Nonlinear Structural Model Updating Based On The Deep Belief Network

With the development of finite element calculation technologies and theories,many significant advances have been made in structural model updating.However,most model updating techniques ignore the extensive nonlinearity of existing structures.Therefore,it is necessary to study the nonlinear structural model updating method.In this thesis,a nonlinear structural model updating method based on the Deep Belief Network(DBN)is proposed.The proposed method aims to solve the complex optimization process in the existing methods.Firstly,the mono-component of the nonlinear structural vibration responses is extracted by the Variational Mode Decomposition(VMD)method or the Discrete Analytic Mode Decomposition(DAMD)method.Then,the instantaneous parameters of the mono-component are identified by the Hilbert transform(HT).The instantaneous parameters after principal component analysis are then taken as the input,and the nonlinear model parameters are considered as the output.The DBN is applied to fit the complex mapping relationship between the input and the output.Finally,the updated nonlinear model parameters can be identified directly by feeding the instantaneous parameters of the measured vibration responses after principal component analysis into the well-trained DBN.Three nonlinear structural models,which are a local nonlinear joint model,a BoucWen frame model,and a complex Giuffré-Menegotto-Pinto frame model,are numerically simulated to verify the effectiveness of the proposed method.In the numerical examples,the effects of different signal decomposition algorithms,load types,noise levels,and nonlinear types on the modified results are studied.The numerical results show that different signal decomposition algorithms have little effect on the updated accuracy of the proposed method.The three different nonlinear structural models under seismic excitation and harmonic excitation can be updated effectively by the proposed method.Moreover,the proposed methodology has a relatively good noise resistance.Finally,the shaking table test data of a 1/4-scale three-story steel frame under seismic load and a transmission structure under the harmonic load are applied to verify the theory and method of this paper.The steel frame experimental structure is equipped with a piezoelectric friction damper(PFD)at the bottom,so that the experimental structure model can be updated by modifying the relevant parameters of the PFD model.Moreover,the nonlinear model of the transmission structure is assumed to be a complex nonlinear model with three joints.The nonlinear model parameters are successfully identified through the measured vibration responses and the proposed method,and the nonlinear model of the experimental structure can be modified.The experimental results show that the proposed method can update the nonlinear structural model relatively effectively and accurately.