Damage Assessment Using Neural Network and Genetic Algorithm
Kayvan Saberi-Haghighi, Mohsen Ghafory-Ashtiany, and Caro Lucas
In this paper a damage assessment procedure has been proposed based on backpropagating feedforward neural network simulators and a genetic algorithm identifier. Damage assessment is performed in two steps. First, neural networks are utilized to locate possible damage states associated with the changes in vibration signature. Second, genetic based identification procedure has been applied to evaluate the dynamic parameters of the structure at damaged locations. The stiffness of the damaged parts of the structure has been identified by the genetic algorithm such that the difference between analytically predicted and experimentally observed response is minimized throughout the response time history. The amount of stiffness reduction is assumed as the degree of damage. To verify the performance of the proposed scheme, the location and degree of damage in computer-simulated linear and nonlinear structures has been detected. Also to investigate the performacne of the proposed method in conjunction with real data, experiments on a 12 scale model of a four-story steel structure has been performed.