一种基于改进堆栈自动编码器的航空发电机旋转整流器故障特征提取方法

A Fault Feature Extraction Method of Aerospace Generator Rotating Rectifier Based on Improved Stacked Auto-encoder

提出一种基于灰色关联度分析优化堆栈自动编码器的故障特征自适应提取方法,并用于航空发电机的旋转整流器二极管故障诊断中。首先,采集发电机交流励磁机励磁电流信号;其次,借助灰色关联度和深度学习理论对堆栈编码器网络进行训练学习,以确立其较优的网络结构,通过该网络可以自适应地从励磁电流信号中提取故障特征;训练完毕,借助于支持向量机(support vector machine,SVM)分类器实施故障诊断。对所提方法与快速傅里叶变换方法进行了仿真和物理实验,并对分类性能进行比较。结果表明,所提方法自动化程度高,自适应性能好,所提取的特征用SVM评估可以取得很好的分类效果。

This paper proposed a fault feature extraction method based on the stacked auto-encoder （SAE）, which is optimized by the grey relational analysis （GILA）. This method can extract fault features from raw data adaptively, and this method can be applied to fault diagnosis of rotating rectifier diodes in aerospace generator. First, filed current of aerospace generator excitation is collected. Second, the deep learning theory, combined with the grey relational analysis, is adopted to train the auto-encoder for achieving a good network structure of stack auto-encoders, which can extract the fault features adaptively from the generator current data information. Finally, fault diagnosis can be implemented with the support vector machine classifier. The performances of the presented method were compared with fast Fourier transform （FFT） method through simulations and physical experiments. The experiment results showed that the presented fault extractor is automatic and adaptive, and the achieved features with this method can be evaluated ideally with the support vector machine classifier.