基于近似模型的车身气动外形优化  

Aerodynamic shape optimization of automotive body based on approximate model

作　　者：高静[1] 杨志刚[1] 李启良[1] 

机构地区：[1]同济大学上海地面交通工具风洞中心,上海201804

出　　处：《计算机辅助工程》2014年第1期1-6,共6页Computer Aided Engineering

基　　金：国家重点基础研究发展计划("九七三"计划)(2011CB711203);上海市科学技术委员会重点实验室计划(11DZ2260400)

摘　　要：以英国汽车工业研究协会(Motor Industry Research Association,MIRA)阶背模型为基本模型,用参数化建模方法建立其纵向对称面的二维模型.运用优化拉丁超立方方法对每组参数化方案生成600组样本点;将MATLAB与Gambit结合,自动快速生成其网格模型;用FLUENT计算每个样本点的气动阻力.建立径向基神经网络(Radial Basis Function Neural Network,RBFNN)近似模型,以阻力最小为优化目标,采用多岛遗传算法优化外形参数;对优化后的结果进行数值模拟,结果表明阻力减少31.9%.三维验证结果表明:二维优化结果不能完全代表三维结果,直接进行三维优化设计的效果更好.Based on the notchback model of British Motor Industry Research Association （MIRA）, a 2D model of longitudinal symmetric plane is built by parametrization modeling method. The optimized Latin hypercube method is used to generate 600 series of sample points for each parametrization scheme. The mesh model is automatically and quickly generated by the combination of MATLAB and Gambit and FLUENT is used to calculate the aerodynamic drag value of each sample point. An approximate model of Radial Basis Function Neural Network （RBFNN） is built; taking the minimum aerodynamic drag as the optimization goal, the shape parameters are optimized by multi-island genetic algorithm. The optimization results show that the drag is reduced by 31. 9%. The 3D verification results indicate that the 2D optimization results can not fully represent 3 D results and the results are better by directly performing 3 D optimization design.

关 键 词：车身 参数化建模 气动优化 优化拉丁超立方 径向基神经网络模型 多岛遗传算法 

分 类 号：U463.82[机械工程—车辆工程] U461[交通运输工程—载运工具运用工程]