ABSTRACT

Meta-Model-Based Design Optimization (MBDO) has been the mainstream method to achieve optimization of the Constrained Black-box system. Typical methods such as Efficient Global Optimization (EGO) emerged as one of the most promising methods for expensive simulators (Jones et al. 1998). EGO is based on the approximation of responses using a Gaussian process model so that the variance of the prediction is available over the whole design space. Expected Improvement (EI) of the object function can also be assessed; hence, the optimum value can be found by maximizing the EI with a global search. Recently, EGO has also developed Efficient Global Reliability Analysis (EGRA) [Bichon BJ et al. 2008]. The EGRA is based on expectation improvement. To improve

1 INTRODUCTION

In recent years, RBDO has been studied to design safer and more reliable products, which take the uncertainty of loadings, material properties, environment, and other parameters into consideration. Reliability analysis aims to quantify the uncertainty of the output according to the uncertainty of the system inputs, environment, system itself, and so on. On the contrary, the relationship between inputs and outputs becomes increasingly complicated, and mostly they are black-box systems in the practical engineering application, which makes the aforementioned relationships highly nonlinear and not able to be characterized with explicit functions. Therefore, it is very important do more research on the RBDO problems of black-box systems.