Structural damage identification using pole/zero dynamics in neural networks

A method utilizing both poles and zeros in a backpropagation neural network was developed for structural damage identification. The effectiveness in structural damage was demonstrated by employing Kabe's problem. The damage conditions of the system with closely spaced natural frequencies were successfully identified. The discrepancy between the network identification results and desired output was within 1.7%. One or few measurements could be selected in modal testing for the change of pole/zero as the input data of a diagnosis network.