Abstract: For the purpose of solving the non-normal distribution multiple quality characteristics variation problem under temperature noise and design parameters randomness,a robust design method for Microelectro Mechanical Systems (MEMS) product by generalized linear model and Multi-objective Evolutionary Algorithm based on Decomposition (MOEA/D) was presented.The dual-response surface was integrated with the generalized linear model to construct mean-variance model,which overcame the limitations of complex high-dimensional nonlinear model information distortion by non-normal transformation method.It showed more rigor and effectiveness than previous research.On this basis,a MOEA/D strategy was proposed to tackle the optimization problem with multiple quality characteristics,discretization and temperature noise.A numerical example showed that the effect of temperature noise and design parameter randomness on the non-normal quality characteristics could be mitigated.It was effective to solve the multiple quality characteristics decision making under uncertainty.

Key words: microelectro mechanical systems product, non-normal multiple quality characteristic, generalized linear model, multi-objective evolutionary algorithm based on decomposition, robust design

摘要： 针对温度噪声和设计参数随机性下的微机电系统(MEMS)产品非正态多质量特性波动问题,提出一种基于广义线性模型和基于分解的多目标进化算法(MOEA/D)的稳健设计方法。在样本数据建模阶段将双响应曲面引入广义线性模型中,来构建MEMS产品非正态多质量特性均值模型和方差模型,克服以往研究中对非正态转换方法下的复杂高维非线性模型信息失真的局限性,从理论上体现了MEMS产品非正态质量特性处理方法的严密性和有效性。在此基础上,提出MOEA/D优化策略,解决多质量特性冲突、离散和温度噪声条件下的模型求解问题。通过算例证明所提方法可以降低温度噪声和设计参数随机性对MEMS产品非正态质量特性的影响,同时表明该方法解决不确定性环境下的多质量特性决策问题是有效的。

关键词: 微机电系统产品, 非正态多质量特性, 广义线性模型, 基于分解的多目标进化算法, 稳健设计