Resumen
Optimizing the serviceability of highway bridges is a fundamental prerequisite to provide proper infrastructure safety and emergency responses after natural hazards such as an earthquake. In this regard, fragility and resilience assessment have emerged as important means of describing the potential seismic risk and recovery process under uncertain inputs. Generating such assessments requires estimating the seismic demand of bridge components consisting of piers, deck, abutment, bearing, etc. The conventional probabilistic model to estimate the seismic demands was introduced more than two decades ago. Despite an extensive body of research ever attempting to improve demand models, the univariate demand model is the most common method used in practice. This work presents a comprehensive review of the evolution of demand models capturing machine-learning-based methodologies and their advantage in comparison to the conventional model. This study sheds light on understanding the existing demand models and their associated attributes along with their limitations. This study also provides an appraisal of the application of probabilistic demand models to generate fragility curves and subsequent application in the resilience assessment of bridges. Moreover, as a sound reference, this study highlights opportunities for future development leading to enhancement of the performance and applicability of the demand models.