Study On Experimental Design Optimization Method Of Multi-response Problems Based On Principal Component Analysis

This thesis mainly studies the methods of robust design optimization formulti-response problems. A multi-response robust design method based on principalcomponent analysis is proposed, by considering both the mean and variance ofmultiple responses, the proposed method can achieve the robustness.The thesis introduces some basic theories and methodologies of principalcomponent analysis method and multi-response experimental design. Through theanalysis and contrast of some multi-response optimization methods and some robustdesign methods, the main indicators, such as the economics of the process, thecorrelation between responses, the predictive ability, the robustness and so on areproposed.The common multi-response optimization methods don’t consider the robustness,and the common robustness design methods manly aim at single responseoptimization problems. By improving the traditional principal component analysismethod, a robust design method both considering the mean and variance of multipleresponses are proposed in this thesis.The improved method optimizes the mean and variance simultaneously bycombining dual response surface method and principal component analysis method.Firstly, the regression equation of mean and standard deviation, the location anddispersion effect values of multiple responses are calculated. Secondly, the locationand dispersion effect values are processed by principal component analysis to get theprincipal component scores. Finally, the multi-response performance index, which isthe weighted sum of principal component scores is introduced.Case study shows that the proposed method successfully achieves the robustnessof multi-response optimization problems.