Online probabilistic model class selection and joint estimation of structures for post-disaster monitoring

Online selection of the appropriate model and identifying its parameters based on measured vibrational data are among the challenging issues in dynamic system identification. After a severe earthquake, quick monitoring and assessment of structural health status play a crucial role in effective critical risk management for the building owners and decision-makers. The Bayesian multiple modeling approach is a suitable tool for optimal model class selection, which is used in this article mainly for improving data fitting precision, decreasing dimensions of structural unknown vector through removing unnecessary parameters, detecting the occurrence and type of predominant phenomenon related to degradation and pinching, and finally increasing stability and convergence rate of the used identification algorithm. In this study, a reliable and effective time-domain method is proposed for online probabilistic model class selection and joint estimation of structures using the central difference Kalman filter and Bayesian multiple modeling approach. Also, in contrast to existing studies, the performance and efficiency of the proposed algorithm are investigated in nonlinear system identification with a large number of unknowns. The proposed algorithm is implemented on three moment-frame buildings with 8, 54, and 60 unknowns. In addition, a numerical example is provided to analyze the case in which the exact model is not among the plausible models. To verify the performance of the hybrid identification method, robust simulations with synthetic measurement noises and modeling errors were generated using the Monte Carlo random simulation method. The result shows the method can be used to simultaneously select the optimal model class and identify its unknown states and parameters in an online manner.

Automated summary

The study proposes a Bayesian multiple modeling approach for dynamic structural identification, aiming to improve data fitting precision, reduce dimensionality of structural unknowns, and enhance stability and convergence rate of the identification algorithm. Experimental validation on nonlinear systems with a large number of unknowns demonstrates the effectiveness and performance of the method.