| With China’s rapid economic development,the country’s investment in infrastructure construction is constantly increasing and the bridge engineering has also achieved unprecedented development.However,because of environmental erosion,material aging and other factors,bridges will inevitably be damaged during the service period which affects the service performance and safety performance of bridge structures.Therefore,it is very important to know the structural health state from the structural response timely and accurately.In this paper,combined with the current popular deep learning technology,the bridge damage identification problem is studied by introducing the deep belief network.The main work and conclusions are as follows:(1)The research status of bridge damage identification technology is reviewed,the damage identification technology based on dynamic characteristic parameters,neural network and deep learning is introduced,the advantages and limitations of existing technology are analyzed,and the application of deep belief network in bridge damage identification is further selected.(2)The basic principle of the deep belief network is expounded,Taking the simple beam as an example,considering vibration mode difference as the damage index,it is proved that the multi-layer restricted Boltzmann machine has good characteristic extraction ability.(3)Taking a two-span continuous beam as an example,the modal curvature difference is used as the damage index and the proposed method is applied to identify the damage location and damage degree of the structure.The research results show that the deep belief network is better than the traditional BP neural network in the identification of damage location and damage degree.In the case of fewer measuring points,the deep belief network also shows strong robustness.(4)The deep belief network is applied to the damage identification of a concrete filled steel tube(CFST)arch bridge and the identification process is divided into three stages: damage alarming,substructure identifying and damage position fixing.The results show that the deep belief network can complete the identification tasks of these three stages well,but when the noise increases and the damage position increases,the ability of the proposed method extraction features decreases and the identification accuracy decreases.(5)Based on the three-layer frame test of Alamos Laboratory,linear damage and nonlinear damage identification of the structure are realized directly by using the frequency function as the input of the deep belief network.The results show that: the proposed method can achieve 100% complete identification for linear damage of structure,while for the nonlinear damage,the proposed method can only achieve 83.6%identification accuracy.For the two types of damage,better identification results are obtained when the frequency function obtained by the top and bottom sensors is used as the network input.In summary,it is feasible to apply the deep belief network to the field of bridge damage identification.The proposed method can improve the identification accuracy well and has a broad prospect of engineering application. |
