Intelligent Prediction of Prestressed Steel Structure Construction Safety Based on BP Neural Network
Round 1
Reviewer 1 Report
This paper deals with an innovative topic concerning the special issue of the journal.
Each step of the research is well described, but I have to report two weakness:
- it is not clear how BIM has improved this research, and how much weight it has had in this research; it seems that the MIDAS software is the only one to be used; MIDAS is BIM-ready, so how did the authors make the two software interact?
- the structure of the paper — or the sequence of its sections and their content — does not follow the "Research Manuscript Sections" as indicated in the "Instructions for Authors" < https://www.mdpi.com/journal/applsci/instructions >
If the authors could organize the paper according to the "Research Manuscript Sections”, they would most likely be able to clarify the weight that BIM has certainly had in their research.
Author Response
Response to Reviewer 1 Comments
Dear reviewer:
First of all, thank you very much for your affirmation of the research content of the article. Thanks for your suggestions and questions. Your opinions are of great help to improve the level of the paper. I made the following responses to your suggestions and questions:
Point 1: it is not clear how BIM has improved this research, and how much weight it has had in this research; it seems that the MIDAS software is the only one to be used; MIDAS is BIM-ready, so how did the authors make the two software interact?
Response 1: Thank you for your suggestion. BIM is the basic modeling software in the research process. At the same time, according to the collaboration of BIM software, we modify the properties of components in BIM model to obtain the form of structure directly. The establishment of BIM model provides design parameters for the prediction process. The BIM model is imported into MIDAS, and the corresponding material properties are added in the finite element model to analyze the mechanical properties, so as to obtain the mechanical parameters of structural safety. The integration of BIM and MIDAS provides data support for structural safety performance prediction. The relationship between the two is supplemented in lines 246-250 and 262-264 of the article. At the same time, Figure 3 is also improved in order to better realize the role of BIM. In short, BIM has played a role in the establishment of geometric models in the study. The structural form can be directly obtained by modifying its properties, providing design parameters and supporting the establishment of finite element models. Based on BIM model, a digital twin model is formed and combined with BP neural network, a structural safety prediction method is formed. The integration of BIM and other technologies has promoted the development of the intelligent level of the construction industry, and also conforms to the theme of the special issue.
Point 2: the structure of the paper — or the sequence of its sections and their content — does not follow the "Research Manuscript Sections" as indicated in the "Instructions for Authors" < https://www.mdpi.com/journal/applsci/instructions >
Response 2: Thank you for your suggestion. According to the format, we adjusted the chapters of the article. This makes the idea of the article clearer and improves the discussion of the results of the article (lines 406-413 and 432-437).
Point 3: If the authors could organize the paper according to the "Research Manuscript Sections”, they would most likely be able to clarify the weight that BIM has certainly had in their research.
Response 2: Thank you for your suggestion. Through the modification of the format, the analysis of BIM is supplemented, and a complete explanation is given in section 2.2.1 of the article.
Finally, thank you again for your guidance. I hope I can finish an excellent paper with your guidance and help, and sincerely hope that my paper can be published in your journal.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The manuscript with reference number applsci-1554795 and title ‘Intelligent prediction of prestressed steel structure construction safety based on BP neural network’ by the authors H. Zhu and Y. Wang is a concept paper aimed at providing a methodology to assess structural safety based on the design/mechanical parameters and working conditions of a structure.
The paper intends to integrate the use of neural networks with digital twin models. This is a positive point for the study because it discusses the applicability of advanced models for assessing structural safety that to date have not been adopted systematically, and little research has been conducted to assist in establishing their use in the construction section. Moreover, one can accept that the results could be of some interest and relevance beyond the specific structural typology (i.e. prestressed steel structures) since similar systems can be applied on many other types of structures.
The study seems to be problematic in the sense that the novelty of the paper is very limited from the methodological point of view, as the paper adds very little to the existing state-of-the-art in constructions. Overall the benefits to the broader readership of Applied Sciences from this paper are minimal. Some issues regarding the clarity and coherence of the manuscript are pointed out in the list of comments found below. Notwithstanding the novelty issues, a weak element of the paper is the presentation of the proposed methodology, which seems to be very general and theoretical. For example, the Authors fail to detail the mechanical rules governing the structural response (in deformation or stress terms) of the example structure to external loading.
The paper has been written generally well, but in some parts carelessly; consequently, some passages were superfluous and challenging to read, while mention to some critical points was inadequate or completely missing. For instance, the introduction of the paper provides much information on prestressed steel cable structures, which could have been avoided. Also, in many parts reference is made to existing literature published in the Chinese language, which is not accessible to the broader readership of Applied Sciences.
Finally, the Authors fail to demonstrate how to integrate their mechanical models with the neural network that assesses the safety of a structure. As such the paper appears to be incomplete. The Reviewer advises the Authors to revise their study, elaborate more on the numerical models they use, and provide specific guidelines on how to efficiently use the tools they mentioned in their manuscript to simulate structural response and assess safety. In it's current state, the manuscript has little new information to offer, as evident by the Authors’ poor commentary in the Conclusion section.
Other comments:
1) In the Abstract, the Authors could include a couple of sentences explaining the novelty of their study.
2) In the Introduction, the background provided in the first two paragraphs is not relevant to the work described in the manuscript. The Authors could discard repetitive sentences and organize previous studies in a clear manner.
3) The Authors should provide more information on the numerical structural models they employed to assess the response of the structure to external loading.
4) It is not clear to the Reviewer how the Authors combined digital twin models and neural networks to assess the safety of the structure. Please, elaborate on the details of this integration and how such a technique should be used by a reader interested in generating similar prediction models.
5) The utility of Figure 7 is questionable, and any information provided therein is hard to appreciate. Could you please elaborate more on the data presented in this figure?
Author Response
Response to Reviewer 1 Comments
Dear reviewer:
First of all, thank you very much for your affirmation of the research content of the article. Thanks for your suggestions and questions. Your opinions are of great help to improve the level of the paper. I made the following responses to your suggestions and questions:
Point 1: The study seems to be problematic in the sense that the novelty of the paper is very limited from the methodological point of view, as the paper adds very little to the existing state-of-the-art in constructions. Overall the benefits to the broader readership of Applied Sciences from this paper are minimal. Some issues regarding the clarity and coherence of the manuscript are pointed out in the list of comments found below. Notwithstanding the novelty issues, a weak element of the paper is the presentation of the proposed methodology, which seems to be very general and theoretical. For example, the Authors fail to detail the mechanical rules governing the structural response (in deformation or stress terms) of the example structure to external loading.
Response 1: Thank you for your suggestion. The focus of this study is how to predict the mechanical parameters of the structure through the design parameters of the structure so as to realize the intelligent analysis of the safety state in the construction process. This provides a basis for structural safety control. In line 373-386, the analysis process of mechanical parameters driven by neural network is supplemented. In the case of self-weight, the number of hidden layer nodes in the neural network model for vertical displacement and cable stress is trained. The construction condition considered in the construction process of the structure is 0.9 * constant load + 1.5 * wind load. In the research process, two aspects of mechanical parameters data are formed. On the one hand, the design parameters of the structure are modified by twin model, and the mechanical parameters of the structure are obtained by arranging the construction conditions. The resulting data is regarded as the data set of neural net-work. In the neural network, the design parameters of the twin model are used as the input layer and the mechanical parameters are used as the output layer. Thus, the re-liable prediction model is obtained by training, and finally the mechanical parameters of the test set are generated to predict the safety of the structure. On the other hand, the design parameters of the test set are used to form a realistic construction structure, and the construction conditions are set at the site to obtain the corresponding mechanical parameters. Figs. 10 and 11 show the mechanical parameters corresponding to different design parameters in the specific test process. The design parameters are combined in Table 1. The final test found that the structural model studied was in a safe state under working conditions. It is important to accurately obtain the mechanical parameters of the structure under different design parameters in the prediction model.
Point 2: The paper has been written generally well, but in some parts carelessly; consequently, some passages were superfluous and challenging to read, while mention to some critical points was inadequate or completely missing. For instance, the introduction of the paper provides much information on prestressed steel cable structures, which could have been avoided. Also, in many parts reference is made to existing literature published in the Chinese language, which is not accessible to the broader readership of Applied Sciences.
Response 2: Thank you for your suggestion. In the introduction of the article, according to your proposal, the related forms of prestressed steel structure are deleted, and the analysis of structural safety and the introduction of intelligent algorithm are highlighted. The focus of this study is how to predict the mechanical parameters of the structure through the design parameters of the structure so as to realize the intelligent analysis of the safety state in the construction process. Additional explanations are provided in the introduction ( lines 73-75 ). In addition, I changed some references to relevant literature in international journals.
Point 3: Finally, the Authors fail to demonstrate how to integrate their mechanical models with the neural network that assesses the safety of a structure. As such the paper appears to be incomplete. The Reviewer advises the Authors to revise their study, elaborate more on the numerical models they use, and provide specific guidelines on how to efficiently use the tools they mentioned in their manuscript to simulate structural response and assess safety. In it's current state, the manuscript has little new information to offer, as evident by the Authors’ poor commentary in the Conclusion section.
Response 3: Thank you for your suggestion. In Section 2.2.1, the establishment of the digital twin model is supplemented, and Figure 3 is modified. The digital twin model mainly includes four levels : geometry, physics, behavior and rules. In order to improve the reference value of this research method, the application of digital twin model and neural network is supplemented in the case application section ( line 406-413 ). The construction conditions are set in the twin model to extract the training set and test set data required by the neural network. The structural safety prediction model is formed by BP neural network, and the mechanical parameters characterizing structural safety are obtained, so as to intelligently analyze the change of safety performance under construction conditions. This research method realizes the prediction of structural construction safety and provides a basis for structural safety control. At the same time, the conclusion of the article also discusses and summarizes the integration of the two applications.
Point 4: In the Abstract, the Authors could include a couple of sentences explaining the novelty of their study.
Response 4: Thank you for your suggestion. In the summary, the article combs, clear key innovation points. Firstly, the correlation mechanism between design parameters and mechanical parameters is established. The relationship between design parameters and mechanical parameters is captured by BP neural network. Driven by digital twinning, data samples are provided for the neural network. The structure safety intelligent prediction framework and theoretical method are formed by BP neural network and digital twin. The safety performance of the structure is analyzed according to the predicted mechanical parameters, and the basis for safety control is provided. Finally, the theoretical method is applied to the structural model to verify its feasibility.
Point 5: In the Introduction, the background provided in the first two paragraphs is not relevant to the work described in the manuscript. The Authors could discard repetitive sentences and organize previous studies in a clear manner.
Response 5: Thank you for your suggestion. In the introduction part, the main forms of prestressed steel structure are deleted. In the introduction part, the importance of structural safety assessment is analyzed, the current status of safety analysis is analyzed, and the intelligent prediction method is summarized. Also highlighted the focus of this study. In the construction process, how to predict the mechanical parameters of the structure through the design parameters of the structure so as to realize the intelligent analysis of the safety state. Next is the application analysis of intelligent algorithm in civil engineering. Finally, the architecture of this article is introduced.
Point 6: The Authors should provide more information on the numerical structural models they employed to assess the response of the structure to external loading.
Response 6: Thank you for your suggestion. In the case analysis section ( lines 373-386 ), the description of the structural safety performance prediction model is supplemented. In the case of self-weight, the number of hidden layer nodes in the neural network model for vertical displacement and cable stress is trained. The construction condition considered in the construction process of the structure is 0.9 * constant load + 1.5 * wind load. In the research process, two aspects of mechanical parameters data are formed. On the one hand, the design parameters of the structure are modified by twin model, and the mechanical parameters of the structure are obtained by arranging the construction conditions. The resulting data is regarded as the data set of neural net-work. In the neural network, the design parameters of the twin model are used as the input layer and the mechanical parameters are used as the output layer. Thus, the re-liable prediction model is obtained by training, and finally the mechanical parameters of the test set are generated to predict the safety of the structure. On the other hand, the design parameters of the test set are used to form a realistic construction structure, and the construction conditions are set at the site to obtain the corresponding mechanical parameters. Finally, the feasibility of this research method is fully verified by the comparison between Figs. 10 and 11.
Point 7: It is not clear to the Reviewer how the Authors combined digital twin models and neural networks to assess the safety of the structure. Please, elaborate on the details of this integration and how such a technique should be used by a reader interested in generating similar prediction models.
Response 7: Thank you for your suggestion. In the chapter of the fusion mechanism of BP neural network and digital twin, the establishment of digital twin model is supplemented, and Fig. 3 is modified. The digital twin model mainly includes four levels : geometry, physics, behavior and rules. In order to improve the reference value of this research method, the application of digital twin model and neural network is supplemented in the case application section ( line 406-413 ). The construction conditions are set in the twin model to extract the training set and test set data required by the neural network. The structural safety prediction model is formed by BP neural network, and the mechanical parameters characterizing structural safety are obtained, so as to intelligently analyze the change of safety performance under construction conditions. This research method realizes the prediction of structural construction safety and provides a basis for structural safety control.
Point 8: The utility of Figure 7 is questionable, and any information provided therein is hard to appreciate. Could you please elaborate more on the data presented in this figure?
Response 8: Thank you for your suggestion. The description of Figure 6 is supplemented in lines 355-364. In the process of neural network training, the fewer iterations, the higher the calculation efficiency. At the same time, the corresponding prediction accuracy is also higher. According to figure 6, for the prediction of vertical displacement and lower chord cable stress, the number of hidden layer nodes respectively takes 11 and 6 corresponding to the least number of iterations, and the accuracy of the model is the highest. Thus, the number of nodes corresponding to the neural network for different mechanical parameters in the prediction of structural safety performance is determined.
Finally, thank you again for your guidance. I hope I can finish an excellent paper with your guidance and help, and sincerely hope that my paper can be published in your journal.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
I thank the authors for accepting my suggestions.
I think the relationship between this paper and the journal’s special issue is clearer now.
I point out to the authors and the editorial staff to better organise the graphic layout of the paper, because it appears non-harmonious. The graphic hierarchy among the titles, texts, mathematical formulas, figure captions are not clear, in some parts of the paper.
Thank you.
Author Response
Thank you for your suggestion. There are relevant markers in the previous revision, which hinders your reading to express regret. This revision was revised according to the standard format of journal articles. The revised version chart is neat and meets the requirements of reading. Finally, thank you for your recognition of the article.
Reviewer 2 Report
The revised manuscript is clear and presented in a sufficiently well-structured manner. It includes a sufficient number of references, all current (mainly within the last five years) and relevant to the study. The figures and tables are appropriate as they properly show the data and are easy to interpret and understand. The data is interpreted appropriately and consistently throughout the manuscript. The conclusions are consistent with the evidence and arguments presented.
Author Response
Thank you very much for your approval of the article. It was at your suggestion that I finished writing the article. In the following research process, I will continue to study the subject of intelligent construction of structures.
