Application of Convolutional Neural Network for Fault Diagnosis of Bearing Scratch of an Induction Motor†
, Kenichi Yatsugi 1, Yukio Mizuno 1 and Hisahide Nakamura 2Round 1
Reviewer 1 Report
The paper relates the use of the Convolutional Neural Network for fault diagnosis on induction motors. This is a theme highly recurrent in the literature, and there are a lot of works on this topic.
The introduction presents a good review of related works. However, the authors did not get to differentiate the paper's contribution when compared with this analysis. Therefore, it is necessary to refine the paper's goals definition and their contribution to the research field.
Besides that, the whole method presentation describes a general CNN. Reading sections 2 and 3, there is no evidence of contribution when compared with those existent in the literature. Finally, the paper describes a general CNN solution.
Reading subsection 4.3, I understand that the CNN needs to be trained for each kind of scratch to be identified, and the method is not general. Is it true?
Besides this general analysis, I have some minor comments.
In the Abstract, it is said, "Practically, versatility is an important feature of fault diagnosis methods." This is not an obvious statement, and it deserves an explanation.
In line 157, the authors argued that they developed the equipment for measurement. Is this artifact software, or does it include software? Is the code available? This information is essential due to reproducibility aspects.
The acronyms C2S and C4S were not defined.
In subsection 4.1, it is necessary to detail the experimental conditions in each technique used to compare with the proposed approach to guarantee some validation in the analysis made.
Finally, the paper needs to refine the presentation of its goals and contribution. Besides that, the CNN presentation needs to detail fundamental aspects of its application in this problem.
Author Response
Dear Reviewer,
The authors thank the reviewer for reviewing the manuscript and giving valuable comments and suggestions for the betterment of the manuscript.
The manuscript is revised thoroughly to reflect your comments and suggestions. The modified portion of the manuscript is written in red characters. The authors also changed the title of the manuscript to precise the content.
Please find the reply as an attached file
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The proposed CNN-based diagnostic method can differentiate a scratched bearing from a healthy one by solving the overlapping of features.
- Please clarify the innovation of the proposed method.
- How to select the structure and parameters of the model?
- The format and font of the pictures in the manuscript should be consistent. (Figure 9, Figure 10)
Author Response
Dear Reviewer,
The authors thank the reviewer for reviewing the manuscript and giving valuable comments and suggestions for the betterment of the manuscript.
The manuscript is revised thoroughly to reflect your comments and suggestions. The modified portion of the manuscript is written in red characters. The authors also changed the title of the manuscript to precise the content.
Please find the reply in the attached word file.
Author Response File: Author Response.pdf
Reviewer 3 Report
Dear authors,
- The title of the manuscript is too long, I suggest changing it,
- Line 15, check and revise " have been progressively increasing",
- Line 16, delete "and" before transportation and put ",",
- Line 22, using a previously proposed method, which method you mean,
- Line 30, delete " failure [1]",
- Can you explain Lines 125-128, mathematically,
- What are the equations of "the weight matrix and weight parameters",
- Lines 169-173, are the values of ss length, width and depth, optimal values?,
- Line 197, can you explain the using of fast Fourier transform under various conditions , mathematically,
- Line 234, explain why differentiating the bearing conditions was difficult,
- Line 329, write the equation of the error performance and explain the way to decrease the error performance value,
- Line 368, check and revise the sentence " The results are summarized in Table 2, which shows".
Author Response
Dear Reviewer,
The authors thank the reviewer for reviewing the manuscript and giving valuable comments and suggestions for the betterment of the manuscript.
The manuscript is revised thoroughly to reflect your comments and suggestions. The modified portion of the manuscript is written in red characters. The authors also changed the title of the manuscript to precise the content.
Please find the reply in an attached word document.
Author Response File: Author Response.pdf
Reviewer 4 Report
I do not have serious problems with the paper content, only two aspects:
- Boosting ensembles is key today to make more sound, as it was defined and demonstrated by Smart optimization of a …Journal of Manufacturing Systems 48, 108-121 and you did not mention it.
- Conclusions are too short, please define better and give them as points.
- Future work: give more details, what faults are you aiming at? Check works in MDPI, such as https://doi.org/10.3390/ma14247746
Author Response
Dear Reviewer,
The authors thank the reviewer for reviewing the manuscript and giving valuable comments and suggestions for the betterment of the manuscript.
The manuscript is revised thoroughly to reflect your comments and suggestions. The modified portion of the manuscript is written in red characters. The authors also changed the title of the manuscript to precise the content.
The authors reply to comments one by one in the attached word document.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Recovering my general analysis of the original manuscript, I admit that the authors tried to detail the paper's goals by comparing their contributions with those in the related works. However, I understand that:
- "The present paper aims to comprehensively discuss the potential of a proposed convolutional neural network (CNN)-based method in bearing fault diagnosis." They did not make the discussion. Once there is no analysis of the conditions when this CNN is applicable and when it is preferable to the other methods presented.
- "To avoid the low detection percentage, and overlapping effects, both he hidden layer and pooling layers are optimized, and epoch is performed." The paper did not present this optimization process. How was it implemented? What techniques were used? It is just a general model without the application specificities when showing the CNN.
- "...the proposed method has an advantage of parameter optimization, tuning of hidden layer and auto-encoder is attached, which makes the system also possible for unsupervised learning algorithm." The paper did present any discussion or validation of this statement.
Besides that, the new abstract version is poor compared with the original one. The problem specificity was lost, and it is impossible to understand when the proposed method is applicable.
On the other hand, all the minor suggestions were solved in the new version of the manuscript.
Author Response
Dear Reviewer,
The authors thank the reviewer for reviewing the manuscript again and giving the valuable comments and suggestions for the betterment of the manuscript.
The manuscript is revised thoroughly to reflect your comments and suggestions. The modified portion of the manuscript is written in red characters.
The authors reply to comments one by one as shown below.
Author Response File: Author Response.pdf
