Structural health monitoring and condition assessment of bridge structures

Abstract

Engineering structures continuously accumulate damage during their service life due to material degradation, human error, and unexpected catastrophic events. Such damage adversely affects the safety and performance of structures. The work described in this thesis is concerned with condition assessment of bridge structures using long-term monitoring data, including vibration-based damage detection and structural reliability evaluation, modelling of the temperature-frequency correlation, and condition assessment of bridge expansion joints. A crucial issue in vibration-based damage detection is the treatment of the ill-conditioned and noisy system of equations, and it is pursued in this study by two numerical regularization methods, namely Tikhonov regularization and truncated singular value decomposition. Three approaches, including the L-curve method, generalized cross validation and minimum product criterion, to selecting the regularization parameters are presented. The performances of the two regularization methods with three regularization-parameter-selection approaches are rigorously examined and assessed through numerical studies of a truss bridge using both noise-free and noisy 'measurement' data. Minimum product criterion is shown effective and robust in the selection of appropriate regularization parameters for these two regularization methods. In order to take into account the uncertainty in the measured modal parameters, a novel method for stochastic FE model updating is proposed. The proposed method follows a two-stage model updating scheme. The first stage refers to the identification of probability density functions (PDFs) of updating parameters based on measured random modal parameters and the second stage deals with the determination of posterior PDFs of structural parameters from the identified PDFs and the prior PDFs of structural parameters. An improved perturbation method and the Monte Carlo simulation (MCS) method are used to perform the first-stage updating. At the second stage updating, the first-stage updating results are combined with the prior distribution of updating parameters (if available) by means of Bayesian theory to achieve the posterior distribution. Two numerical examples are provided to demonstrate and verify the proposed method. Three types of uncertainty in modal parameters are considered, and the updating parameter statistics is obtained using the improved perturbation method and verified by the MCS method for each type of uncertainties. The numerical studies show that the perturbation method generates satisfactory model updating results in the case of low uncertainty however the results may be less accurate in the case of high uncertainty. Using the stochastically updated FE model, structural reliability theory is applied to determine the failure probability and reliability index for the predefined limit state. With the obtained failure probability and reliability index, rational inspection and maintenance strategies can be laid down according to the correspondence between reliability index and required maintenance action established by other researchers. Such a systematic procedure bridges the gap currently existing between structural health monitoring technologies and bridge maintenance and management exercises, and the procedure is also capable of taking into account the uncertainty to make a decision on inspection/maintenance strategies. Following this approach, structural health monitoring is able to provide quantitative information regarding bridge inspection and maintenance. The proposed approach is demonstrated through numerical studies with respect to the nominal, updated, and actual models of two truss bridges. A combined method of principal component analysis (PCA) for feature extraction and support vector regression (SVR) for data-based statistical learning is proposed to characterize the correlation between modal frequency and temperatures using one-year monitoring data from the cable-stayed Ting Kau Bridge. The well-defined nature of temperature effects on modal parameters makes it possible to discriminate abnormal modal change caused by structural damage from normal modal change due to temperature variation. Research is focused on the optimal selection of predominant features and SVR hyper-parameters to achieve correlation models with good generalization capability. The performance of the formulated SVR models with the hyper-parameters determined by a grid search method with cross validation and a heuristic method, respectively, is examined. Both the 'dynamic' regression model taking into account thermal inertia effect and the 'static' regression model without considering thermal inertia effect are formulated and compared. Additionally, the proposed method is compared with the method directly using measurement data to train SVR models and the multivariate linear regression (MLR) method. A procedure for the assessment of bridge expansion joints making use of long-term monitoring data is developed. Based on the measurement data of expansion joint displacement and bridge temperature, the normal correlation pattern between the effective temperature and thermal movement is first established. Alarms will be raised when a future pattern deviates from this normal pattern. The extreme temperatures for a certain return period are derived using the measurement data for design verification. The annual or daily-average accumulative movements experienced by expansion joints are then estimated from the monitoring data for comparison with the expected values in design. The proposed procedure is applied to the assessment of expansion joints in the Ting Kau Bridge with the use of one-year monitoring data. In summary, the research described in this dissertation involves the development of a systematic approach from statistical identification of structural parameters to assessment of component reliability and condition based on long-term monitoring data. This approach enables structural damage identification and monitoring-based reliability assessment to be explored both in a probabilistic framework taking into account uncertainty and randomness inherent in measurement data and structures. Following this approach, structural health monitoring technology can provide quantitative information for bridge managers to enable decision making on the optimization and prioritization of bridge inspection and maintenance.