For experiments on mechanical products, conventional methods for reducing the experimental effort that is needed to extract information, such as those of Taguchi, can be infeasible because components with the dimensions needed for a standard factorial plan can be prohibitively expensive. Also, many factors for investigation are not directly measurable, as they are derived from the properties of several components and their assembly.
Methods are described for finding optimal plans for such experiments from a minimal sample of measured components, and are illustrated through a pilot investigation of a hydraulic pump.
Important issues are addressed of the robustness of the plans to possible missing data, for example due to product failure, and the need to ensure that the influences of the various different factors can be clearly extracted from the experimental results.
On-going work in the aeronautical industry on products for low-volume manufacturing is also described, where experiments are run on-line so that information can be swiftly obtained without delay in the manufacturing process.