Prediction of the Ultimate Impact Response of Concrete Strengthened with Polyurethane Grout as the Repair Material

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Numeous studies have already been published on the prediction of mechanical properties, so the authors need to highlight the novelty of their work.
Regarding the use of machine learning, the selection and combination of hyper-parameters significantly impact prediction performance. The authors should explain the rationale behind their hyper-parameter choices.
When selecting the LSTM, RF, and WNN models, did the authors consider other potential machine learning models, such as XGBoost (https://doi.org/10.1016/j.conbuildmat.2025.141319)?
In addition to the collected public datasets, did the authors conduct validation and testing on their own experimental dataset?
The authors are expected to explain the sources of prediction error.
How well does the proposed method generalize to unseen datasets? Was it tested on any new datasets?

Author Response

Please see the attachments 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The primary aim of this study is to develop and evaluate explainable deep learning models—particularly Long Short-Term Memory (LSTM), Random Forest (RF), and Wide Neural Network (WNN)—to accurately predict the ultimate strength capacity (Us) of normal concrete (NC) elements strengthened with polyurethane-based polymer grouting (PUG) under static and impact loads. The study seeks to identify the most influential parameters affecting structural performance and to establish a reliable and interpretable model for optimizing repair strategies and ensuring structural serviceability. The authors are required to comprehensively address the following suggestions:

 

1. The manuscript contains numerous grammatical errors. Please review and correct them meticulously.
2. The beginning of the abstract is too general and lacks precision. It would be more effective to start with a more specific and constructive statement to better capture the essence of the research and enhance the overall quality of the abstract. Additionally, conclude with the study’s significance. Keep it concise, informative, and structured within a few sentences.
3. Revise “With the incorporation of ML algorithms, it is feasible to add many parameters that 74 can affect the bond strength of NC substrate with repair materials into the prediction 75 model, facilitating a thorough and exact assessment of the ultimate strength capacity of 76 composite structure under impact loading condition” on page 2.
4. In the final paragraph of the introduction section, it is important to explicitly identify the research gaps based on the literature provided earlier. This will help highlight areas where further investigation is needed, drawing attention to the limitations or unanswered questions in existing studies. Clearly stating these gaps will lay the foundation for the rationale behind the current research. And also describe the research goal precisely. Additionally, clarify the content “The ML objective is to develop advanced deep learning models, including Long Short-Term 82 Memory, Random Forest, and a novel wide neural network, to predict ultimate strength 83 based on experimental data using feature engineering-based model combinations (M1, M2, and M3). “What is the meaning of M1, M2, and M3?
5. The manuscript contains excessive abbreviations. Only the necessary ones should be retained.
6. Ensure the accurate and consistent use of superscripts and subscripts throughout the manuscript. Any improper formatting or irregularities may indicate that the text was generated or modified using AI tools.
7. The introduction lacks sufficient detail. Authors can strengthen it by incorporating relevant references to provide more supporting information. https://doi.org/10.1016/j.rineng.2025.104542
8. The authors should incorporate these reference articles in the introduction, among other pertinent references, to enhance the literature. I anticipate that these references will improve the manuscript's quality.
9. “The dataset used for developing the XAI deep learning models is obtained from the experiments” (page 3). Please explain the XAI deep learning models.
10. Revise Fig. 4.
11. The use of SHAP for feature importance is a strong point; however, the discussion remains surface-level. For instance, while PUG thickness and peak load are highlighted as critical parameters, the interaction effects between input variables or the nonlinear relationships are not discussed. Could the authors extend the SHAP analysis to explore interaction effects or include SHAP dependence plots to better understand how the influence of one input parameter changes with the value of another? This would enhance the interpretability and robustness of the model insights.
12. Clearly describe the methodology of your study in the Methods section with appropriate references. Including a flowchart to illustrate the process will enhance the clarity and overall quality of the paper.
13. For the conclusions, it is so long; please shorten the sentences and summarize the main viewpoints to present the main novelties for the readers to be easy to understand.

Author Response

Please see the attachments 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

• In Table 2, the correlation result between P and λ is -0.73784. This value proves that there is a relatively high linear relationship specifically between these two characteristics. In the subsequent model training, the data of these two features are still adopted. Will this have an impact on the prediction accuracy of the model?

 

• The total sample size of the dataset is lacking in the text. Meanwhile, what is the selection basis for 70% of the original dataset in Figure 6?

 

• The article only provides a brief description of the data preprocessing part. The specific methods and steps of data preprocessing should be elaborated in detail to ensure the reliability of data quality.

 

• Section 4.1 mentions the use of grid search and Bayesian optimization to adjust the hyperparameters of the model, but does not provide the final values of the key hyperparameters. It is recommended to list the best parameter configuration in table form.

 

• There is a problem with the layout order of the pictures. It is suggested that the article be re-examined and modified.

 

• It is suggested that the fonts in Figure 8 and figure 9 be enlarged and the clarity be improved

 

• Section 4.3 points out that thickness (T) has the greatest influence on strength, but its negative correlation is not explained in combination with the mechanical mechanism of materials. It is suggested to supplement the physical explanation.

 

• The font sizes in Figure 11 are not uniform. In addition, it is suggested to add the meaning of the figures in figure 11.

Author Response

Please see the attachments 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I have no other comments.

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for your thorough effort and careful attention to addressing all the comments. Upon reviewing the revised manuscript, I recommend its acceptance.