基于FrFT-CNN的变频电机匝绝缘状态监测  

Inverter-fed Machine Turn Insulation Condition Monitoring Based on FrFT-CNN

作　　者：范瑞天 李豪 FAN Ruitian;LI Hao(School of Electrical Engineering,Shanghai University of Electric Power,Shanghai 200090,China)

机构地区：[1]上海电力大学电气工程学院,上海200090

出　　处：《上海电力大学学报》2024年第5期468-474,共7页Journal of Shanghai University of Electric Power

基　　金：国家自然科学基金(51907116);上海市自然科学基金(22ZR1425400)。

摘　　要：定子绕组匝绝缘劣化是导致变频电机故障的主要原因之一。通过在线监测其状态,可以及时发现潜在的安全隐患,但会面临匝绝缘劣化特性微弱的挑战。为了提高匝绝缘劣化灵敏度,提出了一种利用分数阶傅里叶变换结合卷积神经网络(FrFT-CNN)的变频电机匝绝缘状态监测方法。首先,在分数域内分析了高频开关振荡电流对匝绝缘变化的灵敏性。然后,设计了一种适用于变频电机匝绝缘状态监测的一维卷积神经网络模型。实验结果表明,与传统卷积神经网络方法相比,FrFT-CNN方法显著提高了匝绝缘状态监测的准确率,且兼具稳定性较高的优点。The degradation of stator winding turn insulation is one of the primary causes of inverter-fed machine failures.Online monitoring of the stator winding turn insulation degradation can detect potential safety hazards in time,but faces the challenge of weak turn insulation degradation characteristics.In order to improve the sensitivity of turn insulation degradation,a turn insulation condition monitoring method for inverter-fed machine using fractional-order Fourier transform combined with convolutional neural network(FrFT-CNN)is proposed.Firstly,the sensitivity of the high-frequency switching oscillating current to the turn insulation change is analyzed in the fractional domain.Then,the one-dimensional convolutional neural network model suitable for turn insulation state monitoring of inverter motors is designed.The experimental results show that the FrFT-CNN method significantly improves the accuracy of turn insulation state monitoring compared with the traditional convolutional neural networks method,and also exhibits the advantage of higher stability.

关 键 词：匝绝缘 变频电机 分数阶傅里叶变换 卷积神经网络 状态监测 

分 类 号：TM307[电气工程—电机]