基于卷积神经网络的六相永磁同步电机系统开路和电流传感器故障诊断  

A Fault Diagnosis Method based on Convolutional Neural Network for Open Circuit and Current Sensor Fault of Six-phase PMSM System

作　　者：王志文[1] 郑晓钦[1] WANG Zhiwen;ZHENG Xiaoqin(Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学,山东青岛266071

出　　处：《大电机技术》2025年第2期1-8,17,共9页Large Electric Machine and Hydraulic Turbine

基　　金：国家自然科学基金(U2106217,52277058和52037005)。

摘　　要：为了实现多相永磁同步电机(Permanent Magnet Synchronous Motor,PMSM)系统多种故障状态的准确诊断,进而提高电机驱动系统的可靠性,本文采用基于一维卷积神经网络(one-dimensional Convolutional Neural Networks,1D-CNN)的故障检测算法,该算法能够准确识别六相永磁同步电机系统的正常状态,以及不同位置的开路故障、电流传感器故障(包括增益、漂移、卡滞、断线),并通过原始相电流数据进行状态诊断。首先,分析了电机系统在不同故障状态下的故障特征;其次,基于卷积神经网络原理,在仿真平台中构建一个多层宽卷积核的卷积神经网络,基于六相永磁同步电机特性合理设计其卷积层和池化层的结构,并确定了其相关超参数的值;再次,根据六相永磁同步电机解耦数学模型,搭建六相永磁同步电机仿真系统,并收集其在多种状态下的六相电流数据制作数据集并划分训练集和测试集;最后,将相电流数据集输入卷积神经网络模型进行模型训练并测试,测试结果表明其故障诊断准确度达到了99.98%。In order to achieve precise diagnosis of multiple fault states in multiphase Permanent Magnet Synchronous Motors(PMSM)system and enhance the reliability of motor drive systems,onedimensional Convolutional Neural Networks(1D-CNN)is employed in this paper,the algorithm accurately identifies the normal state of the six-phase PMSM system,as well as open circuit faults,current sensor faults including gain,offset,stuck and loss signal at various positions,using raw phase current data.Firstly,the fault characteristics of the motor system in different fault states are analyzed.Secondly,a convolutional neural network featuring multi-layer wide convolution kernels is constructed on the simulation platform,based on the principles of CNN.The architecture of the convolution and pooling layers is thoughtfully tailored to the unique characteristics of the six-phase PMSM,with determination of relevant hyperparameters.Thirdly,with the decoupling mathematical model of the six-phase PMSM,a simulation system for six-phase PMSM is developed,and the phase current data in different states are collected to create a dataset for training and testing purposes.Finally,the phase current dataset is fed into the 1D-CNN model for training and testing.The test results show that the fault diagnosis accuracy reaches 99.98%.

关 键 词：永磁同步电机 开路故障 电流传感器故障 故障诊断 卷积神经网络 

分 类 号：TM341[电气工程—电机]