Optimization of Design Tolerances Through Response Surface Approximations

Abstract

A methodology whereby the optimal set of design tolerances is assigned to the dimensions of a general mechanical assembly, is developed and tested. The manufacturing cost is minimized, while the design is constrained to a specified probability of meeting functional requirements, called the yield of the design. An analytical relationship for the assembly yield surface is generally unknown, and use is made of response surface approximations in the optimization algorithm. Yield values are determined at design space points through Monte Carlo simulations, seen as the response surface experiments. The methodology is benchmarked on example problems from the literature, and the optimum compares superior to published results.