Structural health monitoring (SHM) and damage detection based on vibrationbased techniques have attracted significant attention in the past. Many of these methods rely on appropriate Finite Element (FE) or other mathematical model of a structure and updating the model using the field data. There are a number of existing methods for updating such models and identifying system parameters such as stiffness and mass based on dynamic parameters response of a structure. These methods can be broadly categorized as Physics-based and data-driven models. Since, the model updating problem is an inverse problem, the Physics-based model often provide nonunique solutions. On the other hand, data-driven models using methods like artificial neural networks (ANN) or Genetic Algorithm (GA) utilizes the differences in the data patterns in the structural response in order to update the system parameters. In this study, the FE models of two bridges, a Pre-Stress Concrete Box (PSCB) girder bridge and a Void Slab Bridge have been constructed and updated using measured vibration data. It should be mentioned that both of the bridges are relatively new and there is no reported damage in them. The bridges were instrumented with accelerometers and the modal vectors and frequencies were determined based on ambient vibration tests. The aim of this research is to compare the physics-based and data driven methods of model updating using the data from ambient vibration tests. While the physics-based method was found to be is more precise than GA and NN in both of case studies, the latter two methods provide the flexibility when the data sets are incomplete and noisy.
doi: 10.12783/SHM2015/156