In this paper, multi-novelty indices are developed to detect the damage region in large-scale cable-supported bridges based on vibration measurement. Following this approach, the bridge is partitioned into a set of structural regions and it is assumed that there are vibration transducers at each region. For each structural region, a neural network based novelty detector is formulated by using the global natural frequencies and the localized modal components measured from the sensors located within this region. The modal flexibility values at the measured nodes are used to train an auto-associative neural network and to obtain a novelty index for each region. The damage region is signaled by the corresponding novelty index that displays drift from the training phase to the testing phase. The applicability of the proposed method for structural damage region identification is demonstrated by taking the suspension Tsing Ma Bridge and the cable-stayed Ting Kau Bridge in Hong Kong as examples, both the bridges being instrumented with a long-term monitoring system.
展开▼