Next Article in Journal
Marchenko Green’s Function Retrieval in Layered Elastic Media from Two-Sided Reflection and Transmission Data
Next Article in Special Issue
Particle Swarm Optimisation in Practice: Multiple Applications in a Digital Microscope System
Previous Article in Journal
Characterization of Berry Skin Phenolic Profiles in Dalmatian Grapevine Varieties
Previous Article in Special Issue
IIoT Malware Detection Using Edge Computing and Deep Learning for Cybersecurity in Smart Factories
 
 
Article
Peer-Review Record

Application of Efficient Channel Attention Residual Mechanism in Blast Furnace Tuyere Image Anomaly Detection

Appl. Sci. 2022, 12(15), 7823; https://doi.org/10.3390/app12157823
by Rihong Wang 1, Ziyu Li 1, Lingzhi Yang 2, Yuming Li 1, Hao Zhang 1, Chuanwang Song 1,*, Mingjian Jiang 1, Xiaoyun Ye 1 and Keyong Hu 1
Reviewer 2:
Appl. Sci. 2022, 12(15), 7823; https://doi.org/10.3390/app12157823
Submission received: 1 July 2022 / Revised: 1 August 2022 / Accepted: 2 August 2022 / Published: 4 August 2022
(This article belongs to the Special Issue Applications of Deep Learning and Artificial Intelligence Methods)

Round 1

Reviewer 1 Report

This paper is framed in the steelmaking industry. In this industry, it is common to use furnaces that work at high pressures and temperatures, and which need to be supervised or tracked to avoid risky operations states that might result in catastrophe situations. This difficult task was traditionally done by experienced workers who observe the raceway from the peephole and judge the combustion state of in the blast furnace. In the light of this circumstances, in this paper the authors propose a novel method to detect abnormal conditions from images of the blast furnace tuyere, based  on improved channel attention mechanism and residual network.  

The document is written in a good English. It is fairly well organized, and it includes an Abstract, a first section containing an Introduction, followed by a Related work, a Porposed methods, an Experiments and a Conclusion sections, respectively, finishing with the listing of References used.

As it is, after a detailed review, in my opinion the manuscript shows enough value, and aiming to improve the quality of the work and without any intent to underrate neither its accuracy nor its contributions, I would like to make the following suggestions and comments, considering that the authors need to expand their manuscript and clarify some issues. My main concerns would be the following:

1.  As for the structure of the manuscript, I would recommed to split up the 'Experiments' section into 'Experiments' and a new 'Discussion of results' section, in order to separate the description of the experiments and the critical analysis of the obtained results. Also, according to the instructions for authors of the Applied Sciences journal: ‘We do not have strict formatting requirements, but all manuscripts must contain the required sections: Author Information, Abstract, Keywords, Introduction, Materials & Methods, Results, Conclusions, […].’. Please, try to adapt your manuscript to this structure.

2.  Regarding the Abstract:
       o  The importance of the problem is not clear. The scope of application is not well presented.
       o  Regarding the architecture used, it is not clear if you propose a new one or you use an architecture that already exists. Please, check also the title.
       o  Highlight in the abstract that the obtained accuracy is using k-fold cross validation. This value of accuracy was not obtained in a test dataset.
       o  Explanations regarding to the dataset used for training the model are needed. You also need to explain that it was unbalanced and how you have fixed these issues.
       o  It is also mentioned that: ‘Compared with the other methods such as SERNet, ResNeXt and repVGG, the proposed method has a better classification effect.’ I consider that you cannot generalize to that level with the analysis you have made. You can only say that this is true with the dataset you have used.

3.  In Lines 30-31, the authors mention ‘Blast furnace steelmaking is a key task in the iron and steel industry, which accounts for more than 95% of global iron production.’. The authors should add some citations which support this affirmation. Same for the catastrophes: if possible add citations.

4.  I consider that an explanation accompanied with an image of the parts of blast furnace could help to frame and better understand the issues to be addressed.

5.  The authors mention that they use  a monitoring camera. Which type of camera? Could they explain what is the set-up? How is it placed in the furnace? It would be helpful to have some images to illustrate this.

6.  The authors mentioned that several catastrophes happened in the past in this kind of furnaces. What is the aim of your methodology? To replace the workers who monitor the furnace or to complement them? These issues must be discussed in detail. In case of an error, who will be responsible? Is this an early-warning system?

7.  It is recommendable to add a paragraph at the end of the Introduction section explaining how the rest of the manuscript is organized.

8.  The beginning of section 2 might result difficult to understand. First, you speak about several parts which are detected by the video system without presenting them before. As commented, it would be of help to have some images and explanations of the furnace and its components.

9.  In Section 3, the architecture used is presented. As commented in the Abstract section analysis above, it is not clear if this is a novel proposal or if they are just applying an architecture which already exists to an specific case. Do you know if someone else have used this architecture before?

10.  In Section 4, the authors introduce a dataset. Is this a public dataset or was it collected by the authors? If it was collected, which were the conditions? How was it labeled? How many instances does it have of each class?

11.  Which software was used for training the model?

12.  I consider that the training parameters of sub-section 4.2 should be presented in a table.

13.  The authors mention that ‘The accuracy of ESERNet model is only 0.28% lower than that of SERnet model, and the number of parameters is reduced by 26.62%, which not only achieves higher classification accuracy, but also greatly reduces the model complexity.’ This is true in this context, with the described training and validation data. I think that you should not generalize: what would happen with new data? Will your model be better than the other ones?

14.  Could the trained architecture be used in other furnaces from other factories with this training? If not, what should be the investment needed to build a new dataset?

15.  You mention the goodness of your approach. However, you do not mention the caveats and limitations. Can you elaborate on these?

16.  As a minor comment, please check line 84: after the comma you should not capitalize the following word.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors have provided an image recognition method for blast furnace tuyere image anomaly detection. The obtained results of the proposed method have high accuracy; however, the following modifications should be considered in the manuscript.

 

-        Kindly explain and emphasize the main contributions of the presented method. Please clarify that if the previous networks are used for anomaly detection or the proposed one?

-        Which data set is used for training the network and also for experiments? The overview of the used images or videos datasets should be depicted in the manuscript.

-        There are several grammatical errors and typos in the text. the manuscript English should be revised carefully.

-        It is stated that the “ESERNet model greatly reduces the number of parameters”. Is this reduction improved the time of the network? The run time of the networks are not compared.

-        The feature extraction method is not discussed.

-        It seems that the SERNet and repVGG networks have better accuracy compared to the presented one. What are the main advantages of the proposed method?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The Authors have submitted a new version of the manuscript that aims to address the issues that were pointed to in the submitted review report.

Firstly, I wish to thank the Authors for their effort in the changes made in the document in response to the indications in my previous review. It would have been helpful, though, that you had pointed to those specific lines where the changes were made.

After reviewing the modifications, I think relevant to point point out that in my opinion some issues still need to be addressed:

*  According to the instructions for authors of the Applied Sciences journal: ‘We do not have strict formatting requirements, but all manuscripts must contain the required sections: Author Information, Abstract, Keywords, Introduction, Materials & Methods, Results, Conclusions, […].’. Please, try to adapt your manuscript to this structure.

*  The abstract still needs to be improved. I still consider that the relevance of the problem, and thus the need for a solution, is not clear. The scope of application is still not well presented, in my opinion.

*  I consider that the authors should make a reflection on the responses to points nr. 6, 9, 14 and 15 in the Discussion section.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

All of my comments are considered in the manuscript and the manuscript can now be accepted after correcting following minor modifications:

-        The refs [19-21], [23], [24], … are not cited in the numerical order. Kindly correct the order of citations in the text.

-        Also, the style of the references at the end of manuscript should be unified and should follow the journal format.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Back to TopTop