Research On Bridge Damage Identification Based On Deep Learning

Bridge is the key of the traffic infrastructure,which plays an important role in convenient transportation and rapid economic development.However,the bridge will inevitably be damaged due to fatigue,overload and earthquake.This will affect the performance of the bridge,and even threaten the safety when the damage is serious.Therefore,it is necessary to detect the bridge damage in time and accurately.In addition,deep learning is an important achievement of machine learning in recent years.Moreover,deep network structure has advantage in recognition performance compared with traditional pattern recognition method.But the research on the application of deep learning in bridge damage identification is still insufficient,so the application of deep learning method in bridges damage identification is studied.The main contents include:1,The research status of deep learning is reviewed,and 3 commonly used methods of deep learning are introduced.Furthermore,the characteristics of different methods is compared,then the stacked denoising autoencoders is selected for further study.2,In view of the characteristics of bridge damage identification,stacked denoising autoencoders suitable for bridge damage identification are built,and a bridge damage identification method based on stacked denoising autoencoders is proposed.3,Taking the simply supported beam as an example,the identification of the damage location and damage degree are realized by the proposed method with the structural acceleration response taken as the damage index.The results indicate that the stacked denoising autoencoders method based on acceleration index has great potential.4,Taking the continuous girder bridge as an example,the identification results of the proposed method and the BP neural network method are compared.The results show that the identification accuracy and noise immunity of proposed method is better than that of BP neural network method,and its advantage is more obvious in damage location identification.5,Based on a shaking table experiment of cable-stayed bridge,the seismic damage identification of the bridge tower is realized by the proposed method.The identification results are in good agreement with the damage state obtained by the observation.6,The static and dynamic identification of bridge damage are realized by the proposed method and BP neural network based on a curve cable-stayed bridge model experiment.Furthermore,the results indicate that the identification performance of the proposed method is better than that of BP neural network,and its advantage is more obvious when the damage index is hard to distinguish the characteristics of different damage patterns or the quality of data is poor.In summary,the deep learning method represented by the stacked denoising autoencoders can improve the accuracy of bridge damage identification,and has a good prospect of application in bridge damage identification.Moreover,further research is need for its extensive application in bridge damage identification.