Review Reports - Anomaly Detection of Liquid Level in Mold during Continuous Casting by Using Forecasting and Error Generation

Round 1

Reviewer 1 Report

The paper proposes "Anomaly Detection of Liquid Level in Mold during Continuous Casting by Using Forecasting and Error Generation".

The manuscript is well presented.

There some issues with the manuscript. They are :

1. The literature review is not exhaustive. Authors need to explain why the literature review is limited.

2. Author need to compare their work with already published results, with respect to accuracy.

Authors mention related work but need to compare specifics with it.

The key contributions of this paper must be supported by gaps in the earlier research.

The statement " Compared to a fixed threshold, the proposed method significantly increases the precision but slightly reduces recall" can be elaborated/detailed in the conclusion/discussion section.

Otherwise, the paper is well written and presented.

Author Response

Dear Sir/Madam:

We sure appreciate your time to review our manuscript entitled " Anomaly Detection of Liquid Level in Mold during Continuous Casting by Using Forecasting and Error Generation" again. Your comments and suggestions are very beneficial for us to improve our work. We have revised our manuscript based on your comments, and have added text and tables to our manuscript to make it more specific and clear. We also have finished proofreading and editing our manuscript to improve its readability. Based on the instructions in editor’s letter, we have uploaded the file of our revised manuscript.

Our responses to your comments are appened one by one. Your comment is first reproduced, and our response is given directly afterward in different color (red).

Point 1: The literature review is not exhaustive. Authors need to explain why the literature review is limited.

Response 1: Because current method is all focused on automatic control phase. The temporal feature of liquid level in mold is diffeernt. In fact, we are unable to find any research that specific on that phase. Most paper don't seprate two different casting phases. But in actual industrial practice, the liquid level in mold needs to reach certain level so that continune casting machine can be activated. It has been explained in paper now.

Point 2: Author need to compare their work with already published results, with respect to accuracy. Authors mention related work but need to compare specifics with it.

Response 2: We expand our experiment by comparing with LSTM-adv and TadGAN. Also we reorganize the expeirment section.

Point 3: The statement " Compared to a fixed threshold, the proposed method significantly increases the precision but slightly reduces recall" can be elaborated/detailed in the conclusion/discussion section.

Response 3: The recall issue now is being discussed in the paper. It was caused by dynamic threshold.

We thank you for the time and effort that you have put into reviewing the previous version of our manuscript. Your comments reveal some shortcomings in our past work. We have carefully considered your comments in the modification process. Your comments also lead us to think of this field deeply, and give us inspiration for future work. All in one words, we are grateful for your comments and suggestions.

We hope that our revised manuscript would be accepted for publication in the Applied Science.

Best regards!

Yours sincerely,

Dr. Xiaojun Wu

Corresponding author:

Assoc. Prof. Xiaojun Wu

School of Software Engineering, Xi’an Jiaotong University, Xi’an 710049, China

Tel: +86-13629206003

E-Mail: xiaojunw@xjtu.edu.cn

Author Response File: Author Response.docx

Reviewer 2 Report

The revised article responds to a new proposal for a method based on RAE which provides the detection of liquid level anomalies in the mould during continuous casting. For this purpose, the investigation is carried out by means of prediction and error generation. The new method has shown a significant increase in accuracy compared to other methods currently in use.

The article needs more elaborate conclusions, and it is also necessary to respond to the major handicap of the proposed method, which is that it slightly reduces recovery, even if it is in future proposed lines of research.

Figure 5 deserves revision as it does not seem to be concrete, and the object should be clarified.

The discussion of the article is not delimited and is confused with the experimental part. The structure of the article is not clear although the results are clear. Please modify it to improve the reader's understanding.

Author Response

Dear Sir/Madam:

We sure appreciate your time to our manuscript entitled " Anomaly Detection of Liquid Level in Mold during Continuous Casting by Using Forecasting and Error Generation". Your comments and suggestions are very beneficial for us to improve our work. We have revised our manuscript based on your comments, and have added text and tables to our manuscript to make it more specific and clear. We also have finished proofreading and editing our manuscript to improve its readability. Based on the instructions in editor’s letter, we have uploaded the file of our revised manuscript.

Our responses to your comments are appened one by one. Your comment is first reproduced, and our response is given directly afterward in different color (red).

Point 1: The article needs more elaborate conclusions, and it is also necessary to respond to the major handicap of the proposed method, which is that it slightly reduces recovery, even if it is in future proposed lines of research.

Response 1: The recall issue has been discussed in the revision paper. We also reorganized the conclusion section, and discussed the future work.

Point 2: Figure 5 deserves revision as it does not seem to be concrete, and the object should be clarified.

Response 2: Figure 5 has been revisioned. We use different color boxes to point out abonormal places.

Point 3: The discussion of the article is not delimited and is confused with the experimental part. The structure of the article is not clear although the results are clear. Please modify it to improve the reader's understanding.

Response 3: We reorganize the experiment section. The experiment results section now includes three sub-sections: Comparison Experiment, Ablation experiment, Parameter Experiment. We believe the revisioned experiment result section can make experimental part clearer.

We hope that our revised manuscript would be accepted for publication in the Applied Science.

Best regards!

Yours sincerely,

Dr. Xiaojun Wu

Corresponding author:

Assoc. Prof. Xiaojun Wu

School of Software Engineering, Xi’an Jiaotong University, Xi’an 710049, China

Tel: +86-13629206003

E-Mail: xiaojunw@xjtu.edu.cn

Author Response File: Author Response.docx

Reviewer 3 Report

The manuscript does not have significant practical innovation

Author Response

Thanks for taking time reviewing the paper.

The reason why we use unsupvervised AE-based TSAD for liquid level in mold anomaly detection is been address in the paper. The two issues remained in liquid level in mold anomaly detection during manual control results in lots of FPs that's why we purpose the method to address and solve the issues.

I hope you find my response helpful. Thanks again.

Reviewer 4 Report

The presented manuscript seems to be interesting for readers of the Applied Sciences journal, it is written in a good manner and suits the requirements of the journal. It can be accepted for publication after minor corrections listed below.

Please respond to the following questions and make necessary revisions.

- What is the main question addressed by the research?

- Do you consider the topic original or relevant in the field? Does it address a specific gap in the field?

- What does it add to the subject area compared with other published material?

- What specific improvements should the authors consider regarding the methodology? What further controls should be considered?

- Are the conclusions consistent with the evidence and arguments presented and do they address the main question posed?

- The "Abstract" section should contain the main achievements of research not general discussion. Re-organization of abstract is needed.

- Abbreviation/ acronyms, should all be defined at their first occurrence in the manuscript.

- All parameters used in formulas must be explained. It is recommended to attach all parameters and abbreviations used in a table at the end of the article.

- The novelty of work at the end of manuscript “introduction” is not sufficient and should be explained more.

- “Yoon et al.[3]” should be changed to “Sarda et al.[3]”

- “Chen et al. proposed Dynamic Graph Anomaly Detection (DyGraphAD) [19],” should be changed to “Li et al. proposed Dynamic Graph Anomaly Detection (DyGraphAD) [19],”.

- The flowchart of the research method should be given. Also, sample coding and specifications should be provided in table.

- In the "Conclusion" section, the authors should present more quantitative data as the main results of the research study rather than just some qualitative data.

- Literature review is not sufficient and authors must review and cite more papers in the field of “Using neural network to predict and improve the properties and structure of steel” and especially newly published ones. Doing this, review and citing the following refs could be helpful:

Neural Network World 4, no. 13 (2013): 351-367.; Measurement 75 (2015): 5-11.

English language of manuscript is acceptable in general. However, it would be much better to improve. Please avoid the unnecessary long sentence.

Author Response

Dear Sir/Madam:

Our responses to your comments are appened one by one. Your comment is first reproduced, and our response is given directly afterward in different color (red).

Point 1: What is the main question addressed by the research?

Response 1: The main question is that 2 issues remain in the liquid level in mold series that makes AE-based TSADs result in many false positives. First is that anomalies usually last longer due to manual control. Second, the abnormal sequence usually shares a similar feature with the normal sequence. All these 2 issues are addressed in the paper.

Point 2: Do you consider the topic original or relevant in the field? Does it address a specific gap in the field?

Response 2: It explores the almost untouched manual control phase before the casting machine is activated. The liquid level in mold is not solidified and is not drawn out by the casting machine in this phase. And the paper designed FEG-AE to address the two typical anomaly features found in liquid level in mold series.

Point 3: What does it add to the subject area compared with other published material?

Response 3: We explored the anomaly detection for liquid level in mold during the manual control phase, and we proposed FEG-AE to detect anomalies in the liquid level in mold.

Point 4: What specific improvements should the authors consider regarding the methodology? What further controls should be considered?

Response 4: We have modified our conclusion section to explain future improvements. Future work will investigate a more robust way to balance the forecasting and error extraction networks to reduce the hyperparameter (swp) effects. Also, a more robust dynamic threshold method is under investigation to improve the recall

Point 5: Are the conclusions consistent with the evidence and arguments presented and do they address the main question posed?

Response 5: We have reorganized our experiment results section. The experiment result shows that our method can effectively address the main questions. The FEG-AE architecture addressed the long-time anomaly issue, and the forecasting network addressed the problem that abnormal sequences could share similar features with normal sequences.

Point 6: The "Abstract" section should contain the main achievements of research not general discussion. Re-organization of abstract is needed.

Response 6: Thank you for raising this suggestion. We have reorganized the Abstract section according to your suggestion.

Point 7: Abbreviation/acronyms, should all be defined at their first occurrence in the manuscript.

Response 7: Abbreviations are now all defined at their first occurrence.

Point 8: All parameters used in formulas must be explained. It is recommended to attach all parameters and abbreviations used in a table at the end of the article.

Response 8: All parameters are explained in the paper. Most of the parameters are context-related, so we left the explanation in the context for better understanding.

Point 9: The novelty of work at the end of manuscript “introduction” is not sufficient and should be explained more.

Response 9: The Introduction section has been reorganized and the purpose and background of the proposed method are clearer. More references have been added in the section.

Point 10: “Yoon et al.[3]” should be changed to “Sarda et al.[3]”.

Response 10: The issue has been fixed.

Point 11: “Chen et al. proposed Dynamic Graph Anomaly Detection (DyGraphAD) [19],” should be changed to “Li et al. proposed Dynamic Graph Anomaly Detection (DyGraphAD) [19],”.

Response 11: The issue has been fixed.

Point 12: The flowchart of the research method should be given. Also, sample coding and specifications should be provided in table.

Response 12: Thanks for raising the suggestion. The flowchart is now provided in Figure 2.

Point 13: In the "Conclusion" section, the authors should present more quantitative data as the main results of the research study rather than just some qualitative data.

Response 13: We have reorganized the “Conclusion” section to present more quantitative data.

Point 14: Literature review is not sufficient and authors must review and cite more papers in the field of “Using neural network to predict and improve the properties and structure of steel” and especially newly published ones. Doing this, review and citing the following refs could be helpful:

Neural Network World 4, no. 13 (2013): 351-367.; Measurement 75 (2015): 5-11.

Response 14: We have carefully read the papers that were recommended previous reviewer, and their research ideas are helpful to this paper. We have cited them as references [3] [11] to improve the introduction and related work section.

We hope that our revised manuscript would be accepted for publication in the Applied Science.

Best regards!

Yours sincerely,

Dr. Xiaojun Wu

Corresponding author:

Assoc. Prof. Xiaojun Wu

School of Software Engineering, Xi’an Jiaotong University, Xi’an 710049, China

Tel: +86-13629206003

E-Mail: xiaojunw@xjtu.edu.cn

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

In my opinion, this manuscript lacked sufficient novelty

Author Response

We sure appreciate your time to review the paper.

Reviewer 4 Report

- The authors have responded to some requests of previous reviewer, but many corrections remain unanswered.

- It is necessary for the authors to answer the reviewers, point by point, to bring the requested items and answer them. It is necessary for the authors to highlight the changed parts in the revised text.

- The quality of the figures is low and their contrast should be increased.

- It is necessary to review and refer to the articles suggested by the previous reviewer.

Neural Network World 4, no. 13 (2013): 351-367.; Measurement 75 (2015): 5-11.

- Reference 21 is not referenced in the text

- The entire section 1. Introduction is given with only one reference. It is necessary to use relevant and up-to-date references

- Referencing in the sentences added in the first revision of the manuscript should be corrected:

TadGAN {Geiger, 2020 #29}: A sliding window of size 10 to calculate the area difference for reconstruction error is used.

LSTM-AE-Advanced (LSTM-AE-ADV) {Kieu, 2018 #104}: use a sliding window size of 4 to perform enrich time series process.

{Geiger, 2020 #29}; {Kieu, 2018 #104}

- The following sentences should be referenced:

Zhou et al. proposed a liquid level in mold anomaly...

Geiger et al. proposed TadGAN...

Minor editing of English language required.

Author Response

Dear Sir/Madam:

We sure appreciate your time to review the new version of our manuscript entitled " Anomaly Detection of Liquid Level in Mold during Continuous Casting by Using Forecasting and Error Generation" again. Your comments and suggestions are very beneficial for us to improve our work. We have revised our manuscript based on your comments, and have added text and tables to our manuscript to make it more specific and clear. We also have finished proofreading and editing our manuscript to improve its readability. Based on the instructions in editor’s letter, we have uploaded the file of our revised manuscript. We highlight all changes in our manuscript by using the “Track Changes” function in MS Word.

Our responses to your comments are appened one by one. Your comment is first reproduced, and our response is given directly afterward in different color (red).

Point 1: The authors have responded to some requests of previous reviewer, but many corrections remain unanswered.

Response 1: I have attached a re-response in the final 2 pages of this letter. We hope you find it useful.

Point 2: It is necessary for the authors to answer the reviewers, point by point, to bring the requested items and answer them. It is necessary for the authors to highlight the changed parts in the revised text.

Response 2: The text change now uses the “Track changes” function in Microsoft Word.

Point 3: The quality of the figures is low, and their contrast should be increased.

Response 3: The quality of the figures is improved, and we have changed the font color in the figures from gray to black to increase the contrast.

Point 4: It is necessary to review and refer to the articles suggested by the previous reviewer.

Response 4: We have carefully read the papers that were recommended previous reviewer, and their research ideas are helpful to this paper. We have cited them as references [3] [11] to improve the introduction and related work section.

Point 5: The entire section 1. Introduction is given with only one reference. It is necessary to use relevant and up-to-date references.

Response 5: We have added more references to Introduction section.

Point 6: Referencing in the sentences added in the first revision of the manuscript should be corrected:

TadGAN {Geiger, 2020 #29}: A sliding window of size 10 to calculate the area difference for reconstruction error is used.

LSTM-AE-Advanced (LSTM-AE-ADV) {Kieu, 2018 #104}: use a sliding window size of 4 to perform enrich time series process

{Geiger, 2020 #29}; {Kieu, 2018 #104}.

Response 6: We have corrected the errors.

Point 7: The following sentences should be referenced:

Zhou et al. proposed a liquid level in mold anomaly...

Geiger et al. proposed TadGAN...

Response 7: We have corrected the error. Zhou et al.’s method is referred by paper [6] and Geiger et al.’s method is referred by paper [23].

We hope that our revised manuscript would be accepted for publication in the Applied Science.

Best regards!

Yours sincerely,

Dr. Xiaojun Wu

Corresponding author:

Assoc. Prof. Xiaojun Wu

School of Software Engineering, Xi’an Jiaotong University, Xi’an 710049, China

Tel: +86-13629206003

E-Mail: xiaojunw@xjtu.edu.cn

Re-response to First Reviewer 4 Comments