Research On Damage Identification Based On Neural Network For Shihe Bridge

The bridge belongs to urban traffic, but it’s material performance and reliability will reduce in the long-term effects of the traffic load and natural disasters, and it will bring big loss once damaged. Therefore, finding the bridge damage timely is particularly important. This paper studies the finite element model, and tests that the method of neural network is able to identify the structural damage. Then a cloud particle swarm optimization is introduced in neural network for structural damage identification and tests the method.Firstly, a finite element model is established on the basis of the drawings and material property of shihe bridge. In order to be close to the sensor measurements of bridge, the model is updated based on orthogonal design by selecting suitable material parameters.Secondly, select damage location and damage factor, then simulate structural damage of different positions and various degrees based on finite element model in order to obtain damage factor. The damage factor will be as input parameters of neural network after subtracted from undamaged factor and normalization processed to distinguish structural damage. The simulation results show that the method of neural network is able to identify the damage location and damage degree in structures, and the accuracy of neural network based on change rate of strain is better than the accuracy of neural network based on modal frequency. Then, in order to identify health condition of shihe bridge, the real-time data measured by sensors in bridge will be as input parameters of trained neural network, the output of neural network will show the damage conditions.Finally, in order to increase the stability of the results, a method of cloud particle swarm optimize neural network is presented in this part, it is avoid the instability of neural network results because of randomness of initial weights, even appear damage misjudgment situation because the results of neural network sink into local optimum. The simulation results show that the method of cloud particle swarm optimize neural network is able to identify the damage location and damage degree in structures, and the results are stable. Then, in order to identify health condition of shihe bridge, the real-time data measured by sensors in bridge will be as input parameters of trained CPSO optimize neural network, the output of CPSO optimize neural network will show the damage conditions.