Leveraging Variable Frequency Drive Data for Nondestructive Testing and Predictive Maintenance in Industrial Systems

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This technical note provides an interesting overview of the non destructive testing methods for the predictive maintenance. The paper is clear, well structured, and well organized. I am suggesting the below comments, which will improve the note further more, and will also help to readers. 

 

1. This technical note discussed the use of deep learning models, but did not discussed the use of traditional machine learning classifiers including supervised, unsupervised, or semi-supervised models. 
2. In most of the cases, the data collected through condition monitoring is the healthy data. The faulty data is very rare in reality. Can you please describe how this data balancing problem will be addressed especially in terms of non-destructive testing and predictive maintenance. 
3. As the Predictive maintenance is about the prediction of the remaining useful life of an equipment. In this technical note, the figures illustrates the condition monitoring. How the predictive maintenance is performed is also missed, as in Figure 5 ( The NDT Dashboard).
4. The PLC, VFD etc if discussed in terms of their computational power, such as which operations can be performed will also improve the quality of this technical note. For example, in terms of the predictive maintenance and condition monitoring, if warning and alarm are implemented at the VFD and PLC level, and if they are okay, then if did not stored what will be its effect. What if only the record or the rate of failure is stored from VFD or PLC for predictive maintenance. 

Author Response

Dear Reviewer 1,

Thank you for your thoughtful review and positive feedback on my manuscript titled "Leveraging Variable Frequency Drive Data for Nondestructive Testing and Predictive Maintenance." I am grateful for your kind words about the clarity, structure, and organization of the technical note, as well as your constructive suggestions for improvement. Below, I address each of your comments and outline the changes made in the revised manuscript to enhance its quality and utility for readers.

Comment 1: "This technical note discussed the use of deep learning models, but did not discussed the use of traditional machine learning classifiers including supervised, unsupervised, or semi-supervised models."

I appreciate your suggestion to include traditional machine learning (ML) classifiers alongside deep learning models. To address this, I have expanded Section 3 ("Data Transformation and Learning") by adding a paragraph (lines 210–225) that discusses the role of traditional ML classifiers in predictive maintenance and NDT. Specifically, I introduced examples such as supervised models (e.g., decision trees, support vector machines [SVMs]), unsupervised models (e.g., k-means clustering, Gaussian mixture models [GMMs]), and semi-supervised approaches, referencing Murphey et al. (2006) and Roulias et al. (2012). I explained how these methods complement deep learning by offering robust solutions for anomaly detection and fault classification, particularly when computational resources are limited or labeled data are scarce. This addition broadens the scope of the discussion and provides readers with a more comprehensive view of ML applications in VFD-based diagnostics.

Comment 2: "In most of the cases, the data collected through condition monitoring is the healthy data. The faulty data is very rare in reality. Can you please describe how this data balancing problem will be addressed especially in terms of non-destructive testing and predictive maintenance."

Thank you for highlighting the critical issue of data imbalance in condition monitoring. I have addressed this by revising Section 3 ("Data Transformation and Learning") to include a dedicated discussion (lines 230–245) on strategies to mitigate the scarcity of faulty data in NDT and predictive maintenance. I described techniques such as synthetic data generation (e.g., using generative adversarial networks [GANs]), oversampling of rare fault instances, and transfer learning from similar applications, drawing on Ali et al. (2020) and Roulias et al. (2012). I also noted how VFDs’ continuous data collection can be leveraged to build robust historical datasets over time, enhancing the training of ML models despite initial imbalances. This revision clarifies how the proposed approach tackles this practical challenge, making it more relevant to real-world industrial scenarios.

Comment 3: "As the Predictive maintenance is about the prediction of the remaining useful life of an equipment. In this technical note, the figures illustrates the condition monitoring. How the predictive maintenance is performed is also missed, as in Figure 5 (The NDT Dashboard)."

I agree that the connection between condition monitoring and predictive maintenance, particularly remaining useful life (RUL) prediction, needed further elaboration. To address this, I have enhanced Section 5 ("Implementation: Real-Time Monitoring with Python") by adding a paragraph (lines 370–385) that explains how the Python-based NDT dashboard can be extended to perform predictive maintenance through RUL estimation. I described the integration of time-series VFD data (e.g., current, torque trends) with ML models like regression-based estimation or deep learning for sequential failure prediction, referencing Ali et al. (2020). Additionally, I included an example RUL metric to add an option to add alongside condition monitoring parameters. I added Figures 6 and 7 to illustrate a sample Python code snippet and its output for RUL calculation, ensuring readers see the practical link between monitoring and prediction.

Comment 4: "The PLC, VFD etc if discussed in terms of their computational power, such as which operations can be performed will also improve the quality of this technical note. For example, in terms of the predictive maintenance and condition monitoring, if warning and alarm are implemented at the VFD and PLC level, and if they are okay, then if did not stored what will be its effect. What if only the record or the rate of failure is stored from VFD or PLC for predictive maintenance."

Your suggestion to discuss the computational capabilities of PLCs and VFDs and their implications for predictive maintenance is well-taken. I have revised Section 4 ("Accessing VFD Data via Industrial Communication Protocols") by adding a paragraph (lines 290–305) that explores the computational power of modern VFDs and PLCs. I explained that many VFDs (e.g., ABB ACS880) and PLCs can perform basic operations like threshold-based warnings and alarms locally, reducing latency in condition monitoring, as noted in Bansal & Dubey (2024). I addressed your specific scenarios by discussing: (1) the effect of not storing alarm data (e.g., loss of historical trends for ML training, limiting predictive accuracy), and (2) the implications of storing only failure rates (e.g., sufficient for basic trending but inadequate for detailed fault pattern recognition). I also noted that offloading complex computations (e.g., RUL estimation) to external systems like the Python setup in Section 5 complements the limited onboard processing of VFDs, enhancing scalability. This addition provides readers with a clearer understanding of hardware constraints and data management strategies.

General Improvements

• Figures and Captions: I refined the captions for Figures 5, 6, and 7 to reflect the new RUL focus and ensure consistency, improving their interpretive value.
• Clarity: I made minor edits throughout to enhance readability, such as simplifying technical terms in the expanded sections and ensuring abbreviations (e.g., "PLC," "RUL") are consistently defined.
• References: I updated the reference list to include additional citations (e.g., Murphey et al., 2006) supporting the new content on traditional ML classifiers.

I believe these revisions address your comments effectively, enhancing the technical note’s depth, practicality, and appeal to readers interested in NDT and predictive maintenance. Please let me know if further adjustments are needed. Thank you again for your valuable suggestions, which have significantly improved the manuscript.

Reviewer 2 Report

Comments and Suggestions for Authors

The introduction can be improved in terms of how NDT is used in industry and the practicality of this application. Is this application supposed to help with IoT companies?

Author Response

Dear Reviewer 2,

Thank you for your insightful feedback on my manuscript titled "Leveraging Variable Frequency Drive Data for Nondestructive Testing and Predictive Maintenance." I appreciate your suggestion to enhance the Introduction regarding the industrial use of NDT and the practicality of my proposed application, as well as your question about its relevance to IoT companies. Below, I address your comment and outline the changes made in the revised manuscript.

Comment: "The introduction can be improved in terms of how NDT is used in industry and the practicality of this application. Is this application supposed to help with IoT companies?"

I agree that the ntroduction could better highlight the industrial context of NDT and the practical benefits of leveraging VFD data, as well as clarify the application’s relevance to IoT frameworks. To address this, I have revised Section 1 ("Introduction") as follows:

1. NDT Use in Industry: I expanded the discussion on NDT’s industrial applications (lines 60–80) to provide a clearer picture of its role across key sectors. Specifically, I elaborated on examples such as aerospace (e.g., detecting fatigue in aircraft components), oil and gas (e.g., monitoring pipeline corrosion), manufacturing (e.g., weld inspections), and power generation (e.g., turbine assessments), supported by references to Gdoutos & Konsta-Gdoutos (2024) and Herlekar et al. (2024). I also added a sentence emphasizing how NDT ensures safety and reliability in these high-stakes environments, setting the stage for the proposed VFD-based enhancement.
2. Practicality of the Application: To underscore the practicality of using VFD data for NDT, I added a paragraph (lines 85–95) detailing how this approach leverages existing VFD infrastructure—widely deployed in industrial automation—to enable real-time, sensorless condition monitoring. I highlighted tangible benefits, such as reduced hardware costs, simplified system complexity, and continuous diagnostics, contrasting this with the periodic, resource-intensive nature of traditional NDT methods. This revision ties directly to the paper’s focus on scalability and cost-effectiveness, as demonstrated in Sections 2 and 5.
3. Relevance to IoT (IIoT) Companies: In response to your question about IoT applicability, I clarified the connection to IoT frameworks in the revised Introduction (lines 100–110). I noted that VFDs, as data-rich edge devices, align with Industrial Internet of Things (IIoT) paradigms by providing real-time operational data that can be integrated into cloud-based SCADA systems or enterprise-level predictive maintenance platforms, as discussed in Section 4. I explicitly stated that this application supports IoT companies by offering a scalable, interoperable solution for smart maintenance, referencing Moens et al. (2020) and the Python-based implementation in Section 5. Additionally, I added a sentence in Section 6 ("Conclusions") (lines 415–420) to reinforce how this approach enhances IIoT-enabled reliability management.

These revisions aim to provide a more comprehensive industrial context for NDT, emphasize the practical advantages of my VFD-based method, and confirm its utility for IoT companies seeking to advance predictive maintenance and condition monitoring. I believe these changes strengthen the Introduction and better align it with the manuscript’s objectives. Please let me know if further clarification or adjustments are needed.

Thank you again for your valuable feedback, which has significantly improved the manuscript.