Investigation of the Electrical Impedance Signal Behavior in Rolling Element Bearings as a New Approach for Damage Detection

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. The illumination is not certain enough and some statements are described vaguely. In the abstract, you should express the final results of the experiment, which could be presented more clearly with specific results. 2. What is the basis for the selection of features in Table 2? In addition, please describe the content of the selected features in the table. 3. How were the parameters selected for the spectral conversion of the data? 4. the pictures of the result displays, such as Figures 9, 10, 11 and Table 3, the font of the information on the axes is too small, please re-generate the pictures. 5. the article lacks comparison algorithm and comparison scheme. 6. in the conclusion, please describe the specific results of the experiment.

Author Response

Dear Reviewer,

thank you for your time to review our article. Your feedback helped us to improve our submission tremendously. In the following, we addressed your comments one by one.

1. The illumination is not certain enough and some statements are described vaguely. In the abstract, you should express the final results of the experiment, which could be presented more clearly with specific results.

The results have been clarified in the abstract, but because of the word limitation, a deep dive into the topic in the abstract is not possible. We extended the abstract as follows:

“In case of pure radial loads, explicit changes in the impedance signal are detectable, which indicate a pitting damage. Under combined loads, the signal changes are detectable as well, but not as significant as under radial load. Damage indicating signal changes occur later compared to pure radial loads, but nevertheless enable an early detection.”

 

2. What is the basis for the selection of features in Table 2? In addition, please describe the content of the selected features in the table.

We see your point and added these description:

“The features are common for investigating vibration signals [8,10,45]. Because these features are state of research in feature engineering for condition monitoring approaches, they are chosen to ensure a comparability to commonly used bearing observation techniques. The content of the features can be found in [8].”

We do not see an additional value in repeating the feature’s content from the main reference [8], which would enlarge the article and will reduce the readability. Therefore, we decided to name the reference where to find the content in detail, but not to include it.

3. How were the parameters selected for the spectral conversion of the data?

“Every 90 seconds, the impedance is measured for 1.5 seconds with a sampling rate of 1 MHz. (…) For the frequency domain transformation using a Fast Fourier Transformation, the measurements with a length of 1.5 seconds are operated by the NumPy FFT algorithm. From these transformed data, features are calculated without using machine learning or feature engineering algorithms.” This was added to the article.

4. the pictures of the result displays, such as Figures 9, 10, 11 and Table 3, the font of the information on the axes is too small, please re-generate the pictures.

The pictures have been revised.

5. the article lacks comparison algorithm and comparison scheme.

Comparison algorithms and schemes are not necessary, because the focus is not investigating a special algorithm or AI approach. We identified features which enable condition monitoring based on the electric impedance independently from algorithms at first. That has been clarified in the text.

6. in the conclusion, please describe the specific results of the experiment.

The conclusion has been revised with the specific results.

Thank you again for your feedback!

Kind regards,

the authors

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Manuscript ID: machines-3062676

Investigation of the electrical impedance signal behavior in rolling element bearings as a new approach for damage detection

 This article investigates the use of impedance-based condition monitoring for rolling bearings by examining impedance signals and their features across various types of rolling bearings under different load conditions. The features are analysed in the time, frequency, and order domains. Impedance measurements were conducted using an alternating current measurement bridge.

 Reviewer's Considerations

 The article's contribution shows that the comparison of impedance signal changes under pure radial and combined loads indicates that rolling bearing impedance is a reliable instrument for condition monitoring. The study concludes that rolling bearing impedance is effective for pitting damage detection, independent of bearing type and load angle.

Despite the reviewer's contribution, the abstract could benefit from a few improvements to enhance clarity and readability.

 Suggestions for improvements:

1) For better reader understanding, please specify the acronyms EHL (on page 2, line 78) and EDM (on page 2, line 81).

EHL: Elasto Hydrodynamic Lubrication

EDM: Electric Discharge Machining

2) Correct the text to the full stop after the references, e.g. page 4, line 128: …bearings operating life. [28,31–37]

bearings operating life [28,31–37].

Line 142;   … is also possible. [42]

3) References

 Please review the incomplete references, including 2, 3, 12, 13, 16, 18, 28, 33 and others as needed.

e.g 28: Tuomas, R.; Isaksson, O. Measurement of lubrication conditions in a rolling element bearing in a refrigerant environment.

 Tuomas, R. and Isaksson, O. (2009), "Measurement of lubrication conditions in a rolling element bearing in a refrigerant environment", Industrial Lubrication and Tribology, Vol. 61 No. 2, pp.

Based on these considerations, the reviewer recommends accepting after minor revisions.

 

 

Author Response

Dear Reviewer,

thank you for your time to review our article. Your feedback helped us to improve our submission tremendously. In the following, we addressed your comments one by one.

1. For better reader understanding, please specify the acronyms EHL (on page 2, line 78) and EDM (on page 2, line 81).

EHL: Elasto Hydrodynamic Lubrication

EDM: Electric Discharge Machining

 

Both acronyms have been specified.

 

2. Correct the text to the full stop after the references, e.g. page 4, line 128: …bearings operating life. [28,31–37]

bearings operating life [28,31–37].

Line 142;   … is also possible. [42]

 

We corrected it.

 

3. References

Please review the incomplete references, including 2, 3, 12, 13, 16, 18, 28, 33 and others as needed.

e.g 28: Tuomas, R.; Isaksson, O. Measurement of lubrication conditions in a rolling element bearing in a refrigerant environment.

Tuomas, R. and Isaksson, O. (2009), "Measurement of lubrication conditions in a rolling element bearing in a refrigerant environment", Industrial Lubrication and Tribology, Vol. 61 No. 2, pp.

We added the missing information.

Thank you again for your feedback!

Kind regards,

the authors

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Dear authors,

The proposed manuscript address to important topic of bearings fault diagnosis of rotating machinery. Authors use know method of impedance-based condition monitoring for rolling bearings. Main effort was put at data collection from test chamber of the bearing test rig. After careful reading of the submitted manuscript, areas have been identified that require the addition of knowledge or improvement of the content.

1.       “using different machine-learning approaches [8–11]” is not described well. It must be extended.

2.       Moreover, “using different machine-learning approaches [8–11]” references are not up to date (2021,2020, 2008,2016). Please refer to up to date publications e.g. 2024 https://doi.org/10.3390/en17091998

3.       EHL is not explained.

4.       Figures 9, 10, 11, 12, 13 and Table 3 are unreadable.

5.       What are the parameters of FFT (e.g. time-domain window length, overlap, window type).

6.       Authors have three research questions which doesn’t evaluate variable velocity. Please extend dataset for variable velocity.

7.       In discussion section directly address to each of three research questions from 1.2.

8.       Motivation and results are not clearly presented. It not clear if more important is new dataset or application of classification method based on time and frequency features. Ther is no confusion matrix of classification.

9.       Main effort should be addressed to dataset collection explanation and visualisation and analysis of each fault. Please give description of each fault with photograph and representational time and frequency analysis.

10.   Section “Data Availability Statement:” is not given. Are the data publicly available?

11.   Compare your dataset with other datasets e.g. CWRU https://engineering.case.edu/bearingdatacenter

In general, the study could provides valuable data of impedance-based condition monitoring for rolling bearings faults. However, to further enhance the scholarly impact, it is recommended to emphasize novel scientific contributions and provide comprehensive technical details.

 

Author Response

Dear Reviewer,

thank you for your time to review our article. Your feedback helped us to improve our submission tremendously. In the following, we addressed your comments one by one.

1. “using different machine-learning approaches [8–11]” is not described well. It must be extended.

Though, this is not the main topic of our research, we extended it as follows: “(…) like classification algorithms or algorithms for a remaining useful life prediction, e.g. convolutional neuronal networks.”

2. Moreover, “using different machine-learning approaches [8–11]” references are not up to date (2021,2020, 2008,2016). Please refer to up to date publications e.g. 2024 https://doi.org/10.3390/en17091998

We added two additional references. But we emphasize that this is not the topic of the article. 

3. EHL is not explained.

The explanation has been given in line 82.

4. Figures 9, 10, 11, 12, 13 and Table 3 are unreadable.

We revised the figures.

5. What are the parameters of FFT (e.g. time-domain window length, overlap, window type).

“Every 90 seconds, the impedance is measured for 1.5 seconds with a sampling rate of 1 MHz. (…) For the frequency domain transformation using a Fast Fourier Transformation, the measurements with a length of 1.5 seconds are operated by the NumPy FFT algorithm. From these transformed data, features are calculated without using machine learning or feature engineering algorithms.” This was added to the article.

6. Authors have three research questions which doesn’t evaluate variable velocity. Please extend dataset for variable velocity.

As mentioned in section 2 and the conclusion, the experiments are fulfilled under stationary condition. Therefore, changes in the velocity are not in the scope of this research. In further research, changing operational conditions will be investigated.

7. In discussion section directly address to each of three research questions from 1.2.

We revised this section and addressed the research questions more clearly.

8. Motivation and results are not clearly presented. It not clear if more important is new dataset or application of classification method based on time and frequency features. Ther is no confusion matrix of classification.

With respect, the motivation and results are presented clearly. We did not investigate e.g. feature engineering approaches or machine learning methods, which has been clarified in the article: “(…) features are calculated without using machine learning or feature engineering algorithms”. Before are investigating this kind of methods, we are researching the signal’s and feature’s behavior by analyzing their changes over the operational time of the bearings.

9. Main effort should be addressed to dataset collection explanation and visualisation and analysis of each fault. Please give description of each fault with photograph and representational time and frequency analysis.

We do not, because this data are still under investigation and not topic of this research.

10. Section “Data Availability Statement:” is not given. Are the data publicly available?

The data will be published at the end of the funding project with all connected data.

11. Compare your dataset with other datasets e.g. CWRU https://engineering.case.edu/bearingdatacenter

We will not do this. Again, we are not investigating vibration data or feature engineering approaches or machine learning methods. The CWRU dataset is well known to use, but has nothing to do with our research. We apologize for that.

Thank you again for your feedback!

Kind regards,

the authors

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

Improvements can be still made:

1) Please compare your approach with other approaches with used features (Tab2) in selected by your terms (e.g. sensitivity and repeatability).

2) How this approach (features) are better than other available approach in quantitative manner.

3) Line 218 Wavelet filter is not clearly presented - what wavelet, parameters and which wavelet algorithm was used?

4) Add Figures that presents complex impedance signal (four basis: real part, imaginary part, absolute value and phase angle), which were used to calculate frequency domain features.

5) Add Figures that presents complex impedance signal (four basis: real part, imaginary part, absolute value and phase angle) that depicts research questions:

a) How does the impedance signal change under combined axial and radial load? - Missing Figures of impedance signal change under combined axial and radial load.

b) How does the impedance signal behave for different bearing types in case of point contact? - Missing figures of impedance signal behave for different bearing types in case of point contact.

Author Response

Dear Reviewer,

thank you again for your feedback regarding our submission. In the following, we are going to address your comments.

Please compare your approach with other approaches with used features (Tab2) in selected by your terms (e.g. sensitivity and repeatability).

This is not possible, because we are using electrical impedance data and other approaches use vibration data. Vibration data from other systems are not comparable with data generated at our test rig. Therefore, we cannot address this feedback.

How this approach (features) are better than other available approach in quantitative manner.

From the aspect of signal changes, this approach displays abnormalities in the impedance signal and its features before changes in the vibration observation system of the test rig are detected.  In case of pure radial loads, changes up to 60 hours before the vibration observation stopped the tests could be detected. For combined loads, the impedance signal changes occur up to six hours earlier than the vibration changes. This has been clarified in the manuscript.

Line 218 Wavelet filter is not clearly presented - what wavelet, parameters and which wavelet algorithm was used?

We apologize for this mistake; we did not use a Wavelet filter. We used a low pass filter first order that filters frequencies twice as high as the carrier signal frequency of 20 kHz.

Add Figures that presents complex impedance signal (four basis: real part, imaginary part, absolute value and phase angle), which were used to calculate frequency domain features.

We do not see the additional value of this figure. The features behavior is displayed by the figures provided in the manuscript. A further figure would enlarge the manuscript inadequately with no additional value for the audience.

Add Figures that presents complex impedance signal (four basis: real part, imaginary part, absolute value and phase angle) that depicts research questions:

1. • How does the impedance signal change under combined axial and radial load? - Missing Figures of impedance signal change under combined axial and radial load.

These figures are provided in table 3 and figure 12.

1. • How does the impedance signal behave for different bearing types in case of point contact? - Missing figures of impedance signal behave for different bearing types in case of point contact.

Figure 12 displays the impedance behavior for bearing type 7205 and is already included.

Enclosed you will find a version with the changes marked in yellow.

Thank you again for your time!

Kind regards,

The Authors

Author Response File: Author Response.pdf

Round 3

Reviewer 3 Report

Comments and Suggestions for Authors

Dear authors,

Improvements can be still made:.

1.       “This is not possible, because we are using electrical impedance data and other approaches use vibration data. Vibration data from other systems are not comparable with data generated at our test rig. Therefore, we cannot address this feedback.”

Remark: Therefore describe only yours approach in term of sensitivity and repeatability. For readers it is interesting if this aspect of approach was investigated, if this aspect was not investigated justify it in manuscript so the reader will have broader view at approach sensitivity and repeatability.

2.       “We do not see the additional value of this figure. The features behavior is displayed by the figures provided in the manuscript. A further figure would enlarge the manuscript inadequately with no additional value for the audience.”. Remark: For reader it can be confusing that some data are missing. Especially, figures of raw data are missing from which many different features are calculated. In example Figure 9 depicts feature F2 which use in equation W1 which is F1 feature, both features are frequency-domain features, however full workflow is missing. I would expect to see raw time-domain data, then after Fourier transform the  frequency domain-data from which F2 was calculated. In line 225 is given “length of 1.5 seconds”, therefore representative raw data without fault and with fault can be depicted in time domain, and frequency domain. It doesn’t have to be all variants, just those which are the most representative.

3.       “(…) These figures are provided in table 3 and figure 12. (… )Figure 12 displays the impedance behaviour for bearing type 7205 and is already included.".  Remark: The Table 3 and Figure 12 depicts frequency domain features e.g. F2 of impedance signals. However, research question 2 and 3 are defined for impedance signals not for frequency domain features of impedance signals. The reader can be confused by reading research question 2 and 3, it will be expecting to see how behave the impedance signal, not only how behave the selected features calculated based on impedance signal. Add representative figures that presents complex impedance signal (four basis: real part, imaginary part, absolute value and phase angle) that depicts research question 2 and 3 in theirs present form of impedance signal change/behave (currently the change of selected frequency domain features are depicted).

 

 

Author Response

Dear reviewer,

thank you again for your feedback. In the following, you can find our answers:

Remark: Therefore describe only yours approach in term of sensitivity and repeatability. For readers it is interesting if this aspect of approach was investigated, if this aspect was not investigated justify it in manuscript so the reader will have broader view at approach sensitivity and repeatability.

We cannot understand this remark and will not include it. Sensitivity and repeatability might be interesting for other research topics linked to machine learning, but not for the investigations described in this article. 

Remark: For reader it can be confusing that some data are missing. Especially, figures of raw data are missing from which many different features are calculated. In example Figure 9 depicts feature F2 which use in equation W1 which is F1 feature, both features are frequency-domain features, however full workflow is missing. I would expect to see raw time-domain data, then after Fourier transform the frequency domain-data from which F2 was calculated. In line 225 is given “length of 1.5 seconds”, therefore representative raw data without fault and with fault can be depicted in time domain, and frequency domain. It doesn’t have to be all variants, just those which are the most representative.

We included further figures which show the raw impedance signal.

We do not show the Fourier transformed data, because we do not see an additional value. The data handling process is described in a sufficient way, so no information or data are missing. Adding unnecessary figures enlarges the article in an inadequate way, reduces the readability for the audience and infringes the author guidlines MDPI.

Remark: The Table 3 and Figure 12 depicts frequency domain features e.g. F2 of impedance signals. However, research question 2 and 3 are defined for impedance signals not for frequency domain features of impedance signals. The reader can be confused by reading research question 2 and 3, it will be expecting to see how behave the impedance signal, not only how behave the selected features calculated based on impedance signal. Add representative figures that presents complex impedance signal (four basis: real part, imaginary part, absolute value and phase angle) that depicts research question 2 and 3 in theirs present form of impedance signal change/behave (currently the change of selected frequency domain features are depicted).

The figures are added, if they are necessary. We clarified in research question 2 and 3 that we investigated the features and the raw signal.

Kind regards,

the authors.

Author Response File: Author Response.pdf