An engineering approach to model-class selection for measurement-supported post-earthquake assessment

Abstract

Structural identification of existing structures is a subject of increasing interest in the civil-engineering community because of its potential to use measurement data to enhance asset-management decision making. An important structural-identification application is residual-capacity assessment of earthquake-damaged structures. Known to be potentially slow and subjective, current assessment practices rely mostly on expert-conducted visual inspection. Structural-identification techniques can help overcome shortcomings of visual inspection through improving estimates of residual capacity. Physics-based models are needed to predict structural behavior under future loading (extrapolation). Especially for earthquake-engineering simulations, a large variety of prediction models and techniques exists. While engineers often prefer simplified behavior models for assessment, data-interpretation applications usually involve detailed model classes. Neither choice is appropriate for all situations. This paper contains a proposal for more rational model-class selection than typically employed in current practice. Model-class selection criteria are described and illustrated using two cases. Knowledge of the earthquake signal is identified as the main criterion to select model classes and analysis tools. Displacement-demand predictions are reduced by up to 91% using structural identification techniques and are validated for all tested model classes by observed behavior under aftershocks. Applicability of this model-class selection is most attractive for post-earthquake assessment of residual capacity (not damage detection) where there is a reduced availability of measurement data, such as when there is no continuous monitoring data. This strategy provides useful support to engineers for key decisions related to asset management and structural resilience.