MoBiSea: A Binary Search Algorithm for Product Clustering in Industry 4.0

Round 1

Reviewer 1 Report

This paper presents a Mode Binary Search algorithm specifically designed for low-cost automated industrial productivity data collection systems. The algorithm is introduced and verified through experiments and comparison with actual data. Although the paper provides detailed descriptions, there are still a lot of work that require improvement.

1. Binary Search is a common algorithm. There are no obvious innovations in this paper.

2. A large number of references are about k-means and x-means, yet there is a lack of relevant support for his own work, which undermines the credibility of this algorithm.

3. Although the authors provide the algorithm's pseudo-code, the whole paper lacks sufficient theoretical explanation.

4. The authors have pointed out numerous shortcomings of k-means and x-means in introduction. However, in terms of experiment, using these two methods as comparative methods is not convincing enough. The authors should compare with recently published improved algorithms to demonstrate their superiority.

5. The background of the paper involves Industry 4.0, Internet of Things, Electromagnetic Interferences and so on. However, I have not observed any substantial connection between these high-level backgrounds and the research content or experiments in the paper.

Only very minor corrections are needed

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

This manuscript presents a variation of the binary search algorithm that can monitor changes in production lines automatically. The authors did extensive literature review in related fields and described the proposed algorithm in detail. The method was validated using the plastic thermosealing line for cheese products, and the results show promising performance compared to the k-means and x-means baseline algorithms. I have three comments below:

1. In the section of “Related Studies”, the authors discussed a few applications of k-means and pointed out k-means does not perform well in experiments present in this manuscript. However, it seems that features may be the more important factor to impact the results rather than the k-means algorithm. For example, citation [11] states the performance difference of using k-means with and without sobel operators. Meanwhile, I think k-means and x-means do not fit into the application authors described since the former fits into a clustering problem but the latter is a search problem, which states in line 235-237 that the processed data/signals have the same order as they are generated. My suggestion is to either clarify the reason to choose k-means and x-means as baseline methods further or find a more suitable baseline method to be compared with.

2. In line 196-197, since K-RMS performs better than k-means, why not use K-RMS as the baseline algorithm?

3. In section 3.1., the authors choose the interval size equals to 50 as the best parameter. Since it is the minimum number in the tested range, I’m wondering if choosing a smaller parameter can result in better performance?

 

There are few comments:

1. Line 171 “however, and the results are 100% effective” is not clear. Does this “100% effective” refer to the proposed algorithm or citation [12]?

2. In line 279, the reference is not clear to me. Does the “initial one” means a parameter used in MoBiSea?

3. Table 4 is hard to read. I suggest to highlight the best performed values for each shift.

 

no

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Accept in present form