Semi-controlled experiment plans for improved mechanical engineering designs

The use of planned experiments to improve designs is considered for situations where standard plans, such as orthogonal arrays or response surface plans, are impractical. An approach is described that is of special interest to the design of mechanical products where it is expensive to make prototype components to specified dimensions, and only a given small sample of components of differing dimensions is available. The plan for the experiment must then make the best use of the available components by determining an assembly that will most efficiently explore the expected design parameter region where the design parameters may be derived from the component dimensions. In contrast to conventional methods, the resulting experiment is semi-controlled, rather than fully controlled, as the component dimensions cannot be chosen but the assembly can. An algorithm for finding such plans is described and an example of its implementation in improving the design of an hydraulic pump is presented. The method extends to studies involving some components whose dimensions cannot be designated and hence are obtained from a small sample, and other components that are easily and cheaply constructed to prescribed dimensions. This results in a combination of semi-controlled and fully controlled factors. Efficient plans for such experiments are described and results from their application in practice are discussed