Research On Convolutional Neural Network Model And Adversarial Strategy Theory For Structural Damage Identification

At present,the main challenges in the field of civil engineering in China have changed from the initial construction needs to the management,detection and maintenance needs.Civil engineering structure will inevitably appear damage during its life period.When the damage accumulates to a certain extent,it will pose a threat to the safety and durability of the structure,and even cause engineering accidents,resulting in property losses and casualties.Therefore,the study of structural damage identification during construction and operation has important theoretical value and practical significance.Deep learning theory is an important research result in the field of artificial intelligence in recent years.It has outstanding feature learning ability,adaptive ability and generalization ability,especially in big data processing and pattern recognition.In this thesis,based on deep learning theory,convolution deep network model and adversarial strategy theory for structural damage identification are studied.The main research contents are as follows:(1)This thesis introduces the research background and significance of structural damage identification,and summarizes the traditional structural damage identification methods.Based on deep learning theory,two reliable and efficient structural damage identification methods are proposed.One is the application of Convolutional Neural Network(CNN)in structural damage identification,and the other is the application of Deep Convolutional Generative Adversarial Networks(DCGAN)in structural damage identification.(2)Research on convolutional neural network model for structural damage identification.Based on the theory of CNN network,this thesis expounds the basic principle and structure of CNN network,and then expounds the application principle of CNN network in structural damage identification,and puts forward the basic framework of convolution neural network model in this thesis.Finally,combined with the test beam,the feasibility of CNN network model in structural damage identification is verified,and it is proved that the CNN network model is an efficient identification method.(3)Research on the influence of the randomness of structural parameters on the damage identification of convolution neural network model.In practical engineering,structural parameters such as structural mass,elastic modulus and damping ratio are random,which will affect the damage identification accuracy of CNN network.The results show that the randomness of structural parameters will reduce the identification accuracy of CNN network,and this effect cannot be ignored.(4)Research on the influence of spatial heterogeneity of noise on damage identification of convolution neural network model.In practical engineering,the spatial heterogeneity of noise has a certain influence on the damage identification accuracy of CNN network.The results show that the more uneven the noise in space,the lower the accuracy and stability of CNN network recognition.(5)Research on adversarial strategy of structural damage identification.In this thesis,a damage identification method is proposed.This method utilizes the unsupervised learning ability of DCGAN network,which can automatically extract the characteristic information of structural acceleration response under given conditions,and thus determine whether the structure is damaged.Combined with experiments,it is proved that DCGAN network is a very efficient identification method and has the potential to be popularized in practical engineering applications.