ABSTRACT
The current work extends the previous framework of Liu & Collette (2015) in two ways. First, the optimization is now multi-objective with the width of the uncertainty interval treated as both a design variable and a component of one of the objective functions. This allows the trade space between structural performance and the amount
1 INTRODUCTION
The common practice in engineering design optimization under uncertainty is to identify the sources of uncertainty and then adopt mathematical distribution models to quantify the effects of uncertainties in the design. However, decision making under uncertainty in the early stage design is troubled by the absence of information about exact form and distribution of the uncertainty. In this paper interval uncertainty (Moore 1966) is proposed to describe uncertainty in those situations. In the general case, such uncertainty may be present in both design parameters as well as stochastic or reliability models used to estimate structural adequacy. Such uncertainty can only be reduced by higher-fidelity engineering analysis. Thus, it is natural to inquire which source of uncertainty has the dominant impact on the performance of the final design. This paper presents a surrogate-assisted multi-objective optimization framework to rapidly determine the trade space between the width, or range, of the intervals in the uncertainty measures and the structural performance of the final design. A case study is presented of a box girder design where interval uncertainty exists in both the configuration of the girder as well as the reliability-based strength constraints applied to the girder design. The trade space between width of the uncertainty interval and structural weight is developed both with and without the surrogate modeling technique.
