Investigation of the effect of model uncertainties on structural response using structural health monitoring data

Calibrated mathematical models used to simulate the actual behavior of civil engineering structures provide more realistic results for further decision making processes. However, the effect of modeling uncertainties has to be considered in response predictions in order to increase the reliability and the capability of the updated models. In this study, model uncertainties are quantified in finite element model updating problem by means of fuzzy analysis. The accuracy of response predictions are investigated for different measurement data sets and level of model complexity. Fuzzy Finite Element Model Updating (FFEMU) approach is employed with some constraints applied on the parameter space. These constraints are handled using Genetic Algorithms. In addition, Gaussian Process model is used as surrogate model in order to tackle with cumbersome repetitive model calculations. University of Central Florida (UCF) benchmark structure developed for testing Structural Health Monitoring (SHM) technologies and algorithms is used to demonstrate the applied methodology. All experimental tests have been repeated couple of times in a controlled laboratory environment. The experiments show that the measurement noise coming from sensors is insignificant compared to the modeling errors. Hence, the trade off between the uncertainty amount and the accuracy in predictions for different level of model complexity and measurement data sets is illustrated in the context of this study. According to the results, this trade off should be considered for more reliable calibrated models used to estimate the actual structural response.