A two-level neural network approach for dynamic FE model updating including damping

Abstract

This paper presents a two-level neural network scheme for finite element (FE) model updating in which both the structural parameters and the damping ratios are updated. Considering the fact that in a lightly damped system the damping has only negligible influence on the resonance and antiresonance frequencies of the system, in the first-level updating the model is assumed to be free of damping and the structural parameters are updated using the natural and antiresonance frequencies as the response data. With the updated structural parameters from the above first-level updating, the second-level updating procedure deals only with the damping ratios, using the integrals of frequency response function (FRF) as reference responses. For the selection of a proper response configuration, a sensitivity analysis scheme is proposed, taking into account the carry-over error during the first-level updating in addition to the anticipated error in the measured FRF data. Through a numerical example it is shown that the approach is effective and efficient. It is also shown that by means of a noise injection learning the neural network can acquire considerable noise-resisting ability, resulting in about 50% reduction of the errors in the updated parameters as compared to the anticipated errors from the sensitivity analysis.