This paper addresses the application of Tikhonov regularization method for output-enor-based finite element (FE) model updating, with research emphasis on determining the optimal value of the regularization parameter Tikhonov regularization is applied at each linearization step of the optimization problem arising from model updating to alleviate the ill-conditioning. Three methods, namely the L-curve method (LCM), generalized cross validation (GCV), and minimum product criterion (MPC), are explored to determine the regularization parameter. The performance of the three methods for regularization parameter selection is rigorously examined and assessed by means of numerical studies of FE model updating of a truss bridge using noise-free and noisy 'measurement' data, respectively It is shown that MPC is the most effective and robust in determining the optimal regularization parameter, and the adaptive strategy that allows variable value of the regularization parameter at different iteration steps is more effective and efficient than the fixed strategy using a constant value of the regularization parameter at all iteration steps.
展开▼