摘要
[目的]为准确诊断永磁同步电机(PMSM)的匝间短路和均匀退磁等故障,[方法]搭建永磁同步电机故障模拟试验台架,通过重新绕组电机定子模拟匝间短路,通过更换不同尺寸永磁体模拟均匀退磁故障,在10种特定的转速和负载工况下对匝间短路故障、均匀退磁故障及耦合故障进行模拟。模拟过程采用单通道、多尺度和多输入卷积神经网络(CNN)等3种流行的CNN架构,以及7种主流图像化方式。[结果]结果表明:在3种流行架构中,多输入CNN网络架构在任何情况下均具有最好的诊断效果;相较于时域和频域图像,采用时频域图像的诊断精度较高;大部分情况下,采用大尺寸图像的诊断效果好于小尺寸图像;多输入CNN网络架构的诊断精度随着信号数目的增加而提高,而单通道CNN和多尺度CNN则恰恰相反。[结论]研究成果可为永磁同步电机故障诊断提供一定参考。
[Purpose]In order to accurately diagnose faults such as inter-turn short circuit and uniform demagnetization in permanent magnet synchronous motors(PMSM),[Method]a permanent magnet synchronous motor fault simulation test bench is built,which simulates inter-turn short circuit by rewinding the motor stator and simulates uniform demagnetization faults by replacing permanent magnets of different sizes.Under 10 specific speed and load conditions,inter-turn short circuit faults,uniform demagnetization faults,and coupling faults are simulated.The simulation process adopts three popular convolutional neural network(CNN)architectures:single channel CNN,multi-scale CNN,and multi input CNN,as well as seven mainstream visualization methods.[Result]The results indicate that among the three popular architectures,the multi input CNN network architecture has the best diagnostic performance in any situation;compared with time-domain and frequency-domain images,the diagnostic accuracy of using time-frequency domain images is higher;In most cases,the diagnostic effect of using large-sized images is better than that of small-sized images,the diagnostic accuracy of multi input CNN network architecture improves with the increase of signal number,while the single channel CNN and multi-scale CNN are exactly the opposite.[Conclusion]The research results can provide some guidance for fault diagnosis of PMSM.
作者
闫国华
胡以怀
YAN Guohua;HU Yihuai(College of Mechanical and Automotive Engineering/Hangzhou Bay Automotive Engineering,Ningbo University of Technology,Ningbo 315211,Zhejiang,China;Merchant Marine College,Shanghai Maritime University,Shanghai 201306,China)
出处
《船舶工程》
北大核心
2025年第1期114-123,共10页
Ship Engineering
关键词
永磁同步电机(PMSM)
匝间短路
退磁故障
卷积神经网络(CNN)
信号图像化
permanent magnet synchronous motor(PMSM)
inter-turn short circuit
demagnetization fault
convolutional neural network(CNN)
signal imagenization
