%0 Generic
%D 2013
%T Model Updating Cognitive Systems: Behavioral Selection
%A Gustavo A. Ospina
%K bayes
%K model updating
%K MUCogS
%X Numerical modeling has been a very effective tool for the simulation of structural systems. Modeling helps engineers and designers to make important predictions about the behavior of the system during simulated loading events. Decisions in structural engineering are most of the time based on the results from numerical simulations, especially during retrofit where numerical modeling is most of the time required. However, numerical idealizations of existing structural systems do not match the actual structure. Several authors acknowledge that the reason behind this mismatch between the behavior of numerical models and the actual systems are due to assumptions in the modeling process and uncertainty in the model parameters. Model updating strategies can be used to reduce the uncertainty in the numerical model, resulting in more meaningful results from structural analysis. Several researchers report the performance of model updating strategies in terms of the error between numerical and experimental data after the updating strategy has been implemented, but little work has been done in evaluating the physical meaning of the updated parameters and the capabilities of the numerical model to predict the behavior of the structure after the system has been modified (i.e. retrofit analysis). Furthermore, researchers have acknowledged that model updating can lead to non-unique problems, and propose techniques to identify potential solutions to the model updating problem. However, it is not clear how to select the appropriate solution from a family of solutions. The work described here proposes a methodology for the evaluation of a family of solutions within a probabilistic framework. The methodology proposes a mean for an analyst to incorporate his/her expertise as a probabilistic expression that can be incorporated in the model updating process. Solutions with high probability are more probable to have meaningful parameters. Finally, a benchmark problem is formulated to aid the comparison of model updating techniques that acknowledge the existence of multiple solutions.
%C University of South Carolina, Columbia, SC
%P 116
%G eng