Machine-Learning-Based Consumption Estimation of Prestressed Steel for Prestressed Concrete Bridge Construction

Round 1

Reviewer 1 Report

In this paper, different machine learning methods are studied to predict prestressed steel consumption in prestressed bridge construction. Neural network (NN), regression tree (RT) based model and genetic programming (GP) model are analyzed. The multi-criterion optimization method is used to determine the optimization model based on GP, and the accuracy of the optimization model expressed by MAPE criterion is realized to construct an algorithm model that can be understood and applied by experts involved in the construction of prestressed bridge.The current document had several weaknesses that must be reinforced in order to obtain documentary results equal to the value of publications.

(1) The numbers and annotations for all figures and tables should be centered.

(2) The text in Figure 9's mind map is unclear and should be replaced.

(3) Tables 1, 2, 5, 6, 7, and 8 are not standard enough. The notes for Table 1 should remain.

(4) Formulas on lines 291 and 295 should be numbered.

(5) The document contains a total of 37 references, with only 11 being published within the last 5 years (30%), 7 within the last 5-10 years (19%), and 19 being over 10 years old (51%). Thus, recent references only make up a small percentage. More practical and recent references should be added.

(6) The content of this paper leans more towards the introduction and optimization of the algorithm model using data from bridge concrete construction prestress for training. Therefore, it may be more appropriate to focus on the algorithm model of machine learning.

(7) Compared to the article's content, the number of words is limited. It is necessary to add the general content of the article, the algorithm model used, optimization methods implemented, and the results obtained.

(8) Machine learning in civil engineering applications should be briefly presented (An experimental investigation and machine learning-based prediction for seismic performance of steel tubular column filled with recycled aggregate concrete. Reviews on Advanced Materials Science. Prediction of ultimate condition of FRP-confined recycled aggregate concrete using a hybrid boosting model enriched with tabular generative adversarial networks, Thin-Walled Structures. Prediction of thermo-mechanical properties of rubber-modified recycled aggregate concrete. Construction and Building Materials.).

(9) Descriptions of other scientists' models and their accuracy and data can be reduced.

(10) Highlight the contribution of your work by placing it in context with the work previously done in the same domain to emphasize its novelty.

(11) The limitations of the work should be rigorously assumed and justified.

(12) Chapter 2 Methods: This paragraph introduces too much several algorithm models, and does not mention how to use them in this research, readers will easily feel boring and fail to find the key content. Too much meaning explanation in this respect is slightly different from the research theme of this paper, so it should be reduced.

(13) Chapter 5 Conclusions: It is suggested that the author objectively evaluate and analyze this research method on the basis of comparing the research experiments of other scholars. At the same time, the feasibility and application prospect of the study should also be analyzed.

(14) In the final conclusion, we should put forward the shortcomings of this paper, analyze what problems and difficulties still exist in the experiment to be solved, and how to further study in the future.

Author Response

We would like to thank the Editor for the consideration of our manuscript. The comments we received from the reviewers have been very helpful. We also thank the reviewers for providing questions, suggestions, and insights to improve our manuscript.

We summarize the changes we made to the manuscript and responses to specific reviewer comments. To facilitate reading our responses, we repeat each comment we received, providing our response below. For the reviewers’ and Editor’s convenience, the revised manuscript also uses track changes for the text that has been added and/or significantly modified.

Responses to Reviewer 1:

1. The numbers and annotations for all figures and tables should be centered.

Response: All numbers and notes in the tables are centered.

2. The text in Figure 9's mind map is unclear and should be replaced.

Response: Figure 9, which was not sharp enough, was completely replaced with a better-quality image.

3. Tables 1, 2, 5, 6, 7, and 8 are not standard enough. The notes for Table 1 should remain.

Response: Tables 1, 2, 5, 6, 7, and 8 have been standardized and corrected.

4. Formulas on lines 291 and 295 should be numbered.

Response: Formulas on lines 291 and 294 are numbered. After that, the labels of all the formulas in work were updated.

5. The document contains a total of 37 references, with only 11 being published within the last 5 years (30%), 7 within the last 5-10 years (19%), and 19 being over 10 years old (51%). Thus, recent references only make up a small percentage. More practical and recent references should be added.

Response: In the paper, the context of the application of machine learning methods in construction was pointed out, and the paper was updated with newer references. The reference section was updated. As a result, the following references have been entered:

• Tang, Yunchao, Wang, Yufei, Wu, Dongxiao, Liu, Zhonghe, Zhang, Hexin, Zhu, Ming, Chen, Zheng, Sun, Junbo and Wang, Xiangyu. "An experimental investigation and machine learning-based prediction for seismic performance of steel tubular column filled with recycled aggregate concrete" REVIEWS ON ADVANCED MATERIALS SCIENCE, vol. 61, no. 1, 2022, pp. 849-872. https://doi.org/10.1515/rams-2022-0274.
• Wanhui Feng, Yufei Wang, Junbo Sun, Yunchao Tang, Dongxiao Wu, Zhiwei Jiang, Jianqun Wang, Xiangyu Wang, Prediction of thermo-mechanical properties of rubber-modified recycled aggregate concrete, Construction and Building Materials, Volume 318, 2022. https://doi.org/10.1016/j.conbuildmat.2021.125970.
• Xin-Yu Zhao, Jin-Xin Chen, Guang-Ming Chen, Jin-Jun Xu, Li-Wen Zhang, Prediction of ultimate condition of FRP-confined recycled aggregate concrete using a hybrid boosting model enriched with tabular generative adversarial networks, Thin-Walled Structures, Volume 182, Part B, 2023. https://doi.org/10.1016/j.tws.2022.110318.

6. The content of this paper leans more towards the introduction and optimization of the algorithm model using data from bridge concrete construction prestress for training. Therefore, it may be more appropriate to focus on the algorithm model of machine learning.

Response: The paper provides a broader explanation of the applied algorithms. The focus is on a more detailed explanation of the practical implementation itself. In addition, a broader explanation is given for the applied methodology in work, from the division of data to the actual implementation of the model.

7. Compared to the article's content, the number of words is limited. It is necessary to add the general content of the article, the algorithm model used, optimization methods implemented, and the results obtained.

Response: The results obtained in the paper are explained in more detail. It was processed in more detail on the obtained model in terms of its critical analysis and limitations.

8. Machine learning in civil engineering applications should be briefly presented (An experimental investigation and machine learning-based prediction for seismic performance of steel tubular column filled with recycled aggregate concrete. Reviews on Advanced Materials Science. Prediction of ultimate condition of FRP-confined recycled aggregate concrete using a hybrid boosting model enriched with tabular generative adversarial networks, Thin-Walled Structures. Prediction of thermo-mechanical properties of rubber-modified recycled aggregate concrete. Construction and Building Materials.).

Response: In the paper, the broader context of applying machine learning methods in construction was pointed out, and the paper was updated with new references.

9. Descriptions of other scientists' models and their accuracy and data can be reduced.

Response: The reduction of the content in terms of other authors and the accuracy of the model has been carried out. In addition, the broader importance of the application of machine learning methods in the construction industry was pointed out.

10. Highlight the contribution of your work by placing it in context with the work previously done in the same domain to emphasize its novelty.

Response: In the last paragraphs of the Introductions chapter, the paper's contribution is highlighted in more detail, and the motives for conducting this research and the primary hypothesis are additionally explained.

In addition, in the Conclusions chapter, the contribution of this research to already realized research is highlighted in more detail. In addition, the limitations of this research and potential future research directions are listed.

In addition, for the first time, to the author's knowledge, this method was applied to predict the amount of steel in prestressed bridges. Research has shown that it can be even more accurate than previously widely applied methods for the same problem.

11. The limitations of the work should be rigorously assumed and justified.

Response: As a limitation of this research, it was stated in the Conclusions chapter that all models based on machine learning techniques make predictions based on the data presented to the model in the training phase. The model created in the paper is limited to having the accuracy specified for data having the statistical characteristics specified in Table 3. of model variables.

The limitation was pointed out in terms of the fact that the bridges were built in a narrow geographical area and the consequences that arose from that.

Although the database of prestressed bridges, in this case, is composed of data on 74 completed bridges, it can be considered significant. Expanding the database would result in more data within the training and test data set, which could serve to define the model and its accuracy even more precisely.

12. Chapter 2 Methods: This paragraph introduces too much several algorithm models, and does not mention how to use them in this research, readers will easily feel boring and fail to find the key content. Too much meaning explanation in this respect is slightly different from the research theme of this paper, so it should be reduced.

Response: Part of the material presented in the paper in the chapter 3 Methods has been reduced.

Essential elements have been retained in the rest of the work. A more detailed explanation is given regarding determining the optimal hyperparameter values in the MGGP model.

13. Chapter 5 Conclusions: It is suggested that the author objectively evaluate and analyze this research method on the basis of comparing the research experiments of other scholars. At the same time, the feasibility and application prospect of the study should also be analyzed.

Response: An additional explanation was given related to the problem of estimating resource consumption, which was also considered by other scientists. It was pointed out that in all the research, as mentioned earlier, models based on linear regression and machine learning black-box models which were used significantly more complex. However, the research applied in this paper has, in terms of results compared to previous research, produced a result that is generally better than the other researched models in terms of criteria of accuracy and complexity. Furthermore, the obtained model is straightforward and is in the form of a simple equation.

In addition, compared to most works in the literature, the database for creating the model was more extensive in this one. Therefore, the model that has been defined can be applied to the category of bridges made of prestressed concrete, which according to the defined span variables (three variables) and the width of the bridge, would be within the range of the data used to create the model. In addition, the conclusions chapter rigorously states the potential limitations of the model.

14. In the final conclusion, we should put forward the shortcomings of this paper, analyze what problems and difficulties still exist in the experiment to be solved, and how to further study in the future.

Response: The bridges within the database were realized in a relatively narrow geographical area. In that area, there are no significant differences in load on bridges (seismic load, wind load, etc.). If there are significant differences in model loading during model training, an additional input variable could be introduced into the training data set that would include this. In addition, the developed model should be corrected in the case of the application of prestressing steel with significantly different mechanical characteristics.

The potential broadening of the base would be positive in finding the optimal model. In addition, it was pointed out that in the event of a significant increase in the database in future research, clustering methods could also be applied, where the methodology developed in work would be applied in individual clusters, and separate equations defined for each type of cluster.

Author Response File: Author Response.docx

Reviewer 2 Report

This article shows the possibility of using various ML tools to solving the problem of resources consumption estimation.

The article is well written, good designed and the problem that the authors are investigating is relevant.

In most, the structure of the article meets the classical requirements.

The article title is clear.

Keywords are chosen correctly and match the direction of the research.

However, there are points that could be improved.

Abstract. I recommend that authors redo the abstract after making changes to the body of the article. An abstract is a should give a pertinent overview of the work and should be an objective representation of the article. I recommend briefly describe the background that motivated the authors to this study.

In the introduction, the authors have analyzed the previous studies in this direction quite well. In my opinion, there is not enough justification of the scientific lack and the reasons that motivated the authors to this study.

The introduction section should be strengthened by clearly identifying the research hypothesis, research questions, objectives, motivations, and novelty of the research. It will help the reader to understand the context and purpose of the study better.

The Section Dataset. 

The process of creating a dataset (input data, output data), cleaning, data preparation, divide into training and verification data is poorly described.

How did the authors get the input data? Why did the authors choose such input variables? Please explain and justify.

Why did the authors add Table 3. What is its value for this study?

The MAPE indicator is calculated in %.

Discussion section.

The problem of resources consumption estimation using various ML tools has already been repeatedly considered by scientists. Accordingly, the "Discussion" section should contain a comparison of the results of this article and similar studies by other authors. The authors need to prove the superiority and scientific novelty of this study.

In abstract, the authors claim that "it is feasible to design a model allowing a sufficiently precise forecast of resource consumption". In this case, the article should show how trained models predict output values on the basis of new input data that the model has not yet "seen".

Author Response

We would like to thank the Editor for the consideration of our manuscript. The comments we received from the reviewers have been very helpful. We also thank the reviewers for providing questions, suggestions, and insights to improve our manuscript.

We summarize the changes we made to the manuscript and responses to specific reviewer comments. To facilitate reading our responses, we repeat each comment we received, providing our response below. For the reviewers’ and Editor’s convenience, the revised manuscript also uses track changes for the text that has been added and/or significantly modified.

Responses to Reviewer 2:

1. I recommend that authors redo the abstract after making changes to the body of the article. An abstract is a should give a pertinent overview of the work and should be an objective representation of the article. I recommend briefly describe the background that motivated the authors to this study.

Response: The paper underwent substantial changes following the reviewer's suggestions. As a result, the abstract has now been significantly modified. In addition, the background that motivated authors is also described and given in the paragraphs at the end of the Introduction chapter.

2. In the introduction, the authors have analyzed the previous studies in this direction quite well. In my opinion, there is not enough justification of the scientific lack and the reasons that motivated the authors to this study.

The introduction section should be strengthened by clearly identifying the research hypothesis, research questions, objectives, motivations, and novelty of the research. It will help the reader to understand the context and purpose of the study better.

Response: The introduction chapter now gives an additional explanation of the reasons that motivated the authors to research. Hypotheses raised in work were identified to understand the research's importance and context better. In addition, additional explanations were included in the other chapters of the work, where potential limitations of the defined model were pointed out.

3. The Section Dataset. 

The process of creating a dataset (input data, output data), cleaning, data preparation, divide into training and verification data is poorly described.

How did the authors get the input data? Why did the authors choose such input variables? Please explain and justify.

Response. In this part, additional explanations related to input and output variables are included. It is explained in more detail why those variables were included in the model. The implementation of dividing data into separate data for model creation - training data, and a separate set of data for model testing - test data is explained.

4. Why did the authors add Table 3. What is its value for this study?

Response:  An additional explanation related to Table 3 was included in the paper. In Table 3, the mechanical characteristics of the prestressed steel ropes are listed because the developed model's limitation is that the amount of steel obtained by prediction refers to the prestressed steel of the given characteristics. Therefore, if the defined model is to be applied to the prestressing rope of other mechanical characteristics, the prediction model must take the relationships of the mechanical characteristics in that case.

5. The MAPE indicator is calculated in %.

Response: In all figures and tables, where necessary, changes were made in order to present the numerical value of MAPE correctly.

6. Discussion section.

The problem of resources consumption estimation using various ML tools has already been repeatedly considered by scientists. Accordingly, the "Discussion" section should contain a comparison of the results of this article and similar studies by other authors. The authors need to prove the superiority and scientific novelty of this study.

In abstract, the authors claim that "it is feasible to design a model allowing a sufficiently precise forecast of resource consumption". In this case, the article should show how trained models predict output values on the basis of new input data that the model has not yet "seen".

Response: An additional explanation was given related to the problem of estimating resource consumption, which was also considered by other scientists, where it was pointed out that in all the research as mentioned earlier, models based on linear regression and black-box were used significantly more complex. The research applied in this paper has, in terms of results compared to previous research, produced a result that is generally better than the other researched models in terms of criteria of accuracy and complexity. Furthermore, the obtained model is straightforward and is in the form of a simple equation.

In addition, compared to most works in the literature, the database for creating the model was more extensive in this one. Therefore, the model that has been defined can be applied to the category of bridges made of prestressed concrete, which according to the defined span variables (three variables) and the sirens of the bridge, would be within the range of the data used to create the model. In addition, the conclusions chapter rigorously states potential limitations of the model.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

accept

Reviewer 2 Report

I recommend publish the article in present form