Surrogate Model Theory And Application In The Aircraft Mdo

Surrogate model technology is an important research tool and working process of the aircraft MDO. It can significantly reduce the time consuming in the design and optimization of various disciplines, help the complex MDO optimization process become feasible, and ultimately increase efficiency of the overall design optimization process in condition of the design accuracy. The main research object of this paper is just the surrogate model technology. By researching the improving technology of the existing surrogate models or employing absolute new surrogate models, we can reach a kind of surrogate models which have strong generalization abilities; as a result the research of the following aspects will be carried out in this paper:(1) The available surrogate models were summarized. The advantage and disadvantage of each kind of model was analyzed in order to guide the practical applications, so that surrogate model with best performance will be chosen on condition of the generalization abilities.(2) The benefits and drawbacks of the basic learning algorithm of BP neural network have been analyzed. Owning to the global optimization characteristics of the genetic algorithm, Hybrid learning algorithm combining the gradient algorithm and genetic algorithm of best samples inheritance was put forward. Function test was carried out to validate the learning efficiency and the superiority of the improved BP neural network.(3) The characteristics of the basic Radial Basis Function (RBF) interpolation surrogate model were analyzed. It was pointed out that the RBF interpolation surrogate model is essentially one of the high-dimensional classifier when the interpolation basis functions satisfy the Mercer condition. The Structural Risk principle was introduced into the RBF surrogate model, and the relationship between Structural Risk and the constant parameter of the interpolation basis functions was found, thus it can achieve better generalization ability by choosing the optical const parameter, which is the Structural Risk Minimization based RBF surrogate model. Finally several function tests were carried out to validate the generalization ability of the improved surrogate model.(4) The characteristics of the interpolation-based surrogate model were analyzed. It was pointed out that the main factor affecting the generalization ability of the surrogate model was the oscillation characteristics. In order to quantitatively describe the oscillation feature, Energy Function Method was put forward, which can evaluate the generalization ability of the surrogate model. At last, a number of function test examples were utilized to verify the validity of the Energy Function Method.(5) The characteristics of the Energy Function Method were analyzed, it was pointed out that the Energy Function was essentially the variational problem with the sample point as the constraint. Interpolation function of minimal Energy Function can be derived from the variational problem. Usually the solving method of this problem was finite element method, so that a new surrogate model was derived which was call Finite Element Interpolation surrogate model. The one-dimensional and two-dimensional Finite Element Interpolation surrogate model has been derived in this paper. Finally several numeral example tests ware carried out to verify the generalization ability.(6) The multi-disciplinary design optimization of aerodynamic and structural of the horizontal tail of one aircraft has been researched. The Structural Risk Minimization RBF surrogate model was employed in the aerodynamic optimization, structural optimization and the aerodynamic and structural based MDO; the high efficiency of using the Structural Risk Minimization RBF surrogate model in optimization process has been verified by the applications. The design result shown that the aerodynamic and structural based MDO tend to have a more optimal result than the single discipline based optimization.