Research On Bridge Infrastructures Damage Identification Based On Deep Learning

Bridge structure plays an important role in the transportation industry of connecting east and west,connecting north and south.However,after the stage of pursuing economic development,various micro and macro damages have been caused to the bridge structure due to the adverse factors such as vehicle overload,overdue service,environmental corrosion,vehicle collision and fire.Therefore,with the emergence and aggravation of damage,the bridge structure will face serious safety risks,which is not only related to the economic development of the local region,but also related to the safety of the lives of the majority of people,which can not be ignored.Therefore,the bridge health monitoring system appeared,which can monitor the state of the bridge structure online.As the core part of bridge health monitoring system,the importance of bridge damage identification method is self-evident.The bridge health monitoring system obtains a large number of data through the acquisition instrument.By exploring the deep relationship between the damage state and the data and the corresponding law,the state of the bridge structure can be determined in real time.Therefore,the superior ability of deep learning based on big data is reflected here,which has advantages over the traditional identification methods.This paper studies the application of deep learning theory in bridge structure damage identification.The main research contents include:(1)The literature on bridge health monitoring and bridge structure damage identification methods are reviewed,the advantages and disadvantages of various identification methods are analyzed,and the limitations of existing methods are pointed out;(2)This paper introduces the knowledge of deep learning,focuses on three common deep learning models,and finally determines the basic model of the method proposed in this paper;(3)In this paper,a method for bridge structure damage location identification based on stacked denoising auto-encoder is proposed.On the identification result of simply supported beam bridge and continuous girder bridge,with existing machine learning method based on the traditional BP neural network and support vector machine(SVM),compared to the proposed approach,recognition accuracy is much higher than the two,and have good ability to resist noise,have applied to practical engineering in the future;(4)In this paper,a method for bridge structure damage location identification based on deep belief networks is proposed.On the identification results of single damage unit in simply supported beams,with existing machine learning method based on the traditional BP neural network and support vector regression,the proposed method has good recognition accuracy,but do not have antinoise performance;(5)In this paper,a method for bridge structure damage location identification based on deep belief networks is proposed.On the identification results of multiple damage units in continuous beams,with existing machine learning method based on the traditional BP neural network and support vector regression,this paper puts forward the methods to identify accurate rate showed a trend of decline,does not have the ability to resist noise.At the same time,the accuracy of this method is lower than that of the traditional method.The identification results of this paper show that the stack-de-noising auto-encoder based damage location identification method of bridge structure can effectively judge the damage location of simply supported beam bridge and continuous beam bridge,which has a promising prospect.The depth belief network based bridge structure damage degree identification method can effectively judge the damage degree of simply supported beam bridge,has the prospect of application.However,due to the limitation of the special structural system and samples of continuous beam Bridges,the damage degree identification method based on the depth belief network is not suitable for the judgment of the damage degree of continuous beam Bridges,so the researchers need to further study other feasible methods.