BOUNDED-BUT-UNKNOWN UNCERTAINTIES IN DESIGN OPTIMIZATION

In the present paper, an adopted way to deal with uncertainties in the design is to find the combination of uncertainties which yield the worst design. This can be done by utilizing a description of uncertainty via bounds on the uncertainty variables. The focus is placed on analytic, asymptotic first-order approximations if uncertainty is small, the basis for which are sensitivities with respect to uncertainty variables. It is shown that a natural norm for the sensitivity emerges. If uncertainty is large, the worst uncertainties should be found by doing a more elaborate search using an anti-optimiza tion process. These alternative ways of dealing with uncertainty variables are embedded in a structural optimization setting using the Multipoint Approximation Method as optimization tool. Moreover, a method is proposed to construct a more efficient anti-optimization. The basis is the use of response surface techniques.