Multi-stage identification scheme for detecting damage in cable-stayed Kap Shui Mun Bridge

This study aims to develop a multi-stage scheme for damage detection for the cable-stayed Kap Shui Mun Bridge (Hong Kong) by using measured modal data from an on-line instrumentation system, and to perform a damage-identification simulation based on a precise three-dimensional finite element model of the bridge. This multi-stage diagnosis strategy aims at successive detection of the occurrence, location and extent of the structural damage. In the first stage, a novelty detection technique based on auto-associative neural networks is proposed for damage alarming. This method needs only a series of measured natural frequencies of the structure in intact and damaged states, and is inherently tolerant of measurement error and uncertainties in ambient conditions. The goal in the second stage is to identify the deck segment or section that contains damaged member(s). For this purpose, the bridge deck is partitioned into 149 segments defined by 150 sections, and normalized index vectors derived from modal curvature and modal flexibility are presented for damage localization. The third stage consists in identifying specific damaged member(s) and damage extent by using a multi-layer perceptron neural network. Only the structural members occurring in the identified segment are considered in the network input, and the combined modal parameters are used as the input vector for damage extent identification.