Neural network based damage detection using substructure technique

摘要

Many researchers have been studying the feasibility of using Artificial Neural Networks (ANN) in structural health monitoring and damage detection. It has been proven by both numerical simulation and laboratory test data that ANN can give reliable prediction of structural conditions. The main drawback of using ANN in structural condition monitoring is the requirement of enormous computational effort. Consequently almost all the previous work described in the literature limited the structural members to a small number of large elements in the ANN model. This may result in the ANN model being insensitive to local damage, especially when this local damage is small. To overcome this problem, this study presents an approach to detect small structural damage by using ANN progressively. It uses the substructure technique together with a two-stage ANN to detect the location and extent of the damage. It starts by dividing the structure into a few substructures. The condition of each substructure is examined. Those substructures with condition change identified are further subdivided and their condition examined. By doing this progressively, the location and severity of low level structural damage can be detected. Modal parameters such as frequencies and mode shapes are used as the input to the ANN. To demonstrate the effectiveness of this approach, a two-span continuous concrete slab structure is used as an example. Different damage scenarios are introduced by reducing the local stiffness of the selected elements at different locations along the structure. The results show that this technique successfully detects simulated damage in the structure.