Health monitoring and inspection of large-scale infrastructures have posed critical challenges to urban societies. This study presents a comprehensive Bayesian paradigm covering various aspects of structural identification, from characterizing and verifying analytical models to model-based identification of structural dynamics and applied excitations from incomplete noisy measurements. It focuses on data-based noise calibration for robust uncertainty quantification and propagation of response quantities and dynamical parameters. Relying on the capability of the Hierarchical Bayesian modeling approach in dealing with multiple data sets, a novel Finite element model updating method is suggested considering the modal features errors and variability in structural features across different d...[ Read more ]

Health monitoring and inspection of large-scale infrastructures have posed critical challenges to urban societies. This study presents a comprehensive Bayesian paradigm covering various aspects of structural identification, from characterizing and verifying analytical models to model-based identification of structural dynamics and applied excitations from incomplete noisy measurements. It focuses on data-based noise calibration for robust uncertainty quantification and propagation of response quantities and dynamical parameters. Relying on the capability of the Hierarchical Bayesian modeling approach in dealing with multiple data sets, a novel Finite element model updating method is suggested considering the modal features errors and variability in structural features across different data sets. Once an accurate representation of systems is established, a Bayesian filtering strategy is proposed to identify the latent states, applied excitations, and abrupt changes in structural parameters. The precision and reliability of Bayesian filters are refined through a fully Bayesian framework featuring the Expectation-Maximization algorithm (BEM) through simultaneous identification of unknown quantities of interest and calibration of noise parameters. Additionally, a Bayesian steady-state algorithm is suggested for the initial calibration of noise covariance matrices to remove user bias from noise covariance matrices, enabling a robust identification of uncertainties. The efficacy of BEM in solving coupled input-state-parameter and joint input-state estimation problems with and without knowledge of the spatial distribution of the applied excitations is demonstrated through numerical and experimental examples, revealing the importance of stabilizing Bayesian estimators by considering pseudo-input observations or utilizing the Gaussian Process Latent Force models.