A Wind Turbine Fault Diagnosis Method Based on Siamese Deep Neural Network

doi:10.16182/j.issn1004731x.joss.21-0261

Abstract

Abstract:

In order to effectively extract the fault features of time series data in supervisory control and data acquisition (SCADA), considering the advantages of one-dimensional convolutional neural network (1-D CNN) for extracting local time series features and the advantages of long-term memory (LSTM) which can extract long-term dependent features, a method for fault diagnosis of wind turbines based on 1-D CNN-LSTM is proposed. To solve the problem of the scarcity of fault samples of wind turbines based on the siamese network architecture, a wind fault diagnosis method based on siamese 1-D CNN-LSTM is proposed. The proposed siamese 1-D CNN-LSTM method relies on a small amount of sample data to effectively extract the fault features of the wind turbine. The results show that 1-D CNN-LSTM is better than other existing deep learning methods. When the training samples are insufficient, the proposed siamese 1-D CNN-LSTM can significantly improve the fault diagnosis results.

Key words: wind turbine, fault diagnosis, deep learning, siamese neural network, few-shot learning

CLC Number:

TM315
TP391

Jiarui Liu, Guotian Yang, Xiaowei Wang. A Wind Turbine Fault Diagnosis Method Based on Siamese Deep Neural Network[J]. Journal of System Simulation, 2022, 34(11): 2348-2358.

Figures/Tables 13

Fig. 1

Fig. 2

Fig. 3

Table 1

Fig. 4

Table 2

Table 3

Table 4

Fig. 5

Fig. 6

Fig. 7

Fig. 8

Fig. 9

References 17

[1]	陈雪峰, 郭艳婕, 许才彬, 等. 风电装备故障诊断与健康监测研究综述[J]. 中国机械工程, 2020, 31(2): 175-189.
	Chen Xuefeng, Guo Yanjie, Xu Caibin, et al. Review of Fault Diagnosis and Health Monitoring for Wind Power Equipment[J]. China Mechanical Engineering, 2020, 31(2): 175-189.
[2]	Wang J, Qiao W, Qu L. Wind Turbine Bearing Fault Diagnosis Based on Sparse Representation of Condition Monitoring Signals[J]. IEEE Transactions on Industry Applications (S0093-9994), 2019, 55(2): 1844-1852.
[3]	Liu Z, Wang X, Zhang L. Fault Diagnosis of Industrial Wind Turbine Blade Bearing Using Acoustic Emission Analysis[J]. IEEE Transactions on Instrumentation and Measurement (S0018-9456), 2020, 69(9): 6630-6639.
[4]	邓子豪, 李录平, 杨波, 等. 基于SCADA数据特征提取的风电机组偏航齿轮箱故障诊断方法研究[J]. 动力工程学报, 2021, 41(1): 43-50.
	Deng Zihao, Li Luping, Yang bo, et al. Research on Fault Diagnosis Method of Wind Turbine Yaw Gearbox Based on SCADA Data Feature Extraction[J]. Journal of Power Engineering, 2021, 41(1): 43-50.
[5]	余萍, 曹洁. 深度学习在故障诊断与预测中的应用[J]. 计算机工程与应用, 2020, 56(3): 1-18.
	Yu Ping, Cao Jie. Deep Learning Approach and Its Application in Fault Diagnosis and Prognosis[J]. Computer Engineering and Applications, 2020, 56(3): 1-18.
[6]	Lecun Y, Bengio Y, Hinton G. Deep learning[J]. Nature (S0028-0836), 2015, 521(7553): 436-444.
[7]	吴定会, 祝志超, 韩欣宏. 基于BO-SDAE多源信号的风电机组轴承故障诊断[J]. 系统仿真学报, 2021, 33(5): 1148-1156.
	Wu Dinghui, Zhu Zhichao, Han Xinhong. Wind Turbine Bearing Fault Diagnosis Based on BO-SDAE Multi-source Signals[J]. Journal of System Simulation, 2021, 33(5): 1148-1156.
[8]	Ince T, Kiranyaz S, Eren L, et al. Real-Time Motor Fault Detection by 1D Convolutional Neural Networks[J]. IEEE Transactions on Industrial Electronics (S0278-0046), 2016, 63(11): 7067-7075.
[9]	Jiang G, He H, Yan J, et al. Multiscale Convolutional Neural Networks for Fault Diagnosis of Wind Turbine Gearbox[J]. IEEE Transactions on Industrial Electronics (S0278-0046), 2018, 66(4): 3196-3207.
[10]	Lei J, Liu C, Jiang D. Fault Diagnosis of Wind Turbine Based on Long Short-term Memory Networks[J]. Renewable Energy, 2019, 133(4): 422-432.
[11]	赵洪山, 闫西慧, 王桂兰, 等. 应用深度自编码网络和XGBoost的风电机组发电机故障诊断[J]. 电力系统自动化, 2019, 43(1): 81-90.
	Zhao Hongshan, Yan Xihui, Wang Guilan, et al. Fault Diagnosis of Wind Turbine Generator Based on Deep Autoencoder Network and XGBoost[J]. Automation of Electric Power Systems, 2019, 43(1): 81-90.
[12]	魏书荣, 张鑫, 符杨, 等. 基于GRA-LSTM-Stacking模型的海上双馈风力发电机早期故障预警与诊断[J]. 中国电机工程学报, 2021, 41(7): 2373-2383.
	Wei Shurong, Zhang Xin, Fu Yang, et al. Early Fault Warning and Diagnosis of Offshore Doubly-fed Wind Turbines Based on GRA-LSTM-Stacking Model[J]. Proceedings of the Chinese Society of Electrical Engineering, 2021, 41(7): 2373-2383.
[13]	罗会兰, 陈鸿坤. 基于深度学习的目标检测研究综述[J]. 电子学报, 2020, 48(6): 1230-1239.
	Luo Huilan, Chen Hongkun. A Survey of Research on Target Detection Based on Deep Learning[J]. Acta Electronica Sinica, 2020, 48(6): 1230-1239.
[14]	Kingma D P, Ba J L. Adam: A Method for Stochastic Optimization[C]//3^rd International Conference on Learning Representations, ICLR 2015 Conference Track Proceedings. New York: ACM, 2015: 1-15.
[15]	Bromley J, Guyon I, Lecun Y, et al. Signature Verification Using a “Siamese” Time Delay Neural Network[C]//Advances in Neural Information Processing Systems 6, 7^th NIPS Conference. Cambridge, MA: MIT, 1993: 669.
[16]	Liu X, Zhou Y, Zhao J, et al. Siamese Convolutional Neural Networks for Remote Sensing Scene Classification[J]. IEEE Geoscience and Remote Sensing Letters (S1545-598X), 2019, 69(9): 6630-6639.
[17]	Yin H, Wang Y, Ding X, et al. 3D LiDAR-Based Global Localization Using Siamese Neural Network[J]. IEEE Transactions on Intelligent Transportation Systems (S1524-9050), 2019, 21(4): 1380-1392.

参数	取值	参数	取值
卷积层1	1×5@64, S=1	softmax	Nf
池化层1	1×2, S=2	激活函数	ReLU
卷积层2	1×3@32, S=1	batch size	32
池化层2	1×2, S=2	迭代次数	50
LSTM	64	优化器	Adam
全连接层	80	学习率	0.01
drop out	0.3

数据集A	数据集B
风速(m·s^-1)	风速(m·s^-1)
桨距角(º)	桨距角(º)
偏航位置(º)	内部功率(kW)
网侧功率(kW)	功率因数(%)
无功功率(var)	频率(Hz)
发电机转速(rpm)	发电机电压(V)
环境温度(℃)	发电机输出(kW)
机舱温度(℃)	发电机转速(rpm)
发电机转子温度(℃)	齿轮箱轴承温度(℃)

数据集	数据	训练集(70%)	测试集(30%)	子训练集1(5%)(支撑集)	子训练集2(1%)(支撑集)
A	1 261 (正常：631，结冰：630)	883	378	57 (114)	11 (22)
B	10 267 (轴承异常：1 768，通地泄露故障：2 483，叶片损坏：2 618，发电机异常：3 398)	7 181	3 086	359 (718)	71 (142)

数据集	模型	Precision	Recall	F1score	Accuracy
A	LSTM	0.837±0.030 9	0.871±0.025 9	0.853±0.038 7	0.858±0.035 3
	1-D CNN	0.912±0.018 7	0.904±0.026 8	0.904±0.025 1	0.908±0.025 6
	1-D CNN-LSTM	0.922±0.009 2	0.925±0.012 4	0.921±0.016 0	0.923±0.010 1
B	LSTM	0.951±0.014 1	0.934±0.019 4	0.938±0.016 4	0.947±0.023 5
	1-D CNN	0.974±0.008 9	0.966±0.005 9	0.968±0.007 1	0.969±0.008 9
	1-D CNN-LSTM	0.979±0.005 5	0.971±0.003 7	0.974±0.004 1	0.986±0.002 7