In recent years, health monitoring of structure and infrastructure systems has become a valuable source of information for evaluating structural integrity, durability and reliability throughout the lifecycle of structures as well as ensuring optimal maintenance planning and operation. Important advances in sensor and computer technologies made possible to process a large amount of data, to extract the characteristic features of the signals, and to link those to the current structural conditions. In general, the process of data feature extraction relates to solving an inverse problem, in either a data-driven or a model-based type setting. This dissertation explores state-of-the-art hierarchical optimization-based computational algorithms for solving multiscale model-based inverse problems such as system identification and damage detection. The basic idea is to apply optimization tools to quantify an established model or system, characterized by a set of unknown governing parameters, via minimizing the discrepancy between the predicted system response and the measured data.

We herein propose hierarchical optimization algorithms such as the improved artificial bee colony algorithms integrated with local search operators to accomplish this task. In this dissertation, developments in multiscale damage detection are presented in two parts. In the first part, efficient hybrid bee algorithms in both serial and parallel schemes are proposed for time domain input-output and output-only identification of macro-scale linear/nonlinear systems such as buildings and bridges. Solution updating strategies of the artificial bee colony algorithm are improved for faster convergence, meanwhile, the simplex method and gradient-based optimization techniques are employed as local search operators for accurate solution tuning. In the case of output-only measurements, both system parameters and the time history of input excitations can be simultaneously identified using a modified Newmark integration scheme. The synergy between the proposed method and Bayesian inference are proposed to quantify uncertainties of a system.

Numerical and experimental applications are investigated and presented for macro-scale system identification, finite element model updating and damage detection. In the second part, a framework combining the eXtended Finite Element Method (XFEM) and the proposed optimization algorithms is investigated, for nondestructive detection of multiple flaws/defects embedded in meso-scale systems such as critical structural components like plates. The measurements are either static strains or displacements. The number of flaws as well as their locations and sizes can be identified. XFEM with circular and/or elliptical void enrichments is employed to solve the forward problem and alleviates the costly re-meshing along with the update of flaw boundaries in the identification process. Numerical investigations are presented to validate the proposed method in application to detection of multiple flaws and damage regions. Overall, the proposed multiscale methodologies show a great potential in assessing the structural integrity of building and bridge systems, critical structural components, etc., leading to a smart structure and infrastructure management system.