Prediction of Chloride Diffusion Coefficient in Concrete Based on Machine Learning and Virtual Sample Algorithm
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsOverall, the manuscript does not have serious flaws in methodology and analysis. The authors' research can potentially contribute to efficient predicting the chloride diffusion in concrete with various macro- and microfeatures. However, I believe there are a series of key issues that the authors should address before the manuscript is ready for publication.
· Major issues:
1. In the 3.1.1 section, I suggested the authors briefly summarized how the first group data were obtained. If the data were from experiments, please summarize the measure techniques. If the data were simulated, please introduce the simulation methods.
2. The authors expand the second group to 998 data through the GMM-VSG method. Yet, the authors cited only [28] for the data expansion via GMM-VSG. Please cite some other peer-reviewed works to justify the application of this strategy since the data were arguably “manufactured”.
3. In Section 3.1.2, are there any hidden variables such as temperature, humidity, etc. other than the 6 parameters which also affect the chloride diffusion coefficient?
· Minor issues:
1. The 2nd paragraph on Page 2 argues that “idealized and simplified mathematical-physical models cannot well describe the relationship between complex microstructure and macroscopic of concrete”. This is true, but I suggest the authors cite the following paper to support their argument:
a. Weiss, T., Mareš, J., Slavík, M., & Bruthans, J. (2020). A microdestructive method using dye-coated-probe to visualize capillary, diffusion and evaporation zones in porous materials. Science of The Total Environment, 704, 135339.
b. Gao, Y., Pastrana, A. P. C., Manogharan, G., & van Duin, A. C. (2021). Molecular dynamics study of melting, diffusion, and sintering of cementite chromia core–shell particles. Computational Materials Science, 199, 110721.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThis study investigates the impact of concrete microstructure on its macro parameters, specifically focusing on chloride diffusion, by employing machine learning (ML) techniques such as multi-layer perceptron (MLP) and support vector machine (SVM).
Though the study has significant implications and general interest to the readers of the journal, there are several critical flaws in both methodology and results as detailed below.
1. The selection of merely two ML techniques: MLP and SVM, lacks a clear rationale. Prior studies advocate for a spectrum of ML techniques to be evaluated, from basic to complex, to cultivate efficient and robust models. The justification for the exclusive focus on MLP and SVM is not sufficiently grounded in the text.
2. What is the novelty and significance of the study compared to similar works, such as:
Al Fuhaid, A. F., & Alanazi, H. (2022). Prediction of Chloride Diffusion Coefficient in Concrete Modified with Supplementary Cementitious Materials Using Machine Learning Algorithms. Materials, 16(3), 1277.
3. Please, include a graphical depiction of the database, including correlations among input features to significantly enhance comprehension of the data distribution and inter-variable relationships.
4. The discussion pertaining to 10-fold cross-validation is confounded and potentially misleading. It is crucial to clarify that 9 folds are for training and one for validation, not testing.
5. Though hyperparameter optimization is the backbone of any machine learning model, it is not properly conducted in this study, undermining the significance and validity of the results. This is one of the major critical flaws in the methodology of this study. Please, refer to Section 3.2 of doi.org/10.1016/j.istruc.2022.08.023 as well as Section 4.6 and Figure 3 of doi.org/10.1016/j.mtcomm.2022.104461 for further guidance.
6. The paper critically lacks a proper evaluation of the models on both the training and test datasets, which is essential for assessing the models' generalizability and reliability in real-world scenarios. It is imperative that the model validation process is expanded to include a rigorous assessment using both the training and unseen testing datasets to ensure the findings are not artifacts of overfitting. This omission is a fundamental flaw that must be addressed to substantiate the models' predictive capabilities and the overall validity of the research outcomes.
7. The results in the paper clearly demonstrated the inadequacy of the investigated models, as can be clearly observed from the results in Figure 6. Therefore, exploration of more advanced ML models and a comprehensive methodology for hyperparameter optimization is required.
8. The Conclusion section is too brief and does not provide adequate discussion. Significant revisions are required in the conclusions section considering the above comments.
9. The paper does not address the practical application of the developed models, such as the creation of a graphical user interface for wider use. How can others utilize the developed model?
10. Please, discuss the limitations of the current study and recommendations for future work.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe article tackles an important problem for cement-based construction materials, namely chloride diffusion. The authors correctly point out that the subject is related to many parameters, and finding the diffusion coefficient by means of machine-learning techniques would be a great time-saving route. The main subject, however, is represented by the machine-learning methods, and this point should be made clearer in the introduction.
While chloride diffusion seems to be somewhat a random choice of parameter to study, the authors do provide interesting insight into future possibilities.
It would have been a more substantial contribution if more parameters, other than the porosity (i.e. w/c ratio, strength class etc.), would have been taken into consideration.
The authors did show the difference between MLP and SVM, but, for future use, further explanations would be needed, as far as the difference in results is concerned and the deviation of the first two groups from the experiment.
References are appropriate, although more fundamental studies, related to the actual chloride diffusion, could have been cited and analyzed.
The subject is interesting and important. I hope the authors will go further and apply this type of modelling for other parameters of durability.
Comments on the Quality of English Language
I did not mark all the problems I found, but the introduction needs some corrections.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors have addressed my concerns successfully. No further revision is needed.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe comments of this reviewer have been addressed in the revised manuscript.
