Tolerance allocation via simulation embedded sequential quadratic programming

Tolerance allocation is an important problem frequently encountered in the synthesis process, for designers as well as manufacturing engineers. Under the objective of minimizing the manufacturing cost while attaining an acceptable yield, the problem can be formulated as a stochastic program. Owing to the nonlinear nature of the stochastic program, a sequential quadratic programming algorithm is developed to solve the problem. The cumbersome multivariate integration in calculating the yield is approximated by a Monte Carlo simulation and the highly nonlinear yield constraint is supported by some auxiliary constraints. In limited experiments, the proposed method has performed efficiently and robustly. Compared with some previous studies, the designs solved in this paper have smaller manufacturing costs and higher yields, indicating that the proposed method is very promising in solving tolerance allocation problems.