Probability-based uncertainty evaluation through Markov Chain Monte Carlo sampling and response surface technologies

摘要

Inspection of a pre-stressed concrete variable cross-section box girder bridge discovered the phenomenon of padding in expansion joint, corrosion of steel plate and local edge failure in pot type rubber bearing, and cracks of box girder. They are the main sources of structural uncertainty for structural performance evaluation, and how to quantificationally evaluate their influences on bridge performance is important. In this article, an approach using the Markov Chain Monte Carlo sampling technology and the Response Surface Method is proposed to deal with the uncertainty problem. First, a population of finite element (FE) models will be established by sampling the main uncertainty sources through the Markov Chain Monte Carlo technology. Then, the posterior probability of each FE model will be evaluated by using the measured static responses and identified structural dynamic characteristics. Especially, the second order response surface method will be used in this step to improve the computation efficiency. Through the above procedures, probability features of the defined key parameters representing structural uncertainty, including the stiffness of expansion joint, the stiffness of pot type rubber bearing and the elasticity modulus of the box girder will be estimated, which will provide valuable information for reliable structural performance evaluation.