Method for Fault Diagnosis of Track Circuits Based on a Time–Frequency Intelligent Network

Round 1

Reviewer 1 Report

On my opinion the papers is interesting and clearly written. I would suggest the following

Maybe you revise the sentence ” Mathematically, the CWT is represented as follows:” since (1) seems to be not a transform but a definition of a mother signal indexed on a and b. 171

If possible, give some detail regarding the phrase „To ensure a robust analysis, 50 sets of training data are meticulously  selected for each type of fault, amounting to a cumulative total of 1000 sets of fault data.” (271)

I suggest that the author present, if possible, some time domain waveforms for the wavelet transform such that relation (3) be more intuitive.

Author Response

Dear Reviewers,
Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in below and my revisions in the re-submitted files.

Comments 1: Maybe you revise the sentence ” Mathematically, the CWT is represented as follows:” since (1) seems to be not a transform but a definition of a mother signal indexed on a and b. 171

Response 1: Thank you for pointing this out. We agree with this comment. Therefore, we have revised this sentence on page 5, line 171.

Comments 2: If possible, give some detail regarding the phrase „To ensure a robust analysis, 50 sets of training data are meticulously  selected for each type of fault, amounting to a cumulative total of 1000 sets of fault data.” (271)

Response 2: Agree. We have added a description of data set partitioning on page 9, line 274-279.

Comments 3: I suggest that the author present, if possible, some time domain waveforms for the wavelet transform such that relation (3) be more intuitive.

Response 3:  Thank you for your comments, we have added the wavelet transform time domain waveform on page 6, as shown in Figure 3.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper involves the application of continuous wavelet transform (CWT) for signal preprocessing, along with the integration of deep belief networks (DBN) and genetic algorithms (GA) to improve the least squares support vector machine (LSSVM) for intelligent fault diagnosis applied to data-driven track circuits.
The CWT time-frequency representations are fed into the DBN, which performs semisupervised dimensionality reduction and feature extraction. Finally, the GA is employed to improve the kernel function and penalty factor parameters of the LSSVM, thus establishing an DBN-GA-LSSVM diagnostic model.
Validation with data resulting from a simulation model demonstrates the effectiveness of the proposed time-frequency intelligent network model. The achieved accuracy rate on the testing dataset reaches an 99.6%.
The paper presents an interesting study, where an original work is shown, with a right methodology and the manuscript is clear, well organized and structured and the authors have worked exhaustively, obtaining good results.
However, it is opinion of the reviewer that, among others, some suggestions could improve the paper:
- The introduction section should be improved and more references should be added. This section should take into consideration the most relevant studies on the subject to build a complete scientific framework.
- It would be advisable to incorporate a list of acronyms.
- Furthermore, it would be advisable that acronyms do not appear in the Abstract.
- Experimental validation is carried out with data resulting from a simulation model... It would be appropriate to carry out experimental validation with data from a real physical system.
- Section 3.2 of the manuscript should be improved, justifying the decisions made (for example, the type of wavelet function) and the diagrams that appear in Figure 5 should be clearly specified, including their units.
- The conclusions can also be improved.
For these reasons, the reviewer suggests the manuscript for the publication after major revisions.

Author Response

Dear Reviewers,
       Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in below and my revisions in the re-submitted files.

Comments 1: The introduction section should be improved and more references should be added. This section should take into consideration the most relevant studies on the subject to build a complete scientific framework.

Response 1: We agree with this comment. Therefore, We have improved the introduction and added more literature in related fields.

Comments 2: It would be advisable to incorporate a list of acronyms.

Response 2: Thanks for your suggestion, we have added a list of abbreviations after the summary.

Comments 3: Furthermore, it would be advisable that acronyms do not appear in the Abstract.

Response 3: Thank you for pointing this out. We have replaced the professional terms in the abstract with full names.

Comments 4: Experimental validation is carried out with data resulting from a simulation model... It would be appropriate to carry out experimental validation with data from a real physical system.

Response 4: Thank you for pointing this out. We have added the corresponding answer to the third point in the conclusion.

Comments 5: Section 3.2 of the manuscript should be improved, justifying the decisions made (for example, the type of wavelet function) and the diagrams that appear in Figure 5 should be clearly specified, including their units.

Response 5: Thank you for pointing this out. We have added the wavelet time-frequency diagram of the voltage signal in 3.2, and improved the description of Figure 5 and Figure 6.

Comments 6: The conclusions can also be improved.

Response 6: Thank you for your suggestions, we added a third point in the conclusion and improved the first two points.

Author Response File: Author Response.pdf

Reviewer 3 Report

1)    The abstract introduces an innovative approach for jointless track circuit fault diagnosis by combining continuous wavelet transform (CWT), deep belief networks (DBN), genetic algorithms (GA), and least squares support vector machine (LSSVM). Could you elaborate on the specific advantages of integrating these techniques and how they collectively address the limitations mentioned?

2)    It's mentioned that DBN is used for semi-supervised dimensionality reduction and feature extraction. Could you provide more insights into how DBN achieves this and why it is a suitable choice for uncovering fault characteristics in track circuits?

3)    The manuscript indicates the use of genetic algorithms (GA) to improve the kernel function and penalty factor parameters of LSSVM. What specific challenges or limitations in existing approaches do these parameter optimizations aim to address, and how does GA contribute to achieving an optimal diagnostic model?

4)    The manuscript provides a succinct summary of the fault diagnosis methodology involving CWT, DBN, and GA-LSSVM. How does this methodology contribute to addressing the challenges posed by different fault types in jointless track circuit systems? Also, the paper emphasizes the use of CWT for voltage signal data preprocessing. Could you elaborate on the specific improvements or advantages gained by employing CWT in terms of diagnostic precision and handling different fault types?

5)    Considering the achieved accuracy rate of 99.6%, what are the potential practical implications and applications of this comprehensive approach in real-world scenarios? Are there any limitations or challenges that need to be addressed for broader implementation?

6)    The conclusion suggests that the DBN-GA-LSSVM approach significantly improves fault recognition accuracy and reduces working time. What metrics or benchmarks were used to measure this improvement, and how does the proposed approach compare to traditional methods, especially multilayer perceptron neural networks? Furthermore, the paper describes the proposed method as an innovative intelligent strategy for diagnosing track circuit faults. What specific aspects of the methodology contribute to its innovation, and how does it advance the state-of-the-art in track circuit fault diagnosis?

7)    To enhance the quality of the manuscript, you may cite the latest papers of the fault diagnosis in the literature review section of the manuscript as,

a)     Ahmad, S.; Ahmad, Z.; Kim, J.-M. A Centrifugal Pump Fault Diagnosis Framework Based on Supervised Contrastive Learning. Sensors 2022, 22, 6448. https://doi.org/10.3390/s22176448.

b)    Ullah, N.; Ahmad, Z.; Siddique, M.F.; Im, K.; Shon, D.-K.; Yoon, T.-H.; Yoo, D.-S.; Kim, J.-M. An Intelligent Framework for Fault Diagnosis of Centrifugal Pump Leveraging Wavelet Coherence Analysis and Deep Learning. Sensors 2023, 23, 8850. https://doi.org/10.3390/s23218850.

Overall it is good, still some improvement can be done in literature and conclusion part of the manuscript.

Author Response

Dear Reviewers,

       Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in below and my revisions in the re-submitted files.

Comments 1: The abstract introduces an innovative approach for jointless track circuit fault diagnosis by combining continuous wavelet transform (CWT), deep belief networks (DBN), genetic algorithms (GA), and least squares support vector machine (LSSVM). Could you elaborate on the specific advantages of integrating these techniques and how they collectively address the limitations mentioned?

Response 1: Thank you for pointing this out. We agree with this comment. Therefore, we have added an explanation in the last paragraph of the introduction. 

Comments 2: It's mentioned that DBN is used for semi-supervised dimensionality reduction and feature extraction. Could you provide more insights into how DBN achieves this and why it is a suitable choice for uncovering fault characteristics in track circuits?

Response 2: Thank you for pointing this out. We have added an explanation in line 196-202 of paragraph 1 of Section 2.3.

Comments 3: The manuscript indicates the use of genetic algorithms (GA) to improve the kernel function and penalty factor parameters of LSSVM. What specific challenges or limitations in existing approaches do these parameter optimizations aim to address, and how does GA contribute to achieving an optimal diagnostic model?

Response 3: Thank you for pointing this out. We have added an explanation in the last paragraph of Section 2.4.

Comments 4: The manuscript provides a succinct summary of the fault diagnosis methodology involving CWT, DBN, and GA-LSSVM. How does this methodology contribute to addressing the challenges posed by different fault types in jointless track circuit systems? Also, the paper emphasizes the use of CWT for voltage signal data preprocessing. Could you elaborate on the specific improvements or advantages gained by employing CWT in terms of diagnostic precision and handling different fault types?

Response 4: Thank you for your comments. The first question we answer in a way that perfects the conclusion. The second question has been added in 3.2.

Comments 5: Considering the achieved accuracy rate of 99.6%, what are the potential practical implications and applications of this comprehensive approach in real-world scenarios? Are there any limitations or challenges that need to be addressed for broader implementation?

Response 5: Thanks for your comments, we have answered the possible practical engineering significance in the future in the conclusion.

Comments 6: The conclusion suggests that the DBN-GA-LSSVM approach significantly improves fault recognition accuracy and reduces working time. What metrics or benchmarks were used to measure this improvement, and how does the proposed approach compare to traditional methods, especially multilayer perceptron neural networks? Furthermore, the paper describes the proposed method as an innovative intelligent strategy for diagnosing track circuit faults. What specific aspects of the methodology contribute to its innovation, and how does it advance the state-of-the-art in track circuit fault diagnosis?

Response 6: Thank you for pointing this out. We have added an explanation of the method comparison in the third paragraph of Chapter 3.5. Contribute to the development of track circuit fault diagnosis, we answer in the way of perfecting the conclusion.

Comments 7: To enhance the quality of the manuscript, you may cite the latest papers of the fault diagnosis in the literature review section of the manuscript as，

6448. Ahmad, S.; Ahmad, Z.; Kim, J.-M. A Centrifugal Pump Fault Diagnosis Framework Based on Supervised Contrastive Learning. Sensors 2022, 22, 6448. https://doi.org/10.3390/s22176448.
6449. b)    Ullah, N.; Ahmad, Z.; Siddique, M.F.; Im, K.; Shon, D.-K.; Yoon, T.-H.; Yoo, D.-S.; Kim, J.-M. An Intelligent Framework for Fault Diagnosis of Centrifugal Pump Leveraging Wavelet Coherence Analysis and Deep Learning. Sensors 2023, 23, 8850. https://doi.org/10.3390/s23218850.

Response 7: Thanks for your suggestion, we have cited your recommended literature in the third paragraph of the introduction.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The paper presents an interesting study, where an original work is shown, with a right methodology and the manuscript is clear, well organized and structured and the authors have worked exhaustively, obtaining good results.

Some contents of the article have seen corrected and increased, as was suggested to authors in the first review but other suggestions have not been fully resolved.

It is opinion of the reviewer that the next suggestions could improve the paper:

- The introduction section has been improved but it needs to be improved further to build a complete scientific framework.

- A list of acronyms has been incorporated but it must be completed (TU, RBN, CNN, ...).

- Experimental validation is carried out with data resulting from a simulation model... It would be appropriate to carry out experimental validation with data from a real physical system. It is opinion of the reviewer that the authors' justification seems correct, but this causes the scientific work to decrease in quality.

- This reviewer considers it positive to add Figure 3 to the manuscript (“Figure 3. Track voltage time domain signal diagram”), but the parameters of the time signals should be indicated: sampling rate, total measured time, …

- Some Chinese characters appear on line 303 of the new manuscript. This must be corrected.

For these reasons, the reviewer suggests the manuscript for the publication after minor revisions.

Author Response

To Reviewers:

Response: Firstly, the authors thank you for all your comments and directions to our paper, they are of great importance and helped a lot. Our works have been made progress, too. Your advice and instructions are also very helpful to our future works. Please allow me to represent all the authors to send our sincere respect to you. We have studied the comments carefully and have made corrections in the paper and the revised portions are in red font. Thanks again for allowing us to revise our paper.

Comments 1: The introduction section has been improved but it needs to be improved further to build a complete scientific framework.

Response 1: Thank you for pointing this out. We agree with this comment. In the last paragraph of the introduction, we add the content of the research purpose and research questions. ( 98-110 lines ). For your convenience, the revisions are as follows.

Track circuit fault diagnosis steps are generally divided into: signal processing, feature extraction, fault classification. Signal processing refers to the use of various signal analysis methods to analyze and process the state signals collected during circuit operation. CWT can obtain the characteristic signal of fault information [18]. For feature extraction and pattern classification, the deep learning algorithm uses the super feature extraction ability of the deep neural network model, and then uses the classification model to classify the sample data, which is the frontier processing method of fault diagnosis. However, the existing deep learning methods also have defects. For example, Zhang K et al. [19] uses one-dimensional convolution for fault feature extraction. These methods directly process one-dimensional vibration time series signals without considering the frequency domain information in the signals. Therefore, this paper uses CWT to perform time-frequency domain processing on track circuit signals and convert one-dimensional signals into two-dimensional time-frequency images.

18. Zhang K, Ma C, Xu Y, et al. Feature extraction method based on adaptive and concise empirical wavelet transform and its applications in bearing fault diagnosis[J]. Measurement, 2021, 172: 108976.
19. Wang X, Mao D, Li X. Bearing fault diagnosis based on vibro-acoustic data fusion and 1D-CNN network[J]. Measurement, 2021, 173: 108518.

Comments 2: A list of acronyms has been incorporated but it must be completed (TU, RBN, CNN, ...).

Response 2: Agree. We have updated the Table 1. For your convenience, the revisions are as follows.

Table 1. List of abbreviations and acronyms used in this article.

Comments 3: Experimental validation is carried out with data resulting from a simulation model... It would be appropriate to carry out experimental validation with data from a real physical system. It is opinion of the reviewer that the authors' justification seems correct, but this causes the scientific work to decrease in quality.

Response 3:  Thank you for pointing this out. We agree with this comment. However, because the railway department does not have a public real data set and the track circuit has a large experimental cost. At present, the mainstream research still uses simulation data sets to simulate the characteristics and distribution of real data for data analysis and model training.

Comments 4: This reviewer considers it positive to add Figure 3 to the manuscript (“Figure 3. Track voltage time domain signal diagram”), but the parameters of the time signals should be indicated: sampling rate, total measured time, …

Response 4:  Thank you for your comments,We have added sampling rate and sampling duration to the description in Figure 3. For your convenience, the revisions are as follows.

The sampling rate is 2500 Hz, and the sampling time is 10 ms.

Comments 5: Some Chinese characters appear on line 303 of the new manuscript. This must be corrected.

Response 5: Thank you for your correction.We have deleted Chinese characters.

Author Response File: Author Response.pdf