Intelligent Diagnosis Of Hard Landing Based On Flight Training Data

Hard landing frequently occurs in flight training due to various factors,which often brings threats and damages to the safety of pilot student,landing gear,and the single judgment method is easy to miss judgment.Therefore,this paper conducts research on the judgment of hard landing.According to the current research status of hard landing,the intelligent judgment model is used to judge the hard landing in flight training.At present,support vector machine model is commonly used for intelligent judgment of hard landing,and QAR data at grounding time is used.However,the recording interval of Garmin-1000 avionics data is longer than QAR data.Therefore,long short-term memory neural network is proposed for intelligent judgment.From the perspective of process,the characteristics of the hard landing sample data were studied and compared with the judgment results of support vector machine.The whole paper was mainly divided into the following three parts:First: The flight training airport and instrument approach procedure are introduced,the characteristics and key parameters of flight training and landing stage are studied,and the definition of hard-landing of Cessna-172 R model is defined.Then,there are many factors that lead to hard landing.Based on landing characteristics,qualitative analysis is made from glide path,approach speed,leveling and grounding,and quantitative analysis is made during grounding from leveling to grounding and grounding to taxiing,and five parameters for judging hard-landing are obtained.Second: Garmin-1000 avionics data are outliers and missing in the recording process.In this paper,according to the characteristics of flight parameter data,the least square method based on polynomial was used to fit the outliers,and the outliers were identified in combination with interval average residuals,and the identification results were good.After that,the landing stage was identified and the data were extracted by using similar landing curve and the docking time was determined according to the force condition at grounding.Third: Using support vector machine(SVM)and the long short-term memory neural network respectively from the Angle of two grounding and grounding process to intelligent identification of hard landing,and to exclude support vector machine(SVM)model of factor influence on the results,using the grid search method to find suitable coefficient.By comparing the test results of the two models,it can draw from the final result of contrast with hard landing identification model of neural network to establish the length of recognition rate is better than support vector machine(SVM),is more suitable for hard landing in judge Garmin-1000 avionics data.