AI-Based Health Monitoring for Class I Induction Motors in Data-Scarce Environments: From Synthetic Baseline Generation to Industrial Implementation

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for this interesting contribution.

The paper includes detailed research on fault detection and health surveillance of industrial induction motors and issues like data paucity and the cold-start issue. It presents a framework of AI-based health monitoring that uses experimentally determined correlations between motor power rating and vibration signals to produce synthetic baselines, which allow effective fault diagnosis without the prior operational data.

The approach consists of a combination of vibration analysis, electrical signal analysis, harmonic analysis, statistical threshold, and rule-based diagnostic algorithm to effectively identify and classify such faults as imbalance, misalignment, and electrical faults. The practical applicability, robustness, and accuracy of the model are validated in a real industrial setting, and a model can be used as a cost-effective tool to predict maintenance of small-to-medium enterprises.

Some suggestions are:

• Elaborate and clarify how the synthetic baseline generation process will be done to increase the likelihood of reproducibility and interpretation by readers who are not aware of the procedure.
• Provide additional details on the constraints that can be given in terms of data extent and data orientation direction, how they will affect the applicability of the findings.
• Key results of further validation or case studies on other types of motors and configurations would help in further supporting arguments that it is widely applicable in industry.
• With the suggested approach, it may be helpful to introduce a comparative analysis with the current existing models of health index or diagnostic framework to outline the strengths and possible trade-offs of the offered approach.
• Elaborate on the description of the rule-based diagnostic algorithm, especially the weighting and scoring procedures, to increase clarity and make them easier to adopt by practitioners.
• Talk about potential difficulties and remedies to actual-time application of the framework into the industrial settings, such as the computational costs and the location of sensors.
• Discuss the cold-start problem more directly, including information on how the synthetic data generation trades off accuracy with computational efficiency, perhaps with sensitivity analyses.
• Elaborate on future research trends, particularly how long-term monitoring and predictive maintenance should be integrated, to give a better roadmap to be followed when conducting future research.
• Enhance clarity of figures and tables through making all the visual elements in the figures and table to be self explanatory with detailed captions and labelling.
• Check and revise English language in areas where the technical descriptions might be thick and unclear to make them easier to read without loss of scientific integrity.

I hope these comments and suggestions may help improve the quality of the manuscript.

Kindest regards

Author Response

"Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

1. The regression model is built using motor power data from 0.55–5.5 kW, yet conclusions are extended to Class I motors up to 15 kW without experimental verification.

2. The exclusion of the 0.75 kW data point lacks a clearly defined statistical outlier criterion or robustness analysis.

3. Vibration measurements are limited to the vertical direction at the drive end, which may not be sufficient to reliably distinguish certain fault types such as misalignment and structural looseness.

4. The number of measurements for V-belt configurations is significantly lower than for other configurations, potentially affecting model generalization.

5. Fault classification performance is evaluated primarily using synthetically generated fault signals rather than experimentally induced real faults.

6. The industrial validation is limited to a single site and a short monitoring period, restricting conclusions about general applicability.

7. The use of the term “999th percentile” is statistically incorrect and should be replaced with “99.9th percentile.”

8. Equation (13) is ambiguously written as f(x) = p1(x) + p2 and should be explicitly expressed as a linear function (f(x) = p1x + p2).

9. Sections 3.7 and 3.8 repeat the titles “Fault Detection” and “Fault Isolation,” causing confusion in the methodological structure.

10. The diagnostic rules in Table 5 are not expressed in a consistent format, making it difficult to interpret or reproduce the decision logic.

11. The assumption that healthy motors always exhibit dominant 1× rotational frequency components is overly strong and may not hold across all operating conditions.

12. Key signal-processing parameters such as filter characteristics, FFT resolution, and windowing are not fully specified, limiting reproducibility.

13. The decision logic used to determine the final diagnostic result in the industrial case study is not clearly explained.

14. The linkage between synthesized baseline vibration signals and real-world operational variability is not sufficiently justified.

15. A typographical error appears in the phrase “999th percentile,” which is not a valid statistical term.

16. Redundant symbols appear in the phrase “18% to 25% %” and should be corrected.

17. The word “intest” should be corrected to “interest.”

18. Duplicate words such as “indicating indicating” should be removed.

19. Inconsistent spacing and formatting are present in author affiliations and keywords.

20. References to ISO 10816 and ISO 20816 standards should be unified and clearly contextualized.

21.  'Fault detection for point machines: A review, challenges, and perspectives’ also discussed the related problem.

Author Response

"Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

please put list of symbols as nomenclature

please explain Limitation and Innovation

please put ref for all equations

please compare with existing HI-based or autoencoder-based approaches

please add latest ref, post year 2020.

please  benchmarking against at least one baseline ML method

 

 

Author Response

"Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

No other concerns.

Reviewer 3 Report

Comments and Suggestions for Authors

Please accept it