一种基于GAMF-CNN的航空发电机整流器故障诊断技术  

A technique for aerospace generator rectifier fault diagnosis based on GAMF-CNN

作　　者：崔江[1] 周凡 陈永凡 于立 张卓然[1] Jiang CUI;Fan ZHOU;Yongfan CHEN;Li YU;Zhuoran ZHANG(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China)

机构地区：[1]南京航空航天大学自动化学院,南京210016

出　　处：《航空学报》2024年第24期231-242,共12页Acta Aeronautica et Astronautica Sinica

基　　金：国家自然科学基金(51977108)。

摘　　要：针对航空发电机整流器实际故障样本数据较少等问题,提出一种基于格拉姆角积场-卷积神经网络(GAMFCNN)的航空发电机整流器故障诊断技术。首先,采集整流器故障原始信号并进行预处理,将一维时序信号转化为格拉姆角积场图像,从而可以使故障诊断问题转变为图像识别问题;其次,借助深度迁移学习理念,采用卷积神经网络将仿真获取的故障特征知识迁移至缺少故障数据的发电机整流器中;最终,解决小样本数据下的航空发电机整流器故障诊断问题。经实验验证并与现有一些方法对比发现,所设计方法能以较高准确率实现故障二极管的诊断和定位。To address the problems such as less actual sample data of aerospace generator rectifier faults,a fault diagnosis technique based on Gramian Angular Multiply Field-Convolutional Neural Network(GAMF-CNN)is presented.First,original rectifier fault signals are collected and preprocessed,and the one-dimensional time series signals are transformed into GAMF images,so that the fault diagnosis problem can be transformed into an image recognition problem.Second,with the help of deep transfer learning concept,a convolutional neural network is used to transfer the fault feature knowledge obtained from the simulation to the real generator rectifier that lacks fault data.Finally,the aerospace generator rectifier fault diagnosis problem with small sample data is solved.Experimental verification and comparison with some existing methods find that the propsoed method can realize diagnosis and localization of faulty diodes with high accuracy.

关 键 词：航空发电机 整流器 故障诊断 格拉姆角积场 迁移学习 

分 类 号：V240.2[航空宇航科学与技术—飞行器设计]