Date of Award

6-2012

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Legacy Department

Civil Engineering

Advisor

Atamturktur, Sezer

Committee Member

Juang , Hsein

Committee Member

Khan , Abdul

Committee Member

Thompson , Lonny

Abstract

An increasing reliance on complex numerical simulations for high consequence decision making is the motivation for experiment-based validation and uncertainty quantification to assess, and when needed, to improve the predictive capabilities of numerical models. Uncertainties and biases in model predictions can be reduced by taking two distinct actions: (i) increasing the number of experiments in the model calibration process, and/or (ii) improving the physics sophistication of the numerical model. Therefore, decision makers must select between further code development and experimentation while allocating the finite amount of available resources. This dissertation presents a novel framework to assist in this selection between experimentation and code development for model validation strictly from the perspective of predictive capability. The reduction and convergence of discrepancy bias between model prediction and observation, computed using a suitable convergence metric, play a key role in the conceptual formulation of the framework. The proposed framework is demonstrated using two non-trivial case study applications on the Preston-Tonks-Wallace (PTW) code, which is a continuum-based plasticity approach to modeling metals, and the ViscoPlastic Self-Consistent (VPSC) code which is a mesoscopic plasticity approach to modeling crystalline materials. Results show that the developed resource allocation framework is effective and efficient in path selection (i.e. experimentation and/or code development) resulting in a reduction in both model uncertainties and discrepancy bias.
The framework developed herein goes beyond path selection in the validation of numerical models by providing a methodology for the prioritization of optimal experimental settings and an algorithm for prioritization of code development.
If the path selection algorithm selects the experimental path, optimal selection of the settings at which these physical experiments are conducted as well as the sequence of these experiments is vital to maximize the gain in predictive capability of a model. The Batch Sequential Design (BSD) is a methodology utilized in this work to achieve the goal of selecting the optimal experimental settings. A new BSD selection criterion, Coverage Augmented Expected Improvement for Predictive Stability (C-EIPS), is developed to minimize the maximum reduction in the model discrepancy bias and coverage of the experiments within the domain of applicability. The functional form of the new criterion, C-EIPS, is demonstrated to outperform its predecessor, the EIPS criterion, and the distance-based criterion when discrepancy bias is high and coverage is low, while exhibiting a comparable performance to the distance-based criterion in efficiently maximizing the predictive capability of the VPSC model as discrepancy decreases and coverage increases.
If the path selection algorithm selects the code development path, the developed framework provides an algorithm for the prioritization of code development efforts. In coupled systems, the predictive accuracy of the simulation hinges on the accuracy of individual constituent models. Potential improvement in the predictive accuracy of the simulation that can be gained through improving a constituent model depends not only on the relative importance, but also on the inherent uncertainty and inaccuracy of that particular constituent. As such, a unique and quantitative code prioritization index (CPI) is proposed to accomplish the task of prioritizing code development efforts, and its application is demonstrated on a case study of a steel frame with semi-rigid connections. Findings show that the CPI is effective in identifying the most critical constituent of the coupled system, whose improvement leads to the highest overall enhancement of the predictive capability of the coupled model.

Share

COinS