Date of Award

12-2013

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Legacy Department

Civil Engineering

Advisor

Dr. Sez Atamturktur

Committee Member

Dr. Mashrur Chowdhury

Committee Member

Dr. Hsein C. Juang

Committee Member

Dr. Abdul A. Khan

Committee Member

Dr. Calvin L. Williams

Abstract

The Computer simulations are commonly used to predict the response of complex systems in many branches of engineering and science. These computer simulations involve the theoretical foundation, numerical modeling and supporting experimental data, all of which contain their associated errors. Furthermore, real-world problems are generally complex in nature, in which each phenomenon is described by the respective constituent models representing different physics and/or scales. The interactions between such constituents are typically complex in nature, such that the outputs of a particular constituent may be the inputs for one or more constituents. Thus, the natural question then arises concerning the validity of these complex computer model predictions, especially in cases where these models are executed in support of high-consequence decision making. The overall accuracy and precision of the coupled system is then determined by the accuracy and precision of both the constituents and the coupling interface. Each constituent model has its own uncertainty and bias error. Furthermore, the coupling interface also brings in a similar spectrum of uncertainties and bias errors due to unavoidably inexact and incomplete data transfer between the constituents. This dissertation contributes to the established knowledge of partitioned analysis by investigating the numerical uncertainties, validation and uncertainty quantification of strongly coupled inexact and uncertain models. The importance of this study lies in the urgent need for gaining a better understanding of the simulations of coupled systems, such as those in multi-scale and multi-physics applications, and to identify the limitations due to uncertainty and bias errors in these models.

Share

COinS