A multi-criteria decision framework to support measurement-system design for bridge load testing

Due to conservative design models and safe construction practices, infrastructure usually has unknown amounts of reserve capacity that exceed code requirements. Quantification of this reserve capacity has the potential to lead to better asset-management decisions by avoiding unnecessary replacement and by lowering maintenance expenses. However, such quantification is challenging due to systematic uncertainties that are present in typical structural models. Field measurements, collected during load tests, combined with good structural-identification methodologies may improve the accuracy of model predictions. In most structural-identification tasks, engineers usually select and place sensors based on experience and high signal-to-noise estimations. Since the success of structural identification depends on the measurement system, research into measurement system design has been carried out over several decades. Despite the multi-criteria nature of the problem, most researchers have focused only on the information gained by the measurement system. This study presents a framework to evaluate and rank possible measurement-system designs based on a tiered multi-criteria strategy. Performance criteria for the design of measurement systems include monitoring costs, information gain, ability to detect outliers and impact of loss of information in case of sensor failure. Through including conflicting criteria, such as cost of monitoring and information gain, the optimal measuring system becomes a Pareto-like choice that ultimately depends on asset-manager preference hierarchies. Several potential preference scenarios are generated and results are compared using a full-scale test study, the Exeter Bascule Bridge. The framework successfully supports an informed design of measurement systems by providing an extensive set of alternatives, including the best solution defined probabilistically and for specific conditions when other near-optimal solutions might be preferred.