Artificial Neural Network-Based Critical Conditions for the Dynamic Recrystallization of Medium Carbon Steel and Application

Round 1

Reviewer 1 Report

The authors of the paper "Artificial Neural Network based critical conditions for Dynamic Recrystallization of Medium Carbon Steel and Application" have developed several models for the calculation of steel microstructural evolution during hot deformation. The constructed models have shown good accuracy and were realized by finite element simulation. The paper is well written and may be accepted for publication after the modification accordingly following comments:

1.                 The most of references about ANN-based models in the Introduction part are too old. It is recommended to analyze new papers about modeling the alloys’ properties using the ANN approach (e.g., 10.3390/met12091429, 10.3390/ma16031083, 10.1016/j.matdes.2022.110880, etc.).

2.                 The scheme in Figure 7 is hardly applicable for the description of the dynamic recrystallization. It is more appropriate for static processes. During dynamic application, the deformation of all grains (even recrystallized) proceeds simultaneously. It is recommended to remove or modify the provided scheme.

3.                 The authors have used a Zener-Hollomon parameter for the constitutive models. What was the value of effective activation energy in the Zener-Hollomon parameter?

4.                 Experimental points are given in Figure 8 but microstructures were not presented. It is recommended to add microstructure images for the investigated modes.

5.                 Not only DRX fraction but DRX grain size has a significant influence on the properties of hot deformed steel. Why the authors have considered only one microstructural factor?

6.                 A complete match between calculated and experimental DRX values seems to be too strange. Are the presented in the microstructure values measured values of the recrystallized fraction? In another case, it is recommended to determine experimental values of the DRX fraction using EBSD techniques.

7.                 Minor correction:

-                     Figure 3 and Figure 9 present the same information. It is recommended to leave only Figure 9 and move it to the Materials and Method part.

Author Response

Comments and Suggestions for Authors The authors of the paper "Artificial Neural Network based critical conditions for Dynamic Recrystallization of Medium Carbon Steel and Application" have developed several models for the calculation of steel microstructural evolution during hot deformation. The constructed models have shown good accuracy and were realized by finite element simulation. The paper is well written and may be accepted for publication after the modification accordingly following comments:

We appreciate the reviewer's interest in this work and their positive comments, which have helped us improve the manuscript. Below are our responses to their comments.

 

1. The most of references about ANN-based models in the Introduction part are too old. It is recommended to analyze new papers about modeling the alloys’ properties using the ANN approach (e.g., 10.3390/met12091429, 10.3390/ma16031083, 10.1016/j.matdes.2022.110880, etc.).

 

In the introduction from line 19 to line 98, only one reference is made to ANN based flow law, in the introduction which is ref [27] dated 2023. The mention of many other works related to ANN behavior law is not necessary and thus irrelevant. This paper's focus is on DRX (ref 6 to 26) and not behavior law using ANN. Regarding the recommended papers, they all pertain to artificial neural network-based flow laws and their application. This is very similar to the approach used in this paper, which is based on references [27,28,29] published in 2022 and 2023.

 

2. The scheme in Figure 7 is hardly applicable for the description of the dynamic recrystallization. It is more appropriate for static processes. During dynamic application, the deformation of all grains (even recrystallized) proceeds simultaneously. It is recommended to remove or modify the provided scheme.

Figure 7 has been removed in the revised version of the paper, because the process is quite well known and we prefer to remove it if it is confusing for the reader. For a visual representation of microstructure evolution during DRX, the reviewer may refer to Figure 6 in this paper (https://doi.org/10.3390/met9121291).

 

3. The authors have used a Zener-Hollomon parameter for the constitutive models. What was the value of effective activation energy in the Zener-Hollomon parameter?

Thanks for this remark, the value of Q is 437.4kJ/mol.

4. Experimental points are given in Figure 8 but microstructures were not presented. It is recommended to add microstructure images for the investigated modes.

 

One new figure (9) has ben added to illustrate the process. We do not have sufficient space to provide more micrographs.

5. Not only DRX fraction but DRX grain size has a significant influence on the properties of hot deformed steel. Why the authors have considered only one microstructural factor?

We agree that both factors have an effect on the material properties. Nonetheless, as outlined in the paper's introduction, our emphasis is on the significance of utilizing the ANN model in assessing the DRX's critical conditions. Therefore, we conducted only a few experiments (for a single strain rate) to verify the proposed model, and the data were insufficient to determine the model's ability to predict grain size evolution. It is expected that future work will involve the development of additional techniques and consideration of grain size evolution during simulation.

6. A complete match between calculated and experimental DRX values seems to be too strange. Are the presented in the microstructure values measured values of the recrystallized fraction? In another case, it is recommended to determine experimental values of the DRX fraction using EBSD techniques.

The strong correlation between the experimental results and the predictions can be attributed to the lengthy process time of 850 seconds, which almost achieves recrystallization at low strain rates. Regarding the images showcased, the selection was based on the simulation duration. In Abaqus/Explicit, simulating for 850 seconds necessitates a lengthy computation period. Therefore, we opted to augment the strain rate to 0.01/s and diminish the computing time to 85 seconds.

7. Minor correction:

- Figure 3 and Figure 9 present the same information. It is recommended to leave only Figure 9 and move it to the Materials and Method part.

Figure 3 depicts a neural network with a single input and output used to filter the stress/strain curves, while Figure 9 depicts an ANN with three inputs (strain, strain rate, and temperature) and one output used to replace the analytical behavior law.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper mainly discussed prediction of DRX of medium carbon steel using artificial neural network.

 

 

-On page 5: lines150-174, this part look like a introduction. Thus better to move them before ‘2. Materials and experiments’ on page 3

 

-On page 11: regarding Zener-Hollomon parameter Z, comment how to determine the activation energy Q

 

-On page 13: lines 366 and 367, comment how to obtain the training database of 21,030 (maybe 703*30)

Author Response

Comments and Suggestions for Authors This paper mainly discussed prediction of DRX of medium carbon steel using artificial neural network.

We thank the reviewer for their interest in this work and the helpful feedback provided to improve the manuscript.

Specific comments 1) On page 5: lines150-174, this part look like a introduction. Thus better to move them before ‘2. Materials and experiments on page 3

Thanks for the remark, we completely remove that part in the revised version of the paper since it is already detailed in the introduction part.

2) On page 11: regarding Zener-Hollomon parameter Z, comment how to determine the activation energy Q.

The procedure is well described in our recent publication (https://doi.org/10.3390/met13030633).

3) On page 13: lines 366 and 367, comment how to obtain the training database of 21,030 (maybe 703*30).

Yes, you're true, but its 701*30 values because strain is in [0,0.7] with an increment of 0.001, so we have 701 values and of course 30 curves. Clarification has been added to the text.

 

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Review on „Artificial Neural Network based critical conditions for Dynamic Recrystallization of Medium Carbon Steel and Application”

 

The paper is well organized and easy to read. All graphs and are of appropriate quality.

 

Minor corrections:

Page 6, line 200: “This can be Done ! by…”; Please correct the sentence.

Please provide more information about the code (Pyton code) of ANN used for the simulation.

 

Major points of concern:

Line 280: Citation to Avrami model ([23]) is improper. Please use proper literature source.

Citations 33-35 are also improper. Zener-Hollomon parameters and quantities described by equation 10 were defined by different research groups (not by cited publications [33-35]). Please refer to original literature sources.

 

Figure 8. Why experimental points are shown only for one strain rate? Please note that model parameters defined for one particular strain rate will not necessary be valid for other ones. This is particularly true in the case of high stain rates.

Figure 14 is nice but experimental validation is missing.

 

Figure 15 is not convincing. Figure 10b does not show 87.5% DRX fraction. The same id true for figure 10b.

 

Summary: This work requires major revision prior to publication.

Author Response

Review on “Artificial Neural Network based critical conditions for Dynamic Recrystallization of Medium Carbon Steel and Application”
The paper is well organized and easy to read. All graphs and are of appropriate quality.

 

We appreciate the reviewer's interest in this work and their positive comments, which have helped us improve the manuscript. Below are our responses to their comments.

 

Minor corrections:

1) Page 6, line 200: “This can be Done ! by…”; Please correct the sentence.

Thanks, it has been corrected.

 

2) Please provide more information about the code (Python code) of ANN used for the simulation.

The subroutine is programmed using Fortran 77 instead of Python. A paragraph has been added at the end of Section 3.1 to provide further information on this process. For more details, please refer to [28,29].

 

Major points of concern:

 

1) Line 280: Citation to Avrami model ([23]) is improper. Please use proper literature source..

 

Thanks, now corrected in the text.

 

2) Citations 33-35 are also improper. Zener-Hollomon parameters and quantities described by equation 10 were defined by different research groups (not by cited publications [33-35]). Please refer to original literature sources.

 

Corrected in the text

 

3) Figure 8. Why experimental points are shown only for one strain rate? Please note that model parameters defined for one particular strain rate will not necessary be valid for other ones. This is particularly true in the case of high stain rates.

 

We agree that a single strain rate is inadequate to validate the model. However, the simulation uses a different strain rate and the results show a good correlation with the experimental results. If the model was unsuitable, the simulation results would have markedly differed from the experimental ones, but this was not the case, thus demonstrating the validity of the proposed model. To further support the validity of our proposed model, we intend to conduct additional experiments in the future.

 

 

4) Figure 14 is nice but experimental validation is missing.

 

Thanks for appreciating the results. However, we conducted relatively few experiments, and possibly, in the future, we will develop techniques to perform more simulations of this kind.

 

5) Figure 15 is not convincing. Figure 10b does not show 87.5% DRX fraction. The same id true for figure 10b.

 

If we interpreted the reviewer's comment correctly, Figure 8 should have been referenced instead of Figure 10.   And, oops, there is a mistake in the Experimental validation paragraph ! In fact, the strain rate used for the simulations was not 0.01/s but 0.1/s. It has now been corrected in the text and legend of Figure 15. The recrystallization is not completed in this case as reported in Figure 8(c), blue line in the graph at the lower left.

 

6) Summary: This work requires major revision prior to publication.

Thanks for great comments

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have answered previous comments and improved the manuscript. The paper may be accepted for publication.

Author Response

Thank you very much for your comments and your appreciation of our work.

Reviewer 3 Report

The Authors did not answer my my comments 3, 4 and 5 properly.

My major point of concern is that the model is validated for a very narrow set of strain rates. I think that the Authors should show the validity of their model for a wider strain rate spectrum.

Author Response

Answers to the three remaining questions are in the attached pdf document.

Author Response File: Author Response.pdf

Round 3

Reviewer 3 Report

-