Predicting and Evaluating Decoring Behavior of Inorganically Bound Sand Cores, Using XGBoost and Artificial Neural Networks
, Rui Li 1, Florian Ettemeyer 1, Melvin Mariadass 1, Philipp Lechner 2, Wolfram Volk 1,2 and Daniel Günther 1Round 1
Reviewer 1 Report
1. The concluding sentence of the abstract appears rather ambiguous. Please ensure that your findings are clearly stated, either numerically or conceptually, in comparison with previous studies.
2. Your manuscript has an excess of keywords. Please retain only those keywords that bear significant relevance to your study.
3. The introduction seems brief and lacks comprehensive literature review. A detailed analysis of existing literature is essential to establish the context of your study.
4. It appears that there is a substantial amount of content cited from reference [1]. Could you please clarify the main differences between your study and the one referred to in [1]?
5. Section 2 contains some material that would be better suited in Section 1, the introduction. Please consider revising and moving this content.
6. The "Materials and Methods" section is overly long. Please condense it to include only the most pertinent information. The description of the data should be relocated to the "Results" section.
7. The data presentation needs to be more engaging and impactful, as it currently feels overwhelming.
8. For figures such as 13 and 14, concentrate more on the y-axis. In other words, the region between 0 and 0.1, which doesn't contain any actual data, need not be displayed.
9. Citations are typically not included in the "Conclusion" section. It's best to stick to summarizing your findings and highlighting their significance here.
The use of English is acceptable, and no further comments are necessary at this time.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper discusses the production of complex casting parts that require high-strength sand cores, which can be conveniently removed after casting. The approach involves utilizing machine-learning techniques to analyze experimental data and identify the key factors influencing the flexural strength and decoring behavior of inorganic sand cores. By training robust models, the authors have successfully predicted the decoring behavior for various sand-binder systems. These findings are crucial for inorganically-bound sand cores in the complex casting parts sand cores field.
For these reasons, I recommend publication subject to some revisions, as detailed later in this review.
- What is the range of mean MAE achieved by the best models?
- Which features are considered important for predicting the target feature after complexity reduction?
- What is the implication of the signal data having no major influence on decoring behavior?
- Why were extended features like the strengths of the sand core after casting deemed not important by the models?
- How many different random initializations were used for each parameter set during calculation?
- How do the best runs of each parameter set compare to the corresponding mean over all ten runs in terms of RMSE?
- Are the models suited for the validation sand-binder system D? Why?
- How do XGB models compare to ANN models in terms of performance?
- What is the relationship between the number of neurons and the mean RMSE in ANN models?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The overall structure of the paper is appropriate. Some comments are provided to improve the quality of this manuscript. In my opinion, this manuscript should be reconsidered after major revision.
1) The abstract is too short. Please consider adding more details to it.
2) Introduction is too short. The authors should add more related works to this section.
3) The authors should add a subsection regarding description of XGBoost and ANN. You should add some of XGBoost outstanding features. You can refer to the following papers for a short explanation of XGBoost:
[1] “Energy Forecasting in a Public Building: A Benchmarking Analysis on Long Short-Term Memory (LSTM), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost) Networks,” Appl. Sci., vol. 12, no. 19, p. 9788, 2022.
[2] “Modeling industrial hydrocyclone operational variables by SHAP-CatBoost - A ‘conscious lab’ approach,” Powder Technol., vol. 420, p. 118416, 2023, doi: 10.1016/j.powtec.2023.118416.
[3] “Modeling operational cement rotary kiln variables with explainable artificial intelligence methods – a ‘conscious lab’ development,” Part. Sci. Technol., pp. 1–10, 2022, doi: 10.1080/02726351.2022.2135470.
4) The authors stated: "By using an interpretable machine-learning model like XGBoost (XGB)," but XGBoost is not an interpretable model.
5) Please add standard deviations to Table 4.
6) Section 2.6 is unnecessary, please consider removing it.
7) In Eq. (1), use mathematical symbol instead of "abs".
8) The authors should improve the resolution of Figs. 11-16. Please use different colors for various datasets.
9) Please provide more discussion on the results.
10) It is recommended to share a link to the source code to make the project reproducible.
11) No statistical tests were performed in the paper. So, how could we determine whether the results are statistically significant?
The manuscript should be proofread to ensure that there are no grammatical and typing errors.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The author's work has significantly improved compared to their previous submission, and the quality has now reached a level that merits consideration for publication.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
In general, the authors have reflected the comments of reviewers quite well and careful, but they did not address one of my comments.
1) Still no statistical test performed in this work. The authors should perform statistical tests such as Welch's t-test to show that the results are statistically significant. You can find the description of Welch's t-test and how to report the results using this test in the following paper:
“MFRFNN: Multi-Functional Recurrent Fuzzy Neural Network for Chaotic Time Series Prediction,” Neurocomputing, vol. 507, pp. 292–310, 2022, doi: 10.1016/j.neucom.2022.08.032.
It is highly recommended to perform Welch's t-test and report obtained p-values in the manuscript to show that your results are statistically significant.
No comment.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 3
Reviewer 3 Report
The authors have addressed all my comments.
