采用深度学习的异步电机故障诊断方法

A Fault Diagnosis Method for Asynchronous Motor Using Deep Learning

为解决传统异步电机故障诊断方法因电机结构复杂、信号非平稳和机械大数据等因素引起的诊断困难问题,提出一种高效准确的异步电机故障诊断(SDAE)方法。该方法利用堆叠降噪自编码提取信号特征,结合Softmax分类器实现高效准确的电机故障诊断。首先,采集异步电机的整体电流和振动信号,将电流信号与傅里叶变换后的振动频域信号组合构成样本,并做归一化处理;然后,构建堆叠降噪自编码网络,确定网络层数、各隐藏层节点数、学习率等参数;最后,输入训练样本依次训练自编码和分类器,微调整个网络并用测试数据验证网络的优劣。试验结果表明,在合适的参数下采用SDAE方法的异步故障诊断准确率高达99.86%,比传统电机故障诊断方法提升至少6%。

An efficient and accurate fault diagnosis method（SDAE）is proposed to solve the problem that the traditional fault diagnosis method for asynchronous motors has difficulty to diagnose caused by motors＇ complex structure,non-stationary signals and mechanical Big Data.The method extracts signal characteristics based on Stacked Denoising Autoencoder,and Softmax classifier is used to diagnose motor faults efficiently and correctly.Firstly,signals of vibration and current of an asynchronous motor are collected,and then transformed using Fourier Transform.Samples are obtained by combining these two kinds of frequency domain signals and then normalized;Then,the SDAE network is constructed and the layers of network,the number of hidden layer nodes and the learning rate are determined;Finally,the whole network is trained with training samples and is fine-tuned.The network is tested using validation samples and results show that SDAE network gains 99.86% accuracy in diagnosis of motor fault under appropriate parameters,which increases at least 6% compared with the traditional method.