The mathematical model of a heavy machine-tool column is established, and 50 data points of finite element analysis for implicit constrains are sampled by the Latin hypercube design of experimental method. The Kriging model, RBNN ( radial basis neural network) model and second-order PRS ( polynomial response surface) model are then applied to constructing the ensemble of sur-rogates for the implicit constrains. Then its feasible design variables are obtained through global optimization, the mass of the column is reduced and the implicit constrains of deflection, stress and hydraulic constrains are al satisfied. The study shows that ensemble of surrogates is suitable for expensive implicit constrains.%建立了某重型机床立柱部件优化问题的数学模型，采用拉丁超立方试验设计生成50个隐式约束的有限元仿真数据样本。使用样本集分别建立Kriging模型、径向基神经网络和二阶多项式响应面模型，通过近似模型聚合方法来替代隐式约束的有限元计算。结合全局优化算法，得到一组可行的立柱尺寸参数组合，使机床立柱的质量减轻，并满足工作载荷下的挠度、应力和油压约束条件。研究表明：近似模型聚合方法很适合解决高仿真代价的隐式约束问题。
展开▼
