Data-Based Models for Assessment and Life Prediction of Monitored Civil Infrastructure Assets.

In a context of deteriorating civil infrastructure and limited funds available to ensure their functionality and safety, there is a need for accurate assessment methods leading to a better allocation of the available resources and a timely detection of abnormal behaviours. Nowadays, technological advances have enabled the acquisition of reliable monitoring data concerning environmental conditions, loadings and structural responses from civil infrastructure assets. With massive amounts of data generated by monitoring systems, the challenge lies on how to extract relevant information that can be used for an enhanced management of civil infrastructure. Motivated by this, the research presented herein is devoted to the development of data-based models and associated methodologies for monitoring data interpretation, assessment and probabilistic life prediction in the specific area of fatigue reliability of welded joints in orthotropic steel decks. Moreover, it contributes to the definition of temporal and spatial requirements for monitoring campaigns and to the assessment of their cost-effectiveness within the present work's application framework. The proposed data-based models, associated methodologies and analysis are illustrated using the monitoring data from the Great Belt Bridge (Denmark). Polynomial regression models are firstly developed to characterize the correlation patterns between environmental conditions (pavement temperatures), operational loads (heavy traffic counts) and a strain-based performance indicator proportional to S-N fatigue damage at monitored welded joints. Monitoring outcomes are also used to develop time series models for simulating the main actions contributing to the fatigue process under consideration, namely pavement temperatures and heavy traffic counts. A methodology for probabilistic fatigue life prediction is then developed by integrating the different data-based models within an S-N fatigue reliability framework. It is based on Monte Carlo Simulation to account for the uncertainty in random variables (e. g. material properties, fatigue model) and random processes (e. g. traffic, temperature) and estimate the remaining fatigue life of selected welded details. The developed method enables to quantify the effect of different scenarios in terms of changes in pavement temperatures and heavy traffic counts. Moreover, an algorithm based on statistical control charts defined by the prediction bands of the regression models is proposed for the interpretation of new monitoring data and the identification of abnormal behaviours, as part of an envisaged "real-time" assessment. Temporal and spatial requirements for monitoring campaigns are determined on the basis of the quantification of the epistemic uncertainty reduction provided by increasing monitoring datasets within the context given by the developed methodology for probabilistic life prediction. Finally, the benefit of monitoring techniques is assessed at different points in time through a posterior decision analysis. The work presented in this thesis provides a theoretical framework that could be adopted in assessing other structural components under different deterioration mechanisms, hence contributing to a wider and more effective use of monitoring-based techniques for enhanced infrastructure asset management