Abstract
In the process of the radial basis function network (abbreviated as RBFN), two design points are added in each iteration to update the approximated surface, one is optimum design point obtained by former optimization step, and the other is the sparse design point to increase the global precision of approximated constraints surface. Naturally, it is desirable that the added design point is close to the active constraints surface. Without newly performing the structural analysis, it is difficult to find if the added design point is close or not. In this study, the support vector machine (abbreviated as SVM in the following) is tried to apply to discriminate whether the design point is close to the surface or not. SVM is currently attracting attention as a pattern classification procedure with a learning function. This SVM has excellent cognitive abilities and is currently applied in a variety of engineering fields. SVM is also characterized by the ease in which it obtains the distance from the discriminant plane. In this study, the possibility of SVM as the support system of constraint approximation in the process of structural optimization is studied with several numerical simple examples.
