Rolling Bearing Fault Diagnosis across Operating Conditions Based on Unsupervised Domain Adaptation

Round 1

Reviewer 1 Report

 

The authors provide a research method for addressing the challenge of inconsistent distribution of data from different working conditions in the mechanical equipment industry, specifically for rolling bearing fault diagnosis.             

1. The originality of the paper is poor; in fact, the approaches are well known, and many related works can be found in the literature. The motivation for the work is not clear. The authors should clarify their originality (innovation).
2. The authors must make a better effort at referencing the significant papers. There are several papers published dealing with the subject that are not cited. The relative works in the following references can be mentioned in the introduction:

§  https://doi.org/10.1007/s13369-023-08025-y

§  https://doi.org/10.3390/a15100347

§  https://doi.org/10.3390/pr11082440

1. There are many symbols and abbreviations. A list (Nomenclature) should be given.
2. in figure 2, presenting the structure of CNN. The vibration signal is directly fed to the network, where authors talk about extracted features using the FFT transform.
3. In Section 3, the authors talk about three steps, but in the details, they cite five steps?
4. In Section 4, the authors propose a convolutional auto-encoder to extract features, but they do not specify the architecture of the model?
5. Figures 7 and 8 have the same title.
6. As can be seen from Fig. 8, please reformulate.
7. in Figure 19. In the confusion matrix, the xlabel and ylabel are not displayed.
8. Using a public dataset is useful because we can compare our results with state-of-the-art results.
9. The conclusion should contain results and future research directions.
10. The manuscript is not very well organized. The readability needs to be improved.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Faculty diagnosis presented by the authors is a timely and well-developed idea offering new capabilities of the proposed model combining deep learning with domain adaption. The results presented by the study are exciting and might help further the condition monitoring's research frontier. I suggest the following improvements before accepting the manuscript for publication.

 

1. A native English must review the manuscript as it contains many grammatical and spelling mistakes. For example:
1. line 34 "... Some reports...". Be specific. 
2. similarly, line 38, "... Generally speaking fault diagnosis .... and data-driven approaches..." How could the author conclude that there are only two categories for fault diagnosis without presenting a comprehensive review or even referring to one?
3. line 69, "da-ta"
4. line 90, the second line of a new paragraph, says, "implementation of this class of methods." What class of methods?
5. line 65, "... Although these methods have achieved certain results,..." be more specific and state what results you are talking about.
6. line 465, "... unsupervised mi-grate..."
7. and more.
2. Figure 1 is ambiguous. Contrary to the figure, an autoencoder uses the same input as outputs. Also, Eq. 7 uses symbols to present models different from the figure. Similarly, recheck line 164, where the authors say the output of the decoder is X (with a cap ^).
3. Figure 2 does not represent data dimension transformation. Similarly, pictorially illustrate what happens to the data in each model layer compared with the preceding layer. Also, state how you optimize the model architecture and on what criteria. 
4. Recheck the mathematical and the machine learning model representation in depth to present models in an accurate unambiguous manner. 
5. Section 4.1. does not say anything about the reproducibility of results from the experimental platform.
6. Figure 5 illustrates random samples without explanation on why these particular signals were selected, and Figure 6, as well. Signals of each class should be discussed, and reproducibility of the signal. Error in the reproduced signals should also be discussed.
7. Figures 7 and 8 are not clear. Pixel density and the choice of color should be selected, keeping readability in mind.
8. The specificity matrix should also be presented in Figures 11 and 19.
9. How could a section be empty; see section 4.2.2.
10. The generalization of the model is also not discussed.

Extensive editing of the English language is required. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Thanks for the point-by-point responses. The majority of comments have been responded to. The manuscript quality is enhanced based on the revised version. However, comment 10 remains without a response. I suggest the authors conduct a comparative analysis and explain the advantages of the proposed method compared with the existing ones.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

It may be accepted for publication in the present form. 

Author Response

Thank you for your feedback and positive response. We greatly appreciate your consideration of our revisions. We are pleased to hear that the manuscript may be accepted for publication in its present form.

We have worked diligently to address the reviewers' comments and enhance the clarity, accuracy, and overall quality of the paper. If there are any final steps or actions required from our end before the manuscript moves forward, please let us know, and we will promptly fulfill them.

Once again, we express our gratitude for your support and consideration throughout the review process. We are excited about the possibility of contributing our work to your esteemed publication.

Round 3

Reviewer 1 Report

Thanks for the revision. I don't have any further comments. The paper can be accepted for publication.