Coupling polynomial chaos expansions with Gaussian process emulators: An introduction

Diaz De la O, F. A. and Adhikari, S.,
Proceedings of the 27th International Modal Analysis Conference (IMAC-XXVII), February 2009, Orlando, Florida, USA.

Uncertainty quantification is an essential part of model validation for realistic complex engineering systems. In order to incorporate uncertainty in the equations that govern the system’s response, the associated elastic, damping and mass properties can be characterized using a random field model. The solution of these governing equations enables uncertainty quantification and it is thus important to have a means of representing the system’s response. This can be achieved conveniently with a series of orthogonal polynomials known as polynomial chaos. However, a code to implement polynomial chaos can be computationally intensive and its application to systems with an important number of degrees of freedom can be very limited. In this paper, a method based on Bayesian emulators to reduce the computational time of a polynomial chaos implementation is developed. It will be shown that, by carefully selecting a small number of points in the input domain of the polynomial chaos code, a Bayesian emulator can provide an efficient approximation to the system’s response.
BiBTeX Entry
@INPROCEEDINGS{cp53,
    AUTHOR={F. A. DiazDeLaO and S. Adhikari},
    TITLE={Coupling polynomial chaos expansions with Gaussian process emulators: An introduction},
    BOOKTITLE={Proceedings of the 27th International Modal Analysis Conference (IMAC - XXVII)},
    YEAR={2009},
    Address={Orlando, Florida, USA},
    Month={February},
    Note={under review}
}
by Sondipon Adhikari