Parameter Selection in Finite-Element-Model Updating by Global Sensitivity Analysis Using Gaussian Process Metamodel

Abstract

Parameter selection is a key step in finite-element-model updating (FEMU) because it determines whether the task of FEMU is successful or not. The as-built engineering structures are inevitably subject to many sources of uncertainty such as geometric dimension variability due to manufacture process, inherent random variation of materials, and imprecisely known boundary conditions. Uncertainty involving parameters challenges the task of parameter selection in FEMU. In this paper, the powerful global sensitivity analysis (GSA) is proposed to perform parameter selection in FEMU when uncertainty exists. The Monte Carlo simulation (MCS) method is extensively adopted to perform GSA. However, the brute-force MCS method is likely to be unaffordable and impractical because it entails a large number of model evaluations due to its slow convergence. Therefore, the Gaussian process metamodel is used as the surrogate model of the time-consuming finite-element model to ease the heavy computational burden. Gaussian process metamodel is favored here because of its probabilistic, nonparametric features and high capability of modeling a complex physical system. The space-filling Sobol sequence sampling method is utilized to generate the informative training data for establishing the Gaussian process metamodel. Finally, two study cases of the simple flat steel plate and full-scale arch bridge are presented to detail the procedure of employing the proposed GSA method to select parameters for FEMU.