Lightweight Network Bearing Intelligent Fault Diagnosis Based on VMD-FK-ShuffleNetV2

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

I have gone through the manuscript. The study presents a robust and efficient method for rolling bearing fault diagnosis, leveraging the strengths of VMD, FK, and ShuffleNetV2. This approach not only addresses the challenges of noise and data complexity in bearing signals but also offers a practical solution for deploying advanced fault diagnosis models on resource-constrained edge devices. The experimental results validate the method's effectiveness, providing a promising direction for future research in intelligent fault diagnosis.

 

The proposed method integrates VMD for noise reduction, an improved FK algorithm for feature extraction, and the ShuffleNetV2 model for fault classification. The VMD algorithm decomposes the signal into intrinsic mode functions (IMFs), from which the optimal number of IMFs is determined based on energy differences and Pearson Correlation Coefficients. The FK algorithm then processes the denoised signal to generate two-dimensional feature maps for CNN input. The ShuffleNetV2 model is selected for its lightweight architecture, suitable for deployment on edge devices.

Overall paper is well written and well presented. I have some minor comments for improvements:

Similar studies on fault diagnostics have been presented by other authors too.  e.g., 

Petrauskiene V, Pal M, Cao M, Wang J, Ragulskis M. Color Recurrence Plots for Bearing Fault Diagnosis. Sensors (Basel). 2022 Nov 16;22(22):8870. doi: 10.3390/s22228870. PMID: 36433467; PMCID: PMC9693566.

I would suggest authors to look into additional references and improve the introduction part of the paper. 

some Figures e.g. Figure 6, Figure 15 and 17 could be improved. 

 

Thanks,

 

Author Response

Please refer to the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The aim of the study is presented clearly and precisely in the first sentence of the Abstract section.

The aim of the work is related to practical issues, which is well indicated in the second sentence of the introduction section.

The authors attempt to apply a sequence of selected data processing techniques to achieve diagnostic information. Here it should be noted that the data pattern should, in addition to mathematical relationships, be somehow related to the phenomena describing degradation, but at the beginning of the paper the authors do not mention anything about it.

Instead, an artificial map of features is created. Such a methodology can be characterized by the approach: if not with your own idea, then attack the issue with artificial intelligence.

With the help of such a methodology, you can basically prove anything. An unsolvable or extremely difficult problem here is the correlation of successive procedures of the model with experimental data. In principle, we get a finished product that we can experimentally test in its entirety. Partial analyses of subsequent degradation phenomena disappear somewhere.

A signal of an overloaded, healthy and damaged bearings would also be priceless here.

Simulator simulates only what engineer that constructed it know. And this is too low very often.

While in Figure 17 we see a comparison of true and predicted results. In my opinion, the true results come from the measurement of damaged elements - not from the simulator. Because if they come from a simulator, it is simply misleading readers.

Please include photos of the tested elements in your work or do not write about real values, only about simulated values that are consistent with the simulation results. But who needs such results? To write a publication?

 

Minor corrections:

I would like to see more precise - section of fig 6 called Type of bearing failure.

Please give a photo of destroyed bearings and results of their experiment investigations. If not, then this work is oriented too much to simulation to not show simulation in title and abstract.

 

Comments on the Quality of English Language

The aim of the paper is presented clearly and precisely in the first sentence of the abstract section, but this sentence is too long to be easy to read.

 Please correct:

 

Sequence:

“4 Experimental Verification

4.1 Experimental Setup

The fault simulation experiment utilized the Mechanical Fault Simulator produced by Spectra Quest Inc.”

Is experimental setup made on simulator basis? Then it is not experimental but simulation work. I do not like such topic presentation, it misleads me.

 "Liu, **ngchen; Zhou, Q.; Zhao, J.; Shen, H.; **ong, **aolei Fault Diagnosis of Rotating Machinery under Noisy Environment Conditions 657 Based on a 1-D Convolutional Autoencoder and 1-D Convolutional Neural Network. Sensors 2019, 19, 972, doi:10.3390/s19040972."

Author Response

Please refer to the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

In this paper, the authors claim a bearing intelligent fault diagnosis based on VMD-FK-ShuffleNetV2. Mainly, the idea can be interesting for the issues explained in the paper. However, the proposed approach corresponds to a combination of several blocks and it suffers from a lack of clarity, as the authors will see in my comments below.

 

 

Comments

- First of all, from page 3 to page 8, representing the heart of the paper, the authors give just the definitions and the introductions to used methods and algorithms.

- The proposed method (subsection 3.2) needs to be explained/justified more, theoretically and algorithmically. Indeed, in this subsection, the authors give just the flowchart of the proposed approach. It is not sufficient for a scientific paper and the proposed combination have to be justified theoretically (why, how, …)

- The efficiency of the proposed intelligent fault diagnosis drastically depends on the severity of the considered faults (Inner and outer race faults, ball fault, compound fault). The authors have to discuss the impact of this severity on the performances such that: detection delay, false alarm rate and missed detection.

- What is the limitation of the proposed approach?

 

Honestly, the paper must be rewritten clearly and deeply.

 

Comments on the Quality of English Language

Moderate editing of English language required

Author Response

Please refer to the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors took my comments into account correctly. Drawings have been improved, also the charts now have correct axis descriptions, graphical algorithms have been introduced. The paper is ready for publication after changing the designations of the steps of the algorithm (1) (2). so that they do not coincide with the numbering of the equations. This interferes with the reception of the text of the work - the area of lines 129-137.

Author Response

Please refer to the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

My questions have been very partially answered. The authors direct my questions towards perspectives of the work currently being done. Therefore, I consider that the answers given to my questions are not sufficient.

Comments on the Quality of English Language

Extensive editing of English language required

Author Response

Please see the attachment.

Author Response File: Author Response.docx