Classification of Unbalanced and Bowed Rotors under Uncertainty Using Wavelet Time Scattering, LSTM, and SVM

Round 1

Reviewer 1 Report

Authors has modeled a faulty systems in ABAQUS/CAE for fault detection, tool like a long short-term memory (LSTM) network and support vector machine (SVM) model is been utilized. The manuscript is good and has practical significance. However, following question need to be addressed before further processing 

1) Abstract need to be modify so as to include main highlights and results obtained from the present work.

2) Although introduction and literature review are satisfactory, but justification for using various classification model need to be added here. Why only SVM and LSTM is been implemented over other classification AI methods? justify

3) While doing FE model, have you validated the results against experimental work? 

4) Even mesh convergence study need to be added or atleast mentioned in the text

5) Why and how rotational speed (50, 75, ..150 rad/s) is been selected? justify

6) which range of time-domain waves and frequency spectrum is targeted? why chosen this range here? justify

7) Is shaft parameters are mentioned in table 1 is only tested? or have u tried for several other combinations too? if not why?

8) Recommanded to re-write section "2.5 classification" with more clarity and justification with respective to the current work.

9) Accuracy of SVM mentioned on line no 470 is 100%. does it means that your model is overfitted?

10) Recommanded to re-write section "3 result and discussion" with more clarity and justification with respective to the current work. It is difficult to understand the relevance and comparision of SVM and LSTM made.

 

 

 

 

Moderate editing of English language

Author Response

in attached the reply to reviewer 

Author Response File: Author Response.docx

Reviewer 2 Report

Title: Classification of unbalanced and bowed rotors under uncer- 2 tainty using wavelet time scattering, LSTM, and SVM

Following are my observations for improvement of the manuscript.

a.       While dealing with the approaches related SVM and others, we require hyper parameter tuning methods to improve the efficacy of the method. Although I can see the various experimentation part done by authors in Table2. Whether, the parameters are optimal or not, it’s still a questionable aspect. Hence, my suggestion is to incorporate a few metaheuristic algorithms and show a comparison.

You can refer a useful article for this:

Saxena, A., Alshamrani, A. M., Alrasheedi, A. F., Alnowibet, K. A., & Mohamed, A. W. (2022). A Hybrid Approach Based on Principal Component Analysis for Power Quality Event Classification Using Support Vector Machines. Mathematics, 10(15), 2780.

b.      Signal processing technique is an important step, especially while dealing with the mechanical component design we require extra caution. Hence, a detailed investigation should be presented in the section 2.4 along with the clear choice why some specific technique has been chosen.

You can refer a useful article.

Vijayvargiya, A., Gupta, V., Kumar, R., Dey, N., & Tavares, J. M. R. (2021). A hybrid WD-EEMD sEMG feature extraction technique for lower limb activity recognition. IEEE Sensors Journal, 21(18), 20431-20439.

 

c.       A strong editorial check is required for possible grammatical mistakes.

d.      I fell discussion of results can be improved and while applying certain method give fruitful results shall be demonstrated.

Must be improved

Author Response

in attached the reply to reviewer 

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

 

In the current paper, combinations of WTS with LSTM and SVM are employed in the classification of unbalanced and bowed rotor systems.

I have a few observations that can help improve the quality of the paper.

1-     The structure of LSTM as the number of hidden layers, activation function, and optimization algorithm are missing. The detailed information about these is needed.

2-     In order to claim which model is appropriate and robust for this classification problem, the authors need to apply various types of examination of the collected dataset such as:

a- Sensitivities to the number of training samples

b- Parameter Tuning

c- Classification performance evaluation by Error rate, Cohen’kappa, F1-score, recall, precision, etc. Only accuracy was used as the evaluation measure of the models, but this is not a sufficient measure. In addition to accuracy, the F1-score, recall, and precision rates must be calculated and evaluated. You should depict all evaluation results in the table format.

Author Response

in attached the reply to reviewer 

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Accept in present form

Minor editing of English language required

Reviewer 2 Report

Suggestions and incorporated. 

Editorial check is still required. 

Reviewer 3 Report

Dear Authors,

 

You've made all the corrections I suggested. The paper is acceptable.