Title:  Explicit Yield Model (EYM) based Tolerance Optimization Methodology

Sponsor: NIST-ATP

Collaborators: R. Kumar, J. Kong, Y. Zhou, S. Gogoneni (Dimensional Control Systems)

One of the major challenges in statistical tolerance synthesis for complex assemblies is the computation intensity in estimating probability of Key Product Characteristics (KPCs) simultaneously being within specifications or yield of assembled products. In assemblies with multiple KPCs, the yield can only be obtained numerically through simulation techniques such as Monte Carlo algorithms. The existed tolerance synthesis methods require a large number of yield assessments during the iteration for an optimal solution. A new approach is developed for yield model approximation based on uniform design (Fang, 1980), multivariate distribution transformation (MDT), and regression analysis. An explicit yield function can thus be approximated which represents relationship between yield and tolerances. Therefore, the widely used gradient-based optimization methods (such as Sequential Quadratic Programming) can be easily applied and the computational intensity of direct optimization procedures could be avoided. The industrial case study with moderate complexity is presented to illustrate the proposed methodology and compared with the existed tolerance synthesis methods.