Reducing noise amplification effects in the direct updating of nonconservative finite element models

The problem of updating nonconservative F.E. models using an incomplete set of input-output measurements is addressed here. In particular, the paper focuses on the actions required to control and reduce the influence of experimental noise on the identification of the correction factors. These actions may include the exploitation of a priori information about the structure and the selection of working frequencies in order to obtain a low noise to signal ratio together with good system conditioning. But the most promising and influential tool to reduce scatter in the identified correction factors is represented by regularization techniques, such as the truncated Singular Value Decomposition, for which appropriate truncation criteria are sought. The effectiveness of the proposed criteria is checked by simulated experiments with noise polluted data. Noise is added in the time domain both on the input forces and on the responses.