Abstract:
In recent decades, the evaluation of the actual behaviour of in-service bridge structures and their safety throughout their life cycle, by measuring structural responses, has been attracting increasing research efforts worldwide as an alternative to visual inspections or localised nondestructive tests. The vibration-based structural health monitoring methodology is a representative approach, based on the direct relationship of stiffness, mass and damping to the modal properties of a structure. Dynamic properties can thus be used to determine deterioration, damage, or change in structural state. For instance, the reduction in natural frequencies is the most easily observable change in the modal data and investigators have linked it to various structural deterioration mechanisms. However, real-world bridges are exposed to time-varying environmental and operational conditions such as changes in temperature as well as vibration level, which can influence the variability of modal parameters. These effects must be understood and quantified so that changes in vibration response resulting from damage can be discriminated from changes resulting from other factors. There is an abundant literature concerned with the effects of temperature on modal parameters, the quantitative relationships between temperature and modal properties, and data normalisation to account for environmental variability. However, comprehensive explorations of the influence of vibrational magnitude on the variability in dynamic characteristics of bridge structures are limited, because this operational variable is difficult to control in situ. In this research, in order to experimentally identify the amplitude-dependent properties of a bridge, a series of steady-state harmonic forced-vibration tests with different forcing levels were undertaken on the Nelson Street off-ramp bridge, an eleven-span, post-tensioned concrete motorway bridge located in Auckland, New Zealand. Both the natural frequencies and the damping ratios were identified from a series of frequency response functions constructed at different levels of excitation, and were subsequently quantitatively correlated with the displacement amplitude. The influence of vibration response amplitude on bridge structure damage detection was studied by comparing the calculated frequency shifts caused by the damage simulated on a beam element girder numerical model of the bridge with the experimental amplitude-dependent frequency data. Accurate identification of the dynamic characteristics of the monitored bridge structure from the collected response data by using modal parameter identification techniques is essential for obtaining reliable evaluation results by using dynamic-based condition assessment methods. Unfortunately, unlike the ideal test environment in a laboratory, external disturbances and elevated sensor measurement noise is unavoidable at an outdoor testing site, which poses great challenges to extracting weakly excited modes from noisy data in ambient vibration testing. The important issue of the feasibility and reliability of the dynamic tests under the circumstances of relatively weak excitation has rarely been investigated in detail. In this thesis a set of dynamic tests with different excitation sources, which included environmental sources, people jumping, broad-band linear chirp excitation induced by electro-dynamic shakers, and sinusoidal sweeping excitation by eccentric mass shakers, were conducted on the motorway bridge. The mode identifiability, and the accuracy of modal properties identified from relatively weak excitation cases were studied through comparison with a large-capacity shaker excitation case at strong forcing level. In order to investigate the ability of different identification algorithms to capture signal characteristics from the noise contaminated response data, various operational modal parameter identification algorithms in the frequency domain (the peak picking and the frequency domain decomposition methods) and the time domain (data-driven stochastic subspace identification method and Eigensystem realisation algorithm) were employed to extract natural frequencies, damping ratios and mode shapes for each excitation case. Cross-validation between different identification techniques was carried out to determine their efficiency, robustness and accuracy. Finite element (FE) updating, in which the physical parameters of a FE model are calibrated to match the measured modal properties of the structure, also plays a significant role in vibration-based condition assessment, because the updated model parameters can be used to trace the evolution of stiffness deterioration. However, due to the unique characteristics of bridge structures, such as their large size, soil-structure interaction, complex structural member connections as well as uncertain material properties and boundary conditions, hindrances can be encountered to solving the updating problem in highly nonlinear solution spaces, as is the case for real-world bridges. For the present analysis, relatively large differences were observed in the natural frequencies and mode shapes between the results calculated from even a very detailed initial shell-element numerical model of the bridge and the field vibration testing results, due to the complex and uncertain boundary conditions. Manual model refinements through tuning the connection stiffness between the main girder and piers/abutments were proposed for reducing the modelling errors of the initial FE model. Then, two different types of non-linear optimisation algorithms, the subproblem approximation method (SAM) and the first order method (FOM), were utilised to further update the manually tuned FE model. The initial attempt of the FOM captured a local minimum. After the initially assumed values of the updating parameters were perturbed, the FOM was rerun. The consistency of the final FE model updating results between the two methods enabled confirmation of the updating results.