关键词: 设计空间差别处理, 全局最优化, 汽车轻量化设计, 耗时问题

Abstract: A meta-model based design space differentiation method (DSD) was proposed to improve the performance of traditional region elimination methods. In this method, a gradually reduced important region was constructed using the expensive points and the space to be removed in the region elimination methods was defined as the other regions and would also be searched. When the design spaces were divided, the quadratic function (QF) might be used in the search of the two subregions and several new expensive points would be selected for the updates of the QF meta-model both in the important regions and the other regions. Instead of region elimination, fewer new expensive points would be selected in the other regions, which might avoid the traps of the local minima. It is demonstrated by several benchmark functions, the newly proposed method shows its great performance. Then, the proposed method was applied in vehicle lightweight design problems. With the proposed method, the masses of the rear frame are reduced by 7.67kg, reaches the 10.4% of the systems, and its stiffness is also improved. When compared with previously developed hybrid and adaptive meta-modelling(HAM) method, both the search efficiency and accuracy are noticeably increased.

Key words: design space differentiation, global optimization, vehicle lightweight design, time-consuming problems