Visualized Stacked Denoising Auto-Encoder Model for Extracting and Evaluating the State Features of Rolling Bearings
Round 1
Reviewer 1 Report
This paper proposed a visualized stacked denoising auto-encoder for unsupervised extraction and quantitative evaluation of bearing state features. It is of interest. However, the authors should address the following points to further improve the quality.
1. The contributions of this paper should be highlighted at the end of the introduction.
2. Please highlight the advantages and disadvantages of your method. What’s the difference between the VSDAE and SDAE?
3. Why is the FFT used in the proposed VSDAE model?
4. Detailed comparison with prior works should be added. Please give reasoning of the results and a deeper explanation.
5. Literature review on the approaches of fault diagnosis is limited. More recently-published papers in this field should be discussed. The authors may be benefited by reviewing more papers such as 10.1016/j.ymssp.2022.109569.
6. Authors can perform a significant test to show the efficiency of the method.
7. The validation is a big concern in this work. How authors can avoid bias prediction?
Author Response
Thanks for your valuable comments. Please kindly find the attached response file.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In the paper the authors proposed a visualization of a stacked auto-encoder method. They applied the method to experimental data of artificially damaged bearings. The effects of each step of the proposed method are shown.
Here my comments that can improve the quality of the paper:
- Please add the data and code of the bearing
- Please add pictures of the damages for all the defects
- How to change the load in the bearing?
- How the number of nodes in each layer listed in table 1 and 2 has been selected? Is is possible to add a sensitivity analysis of the influence on such number of node on the results? Why a different number of nodes has been selected for time and frequency domains? Why 3 layers has been selected?
- It is better to briefly describe the effect of other values in table 1 and 2 on the results
- What is the effect of the load, speed and damage size in the clusters? Is it possible to classified them or shown their distributions?
Author Response
Thanks for your valuable comments. Please kindly find the attached response file.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The revision has addressed all my issues. The quality has improved a lot after revision. I recommend it for publication.
Author Response
Thanks a lot.
Reviewer 2 Report
The point about the load should be better clarified in the paper. The authors stated that the load is the the motor torque, that it doesn't correspond to a radial force on the bearing. The radial (ora axial) force change the vibration level. In that test rig, the radial force is given by the shaft weigth and a torque can produce a variation of the radial load if there is a misalignment in the coupling between the shaft and the motor.
Please consider this comments and review this point in the manuscript accordingly.
Author Response
Thanks for your valuable comments. The description about the load in the original manuscript is inappropriate. In many bearing failure experiments, radial loads are directly applied to the test bearing. On our experimental bench, the defective bearings are installed in the motor to simulate the actual application. Due to the closed structure of the motor, the radial load cannot be applied to the test bearings. Therefore, a magnetic clutch is used to generate the torque load, which is actually the load of the motor. Since there are two discs on the rotor, the change in torque indirectly affects the radial load of the bearing. Some explains have been added in the revised manuscript.
