A feasibility study on the use of bayesian model updating and vibration prediction for structural diagnostic

Abstract

This paper puts forward a feasibility study on the use of Bayesian model updating and vibration prediction for structural diagnostic when the level of modeling error is relatively high. The proposed method consists of two parts. In the first part, the Markov chain Monte Carlo (MCMC)-based model updating is employed to calculate the posterior PDF of uncertain model parameters conditional a set of measurement and a given model class. Modeling error is the key problem to be addressed in the practical implementation of structural model updating or damage detection. Apart from very simple structures, model updating of real structures is usually not globally and locally identifiable. Therefore, MCMC simulation is employed in the proposed method in generating samples in the important region(s) for the approximation of the posterior PDF. In the second part, the probabilities for the vibrations of the structure to exceed a list of threshold limits (i.e., the failure probabilities) were calculated using the MCMC samples. It is believed that the failure probabilities for the damaged structure are higher than those for the undamaged one. A 3-dimensional scaled transmission tower model was tested under laboratory conditions for verifying the proposed method. To test the robustness in the detection of damage existence, artificial modelling error was introduced to the model class in the numerical case study. The numerical case study results were positive implying the feasibility of the proposed method.