Fault Diagnosis Method of Planetary Gearbox Based on Compressed Sensing and Transfer Learning
Round 1
Reviewer 1 Report
Authors propose a system for diagnosing faults in planetary gearboxes. Two main components of this system are compression/reconstruction of vibration signals and the AlexNet trained, transfer learning based algorithm to detect faults from time-frequency domain vibration signals.
The content is well organized and presented. But my use of English has to improve. In many instances, there are in-complete sentences. Please Get it proofread by native english speaker.
Title should be updated to better represent the content. The phrase “utilizing sensing of compression” doesn’t make much sense.
Can you re-word or clarify the meaning of the sentence between lines 37-38?
Can you use some of the well cited references, such as https://ieeexplore.ieee.org/abstract/document/8962952 in related works?
Sentence(s) on lines 47-49 are in complete.
In the introduction you should elaborate a little bit more on why the compression of signal is critical for overall fault diagnosis.
Please provide reference for statements on lines 64-67.
On line 99, what is DP?
Section 2. Continue Wavelet -> Continuous Wavelet
Line 128: incomplete sentence.
Can you elaborate a bit more on the paragraph on lines 186-188? It is too abstract and hard to make sense of it.
Where is the polling layer in Figure 2?
Please provide reference(s) for statements on lines 222-225.
General comment about section 3.2: Please provide some intuition for choosing AlexNet as choice for transfer learning to use for detecting faults in gearbox signals.
What are N & M in equation 10?
Provide labels for blue and red graphs in figure 9.
Figure 11. It is too hard to see how something is distinct or similar just by looking at the graphs. Can you provide some intuition for your explanations here?
Table 5, 6 & 7: you have to compare the results with traditional methods of fault detection. Can you compare with https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6163465/ or something similar?
Author Response
Please see the attachment
Author Response File: 
Author Response.docx
Reviewer 2 Report
It has been a comprehensive study, and I think it has valuable content. However, the following significant corrections seem necessary to improve the scientific level of the article.
1- The caption of figures and tables should be more informative.
2- Why did the authors use CWT? For acoustics data, STFT represents spectrogram much better.
3- Please use “mag2db.m” function to convert magnitude to decibel. Then show all spectrograms based on decibel. They will show more information regarding acoustics data.
4- The logic of the introduction can be improved. For example, the reasons and significance of applying deep learning and machine learning methods to fault diagnosis or other signals could be introduced. Current progress and critical issues could also be mentioned. The authors can use these articles to edit this section. A hybrid deep-learning model for fault diagnosis of rolling bearings, Collaborative deep learning framework for fault diagnosis in distributed complex systems, Computer aided diagnosis system using deep convolutional neural networks for ADHD subtypes, Current-based bearing fault diagnosis using deep learning algorithms.
5- A reasonable justification should be made about why such algorithms are used. Why do authors think they are appropriate for such an application? What is their main advantage over other methods?
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Thanks to authors for comprehensively addressing all the comments on the previous version.
