Research On Helicopter Flight Condition Recognition Method And Application Based On Support Vector Machine

Due to special flight environment, damage degree of helicopter dynamic component and life components is different in various flight conditions, so helicopter fight condition is correctly identified, which has great importance for helicopter in fault diagnoses and life prediction. Due to the shortage of training data, fight condition recognition rate is low in practical traditional method. To solve above problem, the paper makes thorough study of helicopter flight condition using signal denoising, presort technology and SVM method on basis of a helicopter’s truth data. On this basis, helicopter flight condition recognition software system platform has been built, and health and usage monitoring system(HUMS) is provided core method which has indigenous intellectual property rights of Chinese technology. The primary research work and achievements of this paper are as follows:(1) The traditional methods and base theory was overviewed about helicopter flight condition recognition. Firstly, flight condition recognition methods were detailedly analyzed. Secondly, correlation background and basic theory was told by this task method. Finally, the relevance of helicopter handling basic principle, flight parameter and flight condition was analyzed.(2) A flight condition recognition method based on SVM(Support Vector Machines) was proposed. Firstly, the flight data was treated by clipping, removing the outlier data and average filtering. Secondly, change rate of flight data was obtained by least square line fitting, and data redundancy was decreased by helicopter handling characteristics and data linear independency used to extract the characteristics parameters. Thirdly, flight condition was divided into ten classes by characteristics parameters and SVM classifier was designed for each class. Fourthly, the parameter optimization of SVM kernel function was used by genetic algorithm, so as to improve the identification efficiency. Finally, every SVM classifier was trained by training samples, and all flight condition of helicopter was identified by trained SVM classifier. Many experiments using a helicopter’s actual flight show that compared with RBF neural network and Elman network method. The results indicate that this method could obviously improve the recognition rate in the condition of small sample.(3) Helicopter flight condition recognition software system had been developed. Firstly, the general design was realized according to the needs of HUMS which contained system architecture, interior interface and system interface. Secondly, structural detailedly design which contained presort scheme, trial management, network training, condition recognition and history was realized. Finally, each functional module was tested, and the system integration was implemented according to condition recognition method. The system integrated this method, RBF neural network method and Elman neural network method, which had strong stability and good scalability, providing central module for the HUMS.