MODEL UPDATING ANALYSIS BASED ON BAYESIAN INFERENCE

Resumo

The aim of this work is to estimate unknown system parameters based on observed dynamic
data, e.g. natural frequencies and damping rates. These dynamical inputs are extracted from experimental
modal analyses on a simply supported aluminum beam. Measurements come from three accelerometers
and the input excitation is provided by an impact hammer. The experimental setup is submitted into a
process of assembling and disassembling the beam for the purpose of increase the variability of modal
data. The Young’s modulus and the coefficients of the proportional damping model are considered the
updating variables in this study. The exploration of the posterior density function (pdf) of these unknown
model parameters is performed by a novel Markov Chain Monte Carlo method (MCMC) named Delayed
Rejection Adaptive Metropolis (DRAM). In other words the Bayesian paradigm for inverse problems
is adopted to tackles the structural identification. The impact of two likelihood functions at posterior
parameter distributions is analyzed.