基于图像融合与迁移学习的永磁同步电机驱动器强泛化性故障诊断研究  

A Strong Generalized Fault Diagnosis Method for PMSM Drives With Image Fusion and Transfer Learning

作　　者：李政 汪凤翔 张品佳 LI Zheng;WANG Fengxiang;ZHANG Pinjia(National and Local Joint Engineering Research Center for Electrical Drive and Power Electronics(Haixi Institutes,Chinese Academy of Sciences),Quanzhou 362200,Fujian Province,China;State Key Lab of Security Control and Simulation of Power Systems and Large Scale Generation Equipment(Department of Electrical Engineering,Tsinghua University),Haidian District,Beijing 100084,China)

机构地区：[1]电机驱动与功率电子国家地方联合工程研究中心(中国科学院海西研究院),福建省泉州市362200 [2]电力系统及大型发电设备安全控制和仿真国家重点实验室(清华大学电机系),北京市海淀区100084

出　　处：《中国电机工程学报》2024年第12期4933-4944,I0028,共13页Proceedings of the CSEE

基　　金：国家自然科学基金(面上基金项目)(52277070)。

摘　　要：参数失配、死区效应、采样偏差、传感器故障、编码器故障等会导致永磁同步电机(permanent magnet synchronous motor,PMSM)三相电流包含高次谐波,进而产生转矩脉动与振动噪声。针对这一问题,该文提出一种基于图像融合和深度学习的数据驱动方法诊断电驱系统故障。首先,以永磁同步电机驱动器三相电流信号作为原始数据源,结合仿真和实验建立多源故障数据库。其次,对三相电流信号进行短时傅里叶变换,得到时频域彩色图像,各相提取一种颜色形成灰度图表示单相特征。采用图像融合方法将3张灰度图融合为一张彩色频谱特征图。最后,应用SqueezeNet迁移学习对样本进行训练。实验结果表明,该方法综合故障诊断准确率达到98.63%,说明所提方法能实现系统级的多源故障诊断,并且具有较高的实用性与泛化性,可以有效提高故障诊断的准确率。The dead zone effect of the inverter,the nonlinear characteristics,the parameter mismatch,sampling deviation,and the setting error of the controller will lead to the three-phase current of the permanent magnet synchronous motor(PMSM)containing high-order harmonics,and then result in torque fluctuation and vibration noise.This paper proposes a data-driven diagnosis method based on image fusion and deep learning to solve this problem.First,a multi-source fault database is established based on simulation and experimental data.The three-phase current signal of a PMSM is used as the original data source without the aid of an external testing instrument.The spectrum image is obtained by a short-time Fourier transform.The time-frequency gray images are fused into a color image by the method of image fusion.After classifying and labeling the data,the samples are trained using the transfer learning of SqueezeNet.The test results show that the fault diagnosis accuracy of this method is 98.63%.Compared with the traditional method,the proposed method realizes the system-level multi-source fault diagnosis and has higher practicability.Moreover,the feasibility and generalization of fault diagnosis under the condition of insufficient sample data are realized,and the accuracy of fault diagnosis is effectively improved.

关 键 词：永磁同步电机 故障诊断 图像融合 深度学习 

分 类 号：TM341[电气工程—电机]