一种基于深度卷积神经网络的牵引变流器故障诊断技术

Application Research on Deep Convolution Neural Network Based Fault Diagnosis Technology for Traction Converter

变流器是电力机车牵引系统的关键部件之一,其故障易导致列车运行瘫痪,是机车故障中危害性较大的一类。针对基于专家经验的牵引变流器故障诊断仿真模型和特征选取泛化性较差问题,文章提出一种基于深度卷积神经网络的故障诊断方法,其通过修改Xception模型卷积和池化层结构参数以匹配牵引变流器故障数据并进行训练。实验结果显示,本文所提方法Top-1准确率为0.8422,Top-3准确率为0.9201,表明将深度卷积神经网络用于牵引变流器故障诊断具有较好的鲁棒性和准确性,且通过通道增强后可以提高模型的泛化能力并实现故障分类。

Converter is a key component of traction system in electric locomotive.The fault of converter can easily lead to the paralysis of train operation and is one of the most dangerous failures of electric locomotive.In order to avoid poor generalization of feature selection in expert experience and simulation mode in traction converter fault diagnosis,this paper proposes a fault diagnosis method based on deep convolution neural network.Structure parameters in convolutional and pooling layer in Xception model are modified to match the fault data of the traction converter for model training.Experimental results show that the accuracy of Top-1 is 0.8422 and the accuracy of Top-3 is 0.9201,which indicates that the proposed method is robust and accurate for fault diagnosis of traction converter,and channels enhancing can provide better generalization for model and realize fault classification.