Presentation for the iSight User's Conference
CREDO Presentation: September 12, 2002
Kriging meta-modeling (also known as Design and Analysis of Computer Experiments, DACE), based on spatial interpolation of the data has been implemented in iSIGHT™ (A product of Engineous Software Inc). Meta modeling techniques have been traditionally used in practice when the computational expense of optimization using the original simulation is prohibitive. As the use of meta-models (also referred as approximation models or surrogate models) has grown in popularity, a variety of modeling methods are at the engineer’s disposal. Perhaps the most popular techniques involve polynomial models, typically linear or quadratic functions, created by performing a least squares curve fit to a set of data, where the data consists of one or more dependent response values, along with one or more independent variables. Collectively, these polynomial based models have come to be known as response surface models (RSM) which is a term taken from the statistical literature (Montgomery [1]). Such type of models has seen considerable use in engineering optimization studies due to its computational simplicity and ease of use. This type of model is readily available in iSIGHT™. However quadratic polynomial models have limited accuracy when the response data to be modeled has multiple local optima. Another class of meta-modeling technique employs an interpolation scheme known as Kriging developed in the field of spatial statistics and geostatistics. This class of interpolating models has the flexibility to model response data with multiple local optima by superimposing Gaussian or Exponential functions on an RSM. However this flexibility is obtained at an increase in computational expense and a decrease in ease of use. The current version of iSIGHT™ (V 6.0) does not support this kind of interpolation model. The main purpose of this paper is to demonstrate the implementation of Kriging methodology in iSIGHT™ frame work.