Reliability-based design optimization (RBDO) has been developed to consider uncertainty of input design variables during optimization process. To provide the reliability, reliability index approach (RIA) and performance measure approach (PMA) are often used.[
] However, these reliability analyses usually require extremely expensive computational costs due to many simulation runs. Thus, it is necessary to reduce significantly the number of actual simulation runs during RBDO. Metamodels such as response surface model and kriging model are investigated for this purpose [
Metamodel for computer simulation is often built from space-filling sampling that evenly locates sample points within whole design domain. However, it requires considerably many sample points to approximate probabilistic constraints throughout whole design region when constraints reveal nonlinearity and when feasible region is small compared to whole design region.
In this research, constraint boundary sampling technique is proposed to maximize accuracy and efficiency of metamodel-based RBDO. Constraint boundary sampling is sequentially to locate sample points around constraint boundary by using kriging metamodel and its mean squared error. To verify the proposed method, mathematical examples are performed and their accuracy and efficiency are compared to those obtained from classical space-filling design. Through this study, we learn that RBDO using kriging model under the constraint boundary sampling technique coincides precisely with the exact solutions. Moreover, the efficiency of RBOD is improved so that RBDO using constraint boundary sampling technique can reduced by about 50% compared to conventional RBDO in the number of actual response analysis.