Estimation of Modelling Errors in Structural System Identification

The study presented in this paper builds upon previous research into system identification using multiple models at EPFL. This identification methodology is based on the selection of sets of candidate models such that differences between measurements and their predictions are below a certain threshold value. The threshold value is a function of modeling and measurement errors. This paper presents results from a preliminary study on the estimation of modeling errors in structural system identification. Modeling errors have many sources, such as uncertainty in parameter values and finite element discretization. In addition to defining the threshold value, evaluating errors may also help minimizing errors and choosing better measurement locations. Two types of modelling errors are studied. The first is the error introduced by neglecting the effect of secondary structural elements such as roadway barriers and the second is due to element discretization. A composite truss bridge is used as a case study. Results from the study show that the errors due to exclusion of a secondary structural element may be significant and in such cases, they must be taken into account for system identification. Therefore displacement or rotation measurements are better for structural identification than strain measurements near beam-slab interfaces. The approach used and the results described are expected to contribute to increasing the accuracy of probabilistic studies on full scale bridges.