Next Article in Journal
An Enhanced Multiscale Retinex, Oriented FAST and Rotated BRIEF (ORB), and Scale-Invariant Feature Transform (SIFT) Pipeline for Robust Key Point Matching in 3D Monitoring of Power Transmission Line Icing with Binocular Vision
Next Article in Special Issue
Comparison of LSTM- and GRU-Type RNN Networks for Attention and Meditation Prediction on Raw EEG Data from Low-Cost Headsets
Previous Article in Journal
Artificial Intelligence (AI)-Based Radar Signal Processing and Radar Imaging
Previous Article in Special Issue
Prediction of Sunspot Number with Hybrid Model Based on 1D-CNN, BiLSTM and Multi-Head Attention Mechanism
 
 
Article
Peer-Review Record

Transformer-Based GAN with Multi-STFT for Rotating Machinery Vibration Data Analysis

Electronics 2024, 13(21), 4253; https://doi.org/10.3390/electronics13214253
by Seokchae Lee 1, Hoejun Jeong 1 and Jangwoo Kwon 2,*
Reviewer 1: Anonymous
Reviewer 2:
Reviewer 3:
Electronics 2024, 13(21), 4253; https://doi.org/10.3390/electronics13214253
Submission received: 20 September 2024 / Revised: 28 October 2024 / Accepted: 28 October 2024 / Published: 30 October 2024

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1.The introduction of the article only includes 7 references and does not provide a detailed analysis of the current research status and existing problems in the relevant field, which is not a good introduction.

2.The methods introduced in this article are relatively vague, and some details have not been disclosed, making it unclear about their innovation.

3.The results presented in this article are not detailed enough, and the results of signal processing and deep learning models need to be carefully compared and summarized.

4.The conclusion of the article should be more clear, and it is recommended to write it by item.

5.The latest literature related to this article should be cited.

Comments on the Quality of English Language

English expression should be more standardized. It is recommended not to use the first person, but to use the third person.

Author Response

  1. The introduction of the article only includes 7 references and does not provide a detailed analysis of the current research status and existing problems in the relevant field, which is not a good introduction.

-> We appreciate this observation. We have added more citations to the introduction and strengthened it by including content on augmentation techniques using generative models and information about Transformers. (lines 16-86)

 

  1. The methods introduced in this article are relatively vague, and some details have not been disclosed, making it unclear about their innovation.

-> We thank the reviewer for this feedback. To address this, we have added a figure comparing multi-resolution and single-resolution STFT representations of the same signal at different resolutions, along with related explanations. (lines 239-245)

 

  1. The results presented in this article are not detailed enough, and the results of signal processing and deep learning models need to be carefully compared and summarized.

-> We appreciate this insight. We have expanded the dataset section to include more details on how data acquisition and signal processing were conducted. (Lines 274-286)

 

  1. The conclusion of the article should be more clear, and it is recommended to write it by item.

-> We have restructured the conclusion section into separate paragraphs and strengthened each point by adding quantitative experimental results to support our findings.

 

  1. The latest literature related to this article should be cited.

-> We have expanded our citations to 45, aiming to broaden the understanding of our research through related studies.

We sincerely appreciate all the reviewers' valuable feedback, which has helped us improve the quality and clarity of our manuscript.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This paper presents a transformer-based generative adversarial network (GAN) model featuring a multi-resolution short-time Fourier transform (multi-STFT) loss function to enhance vibration data for rotating machinery. This augmentation aims to facilitate the effective learning of deep learning models and addresses class imbalance in fault diagnosis systems. While the proposed approach shows promise, several key areas require clarification:

1. The authors should clearly articulate the specific research gap addressed by their work, especially in the context of numerous studies employing GANs for data generation. What unique contributions does this research make?

2. There are formatting inconsistencies, particularly on line 245, which disrupt the overall flow and professionalism of the paper.

3. Given the known instability of GAN training—especially with transformer architectures—the authors must explain how their method ensures training stability and convergence.

4. The model's interpretability is not adequately addressed. The authors should incorporate relevant literature on interpretability, such as: 1. IEEE Transactions on Industrial Informatics, 2024, 20(6): 8628–8638 (DOI: 10.1109/TII.2024.3366993) ; Mechanical Systems and Signal Processing, 2023, 202: 110680 (https://arxiv.org/abs/2305.19569), which can help to solve this issue.

5. The authors are encouraged to explicitly list their contributions to clarify the paper's impact.

Comments on the Quality of English Language

There are some grammar and typo errors in the manuscript. Please check the manuscript and revise them.

Author Response

  1. The authors should clearly articulate the specific research gap addressed by their work, especially in the context of numerous studies employing GANs for data generation. What unique contributions does this research make?

-> We appreciate this suggestion. To address this, we have added a figure comparing multi-resolution and single-resolution STFT representations of the same signal at different resolutions, along with related explanations. (lines 239-245)

 

  1. There are formatting inconsistencies, particularly on line 245, which disrupt the overall flow and professionalism of the paper.

-> We sincerely thank the reviewer for pointing this out. The formatting issue has been resolved.

 

  1. Given the known instability of GAN training—especially with transformer architectures—the authors must explain how their method ensures training stability and convergence.

-> We appreciate this insightful comment. We have added a graph showing the Generator and Discriminator losses during training to demonstrate the convergence of our proposed model, along with an accompanying explanation. (lines 320-333)

 

  1. The model's interpretability is not adequately addressed. The authors should incorporate relevant literature on interpretability, such as: 1. IEEE Transactions on Industrial Informatics, 2024, 20(6): 8628–8638 (DOI: 10.1109/TII.2024.3366993) ; Mechanical Systems and Signal Processing, 2023, 202: 110680 (https://arxiv.org/abs/2305.19569), which can help to solve this issue.

-> We are grateful for this valuable suggestion. We have added the recommended citations to address the interpretability aspect of our model. (References 8, 11)

 

  1. The authors are encouraged to explicitly list their contributions to clarify the paper's impact.

-> We thank the reviewer for this recommendation. We have explicitly stated our paper's contributions in the introduction section. (lines 70-86)

 

We sincerely appreciate all the reviewers' valuable feedback, which has helped us improve the quality and clarity of our manuscript.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Journal Electronics (ISSN 2079-9292)

Manuscript ID electronics-3243685

Type Article

Title Transformer-based GAN with Multi-STFT for Rotating Machinery Vibration Data Analysis

Authors SeokChae Lee , HoeJun Jeong , Jangwoo Kwon *

Section Computer Science & Engineering

 

This paper investigated the problem of PHM for general rotating machinery. The authors proposed a transformer-based generative adversarial network (GAN) model to detect and distinguish the vibration problems. Verification is performed based on the actual vibration data collected from real machinery. The problem is important, and the utilized AI based method is up-to-date. The whole structure of the paper is fine, and the topic is quite interesting. I have the following comments for the authors’ consideration.

 

1. The literature review section is too weak. The authors only listed 5-6 works in the Introduction, which is clearly inadequate for readers to understand the current research focus in the research field.

 

2. Since the utilized data in this paper is vibration signals, how did the proposed model process the time-series data? Did the authors extracted some features to represent the time-series vibration signals?

 

3. It is suggested to add an overall structure or framework of the proposed method in Section II.

 

4. What are the in feature and out features mentioned in Table 1? How are they selected or determined?

 

5. In Section 4.2, how is the vibration signal collected? Any specification on the measurement devices?

 

6. The case study results should be mentioned and clarified in the conclusion part.

Author Response

  1. The literature review section is too weak. The authors only listed 5-6 works in the Introduction, which is clearly inadequate for readers to understand the current research focus in the research field.

-> We sincerely appreciate this feedback. We have expanded the introduction with additional citations and included content on augmentation techniques using generative models and information about Transformers. (lines 16-86)

 

  1. Since the utilized data in this paper is vibration signals, how did the proposed model process the time-series data? Did the authors extracted some features to represent the time-series vibration signals?

-> We thank the reviewer for this important question. We have added an explanation clarifying that our proposed model directly inputs the vibration data in the Time Domain. (line 182)

 

  1. It is suggested to add an overall structure or framework of the proposed method in Section II.

-> We appreciate this suggestion and we have added an overall structure diagram (Figure 3) to the methodology section.

 

  1. What are the in feature and out features mentioned in Table 1? How are they selected or determined?

-> We thank the reviewer for this question. Table 1 explains the input and output of the model and its components. We have improved the existing Table 1 for better clarity and added explanations for each item in lines 183-190.

 

  1. In Section 4.2, how is the vibration signal collected? Any specification on the measurement devices?

-> We appreciate this inquiry. We have added information about the sensors in Figure 6, Table 3, and lines 274-275.

 

  1. The case study results should be mentioned and clarified in the conclusion part.

-> We thank the reviewer for this suggestion. We have restructured the conclusion section into separate paragraphs and strengthened each point by adding quantitative experimental results to support our findings.

 

We sincerely appreciate all the reviewers' valuable feedback, which has helped us improve the quality and clarity of our manuscript.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The revised version can be accepted with minor modifications, including English expression, literature analysis in the introduction, and data checking.

Author Response

 

  • The revised version can be accepted with minor modifications, including English expression, literature analysis in the introduction, and data checking.
    • We have made the following corrections to the tense wording within the literature. [page 2 of the introduction].

 

Thank you for your review.

Reviewer 2 Report

Comments and Suggestions for Authors

Accept.

Author Response

Thank you for your review.

Back to TopTop