Research On Feature Extraction Technology For Rotating Rectifier Of Aircraft Generator Based On Learning Network

With the development of multi-electric(all-electric)aircraft,the demand for electric power in aircraft electrical systems is increasing.Brushless generator are becoming the main power supply equipment for aircraft power systems,and their status has become increasingly prominent.Rotating rectifier is a key component of aircraft generator and one of the most prone to fault.In recent years,aviation accidents caused by aircraft generator failures have occurred from time to time.Therefore,it is of great significance and application value to study the fault feature extraction and diagnosis technology of aircraft generator rotating rectifier to ensure the stable operation of the aircraft electrical system.This paper mainly compares two different networks of deep learning and broad learning in the fault feature extraction rotating rectifier,and transplant the deep belief network is ported to the digital signal processor for online fault diagnosis:(1)First,the theoretical research on the conventional fault feature extraction method is carried out.Aiming at the fault feature extraction based on the deep learning method requires large amount of data,training process requires large amount of hardware resources and time and other issues.A method based on particle swarm optimization for fault feature extraction of deep belief network rotating rectifier.This method can adaptively extract fault features.The corresponding digital signal processor system is designed to realize the online fault diagnosis of the method.(2)Second,based on the complexity of the deep learning network structure and the problem that the training consumes a lot of time,a new feature extraction and fault diagnosis method based on broad learning is introduced.Aiming at the characteristics of broad learning,a broad learning method based on grid search method is proposed.(3)Finally,the methods of deep learning and broad learning are applied to an aircraft three-stage generator simulation model and fault experimental platform.The algorithms proposed in this paper are verified with simulation and physical experiments.The optimized deep belief network method was tested online.The results show that the proposed and proposed methods can effectively solve the shortcomings of deep learning in feature extraction and improve the efficiency of feature extraction.