Various optimization methods, such as Rosenbrock method (RBM), simulated annealing method (SAM), genetic algorithm (GA), evolution strategy (ES), are trying to be used in the optimal design of electrical machines and electronic equipment. In order to apply these optimization methods to the practical design of magnetic devices using the finite element method, problems in using the optimization methods, such as acceleration technique of the convergence of optimization method, effect of finite element subdivision on the obtained optimal value, should be investigated.
In this paper, effects of initial value, searching region, convergence criterion, number of elements on the obtained optimal value are examined using the die press model (Problem 25) which is the standard model of TEAM International Workshop. It is shown that the fine mesh having about 1% accuracy of flux density is necessary in order to obtain the optimal dimension having about 1% accuracy. The number of iterations of RBM and SAM can be reduced by using the initial value which is obtained using the experimental design method (EDM). Finally, features of RBM, SAM, ES and combination methods are compared with each other from the viewpoints of obtaining the global optimal value and quick convergence.
抄録全体を表示