Performance of System Identification Methods in Structural Dynamics

Abstract

Structural health monitoring aims at early detection of damage in structural systems and represents a convenient strategy to reduce maintenance and repair costs. An important step in this process is to identify dynamic parameters like frequencies, mode shapes and damping ratios, from accelerations recorded at certain points of the structure. The most widely used method for this is the Peak Picking, based on the Fourier transform. Other methods, originally developed for electronic control, are the subspace identification algorithms formulated in state space. In this paper, the performance of these methods was evaluated when they were applied to a steel beam. Different types of dynamic actions were analysed using numerical results, including free vibrations, harmonic loads and triangular pulse loads. Subsequently, experimental measurements were taken on a physical model subjected to free vibrations and forced vibrations produced by a passing vehicle. It is remarkable that the algorithms work with accuracy in some cases if the input remains unknown, but some information might be necessary to make a correct identification in case of specific types of loads.