Methodology for updating nonlinear structural models through experimental data acquisition

Abstract

Nonlinear numerical models are considered to be more accurate because they furnish a realistic representation of the structure under analysis. Moreover, it is known that different uncertainty sources should be considered when evaluating structure behaviour. Therefore, probabilistic based models, which consider the structural properties as distribution functions, are being implemented. In some situations, experimental data is respectively collected from permanent monitoring systems, nondestructive tests and visual inspection, to control this randomness. Consequently, the developed nonlinear probabilistic model may be updated through the use of a Bayesian inference algorithm. The advantage of this framework is its reduced computational cost and the reduced source of statistical uncertainty. This methodology is validated with a set of reinforced concrete structures, tested up to failure, in a laboratory. These structures are modelled with one nonlinear analysis software. The correspondent probabilistic model is then obtained through the introduction of proper distribution functions. This model is then updated with data collected from complementary characterization tests. Obtained results are used to validate the proposed inference algorithm.