Research On Helicopter Flight Condition Recognition Method And Application Based On Deep Belief Networks

Helicopters have a large number of moving parts and life parts.The degree of damage of these structural parts is closely related to the helicopter flight status.Therefore,obtaining helicopter flight status can provide important basis for fault diagnosis and life prediction of key components.The traditional recognition method can identify the limited variety of helicopter flight states and it is difficult to meet actual needs.To solve this problem,this paper uses shallow learning method RVM and deep learning method DBN to study the helicopter flight state recognition technology.The research work and achievements of this paper are as follows:(1)Outline the background of the topic.Firstly,it studies the methods of helicopter flight status recognition and summarizes the advantages and disadvantages of various methods both at home and abroad.Secondly,it introduces several common methods of deep learning.Finally,using deep learning techniques,the helicopter flight status recognition based on DBN is proposed.(2)Helicopter flight state recognition preprocessing is implemented.Firstly,the wild point and noise are removed by wild point elimination,missing parameter filling and data smoothing;Secondly,the operational characteristics of the helicopter flight state and the correlation between flight parameters are analyzed to extract the sensitive flight parameters;finally,the helicopter of flight status are pre-classified.(3)Research on helicopter flight status recognition method based on RVM.The traditional method of analysis image can only recognize a few flight states related to attitude angles.This paper uses real flight data of helicopter to establish an RVM classification model.Firstly,combining the pre-classification results,the RVM classifiers are designed for each small class;Secondly,the particle swarm optimization algorithm is used to optimize the RVM nuclear parameters,and the flight state recognition rate is improved;finally,the RVM model is trained using the sample set and completed Single-point recognition experiments and full-landing experiments.The experimental results show that this method can identify most of the flight conditions.(4)Using the powerful feature extraction capabilities of deep learning,a method of helicopter flight state identification based on DBN was proposed and implemented.Firstly,the DBN network structure and feature extraction process are introduced.Secondly,based on preprocessing and pre-classification,DBN classifiers are designed for each small class.Secondly,using training samples,experimental methods are used to determine each DBN network;Finally,the determined network structure parameters are input into the model,and the validity of the method is verified by a single point recognition experiment,a full-landing experiments,and a comparative experiment.Experimental results show that compared with RVM method,DBN network can achieve higher recognition rate.