Structural Optimization Using Femlab And Smooth Support Vector Regression

Abstract

In recent years support vector machine (SVM) has been emerging as a popular tool for function approximation. Application of SVM for approximation of mathematical functions and complex engineering analysis has been represented by Palancz et al and Clarke et al, respectively. However the training of the original SVM involves the solution of a quadratic programming (QP) problem. This makes the application of SVM to large problem computationally expensive. To circumvent this difficulty, Lee et al developed a more efficient SVM formulation namely epsilon-SSVR which drastically improved the training efficiency of SVM.

In this research the SSVR is used to build a metamodel for structural optimization. In the proposed method, Quasi Monte Carlo (QMC) technique is used for the selection of training data in the design space. SSVR using a radial basis function kernel is used to build the metamodel for structural optimization. The structural responses are evaluated by a commercial finite element package, FEMLAB (recently renamed as COMSOL). Several structural optimization examples are presented to illustrate the effectiveness of the proposed approach.