Fault Diagnosis Method for Rolling Bearings Based on Two-Channel CNN under Unbalanced Datasets

Round 1

Reviewer 1 Report

A two-channel convolutional neural network model was proposed to address the fault diagnosis with unbalanced data distribution problems. Fast Fourier transform is used to extract the frequency spectrum as the input of the one-dimensional convolution neural network and the generalized S-transform is used to extract the time-frequency image as the input of the two-dimensional convolution neural network. The authors provided the experimental results to show that the proposed method can accurately identify different fault types and better fault diagnosis performance than many deep learning methods on highly unbalanced datasets.

On my opinion, this paper can be accepted for publication.

 

Author Response

Thank you very much for your review of the manuscript.

Reviewer 2 Report

This work is based on the proposal of a two-channel convolutional neural network (TC-CNN) model to address the occurrence of bearing faults.

The proposal is interesting since it is focused to address one of the most common problems that affects rotating machine, however, there are some issues that must be addressed.

1. The manuscript is well structured and well-written, however, it must be avoided to write in first person. Please avoid to write sentence like "we propose"

2. It must be better if figure captions provide a more detailed description about what are showing, i.e. Fig. 4 "Flowchart of the proposed CNN structure used to diagnose...."

3. What is the main difference of this proposal in front of diagnosis approaches based on the used of deep-autoencoders. Please include a brief discussion in the introduction section and please consider the following works: https://doi.org/10.3390/s21175832 ; https://doi.org/10.1155/2021/9790053

4. The proposed method can be applied to the identification of other type of faults by using other type of signatures? Like stators currents and/or sound? If not, please include the general considerations that must be taken into account for obtaining high-performance results as it is reported.

5. The conclusions have to be improved, and please include future work and limitations of the proposal.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

After carefully reading manuscript,there are several queries which need to be address by authors and accordingly revised the manuscript.My comments are as follows :

1. Authors mentioned in pg.2, line 64-69 about the issue of unbalance data.This issues can be easily solved with GAN.In this situation there is no utility of using two channel CNN for fault diagnosis.Kindly address and justify in revised manuscript.

2. It is a well known fact that FFT is useful only for stationery signal but the bearing data are non-stationery in nature. What is the rationale to apply FFT in present study.

3. Authors should include method names in Table 3 for better understanding, and it should reflect in corresponding figures also.

4. In pg.12, authors proposed CWT + 2D CNN to extract time-frequency features. No where it is mentioned on what basis morlet 3 wavelet selected ?. Justification needed for wavelet selection.

5. There are certain recently published literature where different authors addressed the issue of data imbalance specifically in bearing fault diagnosis.It is recommended to include recently published literature in revised manuscript which are as follows :

a. https://journals.sagepub.com/doi/abs/10.1177/09544062211043132

b. https://www.mdpi.com/2076-3417/12/14/7346

c. https://www.mdpi.com/2075-4442/9/10/105

6. Training and testing gives biased results due to random splitting in training and testing of dataset.It is always recommended to apply k-fold cross validation on the dataset.

7. Kindly address the limitations of proposed methodology and future scope in revised manuscript.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

Authors responded the queries of authors well and accordingly modified manuscript.