Circuit Breaker Fault Diagnosis Method Based on IWGAN-GP Model under Imbalanced Data
Wang Zhenyuan1, Chen Lei1, Wan Shuting2, Hu Yuyao1, Jiang Ningbo1
1. School of Electrical Engineering Shandong University of Technology Zibo 255000 China; 2. Hebei Key Laboratory of Electric Machinery Health Maintenance & Failure Prevention North China Electric Power University Baoding 071003 China
Abstract：Accurate and reliable fault diagnosis of high-voltage circuit breakers is critical to ensuring the safety and stability of modern power systems. However, in practical applications, circuit breakers operate under stable conditions for long periods, and actual fault events occur infrequently. As a result, the available fault samples are scarce and highly imbalanced compared to normal-operation data, significantly degrading the performance of data-driven diagnostic models. Conventional resampling strategies, such as oversampling and undersampling, have been used to mitigate imbalance. However, these strategies often lead to overfitting, noise amplification, or the loss of essential information. Generative Adversarial Networks (GANs) have emerged as powerful tools for data augmentation. However, standard GANs suffer from training instability, mode collapse, and limited feature learning capacity when applied to complex vibration signals. Therefore, this paper proposes an improved Wasserstein GAN with gradient penalty (IWGAN-GP) framework, integrated with a convolutional neural network (CNN). In the proposed method, vibration signals collected from a 35 kV CT14 spring-operated circuit breaker are preprocessed using pre-emphasis, windowing, framing, and Fourier transformation. Mel-frequency cepstral coefficients (MFCC) are then extracted to capture both frequency and temporal characteristics of the signals and serve as input to the IWGAN-GP model, which employs CNN-based generator and discriminator architectures. The generator uses transposed convolutional layers to reconstruct high-resolution feature maps, while the discriminator applies convolutional operations to extract local patterns and distinguish between real and generated data. A gradient penalty term is introduced to ensure the discriminator's Lipschitz continuity and eliminate conventional WGANs’ instability and weight-clipping limitations. Tests were conducted using six operating states, including normal conditions and five common fault scenarios: loose base screws, buffer spring fatigue, transmission mechanism fault, closing spring fatigue, and opening spring fatigue. For each fault type, 30 groups of vibration data were collected, while 60 groups were collected for the normal state. The IWGAN-GP model was trained to generate balanced fault datasets, which were then compared with data generated by SMOTE, conventional GANs, and WGAN-GP. Quantitative evaluations using the Inception Score (IS), Fréchet Inception Distance (FID), and Kernel Inception Distance (KID) showed that IWGAN-GP achieved the best balance between diversity and fidelity, with IS of 9.01, FID of 12.96, and KID of 0.69. Visual inspection of the generated MFCC feature maps confirmed that the IWGAN-GP model successfully captured both global patterns and subtle local variations in vibration signals. To validate diagnostic performance, the augmented datasets were used to train CNN classifiers optimized through Bayesian hyperparameter tuning. The model trained on IWGAN-GP data achieved an overall accuracy of 98%, significantly higher than those trained on original imbalanced data, SMOTE, GAN, and WGAN-GP. Moreover, IWGAN-GP training exhibited faster convergence, lower loss, and improved robustness to small- sample limitations. In conclusion, this research establishes IWGAN-GP for fault diagnosis under imbalanced data conditions. By integrating CNN feature extraction with gradient penalty stabilization, the method overcomes the limitations of traditional GAN-based augmentation and enables the generation of high-quality synthetic samples. The results confirm that IWGAN-GP significantly enhances classification accuracy, accelerates training convergence, and improves generalization performance in diagnosing multiple fault states of circuit breakers. These findings offer a promising pathway to advance intelligent monitoring and predictive maintenance for high-voltage power equipment.
王震远, 陈磊, 万书亭, 胡玉耀, 姜宁波. 不平衡数据下基于IWGAN-GP模型的断路器故障诊断方法[J]. 电工技术学报, 2026, 41(12): 4231-4245.
Wang Zhenyuan, Chen Lei, Wan Shuting, Hu Yuyao, Jiang Ningbo. Circuit Breaker Fault Diagnosis Method Based on IWGAN-GP Model under Imbalanced Data. Transactions of China Electrotechnical Society, 2026, 41(12): 4231-4245.
[1] Chen Lei, Wan Shuting, Dou Longjiang.Improving diagnostic performance of high-voltage circuit breakers on imbalanced data using an oversampling method[J]. IEEE Transactions on Power Delivery, 2022, 37(4): 2704-2716. [2] 王钰洁, 赖冬明, 王立军, 等. 基于多任务学习组合模型的万能式断路器故障诊断方法[J]. 高电压技术, 2025, 51(5): 2394-2403. Wang Yujie, Lai Dongming, Wang Lijun, et al.Fault diagnosis method for conventional circuit breakers based on multi-task learning combination model[J]. High Voltage Engineering, 2025, 51(5): 2394-2403. [3] 范兴明, 许洪华, 张思舜, 等. 基于改进蜣螂优化算法深度混合核极限学习机的高压断路器故障诊断[J]. 电工技术学报, 2025, 40(12): 3994-4003. Fan Xingming, Xu Honghua, Zhang Sishun, et al.Fault diagnosis of high voltage circuit breaker based on improved dung beetle optimizer algorithm deep hybrid kernel extreme learning machine[J]. Transac-tions of China Electrotechnical Society, 2025, 40(12): 3994-4003. [4] 高伟, 饶俊民, 全圣鑫, 等. 不均衡小样本下多特征优化选择的生命体触电故障识别方法[J]. 电工技术学报, 2024, 39(7): 2060-2071. Gao Wei, Rao Junmin, Quan Shengxin, et al.Biological electric-shock fault identification method based on multi-feature optimization selection under unbalanced small sample[J]. Transactions of China Electrotechnical Society, 2024, 39(7): 2060-2071. [5] 李弈, 张金龙, 漆汉宏, 等. 基于变分深度嵌入-带有梯度惩罚的生成对抗网络的锂离子电池老化特性建模[J]. 电工技术学报, 2024, 39(13): 4226-4239. Li Yi, Zhang Jinlong, Qi Hanhong, et al.Ageing performance modeling of Li-ion batteries based on variational deep embedding-Wasserstein GAN with gradient penalty[J]. Transactions of China Electro-technical Society, 2024, 39(13): 4226-4239. [6] 赵燕, 朱景伟, 任萍. 基于数据预处理的小样本开关磁阻电机智能方法建模[J]. 电机与控制学报, 2025, 29(2): 85-95. Zhao Yan, Zhu Jingwei, Ren Ping.Modeling of switched reluctance motor by intelligent method with small samples and data pretreatment[J]. Electric Machines and Control, 2025, 29(2): 85-95. [7] 李奎, 王鹏, 胡博凯, 等. 基于机械特性的微型断路器服役状态辨识方法[J]. 电工技术学报, 2025, 40(14): 4642-4654. Li Kui, Wang Peng, Hu Bokai, et al.Service status identification method of miniature circuit breaker based on mechanical characteristics[J]. Transactions of China Electrotechnical Society, 2025, 40(14): 4642-4654. [8] Zhang Han, Xu Tao, Li Hongsheng, et al.StackGAN: realistic image synthesis with stacked generative adversarial networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(8): 1947-1962. [9] Wang Chaoyue, Xu Chang, Yao Xin, et al.Evolu-tionary generative adversarial networks[J]. IEEE Transactions on Evolutionary Computation, 2019, 23(6): 921-934. [10] Tang Shengnan, Yuan Shouqi, Zhu Yong.Deep learning-based intelligent fault diagnosis methods toward rotating machinery[J]. IEEE Access, 2019, 8: 9335-9346. [11] Arjovsky M, Chintala S, Bottou L.Wasserstein generative adversarial networks[C]//Proceedings of the 34th International Conference on Machine Learning, Sydney, 2017: 214-223. [12] Luo Jia, Huang Jinying, Li Hongmei.A case study of conditional deep convolutional generative adversarial networks in machine fault diagnosis[J]. Journal of Intelligent Manufacturing, 2021, 32(2): 407-425. [13] 邵阳, 武建文, 马速良, 等. 用于高压断路器机械故障诊断的AM-ReliefF特征选择下集成SVM方法[J]. 中国电机工程学报, 2021, 41(8): 2890-2901. Shao Yang, Wu Jianwen, Ma Suliang, et al.Integrated SVM method with AM-Relief feature selection for mechanical fault diagnosis of high voltage circuit breakers[J]. Proceedings of the CSEE, 2021, 41(8): 2890-2901. [14] 刘会兰, 许文杰, 赵书涛, 等. 面向高压断路器故障分类的电流-振动信号类聚几何敏感特征优选方法[J]. 电工技术学报, 2023, 38(1): 26-36. Liu Huilan, Xu Wenjie, Zhao Shutao, et al.Optimi-zation method of clustering geometric sensitive features of current vibration signals for fault classification of high voltage circuit breakers[J]. Transactions of China Electrotechnical Society, 2023, 38(1): 26-36. [15] 万书亭, 马晓棣, 陈磊, 等. 基于振动信号短时能熵比与DTW的高压断路器状态评估及故障诊断[J]. 高电压技术, 2020, 46(12): 4249-4257. Wan Shuting, Ma Xiaodi, Chen Lei, et al.State evaluation and fault diagnosis of high-voltage circuit breaker based on short-time energy entropy ratio of vibration signal and DTW[J]. High Voltage Engin-eering, 2020, 46(12): 4249-4257. [16] 郑含博, 胡钧浩, 李金恒, 等. 基于改进实例分割与深度神经网络温度拟合的电力设备热故障诊断研究[J]. 电机与控制学报, 2025, 29(2): 116-125. Zheng Hanbo, Hu Junhao, Li Jinheng, et al.Thermal fault diagnosis of power equipment through improved instance segmentation and deep neural network temperature fitting[J]. Electric Machines and Control, 2025, 29(2): 116-125. [17] Li Guoqiang, Wu Jun, Deng Chao, et al.Parallel multi-fusion convolutional neural networks based fault diagnosis of rotating machinery under noisy environments[J]. ISA Transactions, 2022, 128: 545-555. [18] 梁郑秋, 郝亮亮, 周艳真, 等. 基于卷积神经网络的核电多相无刷励磁系统旋转整流器故障诊断[J]. 电工技术学报, 2023, 38(20): 5458-5472. Liang Zhengqiu, Hao Liangliang, Zhou Yanzhen, et al.Fault diagnosis of rotating rectifier in nuclear multi-phase brushless excitation system based on con-volutional neural network[J]. Transactions of China Electrotechnical Society, 2023, 38(20): 5458-5472. [19] 肖进胜, 申梦瑶, 雷俊锋, 等. 基于生成对抗网络的雾霾场景图像转换算法[J]. 计算机学报, 2020, 43(1): 165-176. Xiao Jinsheng, Shen Mengyao, Lei Junfeng, et al.Image conversion algorithm for haze scene based on generative adversarial networks[J]. Chinese Journal of Computers, 2020, 43(1): 165-176. [20] Li Jie, Skinner K A, Eustice R M, et al.WaterGAN: unsupervised generative network to enable real-time color correction of monocular underwater images[J]. IEEE Robotics and Automation Letters, 2018, 3(1): 387-394.
2026, Vol. 41 
