Improved Subspace-Based Method Applied to Structural Damage Detection

Abstract

Precise and automatic identification of structural properties is a vital step toward detection and localization of damage in structures. However, even simple structures may present a complex dynamical behavior, and a large number of poles for a given frequency interval. This complexity leads often to numerical problems, like spurious modes in the identified model. In this article it is proposed a novel frequency-windowed subspace identification method, suitable to perform reduced-order identification, using band-limited excitation. The proposed methodology is employed to identify natural frequencies and damage detection of an experimental bench. Experimental results show its effectiveness to improve model identification. It is also shown that the frequency-windowing subspace identification is able to enhance damage detection, by generating more accurate damage indexes.