Analysis of a Bayesian Parameter Updating Scheme for Truss Structures Based on Sensitivity and Fisher Information

Truss structures are widely used in actual bridge systems. During a long period of service life, the aging process of the bridges and the influence of the environment cause the measurements to be susceptible to noise during observations, thus reducing the ability of the data to effectively limit the posterior distribution of the model parameters. To overcome these issues, a numerical model based on OpenSees is adopted as the forward model to identify the parameter-response sensitivity matrix using the finite difference method. Under the assumption that the measurements follow a Gaussian distribution, a Fisher information matrix is introduced to characterize the information content of different candidate schemes for ranking and selecting loading and measurement schemes. Then, the Bayesian parameter updates based on the selected representative schemes are performed, while posterior degeneracy for the parameters is calculated using Monte Carlo sampling methods. The efficacy of the selected schemes is demonstrated by comparing the ability to reduce uncertainties and parameter correlations.