|
Activity Number:
|
232
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 10, 2004 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Physical and Engineering Sciences
|
| Abstract - #302017 |
|
Title:
|
Integration of Detailed and Quick Simulations via a Bayesian Synthesis
|
|
Author(s):
|
Zhiguang Qian*+ and Roshan J. Vengazhiyil and C.F. Jeff Wu
|
|
Companies:
|
Georgia Institute of Technology and Georgia Institute of Technology and Georgia Institute of Technology
|
|
Address:
|
School of Industrial and Systems Engineering , Atlanta, GA, 30332,
|
|
Keywords:
|
response surface model ; computer experiments ; design of experiments ; Gaussian process ; meta-model ; finite elements analysis
|
|
Abstract:
|
This talk is motivated by collaborative work on robust topology design of cellular material at Georgia Tech. In simulating the material properties finite elements analysis (FEA) can be done based on different physical-mechanistic models. Typically, a more detailed or accurate model will require longer FEA runs while a simplified or rough model will require quicker FEA runs. They are referred to as detailed and quick simulations respectively. Detailed simulations can take up days of CPU time. While they can provide more accurate results, their number can be limited. On the other hand, many quick simulations can be obtained, though the results are less reliable. A new approach is taken here to combine these sources of data to come up with a meta-model that can be used to describe the relationship between the output of FEA runs and input parameters and for prediction. Since the quick simulations form the bulk of the data, they are used to build a semiparametric model based on Gaussian random functions. This fitted model is then "adjusted" by incorporating the information in the detailed simulations. Real data will be used to illustrate this technique.
|
