Gaussian-Linearized Transformer with Tranquilized Time-Series Decomposition Methods for Fault Diagnosis and Forecasting of Methane Gas Sensor Arrays
Sauvik Biswas
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors introduced a novel fault diagnosis and forecast method for methane gas sensor array based on Gaussian-linearized transformer (GLTrans) with tranquilized time-series decomposition. Compared to the traditional transformer model, this method reduced the computational complexity dramatically from O(N2) to O(N), and the decomposition of continuous temporal signal made the output of sensor array and the transformer more compatible. Equivalence between the GLTrans method and traditional methods was also verified. Results on both fault diagnosis and forecast were evaluated and compared with other time-series models, and the GLTrans method showed better efficiency and accuracy. It is also promising to expand the application to more time-series based sensor systems. Although objective, presentation and results of the work are clear and meaningful, there are several questions that the authors need to answer before it could be recommended for publication.
Specific points to be addressed:
1. The definition of the tranquilizing on the data-series should be further illustrated and meaning of the subscript (n) and (n-1) in Equation 1 needs to be denoted. The description about moving average in Section 2.1 and Algorithm 1 showed different behavior, for fault diagnosis, were only past and current data points applied here? Or the diagnosis system was non-casual, the range from [-k, k] would contain the future data points.
2. Did the input series with length L contain any overlaps with its neighbor segments and what were the criteria for determining the length L?
3. Explain why linear separability can be achieved in higher dimensional space (line 141-144).
4. Proper citations are needed for line 186-188.
5. More introduction about the work principle and parameters of the methane sensor array such as the sampling frequency and the role of independent sensor in the array would be helpful for understanding the fault diagnosis and forecast.
6. Authors mentioned that all the methane sensor data used in the model development were from public dataset, details about the dataset should be introduced and cited. The actual methane sensor setup was also introduced in the Figure 3. Were the practical data captured from the sensor setup applied in this study as well?
7. Follow the previous question, were the fault signals intrinsic in the dataset or generated artificially? Illustration about preprocessing of the dataset would make the presentation more comprehensive.
8. What is the exact meaning for the sensor element in confusion matrix (Figure 7)? Is the type of specific sensor fault or different independent sensors in the array?
9. Was the prediction period length fixed with 1800 points? How this length was determined, and did you evaluate the performance of GLTrans method with different prediction periods length?
10. Test conditions applied in other time-series models could be provided as well for better comparison and evaluation.
Other minor points should be revised:
1. Repeat texts were found in Section 3.3 & 3.4 and 3.6 & 3.7. Delete redundant parts and reorganize the writing.
2. Check all the numbers for Figures and Tables in both main text and supporting material. Figure S4 to S6 could not be found in the supporting and the discussion about Figure S2 and S3 in supporting could not be found too.
3. Add initial S to the equation number in the supporting materials to differentiate with the equations in the main text.
4. Check and correct all typos. (e.g. line 110-111, 197, 237, 268, 365, 440 in the manuscript)
5. Use Italic for all variables mentioned in the manuscript for better clarity.
6. Add fault type label in every subfigure of Figure 4 for better readability.
7. Check English writing and grammar carefully.
Comments on the Quality of English LanguageGrammar of writing should be checked carefully before resubmission. The manuscript also can be further reorganized and polished for better presentation.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsIn my view this may be considered for publication after the following modifications:
1-English in the Manuscript should be thoroughly checked and corrected.
2-In the experiment setup section, what components make up the experimental system for methane sensor arrays, and how do they work together to simulate real-world conditions?
3- The Conclusions section should include:
* A highlight of your hypothesis, new concepts and innovations.
* A summary of key improvements compared to findings in literature
* Your vision for future work
4- "In return, considering that the following references have not been published in journals since 2019, they should be replaced."
39,42-44
1-Could you explain the key modifications made to the traditional self-attention mechanism in the proposed model, and how do these modifications contribute to its effectiveness?
2-In the experiment setup section, what components make up the experimental system for methane sensor arrays, and how do they work together to simulate real-world conditions?
3-What are the primary failure types tested in the fault-diagnosis task, and how do these faults relate to the methane gas sensor's performance?
Comments on the Quality of English LanguageMinor editing of English language required
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsIn this paper authors discuss the relationship between tranquilized time-series decomposition method for fault diagnosis in transformers. They have shown the methane sensors behavior of the system loading conditions using tranquilized time series decomposition method. This paper is publishable in the presented form and the reasons are as follows;
This is a specialized paper, which only experts in the application area will be able to understand. However, since its narrative style is very good, it will attract the attention of readers who are not experts in the field. Materials, method and results are well presented in this manuscript. My suggestion is to simply increase the resolution of figure 7.
My final opinion is that the manuscript can be accepted after minor revision.
Comments on the Quality of English Language
There are not grammar errors in the paper.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThe topic is interesting and the mathematical description proposed methodology is good. The result is also compared with other works. However, there are many scope for improvement as follows:
1. Type of faults, reasons & effects of faults and fault under consideration are not well explained.
2. Text quality of Fig 7 is poor.
3. Please also mention limitations of the method.
4. Provide all parametrical constraints that should be considered when you are applying in real environment.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 5 Report
Comments and Suggestions for Authors1. Abstract: '.... time complexity of O(N2) and ...' is it N2 or 2N or N square?
2. Review different time series of the following methods
10.1109/TIM.2020.3044517 10.1016/j.epsr.2023.109526 10.3390/app120316753. In time-series decomposition, generally the different frequency components of time series are decomposed. But here it is not like that. Authors need to address more on the proposed decomposition method for clarity.
4. What is the sampling rate of data retrieve?
5. Mention the accuracies in the confusion matrix.
6. Apart from accuracy, there are more performance parameters like recall, F-measure, precision for a confusion matrix. Address them in the paper.
7. Segregate the introduction into the subsection like A. Motivation of the Work, B. Literature Review and C. Contribution and Novelty. Highlight the proposed method and its advantages over earlier used method more in the Contribution section.
Comments on the Quality of English LanguageNA.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThank you for your response. Most of your explanations make sense to me and it can be recommended for publication now.
About the Comment 6 & 7, I strongly recommend you to add the source and description about the dataset in the Methods or Supplementary Information. It is reasonable to denote the source and illustrate how the work is performed , and also acknowledge others' contribution to your work.
Comments on the Quality of English LanguagePlease check the manuscript carefully before final submission.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsAuthors have replied in right direction, no more comments
Author Response
Authors have replied in right direction, no more comments
Reviewer 5 Report
Comments and Suggestions for Authors1. Segregate the introduction into the subsection like A. Motivation of the Work, B. Literature Review and C. Contribution and Novelty. Highlight the proposed method and its advantages over earlier used method more in the Contribution section.
2. Review different time series of the following methods
10.1109/TIM.2020.3044517
10.1016/j.epsr.2023.109526
10.3390/app12031675
3. What is the sampling rate of data retrieve?
Author Response
Please see the attachment
Author Response File: Author Response.pdf
