<title>Damage detection of bridgelike structures using neural networks</title>

Abstract

It is well known that the static and dynamic structural response of materials can indirectly indicate the health of structural systems. The changes in natural frequencies, mode shapes, and stiffness matrices due to damage are utilized for determination of occurrence, location and extent of damages. In recent years, many researchers have developed global damage detection algorithms using structural modal response. However most of these methods are off-line techniques based on frequency domain data. In this paper we have proposed real-time damage detection methods based on time domain data. In this method damages in the structure can be detected while the structure is kept on its regular use. The algorithm determines reduction in stiffness and/or damping of the structural elements, while assuming that the mass of the structure does not vary due to damage. This algorithm is based on the state space representation of the structure, which is identified from the time domain data. We have also determined a linear transformation matrix for converting the identified model into a state space representation based on physical coordinates of the structural system.The self-organization and learning capabilities of neural networks can be effectively used for structural damage detection purpose. In this paper a hybrid method for the damage detection has been proposed by combining the features of best achievable eigenvector method and neural network classification techniques for detection of location and extent of damage in the structural systems. The feasibility of the proposed method is verified by using simple three-bar truss structure and a cantilever beam test article.