No abstract

No abstract

When planning experiments to examine how product performance depends on the design, manufacture and environment of use, there are invariably too few resources to enable a complete investigation of all possible variables (factors). We have developed new algorithms for generating and assessing efficient two-stage group screening strategies which are implemented through a web-based system called GISEL. This system elicits company knowledge which is used to guide the formulation of competing two-stage strategies and, via the algorithms, to provide quantitative assessment of their efficiencies. The two-stage group screening method investigates the effect of a large number of factors by grouping them in a first stage experiment whose results identify factors to be further investigated in a second stage. Central to the success of the procedure is ensuring that the factors considered, and their grouping, are based on the best available knowledge of the product. The web-based software system allows information and ideas to be contributed by engineers at different sites and allows the experiment organizer to use these expert opinions to guide decisions on the planning of group screening experiments. The new group screening algorithms implemented within the software give probability distributions and indications of the total resource needed for the experiment. In addition, the algorithms simulate results from the experiment and estimate the percentage of important or active main effects and interactions that fail to be detected. The approach is illustrated through the planning of an experiment on engine cold start optimization at Jaguar Cars.

When planning experiments to examine how product performance depends on the design, manufacture and environment of use, there are invariably too few resources to enable a complete investigation of all possible variables (factors). Here we describe a web-based system for eliciting company knowledge into an efficient two-stage group screening method. The method investigates the effect of a large number of factors by grouping them in a first stage experiment whose results identify factors to be further investigated in a second stage. Central to the success of the procedure is ensuring that the factors considered, and their grouping, are based on the best available knowledge of the product. We present a web-based software system that allows information and ideas to be contributed by engineers at different sites and uses these expert opinions to guide decisions on the planning of group screening experiments. The software includes elements that predict the total resource needed for the experiment It also simulates the results of the experiment and estimates the likely percentage of important or active factors that fail to be detected. The approach is illustrated through the planning of an experiment on engine cold start optimization at Jaguar Cars.