Application of artificial neural networks in vibration based damage detection

Abstract

Vibration based damage identification (VBDI) techniques rely on the fact that damage in a structure reduces its stiffness and alters its global vibration characteristics. Measurement of changes in the vibration characteristics can therefore be used to determine the damage in the structure. The VBDI technique does not depend on a-priori information about the damage site; the vicinity of the damage need not be accessible; and often a limited number of sensors can be used to localize and quantify the damage. Unfortunately, most of the available damage identification algorithms fail when applied to practical structures due to the effect of measurement errors, uncertainties induced by environmental and boundary condition, the need to use incomplete mode shapes, mode truncation, and the non-unique nature of the solutions. Damage detection based on changes in modal characteristics can be treated as a pattern recognition problem. Artificial neural networks provide an ideal means of obtaining a solution to such a problem. This paper presents a new robust two-step algorithm for detecting the location and magnitude of damage. The technique uses principles of structural dynamics and artificial neural networks. A modal energy based vibration property, known as the damage index vector, is used as the input to the network. The proposed algorithm is used to detect simulated damage in a simple finite element model of a slab and girder bridge. The result shows that the proposed algorithm is quite effective in identifying the location and magnitude of damage, even in the presence of measurement errors in the input data.