The Studies Of Semi-parametric Approach To Robust Parameter Design

Robust parameter design is selected parameters of the system to reduce the impact of the noise factors in the output process. It determines the product close to the target performance while have a small fluctuations. The goal of RPD is to choose the controllable variables level and reduce the noise factor’s influence on the system. Based on semi-parametric method we study the mean and variance models of quality characteristics. Choosing the optimal factor levels makes the average close to the target and the variance fluctuations as small as possible.Firstly, the paper studies robustness design problem by the parameters of the response surface method, and analyzes factors selection of regression models. Also modeling of the mean and variance of the quality characteristics has been done. We execute the efficient optimization routine using the genetic algorithm and the squared error loss.Secondly, considering futurities of impact factor and complex situations, we may select a non-parametric method to robust design. Using local polynomial regression of the non-parametric shows the impact of various factors on the response output results. Using variable coefficients of the non-parametric regression method to reduce the dimension problems, which will be more flexible and effective for mean and variance model fitting.Thirdly, semi-parametric approach which combines parameter and non-parametric methods can improve the mean and variance models. The study of printing ink example has analyzed the mean, variance and squared error loss values when the optimal factor levels by three RPD methods.Finally, the simulation shows the design robustness applied by the parameter methods, nonparametric methods and semi-parametric methods. This paper prove that the semi-parametric method is very effective in most cases.