Auto-associated memory neural network method of structure damage position detection using mode coding

This paper discusses the algorithms of the auto-associated memory neural network and presents a novel approach for structural damage detection which is based on the auto-associated memory neural network. The training patterns are different modal vectors of the structure when structural damage happens in different locations. In order to make use of the auto-associated memory neural network to identify structural damage location effectively, a totally new coding method is presented which coverls the modal vectors of structures into code before training the neural network. This approach has eminent convergence properties and does not have to get stuck in local minima as compared with the BP neural network. In addition, a reliability analysis method on the basis of the theory of vector distance is developed to confirm the effectiveness of detection results. The example of a cantilever beam is given to demonstrate and verify the presented approach and it is found that the damage identification method based on the auto-associated memory neural network is effective.