Structural Damage Identification Based On Convolutional Autoencoder Neural Network

As an important infrastructure,in recent years,bridge engineering is developed and constructed in an efficient,sustainable,and steady trend.However,under the long-term effect of various loads,the structural safety of the bridge will be generally destroyed.If the structural damage is not diagnosed and repaired in time,it will cause massive casualties and huge economic losses.Therefore,it is urgent to establish an early warning mechanism for structural damage.Since the rapid development of structural health monitoring(SHM),related researchers have proposed many methods for structural damage identification,and achieved some positive results.However,few researches focus on the real time identification of structural damage occurrence time,which is urgent and important in real time SHM.A timely and accurate determination of the instant of structural damage occurrence can provide timely information feedback for the early warning of structural damage and gain valuable time for structural repair and maintenance.In recent years,deep learning has been successfully applied in the field of SHM due to it has the advantages of automatic feature extraction.In this thesis,based on the method of convolutional autoencoder neural network in the field of deep learning,the following aspects are specifically studied for the identification of the structural damage occurrence time:(1)After the comprehensive analysis of the status of structural damage detection methods in the world,the problems and challenges of current SHM are summarized,and then an idea of damage detection using convolutional autoencoder neural network in the field of deep learning is proposed.(2)The theory of convolutional autoencoder neural network is introduced.Then a method of identification of structural damage occurrence time based on the convolutional autoencoder neural network is proposed.In this thesis,according to the characteristics of the sequential acceleration response data,the neural network model is established.This method is a data-driven method that it does not rely on the structural model.It has the ability to complete the adaptive feature extraction of abnormal behaviour of the signal and intelligent diagnosis of the health status of the structure.(3)Numerical simulation and experimental verification are applied on a simply supported beams under environmental excitation.Under the big data frame,the deep network model is used to characterize the complicated mapping relationship between response signal and health status,therefor the adaptive damage feature extraction under different damage conditions,accurate judgement of the health status,and the identification of damage occurrence time are fulfilled.The results of numerical simulation and experiment demonstrate that the proposed method can successfully be used for real-time and continuous health monitoring of bridge structure.