Statistics (Student) Seminar

SPEAKER:  Matthias H.Y. Tan

ABSTRACT:

Although traditional quality improvement methods have been successfully applied in numerous contexts, they are not applicable to many modern manufacturing systems. Two recent manufacturing trends that pose significant challenges to the quality engineering community are production of small batches of high value products and the use of computer simulations for product design. This talk consists of two parts.

In the first part, I will present generalized selective assembly, which is a quality improvement method for small batches of high value products. Selective assembly has traditionally been used to achieve tight specifications on the clearance of two mating parts. We develop a generalized version of selective assembly, called GSA, for improving the quality of assemblies with any number of components and any known assembly response function. Two variants of GSA are introduced: direct selective assembly and fixed bin selective assembly.

In the second part, I will present a statistical methodology for robust design with computer experiments. Gaussian process (GP) models are widely employed in computer experiments to model quality characteristics as functions of control and noise factors. These models enable the average loss to be estimated without time-consuming simulations. However, robust design optimization performed as if the GP predictor were the true response function can give misleading results. We propose expected quadratic loss criteria that take into account uncertainty about the true function, and methods based on the Lugannani-Rice saddlepoint approximation for constructing accurate credible intervals for the average quadratic loss.

Contact: mtan6@gatech.edu