权衡产品质量损失的相关多响应稳健参数设计与优化

Balancing the Weights about Product Quality Loss Terms for Correlated Multi-response Robust Design and Optimization

本文回顾了质量损失函数方法的演变过程,描述了多元期望质量损失函数方法是由分别刻画多响应的最优性、稳健性与预测性能等三个质量损失项以累加形式构成,分析了基于累加形式的多元期望质量损失函数方法的不足,提出了基于乘积指数加权形式的多元期望质量损失函数方法,采用了似然不相关回归技术对模型进行联合参数估计以考虑多响应之间的相关性,考虑了三个质量损失项的权重系数组合变化对相关多响应系统整体质量损失的影响,并结合实例进行分析研究,结果表明本文方法不仅能够得到使实际相关多响应系统更为稳健的全局最优解,而且能够权衡三个质量损失项对实际相关多响应系统整体质量损失的主次作用.

The evolution of quality loss function approach is reviewed in the paper,and the multivariate expected quality loss function approach is described that the method is composed of three quality loss terms,which represent the optimization and robustness and quality of predictions for correlated multi-response,respectively.The shortcomings of the multivariate expected quality loss function approach based on the accumulation form are analyzed,and a new multivariate expected quality loss function approach based on the product index weighted form is proposed in the paper.The seemingly unrelated regression technique is adopted for the joint parameter estimation about the new multivariate expected quality loss function approach,in order to consider the correlation among the multiple responses,and the influence to the integral quality loss of correlated multi-response system on account of the weight coefficient combination change of three quality loss terms.At last,the specific example is used to illustrate the effectiveness of the proposed approach.The result shows that the proposed method can not only produce more robust global optimal solutions for multi-response robust parameter design,but also consider the three quality loss terms to the primary and secondary effects of correlated multi-response system.