Study On Design Of Experiment For Constrained Condition

Experimental design which is wildly used in the improvement phase of Six Sigma project (DMICA), is a technical tool based on statistical theory to arrange experiments comprehensively, economically and scientifically. Setting the input variables of a process purposely, we may observe and identify corresponding changes in the output response. With the popularity of Six Sigma management, the research and application of DOE has aroused the great concern of industry and academia. However, there have been few of study about constrained condition. In this dissertation, areas restrictions, randomization restrictions and resource restrictions were under research in order to expend the range of DOE.This paper opened with some basic concepts and traditional methods of design of experiments, and then it proposed a method of design, analysis and optimize for constrained experimental design. The research included the following aspects.First of all, the restrictions among controllable factors in design of experiments (DOE) were discussed. In consideration of the restrictions between the two factors, a viable solution was provided for DOE when there were restrictions between factors by importing free factors and treating incomplete data in experiment, which converted an irregular-shaped region of experimentation to a regular-shaped region of experiment. Through this treatment the experiment with restrictions between factors was converted to a typical DOE, and the experimental point would not be selected from unallowed points.Secondly, for this kind of design of experiments (DOE) that there was hard-to-change (HTC) factor among related factors, a viable solution was proposed for split-plot design containing related factors by combining split-plot design and nested effects modeling. By appealing to split-plot design, the problem that experimental order couldn't be randomized with HTC factors was resolved. The problem of more correspondence of the experimental region to industrial practice was settled by using the nested effects modeling. It offered solutions and methodology for the existence of HTC factor among related factors in design of experiments (DOE).Thirdly, for the existence of related factors in split-plot, an approach for split-plot design with related factors was proposed. It built nested effects model in split-plot structure and modified the standards face-centered cube design. With the adjustment of the split-plot design matrix, a second-order model was constructed for split-plot with related factors, and the equivalence was also guaranteed between ordinary least squares and generalized least squares estimates. Therefore it solved that how to build the RSM for related factors in split-plot.At last, it introduced some data mining techniques aiming at the factor selection problem in experiment design. Historical data have been used here more efficiently to support factor selection. It also reduced the experimental complexity and cost, and make up the lack of supports from quantitative analysis for factor selection.