Comprehensive study of typical metamodel methods applied in aircraft multidisciplinary design optimization

Metamodels have been widely applied in aircraft multidisciplinary design optimization (MDO) to alleviate the computation burden and improve the optimization efficiency. At present, there are various metamodel methods available, such as response surface method, Kriging model, moving least squares method, radial basis function, neural networks and so on. However, it is difficult to confirm which metamodel method is more promising, which is also an interesting question puzzled most designers. In this article, by using a series of numerical test problems with different inherent features, a comprehensive study is employed to compare five typical metamodel methods commonly applied in aircraft MDO under multiple metrics. In term of the comparison results, the conclusions are drawn and recommendations for selecting suitable metamodels in aircraft MDO practice are also summarized.