A Deployment Method for Motor Fault Diagnosis Application Based on Edge Intelligence
Abstract
:1. Introduction
- (1)
- A motor fault diagnosis system architecture tailored for edge intelligence environments, utilizing edge computing technology to achieve the real-time monitoring of servo motors, thereby enhancing diagnostic response speed and accuracy. This offers a novel motor fault modeling and management methodology in edge environments.
- (2)
- A lightweight model approach based on knowledge distillation, enabling the efficient deployment of the MSCNN-LSTM-Attention model on edge devices while balancing high precision with resource efficiency.
- (3)
- Integrating post-training quantization techniques to further refine the student model, significantly reducing computational resource consumption and storage requirements and ensuring the efficient inference of the quantized model on edge devices.
- (4)
- Empirical research conducted on edge devices that validates the diagnostic accuracy and real-time performance of the lightweight model under resource-constrained conditions, demonstrating the practical value of the approach in industrial field applications.
2. Related Work
2.1. Application of Deep Learning in Fault Diagnosis
2.2. Development of Edge Computing in Industrial Applications
2.3. Application of Knowledge Distillation and Model Quantization Techniques
3. CNC System Architecture and Optimization Methods Based on Edge Intelligence
3.1. Overall System Architecture
3.2. Information Model Structure of Edge-Intelligent CNC System
3.3. Lightweight Fault Diagnosis Algorithm Based on Knowledge Distillation and Model Quantization
4. Optimization and Deployment Methods of Deep Learning Models
4.1. Model Construction
Algorithm 1 Fault diagnosis model construction process |
Input: X: Input vibration signal matrix of size Output: : Predicted fault type probabilities Steps:
|
4.2. Knowledge Distillation and Student Model Construction
- 1.
- Teacher Model and Soft Label Generation
- 2.
- Student Model Design
- 3.
- Student Model Training
Algorithm 2 Student model knowledge distillation process |
Input: X: Training dataset Y: Hard labels T: Trained teacher model : Weight parameter for balancing loss Output: S: Trained student model Steps:
|
4.3. Model Quantization and Edge Deployment
- 1.
- Quantization Overview
- 2.
- TensorFlow Lite Quantization Implementation
- Model conversion: First, the trained student model is loaded and then converted using the TensorFlow Lite Converter. The model’s weight parameters are quantized to integer space:
- Quantization strategy: During model quantization, TensorFlow Lite provides default optimization options, primarily by reducing the bit width of floating-point numbers to lower the precision restoration:The objective is to ensure that the quantized model output closely approximates the original model, thus maintaining accuracy while reducing precision.
- Generate quantized model: After quantization, a TensorFlow Lite formatted quantized model is obtained, suitable for edge devices, significantly reducing memory usage and enhancing computational efficiency:By enabling default optimization during quantization, the inference accuracy of the model remains as close as possible to the original while reducing memory requirements. The quantized model can then be deployed to edge devices, leveraging accelerated computation for inference. Finally, the quantized model is saved as a .tflite file for deployment on edge devices.
- 3.
- Edge Deployment
Algorithm 3 Model quantization and edge deployment process |
Input: S: Trained student model Output: : Quantized model ready for edge deployment Steps:
|
5. Experimental Design and Validation
5.1. Construction of Experimental Platform
5.2. Experimental Procedure
5.2.1. Teacher Model Training
- 1.
- Data Preparation
- 2.
- Measurement and Data Collection
- 3.
- Model Construction
- 4.
- Model Hyperparameter Settings
- 5.
- Model Training and Configuration
- 6.
- Performance Evaluation
5.2.2. Knowledge Distillation and Student Model Training
- 1.
- Model Construction and Knowledge Distillation Process
- 2.
- Hyperparameter Settings
- 3.
- Model Training and Configuration
- 4.
- Performance Evaluation and Comparative Analysis
- 5.
- Experimental Conclusion
5.2.3. Quantization and Edge Deployment Analysis of the Student Model
- 1.
- Model Quantization and Inference Time Comparison
- 2.
- Memory Usage Comparison
- 3.
- CPU Utilization Comparison
- 4.
- Advantages of the Quantized Model and Comprehensive Evaluation
5.2.4. Robustness Analysis and Visualization Results Under Noisy Conditions
- 1.
- Accuracy Comparison Analysis
- 2.
- Confusion Matrix Analysis
- 3.
- Noise Experiment Evaluation
5.3. Deployment and Implementation of Fault Diagnosis Application
- 1.
- Application Design and Implementation
- 2.
- Edge Device Deployment and Application
- 3.
- Advantages and Practicality of the Application System
5.4. Summary of Algorithms and Comprehensive Evaluation
6. Summary and Outlook
6.1. Conclusions
6.2. Limitations and Future Work
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Conflicts of Interest
References
- Lai, Y.H.; Zhang, Y.C.; Fang, L.; Wang, C.S.; Perng, J.W. Fault diagnosis of motor based on low cost edge computing technology. In Proceedings of the IEEE 2020 7th International Conference on Information Science and Control Engineering (ICISCE), Changsha, China, 18–20 December 2020; pp. 635–638. [Google Scholar]
- Huo, J.; Ji, H.; Yang, P. Research on sensorless control system of permanent magnet synchronous motor for CNC machine tool. In Proceedings of the IEEE 2021 40th Chinese Control Conference (CCC), Shanghai, China, 26–28 July 2021; pp. 1592–1595. [Google Scholar]
- An, K.; Lu, J.; Zhu, Q.; Wang, X.; De Silva, C.W.; Xia, M.; Lu, S. Edge solution for real-time motor fault diagnosis based on efficient convolutional neural network. IEEE Trans. Instrum. Meas. 2023, 72, 3516912. [Google Scholar] [CrossRef]
- He, C.; Han, P.; Lu, J.; Wang, X.; Song, J.; Li, Z.; Lu, S. Real-time fault diagnosis of motor bearing via improved cyclostationary analysis implemented onto edge computing system. IEEE Trans. Instrum. Meas. 2023, 72, 3524011. [Google Scholar] [CrossRef]
- Lu, J.; An, K.; Wang, X.; Song, J.; Xie, F.; Lu, S. Compressed Channel-Based Edge Computing for Online Motor Fault Diagnosis with Privacy Protection. IEEE Trans. Instrum. Meas. 2023, 72, 6505112. [Google Scholar] [CrossRef]
- Li, H.; Hu, G.; Li, J.; Zhou, M. Intelligent fault diagnosis for large-scale rotating machines using binarized deep neural networks and random forests. IEEE Trans. Autom. Sci. Eng. 2021, 19, 1109–1119. [Google Scholar] [CrossRef]
- Zhang, H.; Ge, B.; Han, B. Real-time motor fault diagnosis based on tcn and attention. Machines 2022, 10, 249. [Google Scholar] [CrossRef]
- Hakim, M.; Omran, A.A.B.; Inayat-Hussain, J.I.; Ahmed, A.N.; Abdellatef, H.; Abdellatif, A.; Gheni, H.M. Bearing fault diagnosis using lightweight and robust one-dimensional convolution neural network in the frequency domain. Sensors 2022, 22, 5793. [Google Scholar] [CrossRef]
- Wang, Q.; Jin, G.; Li, Q.; Wang, K.; Yang, Z.; Wang, H. A Fast and Energy-Saving Neural Network Inference Method for Fault Diagnosis of Industrial Equipment Based on Edge-End Collaboration. In Proceedings of the 2021 IEEE 11th Annual International Conference on CYBER Technology in Automation, Control, and Intelligent Systems (CYBER), Jiaxing, China, 27–31 July 2021; pp. 67–72. [Google Scholar]
- Wang, H.; Li, Y.F. Robust mechanical fault diagnosis with noisy label based on multistage true label distribution learning. IEEE Trans. Reliab. 2022, 72, 975–988. [Google Scholar] [CrossRef]
- Huang, T.; Fu, S.; Feng, H.; Kuang, J. Bearing fault diagnosis based on shallow multi-scale convolutional neural network with attention. Energies 2019, 12, 3937. [Google Scholar] [CrossRef]
- Lu, Z.; Liang, L.; Zhu, J.; Zou, W.; Mao, L. Rotating Machinery Fault Diagnosis Under Multiple Working Conditions via A Time Series Transformer Enhanced by Convolutional Neural Network. IEEE Trans. Instrum. Meas. 2023, 72, 3533611. [Google Scholar] [CrossRef]
- Chen, H.; Cen, J.; Yang, Z.; Si, W.; Cheng, H. Fault diagnosis of the dynamic chemical process based on the optimized CNN-LSTM network. ACS Omega 2022, 7, 34389–34400. [Google Scholar] [CrossRef] [PubMed]
- Zhang, H.; Xu, B. Chemical Process Fault Diagnosis Method Based on Deep Learning. J. Phys. Conf. Ser. 2023, 2637, 012018. [Google Scholar] [CrossRef]
- Fu, L.; Yan, K.; Zhang, Y.; Chen, R.; Ma, Z.; Xu, F.; Zhu, T. EdgeCog: A real-time bearing fault diagnosis system based on lightweight edge computing. IEEE Trans. Instrum. Meas. 2023, 72, 2521711. [Google Scholar] [CrossRef]
- Qian, G.; Lu, S.; Pan, D.; Tang, H.; Liu, Y.; Wang, Q. Edge computing: A promising framework for real-time fault diagnosis and dynamic control of rotating machines using multi-sensor data. IEEE Sens. J. 2019, 19, 4211–4220. [Google Scholar] [CrossRef]
- Ding, A.; Qin, Y.; Wang, B.; Jia, L.; Cheng, X. Lightweight multiscale convolutional networks with adaptive pruning for intelligent fault diagnosis of train bogie bearings in edge computing scenarios. IEEE Trans. Instrum. Meas. 2022, 72, 3502813. [Google Scholar] [CrossRef]
- Yang, C.; Lai, Z.; Wang, Y.; Lan, S.; Wang, L.; Zhu, L. A Novel Bearing Fault Diagnosis Method based on Stacked Autoencoder and End-edge Collaboration. In Proceedings of the IEEE 2023 26th International Conference on Computer Supported Cooperative Work in Design (CSCWD), Rio de Janeiro, Brazil, 24–26 May 2023; pp. 393–398. [Google Scholar]
- Zhang, Y.; Pang, C.; Yang, G. A real-time computation task reconfiguration mechanism for industrial edge computing. In Proceedings of the IECON 2020—The 46th Annual Conference of the IEEE Industrial Electronics Society, Singapore, 18–21 October 2020; pp. 3799–3804. [Google Scholar]
- Huang, Q.; Han, Y.; Zhang, X.; Sheng, J.; Zhang, Y.; Xie, H. FFKD-CGhostNet: A novel lightweight network for fault diagnosis in edge computing scenarios. IEEE Trans. Instrum. Meas. 2023, 72, 3536410. [Google Scholar] [CrossRef]
- Hinton, G. Distilling the Knowledge in a Neural Network. arXiv 2015, arXiv:1503.02531. [Google Scholar]
- He, F.; Liu, C.; Wang, M.; Yang, E.; Liu, X. Network lightweight method based on knowledge distillation is applied to RV reducer fault diagnosis. Meas. Sci. Technol. 2023, 34, 095110. [Google Scholar] [CrossRef]
- Zhong, H.; Yu, S.; Trinh, H.; Yuan, R.; Lv, Y.; Wang, Y. A lightweight GAN-based fault diagnosis method based on knowledge distillation and deep transfer learning. Meas. Sci. Technol. 2023, 35, 036103. [Google Scholar] [CrossRef]
- Xu, G.; Liu, M.; Jiang, Z.; Shen, W.; Huang, C. Online fault diagnosis method based on transfer convolutional neural networks. IEEE Trans. Instrum. Meas. 2019, 69, 509–520. [Google Scholar] [CrossRef]
- Zhang, Z.; Wang, N.; Wu, H.; Tang, C.; Li, R. MR-DRO: A fast and efficient task offloading algorithm in heterogeneous edge/cloud computing environments. IEEE Internet Things J. 2021, 10, 3165–3178. [Google Scholar] [CrossRef]
- Zhao, Z.; Barijough, K.M.; Gerstlauer, A. Deepthings: Distributed adaptive deep learning inference on resource-constrained iot edge clusters. IEEE Trans. Comput.-Aided Des. Integr. Circuits Syst. 2018, 37, 2348–2359. [Google Scholar] [CrossRef]
- Tang, S.; Chen, L.; He, K.; Xia, J.; Fan, L.; Nallanathan, A. Computational intelligence and deep learning for next-generation edge-enabled industrial IoT. IEEE Trans. Netw. Sci. Eng. 2022, 10, 2881–2893. [Google Scholar] [CrossRef]
- Wang, Y.; Wang, Y.; Cai, J.; Lee, T.K.; Miao, C.; Wang, Z.J. SSD-KD: A self-supervised diverse knowledge distillation method for lightweight skin lesion classification using dermoscopic images. Med. Image Anal. 2023, 84, 102693. [Google Scholar] [CrossRef] [PubMed]
- Yang, Y.; Diao, W.; Rong, X.; Li, X.; Sun, X. Dynamic interactive learning for lightweight detectors in remote sensing imagery. IEEE Trans. Geosci. Remote Sens. 2022, 60, 5632514. [Google Scholar] [CrossRef]
- Zhang, J.; Lei, J.; Xie, W.; Li, Y.; Yang, G.; Jia, X. Guided hybrid quantization for object detection in remote sensing imagery via one-to-one self-teaching. IEEE Trans. Geosci. Remote Sens. 2023, 72, 5632514. [Google Scholar] [CrossRef]
- Yang, Y.; Sun, X.; Diao, W.; Li, H.; Wu, Y.; Li, X.; Fu, K. Adaptive knowledge distillation for lightweight remote sensing object detectors optimizing. IEEE Trans. Geosci. Remote Sens. 2022, 60, 5623715. [Google Scholar] [CrossRef]
- Joung, B.G.; Nath, C.; Li, Z.; Sutherland, J.W. Bearing anomaly detection in an air compressor using an LSTM and RNN-based machine learning model. Int. J. Adv. Manuf. Technol. 2024, 134, 3519–3530. [Google Scholar] [CrossRef]
- Kim, E.; Bui, T.; Yuan, J.; Mouli, S.C.; Ribeiro, B.; Yeh, R.A.; Fassnacht, M.P.; Jun, M.B. Online real-time machining chatter sound detection using convolutional neural network by adopting expert knowledge. Manuf. Lett. 2024, 41, 1386–1397. [Google Scholar] [CrossRef]
Fault Type | Time-Domain Features | Frequency-Domain Features |
---|---|---|
Normal | Uniform amplitude, minimal fluctuation | Low-frequency focus, uniform high-frequency. |
Broken bar | Significant low-frequency fluctuation | Enhanced low-frequency energy, weak harmonics |
Eccentricity | Periodic fluctuation, uneven amplitude | Prominent low-frequency harmonics, distinct peaks |
Short circuit | High-frequency, irregular fluctuation | Enhanced high-frequency energy, complex spectrum |
Inner race fault | Strong impacts, consistent period | Sharp high-frequency peaks, clear harmonics |
Outer race fault | Weaker impacts, unstable fluctuation | Prominent high-frequency peaks, weaker harmonics |
Type | Time Domain | Frequency Domain |
---|---|---|
Normal | a | b |
Broken bar fault | c | d |
Eccentric fault | e | f |
Inter-turn short circuit | g | h |
Inner race bearing fault | i | j |
Outer race bearing fault | k | l |
Hyperparameter | Value |
---|---|
Optimizer | Adam |
Loss function | Cross-entropy |
Learning rate | 0.001 |
Batch size | 64 |
Number of epochs | 100 |
Normalization method | MinMaxScaler |
Number of LSTM units | 64 |
Number of convolutional filters | 64 |
Kernel size | 3 and 5 |
Pooling size | 4 |
Hyperparameter | Value |
---|---|
Optimizer | Adam |
Loss function | Soft label cross-entropy + hard label Cross-entropy |
Learning rate | 0.001 |
Batch size | 64 |
Number of epochs | 50 |
Normalization method | MinMaxScaler |
Number of convolutional filters | 32 |
Kernel size | 3 |
Number of LSTM units | 32 |
Attention layer | Self-attention mechanism |
Balancing coefficient () | 0.5 |
Metric | Original Model | Quantized Model |
---|---|---|
Average inference time (s) | 0.0782 | 0.0165 |
Standard deviation of inference time (s) | 0.1960 | 0.0109 |
Throughput (samples/s) | 12.79 | 60.66 |
Average memory usage (MB) | 501.67 | 502.38 |
Average CPU utilization (%) | 0.17 | 0.01 |
Model size (MB) | 0.14 | 0.02 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Zhou, Z.; Qiao, Y.; Lin, X.; Li, P.; Wu, N.; Yu, D. A Deployment Method for Motor Fault Diagnosis Application Based on Edge Intelligence. Sensors 2025, 25, 9. https://doi.org/10.3390/s25010009
Zhou Z, Qiao Y, Lin X, Li P, Wu N, Yu D. A Deployment Method for Motor Fault Diagnosis Application Based on Edge Intelligence. Sensors. 2025; 25(1):9. https://doi.org/10.3390/s25010009
Chicago/Turabian StyleZhou, Zheng, Yusong Qiao, Xusheng Lin, Purui Li, Nan Wu, and Dong Yu. 2025. "A Deployment Method for Motor Fault Diagnosis Application Based on Edge Intelligence" Sensors 25, no. 1: 9. https://doi.org/10.3390/s25010009
APA StyleZhou, Z., Qiao, Y., Lin, X., Li, P., Wu, N., & Yu, D. (2025). A Deployment Method for Motor Fault Diagnosis Application Based on Edge Intelligence. Sensors, 25(1), 9. https://doi.org/10.3390/s25010009