Damage Identification Based On The Loaded Frequency Changes And Deep Learning For Bridge

Under the comprehensive action of various factors,the bridge will inevitably be damaged during the service phase,which will reduce the safety and reliability of the bridge.It is of great practical significance to get bridge status information in time.In this paper,the damage identification method based on the loaded frequency changes and the stacked auto-encoder networks is studied.The main research contents are as follows:(1)The methods of bridge damage identification are reviewed.The methods are summarized from local and global identification aspects.The methods based on structural dynamic characteristics are introduced from two aspects of signal processing methods and identification methods,and its advantages and limitations are analyzed.(2)The theory of deep learning is introduced.Three common deep learning networks are introduced.Given the characteristics of each network,the Stacked Auto-Encoder(SAE)is selected to identify the damage of bridge in this paper.(3)The damage indexes based on the loaded frequency changes of bridge is constructed.The principle,parameters and limitations of the damage identification methods based on frequency changes are introduced,which lead to the indexes based on loaded frequency changes.Taking the simply supported beam as the research object,the indexes are analyzed.The results show that the identification indexes based on the loaded frequency changes can reflect the damage location and degree of bridge,which can be used as good inputs for damage identification model.(4)The bridge damage identification model based on SAE is constructed.The test is carried out with simply supported beam,and the anti-noise of the damage model is studied by introducing the Stacked Denoising Auto-Encoder(SDAE).The results show that: the damage identification model can correctly identify the damage location and degree of bridge;Compared with back propagation neural network,SDAE network has higher accuracy and confidence degree under the same noise environment,which has a better anti-noise ability.(5)The bridge damage identification model based on loaded frequency changes and deep learning is applied to a continuous beam bridge.The research of identification for the damage location and degree is carried out,which shows that the identification model has a good ability to identify the damage location and degree of bridge.