Abstract
Modal identification and model updating utilizing vibration data from real structures are of long-standing interest among researchers and engineers in the field of structural health monitoring (SHM). Considering the importance of SHM in civil engineering, this thesis focuses on the theoretical development of the parallel Markov chain Monte Carlo (MCMC)-based Bayesian model updating and model class selection methodology and practical implementation of full-scale ambient vibration tests, modal identification and structural model updating of existing buildings.In terms of theoretical development, this thesis has three major contributions. The first contribution is a newly derived mode shape assembly formulation for practically identifying an optimal global mode shape based on partial mode shapes from a multi-setup ambient vibration test. In the new formulation, the global mode shape can be solved analytically in an iterative manner, and a novel stopping criterion is proposed to guarantee convergence and accuracy.
To solve the model updating problem in unidentifiable cases as well as to improve the efficiency of exploring the parameter space of interest, a newly formulated parallel MCMC-based Bayesian model updating method is presented as the second theoretical contribution. In this formulation, multiple Markov chains can be processed at the same time in a parallel computing environment, providing many samples and thereafter increasing the accuracy of the updated model parameters.
It is well known that the quality of an updated model highly depends on the class of models employed in the model updating process. However, most existing model updating methods assume that a suitable class of models is given at the beginning or that the model class is decided based on the subjective judgement of a user. To tackle this problem, the third theoretical contribution of this thesis is the new formulation of Bayesian model class selection that utilizes the samples generated by the parallel MCMC-based model updating method.
Measuring a qualified set of data from an ambient vibration test is the key step for successful modal identification and model updating. The contribution of this thesis in the practical implementation of full-scale ambient vibration tests is the development of a standard measurement procedure for conducting a multi-setup ambient vibration test. The proposed standard measurement procedure is verified to be efficient via three case studies considered in this thesis: (1) a 14-story factory building, (2) a 9-story coupled building, and (3) a 20-story boat-shaped primary-secondary building. The modal parameters in each case study are identified using the fast Bayesian FFT method, and the global mode shapes are determined using the newly proposed mode shape assembly formulation. To assess the dynamic properties of the target buildings, structural model updating is carried out using the parallel MCMC-based Bayesian model updating method. Interesting features of the dynamic properties of the target buildings are discussed in detail based on the identified results.
| Date of Award | 18 Jul 2017 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Heung Fai LAM (Supervisor) |