ONLINE DETECTION OF STRUCTURAL DAMAGE USING NEURAL NETWORKS

An algorithm for on-line detection of damage to structures caused by g round shaking is presented. Real-time sampled response data are proces sed by templates in the form of ID (for structural identification) neu ral networks, which differentiate damage according to similarity of th e response to those encapsulated in the templates. Numerical examples based on a simple 2-story steel-frame building are used to illustrate the proceedings and to underscore the limitations of the method. The c hallenges of on-line damage detection are discussed in detail to promo te better understanding of how the proposed algorithm has evolved and, in particular, why neural networks are used. Widespread application o f the algorithm, and damage detection in general, depends on the estab lishment of an adequate ID networks database, a Far more daunting task in practice than in the theoretical setting of the paper.