Controlled Cooling Temperature Prediction of Hot-Rolled Steel Plate Based on Multi-Scale Convolutional Neural Network

Round 1

Reviewer 1 Report

The authors presented the article named “Controlled Cooling Temperature Prediction of Hot Rolled Steel 2 Plate based on Multi-scale Convolutional Neural Network” journal however needs to be developed according to following comments:

Abstract: Abstract should be revised:

It is important to exhibit the practical contribution of the proposed method into the sector. Also the novelty of the study along with difference with the previous literature needs to be shown in abstract. In addition to that, numerical findings that show the success of the proposed method need to be added.

Introduction: Introduction needs revisions at some parts:

Presentations of references are wrong. There is no parenthesis of something that differ the citation number is missing. The last parts of this section should be widened with the novelty of the work. The main aim and differences from literature should be indicated.

There is need to add literature studies about the topic.

Methods: Are all figures original? If not, there is a need to make appropriate citations and permissions.

Describe the measurement procedure in more detail. At what point in time? How is the measuring setup set up? How many repetitions of measurements? What statistical methods are used to process experimental results? Describe the experimental stand in more detail. What method of experiment planning is used and why?

It will be useful to add a section of Nomenclature in which to sign all the physical quantities and abbreviations encountered in the article. There are physical quantities in the text and such a section will help to find the description of the necessary element.

Results: Frankly speaking, the results are not enough. Some tables and figures are put with insufficient explanations. No comments, no citations, no details about findings. This cannot be acceptable. Much more explanatory details have to be placed.

There is a need to add discussion section. It is necessary to more clearly show the novelty of the article and the advantages of the proposed method. What is the difference from previous work in this area? Show practical relevance. The article is interesting, but needs to be improved. Authors should carefully study the comments and make improvements to the article step by step. Add 4-5 items of the findings of the study.

References: There is a problem with the total number of references in this paper. Such a popular work should be supported at least 30-40 references. Also the authors have to add MDPI papers if their work fits for the journals content.

The research looks like a research report more than a scientific paper. After revision, it will be analyzed again according to the efforts of authors.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The article is relevant for science and practice, but there are a number of issues that require explanation:

1. Schemas at the Fig.2, Fig.5 doesn't contain connections between "Flatten" and "Fully connected" layers, but this connection obviously should exist.

2. Looks inappropriate to choose convolution layers for article's task (Fig.2). As authors stand, "one-dimensional input matrix (Input) is formed by influencing factors such as plate thickness, plate width, plate speed, water temperature..." (line 141). But convolution layers used to find spatial features inside input data (for example specific visual, audio or time patterns). Do authors stand that exist patterns within "plate thickness-plate width- plate speed- water temperature etc" combination values? But what if we change the order of this chain (e.g. put "water temperature" at the beginning of parameters). In this case "spatial order" within input params will different. So, looks like at the place of convolutional layers better use full-connected layers.

Usage of convolutional layers described for example here: https://www.databricks.com/glossary/convolutional-layer#:~:text=Convolutions%20have%20been%20used%20for,between%20neurons%20of%20adjacent%20layers.

Citation: "[convolution] typically used to recognize patterns present in images but they are also used for spatial data analysis, computer vision, natural language processing, signal processing, and various other purposes".

3. It is not entirely clear how the shown method takes into account rolling parameters, for example:

For the production of cold-rolled plate for deep and especially complex drawing, the following modes are recommended:

-The degree of reduction in the last stand of the hot rolling mill should be higher than the critical one and be 15-20%;

- The temperature of the end of rolling for carbon steels (08kp, 08ps, 08Yu) should be in the range of 860-920 0C, which corresponds to the austenitic state of the steel and, after cooling, leads to fine ferritic grains;

Can the proposed model take into account these recommendations? It don’t clear. I don't see in it:

- taking into account the degree of plate compression;

- accounting for equipment wear;

- temperature regimes and their duration are associated with structural transformations in the plates, but this is not the case in the model either. If the structure of the plates is not optimal, and the strength properties not comply with the standards, then the proposed approach will work, but the engineering result will be opposite from the theoretical one.

4. On fig. 9 does not indicate for which type of plates the mode and model are shown (BP neural network), but this is important. It is not clear how much the mode shown differs from that recommended by the standard.

5. The example shown in fig. 9 should include description of the all metallurgical rolling modes (for example, at different speeds etc.) and show the effectiveness of the proposed approach.

6. The conclusions describe the BP neural network, but there is no description of the value of the proposed approaches for metallurgy. It is necessary to show both the scientific novelty and the practical value of the proposed approach for the metallurgical industry.

7. I recommend to supplement the introductory part of the article. It is necessary to show the relationship between the parameters of rolling modes: rolling speed, temperature, degree of sheet reduction and show methods for their description using CNN. Now this problem is not very substantiated.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

the paper can be accepted after revisions

Reviewer 2 Report

The authors answered all questions. I suggest that they supplement the Introduction with an analysis of a recent article on metallurgical topics:

Konovalenko, I.; Maruschak, P.; Brevus, V. Steel surface defect detection using an ensemble of deep residual neural networks, J Comput. Inf. Sci. Eng., 2022, 22, 014501, DOI:10.1115/1.4051435