source data verification section, the cited source for data [31] does NOT match the data claimed in the section (KAIST data).
8. In the experimental verification (section 5), only one type of bearing was tested (FAG 30205-XL). Including more bearing geometries would strengthen the experiment.
9. In the experiment, only raceway faults are tested, including other types of faults such as lubrification faults would strengthen the experiment.
10. Reducing computational complexity was mentioned multiple times in the methodologies section, however, the experimental section does not list or discuss in detail the computational costs of the proposed model.

Author Response

Dear Reviewer,

It is a great honor and pleasure to receive your valuable review and insightful comments on our manuscript entitled "Fault Diagnosis of Rolling Bearing Acoustic Signal under Strong Noise Based on WAA-FMD and LGAF-Swin Trans-former" Your constructive suggestions have significantly contributed to improving the overall quality and depth of our work, and have also provided us with new directions and objectives for our future research. We are sincerely grateful for your thoughtful guidance.

 

Throughout the revision process, we have carefully studied and addressed each of your comments with the utmost diligence and respect. We fully recognize that every excellent academic paper benefits greatly from the rigorous scrutiny and insightful feedback of experienced experts like yourself. Your suggestions have helped us optimize the model design, refine the experimental validation, and enhance the clarity and scientific rigor of the paper.

 

If you have any further comments or suggestions, we would be truly grateful to receive them. We remain committed to addressing any issues to the best of our abilities, in pursuit of the highest academic standards. Under your guidance, we sincerely hope this work will evolve into a more robust and valuable contribution to the field.

 

Thank you once again for your time, support, and professional insight. We genuinely appreciate the opportunity to revise our work under your careful review.

 

With deep respect and sincere gratitude.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Congrats to the authors.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have effectively addressed all my review comments. No further revisions are necessary.

Reviewer 4 Report

Comments and Suggestions for Authors

The authors have revised the manuscript in accordance with the comments. It can be accepted as it is.