Detection of Diffusion Interlayers in Dissimilar Welded Joints in Processing Pipelines by Acoustic Emission Method
, Vladimir Bardakov 1, Dmitrij Arzumanyan 1, Sergey Ushanov 1, Marina Karpova 1, Egor Lepsheev 1,2 and Sergey Elizarov 2
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper considers the application of neural networks to detect diffusion interlayers in dissimilar welded joints using the acoustic emission (AE) method. Before the publication of this paper, the author needs to further consider the following issues:
1. The title and introduction require further revision by the author. Both the title and introduction suffer from a lack of emphasis on key points. The author should highlight the features of this paper (or its innovative aspects) in the title. There are already a large number of articles on the application of neural networks in the study of acoustic emission for structural defects; therefore, in the introduction, the author should summarize the research findings of previous scholars, emphasize the differences between this paper and previous research methods and results, in order to highlight the innovation of this paper.
2. In Figure 1, is "Acoustic waveguide smulation" a typographical error for "Acoustic waveguide simulation"?
3. On line 172, why is a 100...300 kHz digital filter and a 32 dB threshold used to suppress noise?
4. Why is acoustic emission (AE) simulation necessary? Using laboratory data seems to better illustrate the issue; what is the role of numerical simulation in this paper?
5. The numbers and letters in Figure 11 are unclear; please correct them.
6. The last part of the text is labeled as "4. Discussion." The authors should clearly define which part contains the specific conclusions, rather than ending with "Discussion."
Author Response
Dear Reviewer!
Thank you for your valuable critical and detailed analysis of our paper. We made a number of changes to the manuscript (marked areas in the revised version), and also presented below our response on your comments and suggestions.
Please, see our comments in the attached file.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors1. What are the innovations and main contributions of this study compared to other studies in the existing literature? The authors need to further elaborate on the innovativeness of this study.
2. How to ensure that the training dataset used is representative of the variety of conditions that may be encountered in real industrial processes?
3. Does the neural network model presented in the article have the ability to generalize across different types or conditions of welded joints? Is there any validation associated with it?
4. Among the many possible features, why was wavelet kurtosis chosen as the key feature? Are there any studies comparing it with other features?
5. The article mentions that the classification effect is significant at signal-to-noise ratios less than -5 dB, how does the model perform at higher signal-to-noise ratios?
6. How robust is the model if a different type of noise than the training data is encountered in a real application? Are there any corresponding tests?
7. Have the models in the article been validated in a real industrial environment? What are the results?
8. How computationally efficient are the neural network models in real-world applications? Are they suitable for real-time or near real-time monitoring?
9. Neural networks are often considered as “black box” models, how can the predictions of the models be interpreted?
10. In the process of model training, the combination of hyper-parameters has a great impact on the model performance, how to determine the best combination of hyper-parameters?
Comments on the Quality of English Language1. What are the innovations and main contributions of this study compared to other studies in the existing literature? The authors need to further elaborate on the innovativeness of this study.
2. How to ensure that the training dataset used is representative of the variety of conditions that may be encountered in real industrial processes?
3. Does the neural network model presented in the article have the ability to generalize across different types or conditions of welded joints? Is there any validation associated with it?
4. Among the many possible features, why was wavelet kurtosis chosen as the key feature? Are there any studies comparing it with other features?
5. The article mentions that the classification effect is significant at signal-to-noise ratios less than -5 dB, how does the model perform at higher signal-to-noise ratios?
6. How robust is the model if a different type of noise than the training data is encountered in a real application? Are there any corresponding tests?
7. Have the models in the article been validated in a real industrial environment? What are the results?
8. How computationally efficient are the neural network models in real-world applications? Are they suitable for real-time or near real-time monitoring?
9. Neural networks are often considered as “black box” models, how can the predictions of the models be interpreted?
10. In the process of model training, the combination of hyper-parameters has a great impact on the model performance, how to determine the best combination of hyper-parameters?
Author Response
Dear Reviewer!
On behalf of the all authors, I thank you for taking the time to carefully work on our paper. In turn, we also tried to give detailed answers to your questions and make appropriate corrections to the manuscript.
Please, see our comments in the attached file.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsI have no other comments.
Comments on the Quality of English LanguageI have no other comments.
