Press Casting Quality Prediction and Analysis Based on Machine Learning
Round 1
Reviewer 1 Report
Summary
In traditional factory quality management, the sampling inspection is mostly used in the manufacturing process. When a lot of defectives are drawn out, many defectives are produced, which must be rejected, resulting in waste. If the defective is not drawn out in the course of sampling inspection, the defective would be delivered to the customer, lead- ing to return of goods and loss of business reputation. The present study proposed a quality prediction model for aluminum alloy press casting quality prediction. The manufacturing condition can be monitored instantly in the manu- facturing process. If defectives are produced continuously, the machine can be shut down instantly to avoid producing lots of defectives. The pressure data of press casting are col- lected in the press casting process. The Savitzky-Golay performs data smoothing and the data dimension is reduced by PCA. The Random forest, DNN and XgBoost are used for modeling. The model has satisfactory predictive ability. The result of this study can be introduced into actual production process for real-time quality monitoring, so as to enhance the production quality and productive efficiency effectively. The result of this study can be introduced into actual production process in the future, under the same environmental parameter setting. The method of this study is used in other products to implement real-time quality monitoring, so as to enhance the production quality and productive efficiency in the future, as well as to reduce the manpower and time costs and the rejection of defectives.
Comments: Paper reads good, however few changes are required before publication.
1) Include abbreviations as table.
2) Indicate captions of figures and tables clearly, such that they are self-contained.
3) Some figures are not in high quality, please redo them.
4) Some figures are not your own, such as neural network. Please cite from where you have taken.
Author Response
Dear Reviewer 1,
Thank you very much for your encouragement. We appreciate your constructive comments very much, and really hope our study can make a contribution to the literature.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
It is impossible to understand from the manuscript if there is any novelty. At the moment it seems that the authors have applied some algorithms into some data. The whole manuscript is about explaining how some supervised algorithms and evaluation methods work.
It's not clear what the novel algorithm is neither the evaluation and comparison with other methods. Some random algorithms are mentioned in a figure only. The paper is full of equations but it's not clear what is novel and what are the steps/relationships between those.
Author Response
Dear Reviewer 2,
Thank you very much for your suggestion. We appreciate your constructive comments very much, and really hope our study can make a contribution to the literature.
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors of submission electronics-1732835 present an interesting work on the quality prediction of press casting by machine learning. The material that they focused on is aluminium alloy. To perform analysis they used data provided by the casting machine, such as pressure value. In their workflow, the authors used several algorithms. They used Principal Component Analysis to reduce data dimension, to perform data smoothing they used the Savitzky-Golay filter. To train model the used several approaches, involving the use of deep neural networks, XGBoost and random forest. They performed a literature review, which is short and not very accurate given the size of the study. The problem description is presented very briefly. When using the algorithms of other researchers, such as deep neural networks, XGBoost and random forest, you should clearly indicate what is your unique contribution. A proper summary section is desirable. The results of the analysis are described in detail, but not properly commented on. The conclusions part does not provide any meaningful pieces of information. Presenting such a sophisticated method requires a closer look at the available solutions and at least a small comparative analysis with other approaches. The abstract is not a place for a long introduction, it should contain only necessary information. Overall, the paper is not well-structured, illegible and noncoherent. The method has the potential to be used in engineering practice. Several points need to be clarified. I would recommend a major revision.
General comments:
§ The abstract is too long and should be rewritten. Please pay special attention to your unique contribution and details that are essential for the comprehension of your method;
§ It would be best if you focused on emphasising what your unique contribution is, it is not clear;
§ The paper contains many stylistic and structural errors. You should try to avoid wordiness and limit the use of passive voice to improve readability. I would recommend proofreading;
§ Your paper lacks a proper summary. Consider adding this summary to chapter five. You can present it as "Summary and Conclusions";
§ You should extend the "Conclusions" part with a focus on your unique contribution. It is an essential part of the manuscript. The conclusions are too short in comparison with the size of the paper and do not provide any solid information apart from the statement that “The model has the satisfactory predictive ability”, which is debatable;
§ You should deliberate more on Figures 12, 15 and 17.
Additional questions:
· Could you describe the boundary conditions of your method?
· Could you deliberate more on the resilience of your method? Why did you decide to use these particular algorithms (DNN, XGBoost and RF)?
Author Response
Dear Reviewer 3,
Thank you very much for your encouragement. We appreciate your constructive comments very much, and really hope our study can make a contribution to the literature.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Authors have addressed all my comments and now the paper can be accepted.
Author Response
Dear Reviewer 1,
Thank you very much for your encouragement. We reviewed it again with minor revisions.
Author Response File: 
Author Response.doc
Reviewer 2 Report
Fine
Author Response
Dear Reviewer 2,
Thank you very much for your encouragement. We reviewed it again with minor revisions.
Author Response File: Author Response.doc
