基于OD与LHS的复杂作用关系制造过程的计算机实验方法

An OD and LHS Based Computer Experiments Approach for Complex Relationship Manufacturing Process

计算机仿真和计算机实验方法是自动化加工过程常用的产品设计和质量优化技术。文章针对输入因子与输出特性之间存在复杂非线性作用关系的制造过程,提出一种基于正交设计(OD)与超拉丁方抽样(LHS)的两阶段计算机实验方法。首先利用OD粗略探知出输出特性变化较大的子区域;而后在该子区域内采用LHS来安排实验点,获取样本集;最后利用Kriging模型建立起过程的全局性模型。理论与仿真研究表明,与传统的LHS设计相比,所提方法的实验点的分布可随输出特性的变化而调整,其拟合模型的预测性能也有较大幅度的提高,说明了方法的有效性。

Computer simulation and computer experiments are the main techniques of product design and quality optimization in automatic manufacturing processes. For the manufacturing processes featured with the complex nonlinear relationship between input factors and output characteristics, this paper proposed a two-stage computer experiments approach based on orthogonal design （OD） and Latin hypercube sam- piing （LHS）. Firstly, OD is used to roughly detect the significant fluctuation domains of the output; Secondly, LHS is used in these domains to obtain the sample set; finally, the global fitting model is set up by Kriging method. Theoretical and simulation studies show that, compare with traditional LHS, the distribution of the experimental points can adjust according to the fluctuations of the output, and the pre- diction performance of the fitting model is increased obviously as well. All of these demonstrate the ef- fectiveness of the orooosed aooroach.