Structural Nonlinear Model Updating Based On The Structural Response Instantaneous Characteristics And Improved Generative Adversarial Network

Engineering structures usually exhibit strong nonlinear characteristics and complex dynamical behaviors when subjected to extreme external loads,and structural model based on the traditional linearization assumptions cannot accurately reflect the complex dynamical behaviors.This thesis proposes a structural nonlinear model updating method based on the instantaneous characteristic parameters of the principal component of the structural dynamic response and improved Generative Adversarial Networks(GAN).The proposed method aims to solve the problems of the existing methods,such as the difficulty in selecting nonlinear characteristic indexes and the complexity of optimization process.Firstly,the structural dynamic response signal is decomposed into principal component with different frequency component by a time-varying filter based on Discrete analytical mode decomposition(DAMD),and the instantaneous characteristic parameters,such as instantaneous frequency and instantaneous amplitude,of the principal component of the structural dynamic response are obtained by combining the Hilbert transform.Then,the instantaneous parameters of the main component of the structural dynamic response are used as the input data of the improved GAN,and the parameters to be updated of the structural nonlinear model are used as the output data.The complex nonlinear relationship between the instantaneous characteristic parameters of the nonlinear structural vibration response and the set of parameters to be updated is then fitted by the improved GAN.Finally,the measured principal component instantaneous characteristic parameters of the structural dynamic response of the actual structure are input into the trained improved GAN,and the modified structural nonlinear model parameters to be identified can be output.The main research contents of the thesis are as follows:(1)A time-varying filter based on DAMD theory is designed.The identification theory of instantaneous characteristic parameters of single-degree-of-freedom time-varying nonlinear structure and multi-degree-of-freedom time-varying nonlinear structure is further derived.The instantaneous characteristic parameters of the vibration response of the local nonlinear joint model are then identified by using the DAMD theory.(2)The traditional GAN and an improved GAN deep learning model are proposed for structural nonlinear model updating.In the improved GAN architecture,the discriminator in the network is enhanced to learn the response relationship features of each measurement point by introducing the surrogate model,and the surrogate model also makes it feasible to use GAN to solve the structural nonlinear model updating problem.To avoid the gradient disappearance problem of the traditional GAN,the information exchange between network layers is enhanced by using jump connections and dense connections.and the mapping relationship between model input response and output parameters is constructed by introducing the combined objective function to achieve network training.The improved GAN is then successfully applied to structural nonlinear model updating,and the principle and correction process of structural nonlinear model updating by using the improved GAN are elaborated.(3)The effectiveness of the proposed structural nonlinear model updating method is verified by two numerical examples of a three-story four-span steel frame structure and a twelve-story nonlinear frame structure under earthquake excitation load.The effects of different noise conditions,different number of instantaneous characteristics and different number of measurement points on the correction effect are also considered,and compared with the structural nonlinear model updating method based on Convolutional Neural Networks(CNN).The results show that the proposed method can effectively construct the mapping relationship between model input response and output parameters,and reduce the inverse problem of structural nonlinear model updating to the positive problem of solving the parameter set,which improves the computational efficiency and accuracy of structural nonlinear model updating.(4)The shaking table test data of a nonlinear cantilevered aluminum beam under earthquake excitation is used to update the structural nonlinear model based on the proposed method,and the applicability and accuracy of the proposed method in the actual engineering structural nonlinear model updating are verified.The experimental results show that the proposed method can effectively modify the structural nonlinear model,and the structural dynamic response calculated by using the modified nonlinear model is in good agreement with the actual measured data of the tested structure.