Towards Developing Big Data Analytics for Machining Decision-Making
Round 1
Reviewer 1 Report
The article discusses a framework consisting of 5 steps for developing big data analytics in manufacturing.
Comments:
- The stated advantage of the proposed framework is its transparency and explainability for a human. This is supported in the text by relying on a step-by-step statistical analysis of the big data instead of using machine learning models, which tend to be black-boxes for human users. However, both approaches require certain skills that potentially will make both complex to interpret for an average employee of a manufacturing company. Thus, the difference and added value of the proposed framework against the statistical analysis of big data or even data mining is not clear. Please elaborate and provide some concrete benefits of the proposed approach. Also, there is a nice article on the comparison of statistics versus machine learning which could be useful for supporting the added value of the proposed framework (https://doi.org/10.1038/nmeth.4642)
- Also, the introduction is rather long with little to add to the discussion on the framework which is expected ot be the main focus of the paper. Hence, it is suggested the authors to condense it, keeping only key information for the reader.
- How control variables are related to evaluation variables? Also what about variables not measured? Are they also handled or considered in the proposed framework?
- Regarding the sentence in line 68, it is not clear the meaning of big data inequality in this context, especially when considering the data come from one manufacturing company. Please elaborate.
- In line 78 digital manufacturing commons are mentioned. It is reffering to the open platform from MIT? Please explain and if yes include the corresponding reference.
- While the literature review is adequate, a comparative analysis of big data frameworks supporting analytics could be useful for the reader. Please consider extending the literature review.
- Abbreviations used in the document need to be referenced in full name once. Please check the document and update accordingly, i.e. BDA in lines 226, 228, 229 and CVs/EVs in lines 231, 232.
- Regarding the sentence in lines 241-243, please provide an example of underlying knowledge that could be of use and could be revealed by performing data analytics on CV-EV centric documents.
- In lines 264-265 the authors state "the system utilizes...and preference of a user". Why propabilistic and fuzzy logic reasoning were preferred? Also, while they both provide a way to express uncertainty they usually do it in a different way. Thus, please explain how they are used in the proposed framework.
- The steps of the framework presented in section 3 are rather generic with high level descriptions provided. Please elaborate more on the specific methods used for each step, the reasoning for their selection and the technologies that can be used for realizing them in practice.
- Looking at figure 4 it is not clear what is its added value to the reader, as a lot of aspects are not presented or discussed in the text. What is included in D1...Dn? What data? How are they collecetd? What is their volume? How is it integrated? Where are they integrated? What API is used? It there a database? a platform? Please update.
- Section 3 up to line 330 could be condensed for the sake of readability. The framework steps are presented in lines 319-326. These could eb included at the end of the introduction leaving Section 3 to focus on the framework methods. Please consider reworking section 1 and 3.
- Looking at section 3, while the aim and benefit of the proposed framework are clearly stated, yet the contribution of the proposed framework to the state of the art and innovation against it is not convincing. What is the current work proposing against the stat of the art? What is its innovation and contribution? Please clarify.
- Section 4 provides a prototype, but lacks technical information. What infrastructure was used? What databases? Was only java used? Why XML was selected and not for example JSON, especially considering JSON is better suited for cloud applications. What statistical methods were used and how were they implemented? Also, what was the processing time and response time of the prototype and what reasoner was used?
- In the case study the authors mention the term digital twin. What is a digital twin in the current study? How is it defined, implemented and used?
- Regarding the following text in lines 676-677 "Unfortunately there is no systematic approach to developing it", one could argue that statistical analysis is one such approach. Please check and clarify.
Author Response
Please refer to the attached file for the response.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The extensive paper sheds light on the topic of big data analysis. The literature overview is successful and contains almost 30 sources alone.
The centrally proposed approach consists of a human-centered BDA framework. Figure 5 represents a successful approach to presenting the challenges of data analysis. Figure 6 shows the framework graphically well. The approach of handling different data sources via a metafile seems sensible. However, the disadvantages of an Excel-based solution, especially for handling large amounts of data and subsequent automation and transmission, should be briefly discussed. This applies in particular to the processing of control data ("field level").
The large scope also poses a challenge in terms of access to the topic. Overall, the approach is to transfer input data in terms of machine settings (e.g. feed, cutting parameters, ...) into a data system and the correlations between these input variables and the results achieved are not new. The approach itself is very systematic. However, it involves a lot of manual effort and the tools developed are very pure. At least an outlook should be shown here as to what an industrialized solution could look like.
In any case, the claim of wanting to develop a BDA system and to compare it with deep learning approaches should be viewed critically. The framework is successful, but the central analysis core is a simple correlation analysis plus data visualization. Even if such an approach can make sense and has advantages - especially with regard to explainability - it also has clear limits. The metafile generation tool presented in Chapter 4.1 can under no circumstances be regarded as a "Big Data Preparation System". The "Data Visualization System" in Chapter 4.3 also represents a simplest prototypical solution for displaying correlations.
In summary, either significantly more must be invested in chapter 4 or chapter 4 can be omitted due to the large scope of the paper and focus on chapter 5. This is significantly better stocky. Overall, the claim to have created a general big data analysis tool should be critically questioned. There is also commercial software here - for example Cornerstone® - with a significantly larger scope of performance, which also consumes fewer (computational) resources and is precisely tailored for the manufacturing processes described.
Some comments:
- Figure 3 is referenced in the text later than it appears
- The later figures, e.g. 13-1-4-15-16 seem to represent screenshots where the text size of the figures is much larger than the figure descriptions
Author Response
Please refer to the attached file for the response.
Author Response File: Author Response.pdf
Reviewer 3 Report
The article deals with the current and important problem of using big data for decision support in manufacturing processes, especially machining of mechanical parts. Both the structure of the paper and the way in which the concept is presented and the solution to the problem do not raise any objections on my part. I believe that the article can be published in the proposed form.
Author Response
Please refer to the attached file for the response.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
No further comment.
I think it can be published as is.
Reviewer 2 Report
Will there are still limits in respect to the scope, the adjustments of the authors are fine and therefore the paper can be accepted.
