Study of Mathematical Programming Methods for Structural Optimization

摘要

Various programming methods for the optimal design of structural systems are studied, including the recursive QP, gradient projection, feasible direction, optimality criteria, sequential linear programming, SUMT, and multiplier methods. Computer codes which have been developed on the bases of these methods are used with existing programs to solve a set of test problems. It is shown that all primal methods, including recursive QP, gradient projection and feasible directions, have serious difficulties in determining step size and hence in being globally convergent. Transformation methods (penalty and multiplier methods) have very strong global convergences. It is found that the gradient of the transformation function can be computed efficiently.