Article Details

Abstract

Aiming at the failure of supervised diagnostic methods caused by the scarcity of fault samples of industrial robots, this paper proposes a zero-shot fault detection method based on Deep Support Vector Data Description (Deep SVDD) and multi-wavelet fusion. Only using vibration signals in normal state, this method first converts signals into multi-view time-frequency maps through a continuous transformation library fused with multiple wavelets to fully extract time-frequency features. Then, a deep convolutional network is used to map time-frequency maps into a high-dimensional space, and the Deep SVDD loss function is optimized to construct a compact minimum hypersphere. During detection, faults are identified by calculating the distance from the features of the sample to be measured to the sphere center and setting a dynamic threshold. Experimental results show that the average detection accuracy of the proposed method for various faults is above 94.5%, which is significantly better than traditional methods, providing an effective solution for intelligent operation and maintenance of industrial robots under zero-sample conditions.

Keywords

fault detection; zero fault samples; deep support vector data description; multi-wavelet fusion; industrial robots

References

1. [1] SABRY A H，AMIRULDDIN U A B U. A Review on Fault Detection and Diagnosis of Industrial Robots and Multi-Axis Machines [J/OL].Results in Engineering，2024，23：102397[2025-09-03].https://www.sciencedirect.com/science/article/pii/S2590123024006522?via%3Dihub.
2. [2] Zhang Hui, Wang Shujuan, Zhang Qingsen, etc Research on Fault Diagnosis Method of Rolling Bearings Based on Wavelet Packet Transform [J]. Vibration and Shock, 2004 (4): 1-8
3. [3] Yu Qianxiang, Li Qing, Li Linlin, etc A review of data-driven fault diagnosis methods for large-scale industrial production processes [J]. Journal of Engineering Science, 2025, 47 (4): 780-793
4. [4] Chai Zheng, Wang Jiaye, Zhao Chunhui, etc Deep Transfer Learning Methods for Typical Supervision Tasks in Industrial Monitoring: Current Status, Challenges, and Prospects [J]. Chinese Science: Information Science, 2023, 53 (5): 821-840
5. [5] REHMAN A U，JIAO W D，JIANG Y H，et al. Deep Learning in Industrial Machinery: A Critical Review of Bearing Fault Classification Methods [J/OL].Applied Soft Computing，2025，171：112785[2025-09-06].https://www.sciencedirect.com/science/article/abs/pii/S1568494625000961.
6. [6] NEUPANE D，BOUADJENEK M R，DAZELEY R，et al. Data-Driven Machinery Fault Diagnosis: A Comprehensive Review [J].Neurocomputing，2025，627：129588[2025-09-12].https://www.sciencedirect.com/science/article/pii/S0925231225002607.
7. [7] SAOULI M E，TOUBA M M，BOUDIAF A. Hybrid Diagnostic Framework for Interpretable Bearing Fault Classification Using CNN and Dual-Stage Feature Selection [J/OL].Sensors，2025，25（20）：6386[2025-09-20].https://doi.org/10.3390/s25206386.
8. [8] Xie Zhiyong, Zhu Juanfen, Zhao Ying, etc Integrated multi-sensor industrial robot fault information collection system [J]. Mechanical Engineering and Automation, 2023 (2): 19-21
9. [9] Hu Fuqing, Chen Huipeng, Cheng Zhe, etc Fault diagnosis method for planetary gearbox based on empirical mode decomposition and deep convolutional neural network [J]. Journal of Mechanical Engineering, 2019, 55 (7): 9-18
10. [10] RUFF L，VANDERMEULEN R，GOERNITZ N，et al. Deep One-Class Classification [C]//Proceedings of the 35th International Conference on Machine Learning.Stockholm：PMLR，2018：4393-4402.

Funding

2024 Jiangxi Provincial Department of Education Science and Technology Project "Intelligent Fault Diagnosis of Industrial Robots Based on Deep Learning" (GJJ2406902); Research on the Construction of Project based Industry Education Integration and School Enterprise Cooperation Talent Training System - Taking Industrial Robot Field Programming Course as an Example "(JD-24-17), a school level project of Jiangxi Mechanical and Electrical Vocational and Technical College in 2024; Key Activity Project of Nanchang Association for Science and Technology in 2025: "Nanchang Lithium Battery Industry 'Intelligent' Manufacturing Upgrade Technology Service Project" (KTXY250002)

About the authors

Luo Zhe (1987-), male, Han ethnicity, from Ganzhou, Jiangxi Province, is a lecturer and director of the Intelligent Control Technology Teaching and Research Office. He is currently pursuing a doctoral degree in industrial robot technology, industrial automation, and intelligent control technology.