Detection of Train Wheelset Tread Defects with Small Samples Based on Local Inference Constraint Network
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsIn the manuscript titled “Detection of train wheelset tread defect with small samples based on local inference constraint network”, the authors proposed a method for detecting tread defects in wheelsets based on a local inference constraint network and predicted the tread defect problem with data from repair shop workshops. The main purpose of the method is to solve the problem of insufficient feature space due to small samples. However, there are still some problems with the content and formatting of the article, and the following are some of my specific suggestions for changes in the article, this manuscript may be accepted after minor modification.
Minor comments:
1. The authors should consider modifying the layout of all figures, including the alignment of the underlying graphic, the size of the graphic, etc., to keep it aesthetically pleasing.
2. What do the elements in the set of formulas (5, 6) represent?
3. Line 173, the description of the softmax function in Figure 3 is missing from the text.
4. The conclusion section needs to be supplemented with a discussion of the methodology of this paper and future research directions.
5. The authors should carefully revise the English expressions in the manuscript.
6. In introduction, it should point out that the cost of detecting faults using the method of images is high, while the cost of detecting them using the method of acceleration is low. From Ref. [doi.org/10.1016/j.measurement.2022.111268, doi.org/10.1016/j.ymssp.2022.109856], it is clear that the wheel fault information contained in the acceleration is more significant when the train speed increases, but it is not the same using image recognition. When the speed of the train increases, the blurring of the captured image will increase the difficulty of diagnosis.
7. Detecting wheel-rail relationship with machine vision, Ref. [doi.org/10.1016/j.ress.2024.110087] also mentions that when the speed is higher, the number of recognized frames cannot meet the requirements, and the authors should consider the problem of real-time diagnosis in the future work.
Author Response
Thank you for your letter and for the reviewers’ comments concerning our manuscript titled “Detection of train wheelset tread defect with small samples based on local inference constraint network”. Those comments are all valuable and very helpful for revising and improving it, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper.
The responses are as following:
Q1. The authors should consider modifying the layout of all figures, including the alignment of the underlying graphic, the size of the graphic, etc., to keep it aesthetically pleasing.
Response: Thanks for your suggestions. We have tried our best to modify the layout of all figures and hope that the correction can meet with approval.
Q2. What do the elements in the set of formulas (5, 6) represent?
Response: Sorry for our negligence. The corresponding representation of all formulas has been modified and improved in the revised manuscript. Meanwhile, the formatting issue of variables in the formulas has been addressed.
Q3. Line 173, the description of the softmax function in Figure 3 is missing from the text.
Response: Sorry for our negligence of the description. We have added the corresponding function description in formula (5) and in Section 1.3.
Q4. The conclusion section needs to be supplemented with a discussion of the methodology of this paper and future research directions.
Response: Thanks for your suggestion. We have added a discussion on the methodology and future research directions in Section 3.
Q5. The authors should carefully revise the English expressions in the manuscript.
Response: Thanks for your suggestion. We have tried our best to review the entire paper and made correction to English expression issues, which marked in the red in the revised manuscript.
Q6. In introduction, it should point out that the cost of detecting faults using the method of images is high, while the cost of detecting them using the method of acceleration is low. From Ref. [doi.org/10.1016/j.measurement.2022.111268, doi.org/10.1016/j.ymssp.2022.109856], it is clear that the wheel fault information contained in the acceleration is more significant when the train speed increases, but it is not the same using image recognition. When the speed of the train increases, the blurring of the captured image will increase the difficulty of diagnosis.
Response: Thanks for your suggestion. We have cited and elaborated on the corresponding detection methods from Ref. [doi.org/10.1016/j.measurement.2022.111268, doi.org/10.1016/j.ymssp.2022.109856] in the introduction, and explained the related issues in Section 3.
Q7. Detecting wheel-rail relationship with machine vision, Ref. [doi.org/10.1016/j.ress.2024.110087] also mentions that when the speed is higher, the number of recognized frames cannot meet the requirements, and the authors should consider the problem of real-time diagnosis in the future work.
Response: Thanks for your suggestions. The problem that the number of recognized frames cannot meet the requirements is one of our future research directions, and the corresponding description has been added in Section 3.
In addition, we have tried our best to improve the manuscript and made some changes. These changes do not affect the content and framework of the paper. We did not list the changes here but marked them in red in the revised paper.
We sincerely appreciate your thorough review and hope that the corrections meet your approval.
Once again, thank you very much for your comments and suggestions.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsIn this paper, a method for detecting defects in the tread surface of train wheelsets based on local inference constraint networks is proposed, and although the experimental results prove the effectiveness of the proposed method, there are also shortcomings in this paper.
1. Figure 3 in this paper is poorly drawn and it is recommended that it be redrawn so that it clearly reflects the structure of the model.
2. Three methods were used to generate the dataset in Section 2.2.1 of this paper, but Figure 4 only gives an example of an image from one of the generation methods, and it is recommended that the graphs generated by all three methods be partially displayed, with appropriate analytical notes added.
3. The analysis of Table 3 in this paper is not consistent with what is presented in Table 3 and it is recommended that it be rechecked and revised.
4. In this paper, Table 4 appears to be exactly the same as Table 3, and the authors are asked to explain why.
5. The analysis of the experimental results for the Table 4 embodiment in Section 2.2.3 of this paper is relatively small, and it is recommended that a detailed analysis of the results be added and that it be clarified which of the added attentional mechanisms in this paper is being used.
6. The model proposed in this paper is only slightly higher than Resnet-50 in terms of recall and accuracy, can it show the effectiveness of the method proposed in this paper? It is suggested to add a more detailed description of the result analysis.
7. How effective is the method proposed in this paper in practical testing? There is a lack of practical validation of the comparative results.
8. There are many problems with the English writing in this paper, and the author is advised to check and revise the grammatical writing throughout the article.
Comments for author File: Comments.pdf
There are some problems with the English writing of this article and the author is advised to check and revise the grammatical writing throughout the article.
Author Response
Dear Editors and Reviewers:
Thank you for your letter and for the reviewers’ comments concerning our manuscript titled “Detection of train wheelset tread defect with small samples based on local inference constraint network”. Those comments are all valuable and very helpful for revising and improving it, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper.
The responses are as following:
Q1. Figure 3 in this paper is poorly drawn and it is recommended that it be redrawn so that it clearly reflects the structure of the model.
Response: Thanks for your suggestion. Figure 3 has been redrawn, and the roles of each module in the model and the relationships between modules have been reinterpreted and explained.
Q2. Three methods were used to generate the dataset in Section 2.2.1 of this paper, but Figure 4 only gives an example of an image from one of the generation methods, and it is recommended that the graphs generated by all three methods be partially displayed, with appropriate analytical notes added.
Response: Thanks for your comment. The generated data has been added to Figure 4 in the revised manuscript, and the corresponding analysis and annotations have been provided in Section 2.2.1.
Q3. The analysis of Table 3 in this paper is not consistent with what is presented in Table 3 and it is recommended that it be rechecked and revised.
Response: Sorry for our incorrect writing. The data in Table 3 mainly explains the ablation experiment of the residual spinal fully connected layer. The errors in this table have been corrected, and the description has been rechecked and modified in the revised manuscript.
Q4. In this paper, Table 4 appears to be exactly the same as Table 3, and the authors are asked to explain why.
Response: Sorry for this mistake. The data in Table 3 is incorrect and has been modified.
Q5. The analysis of the experimental results for the Table 4 embodiment in Section 2.2.3 of this paper is relatively small, and it is recommended that a detailed analysis of the results be added and that it be clarified which of the added attentional mechanisms in this paper is being used.
Response: Thanks for your suggestion. The mechanisms used in the article are explained, and the results analysis is expanded in Section 2.2.3.
Q6. The model proposed in this paper is only slightly higher than Resnet-50 in terms of recall and accuracy, can it show the effectiveness of the method proposed in this paper? It is suggested to add a more detailed description of the result analysis.
Response: Thanks for your suggestion. It is true as the reviewer pointed out, the proposed method outperforms Resnet-50 and other models in terms of accuracy and recall. However, it should be noted that all the model training and testing data in Table 5 are generated data with small samples, and the data generation method is also an innovation of this article. Relevant explanations are provided in Section 2.2.4 and Section 3.
Q7. How effective is the method proposed in this paper in practical testing? There is a lack of practical validation of the comparative results.
Response: Thanks for your suggestion and comment. The data for this article was collected from the site of a certain vehicle depot. The focus is to solve the problem of identifying wheelset tread defects with small samples. For online testing in actual sites, issues such as image blurring and unstable video frames must also be addressed, which is also a research direction that we need to focus on in the future. This part of the content has been supplemented and explained in Section 3.
Q8. There are many problems with the English writing in this paper, and the author is advised to check and revise the grammatical writing throughout the article.
Response: Thanks for your suggestion. We have tried our best to review the entire paper and correct the English expression issues, which are marked in red in the revised manuscript.
Additionally, we tried our best to improve the manuscript and made some changes. These changes will not influence the content and framework of the paper. We did not list the changes here but marked them in red in the revised paper.
We sincerely appreciate your thorough review and hope that the corrections will meet with approval.
Once again, thank you very much for your comments and suggestions.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper discusses wheel-rail rolling contact and the impact on reliability and wear out, where the detection of wheelset tread defect is important for maintenance. The paper presented local inference constraint network to detect defects in particular with small a sample size.
The introduction should be improved with wider and deeper analysis of inference constraint networks.
The methods presented in Fig. 1 could be elaborated with functional block diagram with main steps.
Fig. 3 is not clear, explain in words will be better.
Explain data set size and uncertainty, and confidence level in results.
Comments on the Quality of English Language
English is fine.
Author Response
Dear Editors and Reviewers:
Thank you for your letter and for the reviewers’ comments concerning our manuscript titled “Detection of train wheelset tread defect with small samples based on local inference constraint network”. Those comments are all valuable and very helpful for revising and improving it, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper.
The responses are as following:
Q1. The introduction should be improved with wider and deeper analysis of inference constraint networks.
Response: Thanks for your suggestion. The advantages of the local inference constraint networks are added in Section 0, and the explanation of why local inference constraints can achieve better computational results is provided in Section 1.3.
Q2. The methods presented in Fig. 1 could be elaborated with functional block diagram with main steps.
Response: Thanks for your comment. Figure 1 has been redrawn, and the roles of each module in the model and the relationships between modules have been reinterpreted and explained.
Q3. Fig. 3 is not clear, explain in words will be better. Explain data set size and uncertainty, and confidence level in results.
Response: Sorry for the unclear expression. Figure 3 has been redrawn and the functions of each part in the figure have been explained in revised manuscript.
Additionally, we tried our best to improve the manuscript and made some changes. These changes will not influence the content and framework of the paper. We did not list the changes here but marked them in red in the revised paper.
We sincerely appreciate your thorough review and hope that the corrections will meet with approval.
Once again, thank you very much for your comments and suggestions.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsI think this paper is okay, and I have no further comments.
