基于Vision Transformer的永磁同步电机故障智能诊断  

Intelligent Diagnosis of PMSM Faults Based on Vision Transformer

作　　者：蒋亦悦 卞东石 焦世琪 张晓飞 JIANG Yiyue;BIAN Dongshi;JIAO Shiqi;ZHANG Xiaofei(China Shipbuilding and Ocean Engineering Design and Research,Shanghai 200011,China;College of Electrical and Information Engineering,Hunan University,Changsha 410082,China)

机构地区：[1]中国船舶及海洋工程设计研究院,上海200011 [2]湖南大学电气与信息工程学院,湖南长沙410082

出　　处：《微电机》2024年第10期20-25,共6页Micromotors

基　　金：湖南省杰出青年基金项目(2024JJ2024);国家自然科学基金项目(52077064)。

摘　　要：针对电机运行过程中故障信号数据量少的问题,本文提出了一种基于Vision Transformer的永磁同步电机智能故障诊断方法。该方法首先通过格拉姆矩阵(Gram)、相对位置矩阵(RPM)方法将传感器获取的一维时序信号数据转换为二维图像数据,然后将矩阵图像数据作为ViT-B/16网络的输入进行故障诊断。经过实验验证,该方法能够对永磁同步电机正常、轴承故障、退磁故障等8种状态进行识别和分类,其中使用Gram矩阵图像作为该方法输入的准确率达到99.2%,使用RPM矩阵图像作为输入准确率达到99.6%,均高于AlexNet、VGG16、ResNet等卷积网络的故障分类准确度,证明该方法可有效提高永磁同步电机故障诊断的准确度。Aiming at the problem of small sample data of fault signal during motor operation,this paper proposed an intelligent fault diagnosis method of permanent magnet synchronous motor(PMSM)based on Vision Transformer.Firstly,the one-dimensional time-series signal data acquired by the sensor was converted into two-dimensional Gram matrix and RPM matrix image data by the Gram Matrix(Gram)and Relative Position Matrix(RPM)methods,and then the matrix image data were used as inputs to AlexNet,VGG16,ResNet and Vision Transformer networks for fault diagnosis respectively.After experimental validation,our method successfully identified and classified eight states of the PMSM such as normal,bearing fault,and demagnetizing fault.Using the Gram matrix image achieves an accuracy of 99.2%,while using the RPM matrix image achieves 99.6%.These accuracies are higher than those achieved by convolutional networks such as AlexNet,VGG16,and ResNet in fault classification.This demonstrated our method effectively enhanced the accuracy of fault diagnosis for PMSM.

关 键 词：二维图像 Vision Transformer 电机故障诊断 

分 类 号：TM351[电气工程—电机] TM341