Designs for assembled products

Designing experiments to identify improvements in products that are assembled from several components does not readily fit into conventional design methods, particularly when the components cannot be re-used and some, or all, of the factors of interest cannot be set to prescribed values and may be dependent on several component features. Two types of search algorithms have been developed for finding D-optimal designs for such experiments, a genetic algorithm and an exchange algorithm which is an advance on Sexton, Lewis and Please (2001). The performances of the algorithms have been compared on a number of industrial examples (Sexton, Anthony, Lewis, Please and Keane, 2004).

The exchange-based algorithm has been extended to generate sequential plans. In these sequential procedures components arrive into a pool, as they are manufactured, and the algorithm dictates the next assembly from this pool that will best augment the existing information from earlier assemblies.

A PC-based software system, DEAP, has been written to gather data and to generate plans for experiments on assembled products. The system has been designed to be sufficiently flexible to be used in a variety of industrial settings and has benefited significantly from the feedback from the industrial partners. The methods have been used in a study of sounder design with Hosiden Besson Ltd (Anthony et al, 2003) and also in work on fuel pumps with Goodrich Engine Control Systems.

---