An SVR and single response modeling strategy based RPD approach for multiextreme processes

Abstract

Robust Parameter Design (RPD) is one of the key approaches to improve product quality. The paper proposes an RPD approach for optimizing processes with multiextreme quality characteristics. Firstly, support vector regression (SVR) is selected as the basic fitting model for a concerned process. Secondly, uniform space filling design is used to arrange the levels of control factors and noise factors, and single response modeling strategy is adopted to fit the SVR model. Thirdly, process mean and variance are estimated according to the probability density function of noise factors. Lastly, RPD is achieved by solving the optimal problem. The results of a simulation study for the larger-the-better problem show that, compared with dual response surface methodology, the proposed approach can well reconstruct the multiextreme response surfaces of process mean and variance and obtain global optimal solution of RPD with higher accuracy, all of which demonstrate the superiority of the proposed approach.