Research On Flight Training Quality Evaluation Based On Flight Data

The training of pilot students and the improvement of training quality are the key concerns of the civil aviation industry.In order to solve the subjective rating of flight instructors in traditional flight training evaluation and improve the efficiency of evaluation,this thesis takes the take-off and landing segments of rectangular take-off and landing routes as the research objects,combines the flight parameters recorded by the Garmin1000 avionics system of the Cessna 172 training aircraft,and quantifies the flight training quality of different trainees based on the time series similarity method.After that,the LSTM neural network was used to establish the flight training quality evaluation model.This thesis focuses on the following aspects.1.Sample data extraction of rectangular take-off and landing routes.Aiming at the operational changes of flight trainees in different phases of rectangular take-off and landing routes,a phase division algorithm was firstly designed to extract a total of four phases,including take-off glide,one side,five sides and landing glide;then the required sample data was extracted and analyzed,and the missing values were processed by using Hampel filter function,and finally the data curves were fitted.2.Evaluation scheme of rectangular take-off and landing routes.According to the expert evaluation method and the influencing factors of flight training quality,five flight parameters including track,altitude,airspeed,slope and attitude were selected for research;with reference to the experience of senior flight instructors,the weights of different parameters in different phases were determined,so that a combined subjective and objective evaluation scheme was established.3.A quantitative model based on the similarity of time series.The parameter curves of each trainee are segmented and dimensionally reduced,and then the angular set was used to represent the processed time series.After that,the similarity evaluation value of each trainee was determined by comparing the actual flight parameter curves and target curves of different trainees from a total of three indexes: angular distance measure,time difference measure and function value difference measure.Finally,the results were compared with the expert scores.4.Flight training quality evaluation model based on LSTM.The K-means algorithm was used to cluster all flight trainees according to the similarity rating values of the samples to obtain the flight training quality rating table of all trainees and transformed it into the label data of each trainee;the rating data and label data of the samples were imported into the LSTM neural network as input data for training to obtain the flight training quality classification evaluation model to realize the classification of the test set trainees.The simulation results show that the model established in this paper can quantify the flight training quality of flight trainees by five flight parameters and is consistent with the expert evaluation results.It can accurately classify the flight training quality of flight trainees with an accuracy rate of 80%.The above research can establish a more comprehensive flight training quality evaluation system and provide relevant improvement measures for the relevant personnel of flight training,which can help to further improve the flight training quality management of the whole life cycle of flight trainees and improve the flight training quality of trainees.