Monte Carlo Study of the Effect of Measurement Noise in Model Updating with Regularization

Abstract

Finite-element (FE) model updating aims at the parametric identification of a structure by correcting model parameters in an initial FE model of the structure to reconcile FE predictions with experimental counterparts. However, experimental data inevitably contain a certain level of measurement noise, and the measurement noise will further generate error and uncertainty in updating results. This paper presents a Monte Carlo (MC) simulation study of the effect of measurement noise on updating parameters in FE models updating with regularization, attempting to quantify the distribution functions of updating parameters in face of measurement noise, and evaluating the adequacy of moment-based stochastic FE model updating algorithms. Taking a numerical study of model updating of a simple truss bridge as an example, a series of artificial measurement noise generated with the normal distribution of zero mean and varying variance is introduced into the simulated modal parameters to quantify the effect of measurement noise on updating parameters. The results indicate that the coefficients of variation of the updating parameters are quite different in magnitude, implying different sensitivities of the updating parameters to measurement noise in modal data. In the case of a low level of measurement noise, the updating parameters also comply with normal distributions; in the case of a high measurement noise level, however, the updating parameters are largely nonnormal. The significant deviation from the normal distribution warrants that the low-order, moment-based stochastic FE model updating algorithms may be inadequate in the presence of severe measurement noise.