Assessment of the dynamics and performance of an in-service twelve-span twin bridge using in-situ testing and finite element modelling

Abstract

Highway bridges are among the most critical components of any nation’s infrastructure. They are subjected to dead and live loads along with extreme stresses resulting from natural and man-made hazards. To improve the quality of bridge infrastructures and avoid extensive human and economic losses due to bridge collapse or loss of functionality, the performance of structures in service needs to be properly assessed to maintain safe and efficient operation. To that end, various aspects of vibration based safety and performance assessment of bridge structures are investigated in this study. The results of vibrationbased performance assessment of a twelve-span curved post-tensioned concrete bridge using full-scale testing data and finite element (FE) modelling are presented and discussed. The study revolves around establishment of an experimentally validated baseline FE model essential for performance assessment of a complex bridge. The first part of this dissertation is devoted to experimental operational modal analysis (OMA) of the bridge. Two full-scale, in-situ ambient dynamic test campaigns executed on the bridge during construction and shortly after completion are described. Four outputonly system identification techniques were used to analyse the data obtained in these tests in order to identify the modal properties of the viaduct. These modal identification methods included: i) peak picking (PP), ii) enhanced frequency domain decomposition (EFDD), iii) eigensystem realisation algorithm-natural excitation technique (ERA-NExT), and iv) data-driven stochastic subspace identification (SSI-DATA). A comparative study of these four output-only system identification techniques was conducted to assess their performance. The results showed that ambient-vibration measurements were sufficient to identify multiple structural modes with low natural frequencies. The accuracy and efficiency of the four system identification methods were investigated and compared. Overall, the natural frequencies and mode shapes identified using the different identification methods were found to be in good agreement across the four methods, although the methods entailed varying computational load. However, EFDD gave the highest quality results followed by SSI-DATA. Damping estimates, on the other hand, showed considerable variability between the methods, and within individual methods when applied to different segments of data. Model calibration of the developed preliminary FE model is then presented. A model updating procedure based on the experimental characteristics identified was used for the calibration. The preliminary FE modelling of the bridge was based on the information provided in the design documentation, material testing data and from site inspections. Relatively large differences were observed when the experimentally identified natural frequencies and mode shapes were compared with their analytical counterparts. The response surface (RS) method based on the support vector machine (SVM) was proposed and utilised for the identification of structural parameters related to the stiffness properties of critical elements of the bridge. To aid comparison with the proposed model, a second-order model was also employed for the model updating. A hybrid optimisation procedure based on genetic algorithms (GA) was implemented to find the best set of FE model parameters for minimising the objective function. The objective function was defined by expressing the discrepancy between the measured and analytical response characteristics. Different parameters of the model were calibrated using the proposed procedure to improve correlation between the measured and calculated modal parameters. A discussion and comparison of the model updating results achieved by using the two RS methods follows. The consistency of the final FE model updating results between the two methods enabled confirmation of the updating results. In the final part of the dissertation, a vibration-based method for structural performance assessment is proposed based on FE modelling calibrated using experimental data. Following the proposed approach, one more full-scale, in-situ ambient dynamic test campaign was executed on the Newmarket Viaduct after two years of operation. The measured natural frequencies and mode shapes extracted from vibration data collected at different construction stages were compared and analysed to reveal the response mechanisms. The analysis of the modal data from the two tests, separated by a two-year interval, showed that there was no obvious structural change. The fuzzy c-means method (FCM) was also used to further check and confirm the results. Based on the results, the reference FE model of the bridge did not require updating at this time. Once a change is identified, updating will be applied in order to identify a new structural state. The comparison of the updated model parameters with their baseline values would then be used for identification of parameter changes within the structure. Finally, the updated model was used for evaluation of the bridge structure’s load carrying capacity (LCC). Evaluation of bridge’s LCC comprised the prediction of the deflected shape and girder internal forces due to the dead load and traffic loading. Based on the results, it can be shown that the study bridge has the capability to withstand the design traffic loads with a considerable safety margin. The research work presented and the results obtained in this dissertation will contribute to the development of robust and reliable vibration-based performance assessment practice for large and complex bridges. It was shown that field vibration tests coupled with FE modelling and updating can be reliably used for structural performance assessment over time.