针对汽车乘员约束系统高度非线性且难于求解最优值的特点,提出全局敏感性分析结合混合元模型的优化方法,通过蒙特卡罗模拟在整个设计空间内采样,以元模型代替仿真模型来完成设计参数的敏感性分析,并将分析获得的信息用于混合元模型优化(hybrid and adaptive metamodeling method,HAM),将二阶多项式响应面、Kriging模型、径向基函数三种元模型有机结合,自适应选择最佳的元模型进行寻优.搜索过程中元模型不断更新与重建,逐渐提高关键区域的精度,从而快速寻找到全局最优解.对某工程实例的优化结果表明该方法是有效的.
The current research of complex nonlinear system robust optimization mainly focuses on the features of design parameters, such as probability density functions, boundary conditions, etc. After parameters study, high-dimensional curve or robust control design is used to find an accurate robust solution. However, there may exist complex interaction between parameters and practical engineering system. With the increase of the number of parameters, it is getting hard to determine high-dimensional curves and robust control methods, thus it＇s difficult to get the robust design solutions. In this paper, a method of global sensitivity analysis based on divided variables in groups is proposed. By making relevant variables in one group and keeping each other independent among sets of variables, global sensitivity analysis is conducted in grouped variables and the importance of parameters is evaluated by calculating the contribution value of each parameter to the total variance of system response. By ranking the importance of input parameters, relatively important parameters are chosen to conduct robust design analysis of the system. By applying this method to the robust optimization design of a real complex nonlinear system-a vehicle occupant restraint system with multi-parameter, good solution is gained and the response variance of the objective function is reduced to 0.01, which indicates that the robustness of the occupant restraint system is improved in a great degree and the method is effective and valuable for the robust design of complex nonlinear system. This research proposes a new method which can be used to obtain solutions for complex nonlinear system robust design.
