Gaussian process model-based approach for uncertainty quantification of natural frequencies of bridge

Abstract

Uncertainty in bridge structural parameters will unavoidably lead to uncertainty in natural frequencies of bridge. Uncertainty quantification of natural frequencies of bridge enables to provide more accuarte information of strucutral dynamic characteristics for the bridge design. Mean and variance, two importance order statistics, can be used to characterize the uncertainty of natural frequencies. This paper aims at developing an efficient and reliable appraoch to compute the mean and varaince of natural frequencies of bridge. Bridge is usually greatly complex and large scale, so the traditional Monte Carlo simulation method is infeasible due to the issue of high computational cost. In this paper, Gaussian process model (GPM) is adopted to replace the complex FEM of bridge, aiming at mapping the relationship between the uncertain parameters and natural frequencies. Within GPM framework, the mean and variance can be efficiently obtained. The present GPM-based method is suitable for calculating mean and variance of physical system responses with arbitrary parameter probability distributions. And its feasibility is verified through a benchmark function with analytical mean and variance. Finally, the present GPM-based approach is applied to compute statistics of natural frequencies of Anqing Railway Yangtze River Bridge.