Fault Prediction Method Towards Rolling Element Bearing Based on Digital Twin and Deep Transfer Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript presents the following concerns:

• Abstract section: Please reduce the abstract to <300 words.
• Please add at the end of the Introduction section, 
• Lines 517-522: Authors should provide a clear justification for the chosen network hyperparameters. Although the model achieves an accuracy of 97%, the rationale behind selecting these specific parameters must be explained to ensure the results are both reproducible and scientifically grounded.
• The article lacks a detailed description of the data preprocessing procedure. This information should be included in the results section to provide a clear reference for researchers who may wish to replicate the study.
• Do the authors consider simulating any nonlinear conditions? Including such scenarios would help evaluate how the neural network responds under nonlinear behaviors, providing valuable insight into its robustness and generalization capability.
• I suggest the authors add a Discussion section to provide an interpretation of the results.
• Also, in the new Discussion section, I suggest adding a table comparing the main contributions of the work with those already reported in the prior art.

Author Response

Dear Reviewer and the Editor,

Thanks for your reviews and comments. We carefully studied them and made additions and corrections in the revised version. Below is the response letter, the modifications are highlighted in green in the new version.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper addresses a very interesting issue, with high potential of practical application - fault prediction at roller bearings. The tools used in the presented research are of high actuality - digital twins, deep transfer learning, multi-level construal neural network. The fault prediction method is appropriately described. The results obtained in the included case study are superior to the ones obtained by applying other known methods (based on CNN, MLCNN, DT-CNN). Two observations:

- The vibration frequency obviously depends on the analyzed system characteristics; did you try to explore what happens with your prediction method when the frequency of vibration occured in REB operation closes to resonance domain? Is it possible to encounter this problem?

- The feasibility of method application in practice should be, eventually, assessed in the case of a MLCNN data-driven model built by using real data - the simulation cannot completely & perfectly reproduce the physical world.

Author Response

Dear Reviewer and the Editor,

Thanks for your reviews and comments. We carefully studied them and made additions and corrections in the revised version. Below is the response letter, the modifications are highlighted in green in the new version.

Author Response File: Author Response.pdf