Aircraft Multidisciplinary Design Optimization Approach For

In the process of aircraft multidisciplinary design optimization,every discipline needs many iterations to get the optimal result,and every computation of them needs a lot of time,so it block the progress of optimization seriously.In order to reduce the computational cost without losing accuracy in the result,we construct every discipline's approximation model via approximation methods,and it can reduce the much time that every discipline expends,then the process of aircraft multidisciplinary design optimization could be successful.If we substitute actual model for approximation model,the precision of approximation model is very important. While approximation model is very close to actual model,and the precision of approximation is very high,the result of optimization is significant,and authentic;contrarily,the result of optimization is insignificant,and unauthentic.The main work of the thesis consists of summarizing the characteristics of the approximation methods,compiling the programs of approximation methods,and getting availability of every approximation method according to examples. First,the paper gives the experiment design method for constructing approximation model.If we select experiment points by experiment design method, experiment points may be more uniform and more typical.,and it can prepare good foundation for constructing precise approximation models.Second, the paper summarizes Square Response Surface method, Moving Least Square(MLS) method,Radial Basis Function Neural Network(RBFNN),BP Neural Network and Kriging method,and also summarizes their characteristics. The five methods are tested on a 2-D test function ,a straight airfoil and a sweepback airfoil. For the straight airfoil,the paper constructs its approximation models of structure disciplinary, aerodynamic disciplinary and stealth disciplinary.For the sweepback airfoil,the paper constructs its approximation models of structure disciplinary and aerodynamic disciplinary.Finally the paper drawsconclusions for the five methods' availability. It is found that kriging method is better and more robust for every model,while Radial Basis Function Neural Network is suitable for the model of having many experiment points.The paper presents an algorithm combining Square Response Surface and Radial Basis Function Neural Network to solve the question that RBFNN is often difficult to meet the precision request of approximation model.The method minimizes extended error of the approximation model.It increases the approximation precision,and enhances the flexibility based on having the same sample points.It indicates that the method increases the approximation precision via examples,and it can increase design efficiency and quality in multidisciplinary design optimization(MDO).