Coarse Return Prediction in a Cement Industry’s Closed Grinding Circuit System through a Fully Connected Deep Neural Network (FCDNN) Model

Round 1

Reviewer 1 Report

Authors say that "the contribution of this research lies first in the prediction of Coarse Return for process control in cement plants which is not yet delivered in literature". But, it is not enough argument regarding with the problem and approached proposed in their study. They should also clarify the contributions of the approached (FCDNN) and motivate why they select and apply this already existing deep learning technique for solving such an industrial problem. Do they see real practical benefits to apply deep learning based approach? Because they motivate, this is an important problem in cement industry to support operators in allowing control strategy and corrective actions. In this context, such an difficult deep learning based approach since it requires a lot of hyper parameters decision choices, tuning, training data, etc. how could it be practically efficient solution in real world implementation? This must really motivated very well by the Authors. 

Event-Modeller should be clarified more how that works. What is the benefit of using it? How these six parameters were ranked? Is it based on a statistical test for feature ranking or another type of analysis is applied to rank the features? 

Another point, cement industry has continuous manufacturing process, so the authors applied any data segmentation or any data aggregation method during the preprocessing of data? How do they handle this situation? It is unclear in the paper. As a pre-processing, they just mention removal of outliers based on domain knowledge thresholds. However, it should be also clarified since it is critical phase of an AI based application. 

Regarding with the results given in Table 3, 

what is the hypothesis there? What is the threshold value for satisfying the accuracy? The values on table 3 are the significance values (p-value) or t-test values? More clarification is necessary!

Regarding with the results given in Table 4, 

The values for both algorithms given in table 4 are average deviation? When the Authors compare the values with the proposed model why they did not run a statistical test like they did to compare (predicted and actual coarse return). How do you they say that their proposed model has a better performance in statistically? They should also conduct a statistical test to be able to say that their proposed FCDNN approach has better performance than the others. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

In this paper, a FCDNN model is proposed which predicts the Coarse Return. The topic is interesting and there are elements of scientific contribution, but through the paper it is vaguely expressed. Equally, there are obvious shortcomings in the structure of paper that must be revised before accepting it. The remarks are listed below:

1. No journal template was used, which is a basic prerequisite that must be met before submitting a paper for review.
2. The introduction is extensive but the structure does not provide clear answers to the questions that need to be answered in it. The introduction gives a much clearer theoretical basis of the problem, but the state of the art needs to be strengthened and more clearly highlighted. I suggest dividing the introduction into two sub-chapters: the first sub-chapter should clearly highlight the paper hypothesis and the aims and sub-goals of the paper (Research focus), and the second sub-chapter should give an overview of previous scientific research that has served as a basis for your research (Literature review). The contribution of your paper in relation to the existing literature should be emphasized.
3. Is Figure 1 taken from the existing literature? There is no reference, but it is quite unusual to find it in the part where you talk about the existing theory.
4. See in the journal template how the literature is cited in the text, especially when it refers to quoting several citations at once (in your text [8], [9] and [2] [7]). The way of quoting in the paper is inconsistent, but also inaccurate.
5. Is Figure 3 yours or taken from existing literature?
6. When quoting formulas, the text is followed by two dots, and the formula is followed by an explanation of symbols. Sentences need to be in lower case. Care must be taken to ensure that all taken formulas are referenced.
7. The font size in Figures 7 and 8 must not be larger than the font size in the text.
8. Separate the discussion from the conclusion. In the discussion, it is necessary to define how your research has provided advances in science compared to the existing literature. In the conclusion, it is necessary to state the concluding considerations and give an insight into future research. Are the hypotheses confirmed and are the objectives (goals) met?

Author Response

Please see the attachment. 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Thank you Authors for revising the manuscript by considering all comments and reflecting very well. 

Some minor language editing is necessary. Please check spelling mistakes.  

Author Response

Thank you for the reviewer for his/her comments and feedback, which improved the manuscript quality.  Three spelling mistakes have been detected and modified. 

Reviewer 2 Report

Most of the shortcomings have been corrected. Two more disadvantages:

1. Figure 1 needs to be referenced in the title of the figure not only in the text.

2. Insert Chapter 2 as a subchapter in the introductory part of the paper.

Author Response

Thanks to the reviewer for his/her valuable feedback. The raised concerns have been edited as the following:

1. Figure 1 is referenced in the title of the figure as requested. 
2. The 'Related Works' is moved to Introduction Chapter and under chapter 1 subsection. 

Thank you.