Signal-to-Image: Rolling Bearing Fault Diagnosis Using ResNet Family Deep-Learning Models
Round 1
Reviewer 1 Report
In this paper, a rolling bearing fault diagnosis method based on vibration signal-image conversion and deep neural network model is proposed. The original vibration signal is converted into 2d image as the bearing state feature sample and the fault classification and recognition are carried out by the residual neural network model. The main problems existing in the paper are as follows :
1 ) Transforming the original one-dimensional vibration signal into a time-frequency image and inputting it into a classical neural network model with two-dimensional images as input samples is a common idea for fault diagnosis of rotating equipment, The paper does not make innovative work on signal processing methods or residual neural network structures.
2 ) In this paper, the continuous wavelet transform algorithm is used to transform the one-dimensional vibration signal into time-frequency image. The selection of wavelet basis function in this method has a great influence on the final result. The author does not specify the wavelet basis function and discuss its superiority over other wavelet basis functions.
3 ) This paper demonstrates the superiority of the SE-ResNet structure compared to the ResNet structure through the CWRU data set. However, from the results, the SE-ResNet structure has no significant advantages over other published deep neural network structures. It is hoped that the author will quote other neural network models of the same data set for comparative analysis.
The quality of English writing in the paper is fair.
Author Response
In this paper, a rolling bearing fault diagnosis method based on vibration signal-image conversion and deep neural network model is proposed. The original vibration signal is converted into 2d image as the bearing state feature sample and the fault classification and recognition are carried out by the residual neural network model. The main problems existing in the paper are as follows:
1 ) Transforming the original one-dimensional vibration signal into a time-frequency image and inputting it into a classical neural network model with two-dimensional images as input samples is a common idea for fault diagnosis of rotating equipment, The paper does not make innovative work on signal processing methods or residual neural network structures.
Response: We appreciate your good comment to improve our manuscript. We acknowledge that we have not changed the network model structure. However, we have explored the detection performance of the Resnet network family for rolling bearing defect signals through extensive experiments, and confirmed that the combination of SE module and Resnet152 can achieve the best detection effect with an accuracy of 96.42%. The contribution of this manuscript should be in discussing in detail the performance of the present network family in detecting rolling bearing defect signals, and finding a network model with high detection accuracy, laying the foundation for subsequent deep learning-based rolling bearing defect detection, which has certain engineering practicality.
2 ) In this paper, the continuous wavelet transform algorithm is used to transform the one-dimensional vibration signal into time-frequency image. The selection of wavelet basis function in this method has a great influence on the final result. The author does not specify the wavelet basis function and discuss its superiority over other wavelet basis functions.
Response: Thank you for your good comment. We have added a description and comparative analysis of the wavelet basis functions used, which can be found from line 117 to line 166 and from line 358 to line 378 in the modified version manuscript.
3 ) This paper demonstrates the superiority of the SE-ResNet structure compared to the ResNet structure through the CWRU data set. However, from the results, the SE-ResNet structure has no significant advantages over other published deep neural network structures. It is hoped that the author will quote other neural network models of the same data set for comparative analysis.
Response: Thank you for your excellent comment. We have added section 4.3 in the revised manuscript, which describes the comparative analysis of the proposed method with three existing deep learning models. The results indicate that the proposed method outperforms the other three deep learning models in terms of detection performance.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Recommendation: Major Changes
Manuscript Title: Signal-to-Image: Rolling bearing Fault diagnosis using ResNet family deep learning models
Detailed Comments: The authors proposed a 14 novel Signal-to-Image method, in which the raw signal data is transformed into 2D images using 15 Continuous Wavelet Transform (CWT).
(1) The presentation of this article is of good quality. However, there is still room for further improvement.
(2) Table 1 and Table 2 enrich the content of the article. So change the table of photos to the table format.
(3) There is an error in the format of formula 3. Please make a correction.
(4) The authors should pay attention to the explainable FD method such as Explainable intelligent fault diagnosis for nonlinear dynamic systems: From unsupervised to supervised learning. I do suggest the authors can discuss it in the introduction.
(5) Can this method be adapted to other industrial systems?
Recommendation: Major Changes
Manuscript Title: Signal-to-Image: Rolling bearing Fault diagnosis using ResNet family deep learning models
Detailed Comments: The authors proposed a 14 novel Signal-to-Image method, in which the raw signal data is transformed into 2D images using 15 Continuous Wavelet Transform (CWT).
(1) The presentation of this article is of good quality. However, there is still room for further improvement.
(2) Table 1 and Table 2 enrich the content of the article. So change the table of photos to the table format.
(3) There is an error in the format of formula 3. Please make a correction.
(4) The authors should pay attention to the explainable FD method such as Explainable intelligent fault diagnosis for nonlinear dynamic systems: From unsupervised to supervised learning. I do suggest the authors can discuss it in the introduction.
(5) Can this method be adapted to other industrial systems?
Author Response
Manuscript Title: Signal-to-Image: Rolling bearing Fault diagnosis using ResNet family deep learning models Detailed
Comments: The authors proposed a 14 novel Signal-to-Image method, in which the raw signal data is transformed into 2D images using 15 Continuous Wavelet Transform (CWT).
(1) The presentation of this article is of good quality. However, there is still room for further improvement.
Response: We appreciate your good comment to improve our manuscript. We have revised our manuscript according to the opinions of the reviewers. Thank you again for the comments provided by the reviewers to improve the quality of our manuscript.
(2) Table 1 and Table 2 enrich the content of the article. So change the table of photos to the table format.
Response: Thank you for your good comment. We have modified Tables 1 and 2, which can be seen in the revised manuscript.
(3) There is an error in the format of formula 3. Please make a correction.
Response: Thank you for such good comment. We have corrected formula 3.
(4) The authors should pay attention to the explainable fault diagnosis (FD) method such as Explainable intelligent fault diagnosis for nonlinear dynamic systems: From unsupervised to supervised learning. I do suggest the authors can discuss it in the introduction.
Response: Thank you for such good comment. We have revised the introduction according to your suggestions and added relevant literatures.
(5) Can this method be adapted to other industrial systems?
Response: We appreciate your excellent comment. We think that the method proposed in this article is suitable for detecting one-dimensional signals containing faults in industrial systems.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Dear Author,
In your revised manuscript, you have clearly articulated your research objectives and conclusions, and provided more detailed data analysis and experimental results. Additionally, you have appropriately addressed the feedback I provided and made necessary improvements. After careful review, I am pleased to accept your revised manuscript for publication.
Your English proficiency is satisfactory and meets the standards required for publication.
Reviewer 2 Report
The reviewer appreciates the revision from the authors. Now, my suggestion is accept.
N/A
