A FRAMEWORK FOR METAMODEL-BASED DESIGN: SUBSYSTEM METAMODEL ASSESSMENT AND IMPLEMENTATION ISSUES

Open Access
Author:
Meckesheimer, Martin
Graduate Program:
Industrial Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
August 15, 2001
Committee Members:
  • Timothy William Simpson, Committee Chair
  • Enrique Del Castillo, Committee Member
  • Calyampudi Radhakrishna Rao, Committee Member
  • Russell Richard Barton, Committee Chair
Keywords:
  • design of computer experiments
  • multidisciplinary design optimization
  • metamodeling
Abstract:
The use of mathematical models to analyze and simulate complex real world systems is widely applied in many scientific and engineering domains. The mathematical representation is used to develop several computer modules that interact with each other and capture the input-output relationship of the underlying physical phenomenon. System-level design based on a structure of computational models is difficult because of the need to integrate disparate subsystem models to predict overall system behavior. Since individual modules of such a computer structure are often computationally expensive, different approximation strategies have been developed to provide inexpensive metamodels of the discipline-specific simulation models. The use of metamodels allows a faster analysis than the complex engineering models, but introduces a new element of error and subsequent uncertainty that must be managed. The goal in this thesis is to investigate the issues involved in developing an efficient integration framework to support the use of metamodels as fast analysis tools. A conceptual metamodel-based integration framework is introduced, and the requirements for supporting the use of metamodels are identified. These requirements translate into five distinct research tasks that are addressed throughout the thesis. The research tasks entail exploring alternative metamodeling strategies, developing inexpensive validation methods for metamodel assessment, and analyzing three distinct implementation issues associated with the nature of disciplinary subsystem models, the type of design problems, and the functional forms of model responses.