采用深度学习的永磁同步电机匝间短路故障诊断方法

李垣江　张周磊　李梦含　魏海峰　张懿

摘 要：针对永磁同步电机匝间短路故障诊断方法因结构复杂、样本特征稀少和大数据冗余等因素引起的诊断困难问题，基于生成式对抗和稀疏自编码深度学习网络提出一种高效准确的匝间短路故障诊断方法。该方法通过采集永磁同步电机负序电流和转矩特征信号，利用生成式对抗神经网络完成样本数据扩张，构建鲁棒、多样的训练集合，结合稀疏自编码网络实现高效准确的故障特征分类、诊断。实验表明该方法实现了对永磁同步电机负序电流和转矩特征信号的样本数据扩张，构建了优化深度网络实现高效准确的电机匝间短路诊断。数据显示，采用此方法的匝间短路故障诊断准确率高达99.4%。

关键词：永磁同步电机;匝间短路;稀疏自编码;生成式对抗神经网络;样本扩张;负序电流;故障诊断

DOI：10.15938/j.emc.2020.09.019

中图分类号：TP 277

文献标志码：A

文章编号：1007-449X（2020）09-0173-08

Fault diagnosis of inter-turn short circuit of permanent magnet synchronous motor based on deep learning

LI Yuan-jiang1，2， ZHANG Zhou-lei1， LI Meng-han1， WEI Hai-feng1，2， ZHANG Yi1，2

（1.College of Electronic Information， Jiangsu University of Science and Technology， Zhenjiang 212003， China;2.Changshu Rhett Electric Co.， Ltd.， Changshu 215500，China）

Abstract：

An efficient and accurate fault diagnosis of inter-turn short circuit method was proposed to solve the problem that the fault diagnosis of inter-turn short circuit for permanent magnet synchronous motor has difficulty to diagnose caused by motors complex structure， few sample features and large data redundancy， which is based on the generative adversarial nets and sparse auto encoder network. Negative sequence current and torque characteristic signal of permanent magnet synchronous motor were collected to expand sample data and build robust and diverse training set by generative adversarial nets， and sparse auto encoder network was used to realize efficient and accurate classification or diagnosis of fault features by sparse auto encoder network. Experiments show that the expansion of sample data of the negative sequence current and torque characteristic signals of permanent magnet synchronous motors were realized， and an optimized deep network was built to achieve efficient and accurate diagnosis of the inter-turn short circuit.The experimental results show that the accuracy rate of the fault diagnosis of the inter-turn short circuit method is as high as 99.4%.

Keywords：permanent magnet synchronous motor; inter-turn short circuit; sparse auto encoder; generative adversarial nets; sample expansion; negative sequence current; fault diagnosis

0 引 言

匝間短路是一种典型的、破坏性极强的电机故障。轻微的匝间短路若不加以识别，会导致更为严重的相间短路、单相接地故障。因此，基于深度学习理论研究匝间短路的故障诊断方法对于及时发现电机早期故障、提高系统可靠性和降低维护成本具有重要意义[1-2]。

现阶段，永磁同步电机（permanent magnet synchronous motor，PMSM）匝间短路的识别大多通过各种状态参数（负序特征[3]、振动[4]等参数）判断其运行的状态，确定故障部位和劣化程度。匝间短路引起的负序电流波动、阻抗变化均能有效地表征故障特征，但其易受电源品质、电机结构、固有参数等客观条件的影响。……