Research On Fault Diagnosis Technology Of Aircraft Three-stage Generator Rotating Rectifier

Aircraft generator,as the key equipment in aircraft main power supply system,is becoming more and more important.Faulty aircraft generator can cause aviation accidents,and the research on reliability of aircraft generator has been concerned in recent years.Rotating rectifier is a crucial but fragile part in aircraft generator.Hence,it is meaningful to implement fault diagnosis research of rotating rectifier in terms of practical and economy values,and this can ensure the safety of aircraft flight.This thesis is composed of theoretical analysis and experiments,which are as follows:(1)First,the paper studies on the fault feature extraction method of rotating rectifier of aircraft generator.The fault feature extraction method based on current information is considered.Conventional methods mainly depend on signal processing theories and engineering experience of field experts,and the presented method applied deep learning theory to fault feature extraction of rotating rectifier,and the fault features can be extracted by stacked auto-encoder adaptively.(2)Secondly,the network structure of stack auto-encoder is mainly selected manually,and this thesis proposes a grey stacked auto-encoder,which can combine grey relational analysis with common stacked auto-encoder,and this method can design the network structure automatically.(3)Thirdly,some traditional fault classification methods for rotating rectifier is slow,so extreme learning machine is introduced to faults classification of generator rotating rectifier.The conventional extreme learning machine has some drawbacks,and a new extreme learning machine,which is optimized by genetic algorithm,is proposed in our investigations.(4)Finally,an aircraft three-stage generator model is set up in Matlab/Simulink software,and a test rig for three-stage generator rotating rectifier open-circuit fault is also built for data collection and methods verification.The algorithms proposed in this paper are verified with simulation and physical experiments.