Random Load Identification Method Based On Deep Neural Network

The external dynamic load information of the structure plays an important role in engineering fields of structural design,safety evaluation,structural dynamic optimization and health monitoring.But in practical engineering applications,it is difficult to obtain external load information by direct measurement in most cases,so load identification becomes an effective way to indirectly obtain load information.For linear structures,the research on the identification of deterministic loads such as periodic and impact loads has been relatively mature.However,there are still some problems in the identification of random loads,for instance ill-posedness and ill-conditioned matrix,and there is still a lack of research on the identification of random loads for nonlinear structures.Meanwhile,the development of deep neural network provides a new and effective method for complex load identification.Based on the linear structure and nonlinear structure,this paper applies the deep neural network to effectively identify various forms of random loads.The main research work is as follows:(1)In terms of numerical calculations,this paper establishes a deep neural network model to identify random loads of linear structure,and realizes the identification of random loads for example white noise,trapezoidal spectrum and seismic spectrum.Taking linear cantilever beam structures as examples,the random excitations at structural supports and nodes are identified by deep neural network.The calculation results show that the overall errors of different forms of random load identification are less than 5%,and the correlation coefficient between the predicted spectrum and the real spectrum is greater than 0.99.There is no ill-conditioned problem in the identification process.(2)In terms of experimental verification,this paper takes FRP nonlinear cantilever beam as the experimental model,generates white noise and trapezoidal spectrum random excitation at the supports through the exciter,and get the structural response through NDI acquisition equipment.Considering the coupling effect between excitation and structure,the deep neural network model is constructed to realize the effective identification of random excitation.The experimental results show that the identification of the target spectrum with ideal input is less than 3%,and the identification error of the actual input spectrum considering noise is less than 10%.Therefore,the deep neural network method established in this paper can effectively identify the random load acting on linear and nonlinear systems,without complex analytical process and convenient calculation,as well as the identification accuracy can meet the needs of practical engineering.