In anticipation of the redefinition of the kilogram and to avoid a parallel non-SI dissemination system for mass, an alternative approach to mass measurement is being developed by constructing a magnetically levitated balance for direct comparison of masses in vacuum versus in air. To build such a balance is extremely challenging due to the requirement of a strong and stable magnetic flux providing rigid link between air and vacuum, a proper electromagnetic shielding to eliminate the stray field, and a minimum uncertainty (highest accuracy) in measurement. Clearly, an optimal design solution can be the key between a success and a failure in achieving these requirements. To aid this process of finding an optimal solution, we have resorted to computer-aided modeling and design. Most electromagnetic computer models are solved using finite element based solvers. The commercially available software package that we use to model the forces on the magnetically levitated body components and the effect of electromagnetic shielding is a typical example of such an application. One of the features of this software package is the ability to automate the model design process through the use of scripts. This functionality, along with the ability to directly parameterize models, forms the basis for computer experiments. A computer experiment is a number of runs of the computer model (or codes) with various model parameter inputs.
In this talk we will describe our computer model including its inputs and response; and we will discuss how we design, implement, and analyze computer experiments to predict the response at untried model parameters and to optimize a functional of the response. Issues concerning the verification and validation of our computer model will also be addressed.
Dr. Hung-kung Liu
Statistical Engineering Division
NIST