BAYESIAN UPDATING OF STRUCTURAL MODELS USING DRAM ALGORITHM

摘要

A finite element model updating method is presented for uncertainty quantification of measurement noise.Based on the maximum information entropy,the generalized unbiased prior distribution is deduced and then the theoretical posterior distribution of updating parameters is given.Markov chain Monte Carlo simulation methods(MCMC)is implemented to solve the high dimensional posterior distribution by numerical calculation.To enhance the computation performance of MCMC,two algorithms,namely the adaptive(AM)and delayed rejection algorithm(DR),are introduced into MCMC,in which AM can automatically adjust the variance of the proposal distribution and DR can improve the acceptance probability of new samples.Results from a numeral simulated 5-story plane frame show that the means of updating parameters can converge to the presupposition values promptly,and results from the experiment of a cantilever steel beam show that frequencies of the updated finite element conform to the measured values satisfactorily.