Dynamic radial basis function proxy model and its application in optimization design of reducer

DAI Jia-wei, LUO Wei-lin
(School of Mechanical Engineering and Automation, Fuzhou University, Fuzhou 350108, China)

Abstract: Aiming at the problem of poor global convergence and low computational efficiency of traditional static metamodel in multidisciplinary design optimization, an optimization strategy for dynamic metamodel was proposed by combining trust region and adaptive lower confidence bound. In this strategy, the initial sampling points were selected by optimal Latin hypercube test design method and the radial basis function metamodel was constructed. The trust region method and the lower confidence limit criterion were used to construct the dynamic sample space. According to the accuracy of the agent model, the trust region sampling space was updated. The adaptive equilibrium constant in the lower confidence bound was determined by the error between the predictive value of agent model and response value of real model. The lower confidence bound was optimized by using genetic algorithm. The optimization strategy was verified by using a numerical test problem and the NASA speed reducer optimization design. The results indicate that the proposed method is proved to be effective，and can not only guarantee the optimal solution, but also significantly improve the computational efficiency.

Key words: dynamic agent model（DAM）; radial basis function（RBF）; trust region; lower confidence bound; reducer