Research On Helicopter Flight Condition Recognition Technology Based On Machine Learning

While helicopter flights in different condition,the damage level of life parts and moving parts is different.Therefore,identifying the flight condition correctly is of great significance in life expectancy and malfunction diagnosis of the key parts of the helicopter.In practice,the sample used for helicopter flight condition training are generally small sample,however,traditional neural network methods have poor recognition rates while training samples are few.In view of this problem,this paper uses the binary tree SVM and random forest method to research the helicopter flight condition recognition technology.It aims to improve the recognition rate and recognition speed,which can provide the core technology for the development of helicopter health and use monitoring system(HUMS).The main research work and achievements are as follows:(1)Research and implement the pretreatment of helicopter flight condition identification.It includes preprocessing of data,extraction of sensitive parameter and pre-classification of condition.The preprocessing of data is used to denoise the flight data by using the off-road point,the limiting and median filtering.Then,The least squares method is used to fit the rates of changes of flight parameters,which are treated as the new flight parameters.The denoising experiment verifies the validity of the method.The extraction of sensitive parameter is based on helicopter control characteristics and flight parameters linear correlation,and the validity of the method is verified by the real flight parameter data.Taking advantage of the selected sensitive parameters,35 helicopter flight condition is pre-classified into 10 subclasses.The validity of the pre-classification method is verified by the real flight parameter data.(2)Propose and implement helicopter flight condition identification based on binary tree SVM.Based on the pretreatment of condition recognition,firstly,the binary tree SVM classifier is designed for each subclass.Then,the binary tree SVM is optimized by particle swarm optimization and genetic algorithm,which improves the recognition rate.Finally,Each binary tree SVM classifier is subjected to sample training,and a trained network model is used for helicopter flight condition identification.This paper takes advantage of the real helicopter flight data as experimental data,and compares this method with SVM method and RBF neural network.The results show that the binary tree SVM has significantly improved the helicopter flight condition recognition rate in the case of small sample training.However,the recognition speed of this method is not fast.In view of this problem,the flight condition recognition method is further researched by taking advantage of the characteristics of Strong generalization ability and fast training convergence rate of random forest in the case of small sample.(3)Propose and implement helicopter flight condition identification based on random forest.Based on the pretreatment of condition recognition,firstly,the random forest classifier of each subclass is designed.Then,the random forest is constructed by using the classification regression tree,and each random forest classifier is trained.Finally,Good network model is used for identifying helicopter flight condition.This paper takes advantage of the real helicopter flight data as experimental data,and compares this method with the binary tree SVM method and the RBF neural network method.The results show that the recognition speed of the random forest is better than that of the binary tree SVM and RBF neural network in the case of small sample training.At the same time,the recognition rate of this method is similar to that of binary tree SVM,which is obviously higher than that of RBF neural network.