Modeling for Optimization (MO-OP): Tools for Manufacturing and Design Engineering Problems

Abstract

Abstract In this paper, a new statistical optimization technique is proposed. The technique employs new variance reduction schemes (VRTs). The performance of three standard designs: L27/L27 OA, L54/L27 OA and L243 / L27 OA are studied. These designs, although both orthogonal and balanced, exhibit high variance reduction properties with questionable convergence in very short number of iterations. Four new composite designs are developed, implemented and compared with the standard ones. These designs are known as: 5-, 7-, 9- and 11-point composite L27 OA. The problem of tolerance allocation with optimal process selection is revisited as a case study for simulation. Results indicate the efficiency of these new designs to reduce variances to lower levels than standard designs and better convergence in fraction of experiments. These designs are then integrated in an optimization algorithm previously developed (Gadallah, M.H., 2000). The algorithm is then modified to deal with the least sensitive optimal solutions for standard and composite designs. Particularly, the parameters that affect the algorithm are varied and their effects on performance of algorithm are studied. A standard manufacturing case study is used for analysis and simulation results for the composite designs are also given.