基于Taguchi-BPNN-SEDEA的精密产品多元质量非参数稳健优化

A robust nonparametric approach to optimize multi-quality of precision product based on Taguchi-BPNN-SEDEA method

针对设计参数不确定性和模型结构未知情形下精密产品多元质量波动问题,同时兼顾主体结构对轻量化设计要求,提出一种基于Taguchi-BPNN-SEDEA的多元质量非参数稳健优化方法.首先,通过正交试验设计和有限元分析获取多元质量数值,运用Taguchi方法将多元质量数值转化为信噪比来衡量精密产品稳健性;其次,运用BPNN非参数模型构建多元质量信噪比预测模型,以避免由参数模型设定导致的误差;在此基础上,提出改进的DEA基本模型,采用SEDEA非参数稳健优化方法,将设计参数不确定性下BPNN非参数模型求解问题转化为不确定性条件下复杂多属性决策问题;最后,通过实例表明,所提出的方法能够有效处理设计参数不确定性和模型结构未知并存情况下的多元质量稳健优化问题,从而验证该方法的可行性.

In view of multi-quality variation for precision products with parameter uncertainty and unknown model structure, while considering the demand of main structural on lightweight design, a multi-quality robust design method based on Taguchi couple with back propagation neural network and super efficiency data envelop analysis(TaguchiBPNN-SEDEA) is proposed. Firstly, the multi-quality value is obtained through orthogonal experimental design and finite element analysis. The Taguchi method is adopted to converse the multi-quality value into signal-to-noise ratio(SNR),which can measure the stability of precision product. Then, the multi-quality SNR prediction model is constructed by the non-parametric model of the back propagation neural network(BPNN) to solve errors of model misspecification. On the basis of the improved data envelop analysis, the super efficiency data envelop analysis is proposed to convert the BPNN non-parametric model of the parameter optimization problem into the complex multiple attribute decision making problem under uncertainty. Finally, an example is given to illustrate that the proposed method is feasible and can handle multi-quality robust optimization with the parameter uncertainty and unknown model structure.