Research On Structural Damage Identification Method Based On Covariance Of Covariance Matrix Of Acceleration Responses And BP Neural Network

With the advancement of human technology,there are more and more large-scale infrastructure.As time accumulates,the structure of these large infrastructures will age and damage.From the perspective of safety,it is essential to monitor the structure and accurately assess its health.Analysis of monitoring data requires an effective analytical approach.The neural network has powerful functions,good generalization ability,non-linear mapping ability and high parallelism,which can improve the accuracy of damage recognition.Therefore,it is widely used in structural damage identification.When using neural network methods to identify structural damage,it is important to construct indicators that are sensitive to damage.The Covariance of Covariance Matrix(CoC Matrix)of the acceleration response under white noise excitation is proved to be only related to the modal parameters of the structure(natural frequency,mode shape and damping ratio)but not the time dimension,it can be used as an indicator to identify structural damage.This paper studies the method of combining the CoC Matrix and BP network to identify structural damage.Firstly,under the white noise excitation,the acceleration responses of the structure is obtained and used to calculate the corresponding CoC matrix.Secondly,the CoC matrix is used as the input feature vector for BP neural network to train the network.Thirdly,the trained network is used to identify the structural damage.This article is mainly composed of the following five parts:(1)This paper introduces the research significance and development status of structural damage identification,summarizes the current research methods,and combines the future development trend of damage identification,and proposes damage identification method based on CoC Matrix of acceleration response under white noise excitation and BP neural network.(2)The basic theories of CoC matrix and BP neural network are expounded,and the neural network is used for damage identification.It provides a theoretical basis for the research method.(3)The numerical simulation of the six-steel frame structure is carried out,the basic theory of wavelet packet analysis is expounded,and the proposed method is compared with the wavelet packet energy.The recognition results of steel structure under single layer damage and multi-layer damage are discussed.The maximum average errors of CoC matrix and wavelet packet energy are 5.41% and 44.11%,respectively.The numerical simulation results of the truss structure are added with different levels of noise respectively.The proposed method is compared with the modal indicators,and the influence of the number of points and the learning rate on the recognition results is studied.The numerical examples based on the simple models show the feasibility and effectiveness of the proposed method.(4)This method is applied to the ASCE benchmark model proposed by the American Society of Civil Engineers for verification.First,the structural damage layers are located;Secondly,the damage unit and the damage extent are identified.When performing damage unit and damage extent identification,the accuracy of using CoC matrix and wavelet packet energy are 95.79% and 88.27%,respectively.The experimental data collected by the Columbia University Earthquake Engineering Laboratory is analyzed.The results show that the wavelet packet energy will be misjudged,and the CoC matrix can accurately locate the damage.(5)Finally,the complex finite element model of the Huangpu Bridge of Zhujiang River is used to obtain the acceleration response under different damage conditions.The damages of the pylon,hanger and girder are identified by the proposed method.The damage identification results show that the proposed method is applicable and feasible for damage identification of complex models.