Complete decoupled adaptive robust design optimization

Abstract

Robust design optimization (RDO) can provide an optimum design solution that is relatively insensitive to input uncertainties, thus possesses significant value in the design of engineering structures. Generally, the RDO problem is a nested double-loop optimization process which requires huge computational costs especially for the case with complicated finite element model simulation. In view of this point, a completed decoupled adaptive RDO method based on the Kriging surrogate model is proposed, which can transform the nested double-loop uncertainty optimization process into the traditional deterministic optimization procedure. At the same time, closed-form expressions of the mean and standard deviation of performance functions under different design parameters are deduced and employed in the uncertainty propagation during the design optimization so to alleviate the post processing computational error. Moreover, the metamodeling uncertainty is considered in the established uncertainty propagation technique and an adaptive framework is introduced to improve the computational accuracy of uncertainty propagation so to further guarantee the estimation accuracy of RDO problems. An engineering application is introduced to illustrate the effectiveness of the established complete decoupled adaptive RDO method.