【好文推荐】|永磁同步电机驱动逆变器开路故障机器学习在线诊断方法

中国电工技术学会活动专区

CES Conference

永磁同步电机驱动逆变器开路故障机器学习在线诊断方法

A hybrid diagnosis method for inverter open-circuit faults in PMSM drives

Author：Zeliang Zhang; Guangzhao Luo; Zhengbin Zhang; Xuecheng Tao

DOI: 10.30941/CESTEMS.2020.00023

https://ieeexplore.ieee.org/document/9211089

01

Abstract

  To improve the evaluating process of inverter open-circuit faults diagnosis in permanent magnet synchronous motor (PMSM) drives, this paper presents a diagnosis method based on current residuals and machine learning (ML) models. The ML models are introduced to make a comprehensive evaluation for the current residuals obtained from a state observer, instead of evaluating the residuals by comparing to thresholds. Meanwhile, fault diagnosis and location are conducted simultaneously by the ML models, which simplifies the diagnosis process. Besides, a sampling strategy is designed to implement the proposed scheme online. In experiments, the diagnosis is completed in less than half an electrical cycle after the fault occurred and the computation burden is acceptable to the controller.

02

Innovation

1. The utilization of machine learning (ML) in this paper has two main merits: 1. Instead of setting empirical thresholds, the ML model evaluates the distribution feature of samples in a range of electrical phase; 2. The phase range of input samples can be adjusted conveniently to achieve a rapid or robust diagnosis.

2. To overcome the conflict between the required constant input dimension of ML models and the variable sampling dimension of motor drive, A phase-based sampling strategy is designed to adapt variable-frequency systems to the online-running ML model.

03

Main content

   A.        Fault analysis

  When no faults occur, the inverter has six space voltage vectors and two zero vectors. When open-circuit occurs, the faulty devices are no longer controlled by the PWM signals, incurring changes in the magnitudes and directions of inherent vectors, as well as the synthesized vectors that control the motion of the stator flux.

Fig. 1. Influence of T1open-circuit on u4

  As shown in Fig. 1, taking T1 open-circuit as an example, u4, u5, u6 are forced to change under the T1 open-switch. The amplitudes of u5f, u6f are reduced to √3/2 of the amplitudes of , while u4f is equivalent to a zero vector, as depicted in Fig. 2. The rotating trace of stator flux ψ and the corresponding healthy and faulty vectors in each spatial sector are depicted in Fig. 3.

Fig. 2 Changes inthe amplitudes and directions of the inherent vectors

Fig. 3.  Changes inthe amplitudes and directions of the inherent vectors

B.       Faultdiagnosis

  Asdepicted in Fig. 4, A Luenburger observer is introduced to obtain currentresiduals as diagnosis variables; then, the residual samples are collectedusing the designed sampling strategy; after that, Principal Component Analysisand Support Vector Machine are trained offline based on the samples; finally,the samples of different fault patterns are classified by the machine learningmodels online. The proposed diagnosis method is integrated into the controlcycles and does not cause any influence on motor control.

Fig. 4. Changes in the amplitudes and directions of theinherent vectors.

C.       Experimental verification

  T1open-circuit, T1T3 open-circuit, and T1T4 open-circuit are injected to verifythe effectiveness of the proposed method. Fig. 5 shows the current residualsamples collected using the proposed sampling strategy under T1T3 open-circuit.Channel 1 of the oscilloscope is the A-phase current, and Channels 2 to 4 arethe three-phase current residuals. The proposed strategy can sample currentresiduals with a fixed dimension in a fixed phase range of electrical cycle, tofacilitate offline training and online implementation of machine learningmodels. After the original sample set is collected, the dimension reduction andtraining are performed offline. The resulting training set and six binaryclassifiers are shown in Fig. 6.

 Fig. 5. Samples collected using the proposed strategy

Fig. 6.  The distribution of the trainingset and six trained binary classifiers

  Fig.7 shows the shortest diagnosis time. In Fig. 7, channel 1 is the A-phasecurrent, and channel 2 is the flag indicating the completion of the diagnosis.The High level represents that a fault has been located. As can be seen, thediagnosis was completed in less than half an electrical cycle after the faultoccurred in the experiment.

Fig. 7.  The shortest diagnosis time in experiments

04

Conclusion

  Online implementation of machine learning and offline sampling are twomajor concerns for utilizing machine learning in electric drives. PCA and SVMare introduced in this paper to comprehensively assess the three-phase currentresiduals obtained through a Luenburger observer. Besides, a sampling strategyis designed to implement this method online. Experiments verify that theproposed method can locate faults in less than half an electrical cycle and canclassify the fault patterns whose responses show similarity. The proposedsampling strategy fulfils the requirements of online implementation andimproving the sampling process to reduce the data amount.

引用本文

Z. Zhang, G. Luo, Z. Zhang and X. Tao, "A hybrid diagnosis method for inverter open-circuit faults in PMSM drives," in CES Transactions on Electrical Machines and Systems, vol. 4, no. 3, pp. 180-189, Sept. 2020, doi: 10.30941/CESTEMS.2020.00023.

本文作者

团队简介

西北工业大学稀土永磁电机及控制技术研究所成立于1991年。建所三十年来，面向航空航天、高端装备、新能源等国家重大战略需求，不断推动我国稀土永磁电机、控制驱动、交通电气化及相关行业发展和技术进步。骆光照教授团队长期致力于高性能、高可靠稀土永磁电机及驱动器的研究与设计，团队主要研究方向有：高动态永磁同步电机控制技术、永磁同步电机驱动可靠性提高与余度化技术、电驱动系统实时仿真技术、新能源转换智能控制技术等。

Zeliang Zhang was born in Shaanxi, China, in 1995. He received his B.Eng. and M.Eng.degrees in electrical engineering from Northwestern Polytechnical University(NPU), Xi’an, China, in 2017 and 2020, respectively. He is currently workingtoward the Ph.D. degree at University of York, UK. His research interestsinclude fault-tolerant control and fault diagnosis for electric drives.

Hongliang Wang (M’12-SM’15) received the B.Sc. in Electrical Engineering from Anhui University of Science and Technology, Huainan, China in 2004, and received the Ph.D. degree in Electrical Engineering from Huazhong University of Science and Technology, Wuhan, China in 2011.GuangzhaoLuo(Senior Member, IEEE)received the M.S. and Ph.D. degrees in electrical engineering from theNorthwestern Polytechnical University (NPU), Xi’an, China, in 1998 and 2003,respectively.From 2003 to 2004, hewas a Postdoctoral Research at the University of Federal Defense, Munich,Germany. He is currently a Professor with NPU. He is the Vice Director of theRare Earth Permanent Magnet (REPM) Electric Machine and Control EngineeringCenter, Shaanxi Province. His research interests include advance control theoryof permanent magnet electrical machine, high performance control technology ofpermanent magnet synchronous motor for electric traction and electric vehicle,real-time simulation technology for electrical drive system, and intelligencecontrol of new energy conversion. Dr. Luo received the Second Prize from theChina National Defense Science and Technology Progress Award in 1995 and 2011.

ZhengbinZhang was born in Gansu,China, in 1986. He received the M.S. degree in pattern recognition andintelligent system from Shengyang Aerospace University Sheng’yang, China, in2011 and 2014. He is currently working at Lanzhou wanli AviationElectromechanical CO.LTD.  His researchinterests include motor control and pattern recognition.

XuechengTao was born in Anhui, China,in1993. He received the B.Eng. and M.Eng. degree in electrical engineering fromInner Mongolia University of Technology (IMUT) and Northwestern PolytechnicalUniversity (NPU)，Inner Mongolia and Xi'an, China in 2017 and 2020. His research interestsinclude motor control and controller reliability analysis.

下载论文PDF版，请点击左下角“阅读原文”，访问期刊网站。

推荐阅读

联系我们

CES TEMS编辑部：010-63256823

邮箱：[e-mail]

网址：http://www.cestems.org/

投稿网址：https://mc03.manuscriptcentral.com/tems

官方微信号：CESTEMS

中国电工技术学会

新媒体平台

学会官方微信

电工技术学报

电气技术

学会官方B站

新浪微博

今日头条号