Optimal design of composite structures using efficient metamodels

Abstract

In this paper, an efficient and accurate optimization technique using metamodels was suggested. The main concerns in system optimization are the guarantee of global optimum and the reduction of computational cost. Many researches have been studied, but there is none that satisfied the both objectives. In general, an accuracy of solution conflicts with a computational cost in optimization process. An acquisition of the two goals is very hard, and the counterproposal for this problem is metamodel. Metamodel is an approximation technique for system response, so an acquisition of fast response is possible. Among various metamodels, RSM (Response Surface Method) and MLSM (Moving Least Square Method) were used in order to find an optimum and to approximate the system response respectively. In order to increase the accuracy of RSM regression model, a design domain was divided to many subspaces, and RSM was applied to per subspace. The RSM for subspaces was named as DSRSM (RSM for Divided Subdomain). An efficient and accurate regression model may be constructed using MLSM, and an appropriate experimental design and a reasonable error measure for effective MLSM regression model were suggested and evaluated. The basic perfotinance of the suggested new optimization technique was verified by applying to a test function. The postbuckling analysis for composite stiffened structures is a representative nonlinear analysis that requires a large computational time and is difficult to get analysis results. A stable, accurate, and efficient postbuckling analysis module for the structures was studied. In order to predict the behavior in postbuckling region, progressive failure analysis was inserted and the results were compared with those of experiment. Very large computational time was consumed in postbuckling region, so some accelerating analysis techniques were implemented such as a change of element, a separation of the linear response region, and a termination of analysis afte...