Measurement data interpretation and prediction for optimal management of bridges

For many developed countries, the asset management of aging infrastructure continues to present challenges and risks. Around the world, built-environment assets such as bridges, require functional changes and future-proofing in order to accommodate new and increasing needs of mobility. The assessment of existing bridges cannot be performed using conservative design-stage models. Moreover, measurements of real behaviour may provide additional information to infer previously unknown reserve capacity. Structural identification helps identify suitable models as well as values for the parameters that influence behaviour. Measurements of structural behaviour are often compared with the predictions of finite-element models in order to improve the understanding of real behaviour. Error-domain model falsification (EDMF) is a recently developed approach to structural identification that identifies plausible models, which are compatible with measurements, among a population of model instances. EDMF is easy to understand for practising engineers and can provide accurate parameter identification even when specific knowledge of uncertainty forms is unavailable. The performance of EDMF is affected by the definition of the uncertainty sources, by the level of detail of behaviour models that are employed, and by the quality of data provided by the measurement system. Furthermore, engineers performing identification need support in order to generate the population of model instances and to interpret identification outcomes. Subsequently, updated models can be used to conduct extrapolation tasks, for example, to predict the reserve capacity of existing bridges for deck-widening scenarios. This thesis proposes a range of tools that support practising engineers working on structural identification of existing bridges. An adaptive sampling strategy is developed to improve identification performance while reducing the computation effort. As structural identification methodologies are not sufficiently robust when incorrect measurements are included in the dataset, a new outlier-detection methodology based on the expected performance of sensor identification is proposed. In addition, this thesis describes a result interpretation tool, based on clustering, that helps engineers interact with data to prioritise the various tasks of structural identification in order to support decision-making. Finally, a rigorous framework to assess the reserve capacity of existing bridges with respect to both serviceability and ultimate limit states is illustrated. These proposals are tested and verified within three case studies concerning full-scale bridges. Proposed tools are compared with state-of-the-art methodologies, and results are verified using strong validation approaches. The results demonstrate that the proposed strategies outperform traditional approaches and improve EDMF performance. Moreover, the case studies show that the combined use of advanced models and sensing help exploit reserve capacity sources that are neglected by conservative design-stage analyses.