Date of Completion
The efficiency of sampling remains as one of the major challenges for uncertainty analysis in structural dynamics. In the context of numerical experiments, this difficulty primarily lies in the fact that every single run of a numerical model, which usually is of large scale in modern engineering practice, would be expensive in terms of CPU times and memories. Nonetheless, understanding and predicting dynamic characteristics of structural systems with uncertainties is important for structural design, assessment and control. In order to develop fast and economical sampling techniques for characterizing structural dynamics, we explored two categories of methods from two opposite direction: the first is first-principle-based order reduced simulation and the second is data based statistical emulation. This thesis particularly focuses on component mode synthesis (CMS) and Gaussian processes regression (GPR).The former is a family of finite-element-based techniques that have drawn significant attention in computational structural dynamics, and the latter is a recently developed statistical approach which has proved to be potentially useful in engineering applications. A general free interface CMS formulation, which has been developed by the author, a concise self-sustained description of GPR and two-level GPR, and a series of application examples will be presented in this thesis.
XIA, ZEPING, "Efficient Characterization of Structural Dynamic Responses under Uncertainties: from Order-Reduced Simulation to Data-Driven Emulation" (2012). Master's Theses. Paper 325.