심층 신경망 기법을 이용한 유도탄 공력 하중 예측 모델 개발

Abstract

In the present study, prediction models for aerodynamic loads of missile configurations were developed using multi-layered perceptron. Aerodynamic coefficients were extracted automatically from Missile DATCOM. Sample points for parametric missile shapes were determined using Latin Hypercube Sampling. A multi-layered perceptron was constructed using Tensorflow. A hyperparameter set with minimum Mean Squared Error(MSE) was determined by genetic algorithm. The trained neural network model was also tested for a verification configuration by comparing the true and predicted values. Trained neural network models give accurate results with MSE for test data set between 10SUP-5/SUP and 10SUP-6/SUP, and relative error below 5%.