Damage Identification Of Bridge Structure By Application Of BP Neural Network And Genetic Algorithm

Bridge structures in the service would be damaged in some places caused by repeated traffic loads and overloads, environmental erosion, material age effect, natural disasters, accidents and other man-made disturbances. Damage in the structural members would largely decrease the safety of a bridge. In recent years, damage diagnostic techniques become an active research area in the bridge engineering. A brief review for the existing damage identification methods is made. Different damage identification methods are discussed and compared. Then, a two-step damage identification approach is developed. In addition, a damage identification method based on the genetic algorithm by using static displacement values as the optimization aim is proposed. The validness and effectiveness of two proposed methods are verified by the numerical results.In this paper, a two-span continuous beam which has only one damage location is detected by neural network. Structure's first 6 step frequency rate-of-change is used in the detection. From the process of single damage identification, we can find that the training samples needed will be extremely numerous if there are multiple damage locations. This is a important reason that the neural network can't be applied to actual engineering structures. Hence, two-step damage identification method is developed in this work. In the first step, the damage locations are detected by the change rate of modal strain energy. Then, the damage degree can be accurately calculated by applying the BP neural network. The two-span continuous beam with multiple damage location is detected by this two-step damage identification method. The results shows that the two-step method is truly has very high efficiency in reducing training samples, the method can be used in complex actual engineering structure.The genetic algorithm is usually used in bridge damage identification. We generally use structure's modal data to construct the search target. This paper uses static data instead of dynamic data as the genetic algorithm's search target, because the static date's accuracy is generally higher than dynamic date's. The genetic algorithm is used in damage identification of simply supported beam, clamped beam, and continuous beam. A good result is achieved in this application. 5% of the random noise is added into the displace data in order to consider the influence of noise on this method. However, the results of identification is also very good. This shows that the method is not very sensitive to noise and it is very stable.