Bearing Health Monitoring Using Relief-F-Based Feature Relevance Analysis and HMM

Round 1

Reviewer 1 Report

The paper discusses an important subject. The advances in the data-acquiring process as well as the large extend of systems necessitates having efficient tools, such as the one proposed in this draft. The paper is well written and organized. My comments are as follows:

-While the abstract has aimed to provide an overall overview about the main contribution, there is a need to be revised in such a way that the general reader can grasp the main idea/topic of the draft as well as the main contribution.  

-Despite the fact that a good discussion about the superiority of the proposed framework in provided in terms of the numerical results, discussion about the complexity of the proposed framework and how it compares with the existing techniques are highly recommended. 

-Having a good schematic diagram in the draft would be really helpful. This definitely alleviates the difficulty of going to details of the techniques for the readers. 

- There have been a surge in the application of Machine Learning and Statistical framework to solve the similar problem focused in this paper. The authors are encouraged to include some of the recent papers in the introduction to give a good holistic overview about the existing techniques to general readers:

#  "Rolling element bearing fault diagnosis using convolutional neural network and vibration image." Cognitive Systems Research 53 (2019): 42-50.

# “Bayesian optimization objective-based experimental design”, In Proceedings of the 2020 American Control Conference (ACC 2020), IEEE. 2020.

# "Intelligent fault diagnosis of rotating machinery using improved multiscale dispersion entropy and mRMR feature selection." Knowledge-Based Systems 163 (2019): 450-471.

# "Bayesian optimization for efficient design of uncertain coupled multidisciplinary systems." In Proceedings of the 2020 American Control Conference (ACC 2020), IEEE. 2020.

# “Optimal Finite-Horizon Sensor Selection for Boolean Kalman Filter,” 51st Asilomar Conference on Signals, Systems, and Computers (pp. 1481-1485), Pacific Grove, CA, 2017.

- The format of some the references is not in standard form. These need to be checked and fixed.

-While the authors have emphasized the contribution of the paper throughout the draft, it would be nice to comment how this draft can be compared with the following paper:

#  "Rolling element bearing fault diagnosis using convolutional neural network and vibration image." Cognitive Systems Research 53 (2019): 42-50.

Author Response

We truly recognize all the valuable comments and suggestions offered by
the reviewers. After a careful review, we have adopted all the observations in
our revised manuscript “Bearing fault diagnosis from vibration signals based on
a stochastic feature selection approach”. Our point-to-point responses to the
comments of the reviewers are given in the attached pdf. 

Regards

Author Response File: Author Response.pdf

Reviewer 2 Report

"6205-2RSJEM SKY"

 SKY: ?  (SKF)

"1730 hp, 1750 hp, 1730 hp and 1797 hp"

hp: cannot be; (rpm)

also: 1730 1750 ???? 1797  

 

 

Author Response

We truly recognize all the valuable comments and suggestions offered by
the reviewers. After a careful review, we have adopted all the observations in
our revised manuscript “Bearing fault diagnosis from vibration signals based on
a stochastic feature selection approach”. Our point-to-point responses to the
comments of the reviewers are given in the attached pdf. 

Regards

Author Response File: Author Response.pdf

Reviewer 3 Report

The work concerns a stochastic feature selection approach in process of bearing fault diagnosis with the use vibration signals. The Authors compared obtained results with the works of other authors.

On time-frequency-based analyzes were obtained the best results of feature diversity. It is worth emphasizing that the Authors methodology can identify both the bearing state and the severity level of all the fault listed in the database.

In section 3.1 Database and Processing in rows 194-195, is mentioned about four operation revolution (1730, 1750, 1730 and 1797). Two of them are repeating, was it the Authors' intentional goal??

Figure 2 shows time and frequency waveforms for an undamaged bearing and three different bearing failures. In the case of bearing ball damage, the expected frequency amplitude should reach at least twice as much (0.1), even up to four times as large (0.2). In Autor case it is 0.05. Please Authors look carefully at these analyzes.

In the article, the Authors uses learning curve. Can you determine how looked like the set of learning signals to the set of test signals? How does it affect to the results of the analysis of changing the proportion of signals between these two sets?

Please provide a detailed summary of bearing damage information.

The literature review is done very well to 2016. Not much has been done since 2017 (four publications from 2017, two from 2018 and one from 2019) . Please extension of the review from the last 3 years.

Please analyze the work Zbigniew Kulesza from 2018 and 2019 from journal Nonlinear Dynamics and Shock and Vibrations in which the subject of shaft and turbine diagnostics is discussed based on vibrational signals,  function of correlation and spectral power density. Please also analyze the work of Paolo Pennacchi (2020) from the journal Mechanical Systems and Signal Processing regarding estimation of bearing health state, from 2019 from the Mechanisms and Machine Science journal and from 2018 from the Sensors journal.

Literature references no. 41, 58, 65 require supplementing - they have incomplete data.

Author Response

We truly recognize all the valuable comments and suggestions offered by
the reviewers. After a careful review, we have adopted all the observations in
our revised manuscript “Bearing fault diagnosis from vibration signals based on
a stochastic feature selection approach”. Our point-to-point responses to the
comments of the reviewers are given in the attached pdf. 

Regards

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The paper is well-revised and in my opinion is ready for publication. 

Author Response

Thank you for your comments.

Reviewer 3 Report

The Authors in the sentence wrote "In addition, emergencies of failure with three levels of severity (0.007", 0.014 "and 0.021") and four operational revolutions (1730, 1750 and 1797 [rpm]) ... ", why are only 3 in brackets ??

Currently, Figure 5.3 is less readable than in the previous version of the manuscript. Scaling of values on the vertical axis reduced the readability of the figure. The Authors did not answer the question: why in their graphs the signal amplitude is slightly higher in the case of undamaged (normal) to damaged (ball) bearings? In the case of inner and outer looks fine.

Please complete the literature after 2017 more attentively.

Other answers are satisfactory and explain the asked questions.

Author Response

We truly recognize all the valuable comments and suggestions offered by the reviewers. After a careful review, we have adopted all the observations in our revised manuscript Bearing fault diagnosis from vibration signals based on a stochastic feature selection approach. See attached pdf

Regards 

Author Response File: Author Response.pdf

Round 3

Reviewer 3 Report

The Authors referred to all comments in the reviews and made corrections in manuscript.
Please only to complete the literature on number 16, 17, 37, 38, 47 - it contain incomplete data, paper number, pages etc.

Author Response

Thank you for your comments. We fix the citations according to the available information in Google scholar and the corresponding journal (pdf and .bib updated in the reviewed version). Cites 37 and 38 do not contain volume information (in press version available).

Regards