Reinforcement Of Rigid Frame Arch Bridge Based On Different Neural Networks Parameter Identification And Correction Of Static Finite Element Model

In the current industry,due to various simplification and assumptions,the static and dynamic responses of the initial finite element model of bridge based on the design data have different degrees of difference between the experimental results and those of the reinforced bridges.Based on the previous research work of finite element model modification,genetic algorithm and neural network are very advantageous in dealing with complex nonlinear problems,combining genetic algorithm with BP neural network,a kind of integrated neural network is proposed,and the generation-type antagonism neural network is introduced and reconstructed(Generative Adversarial Networks,gans)is used in the parameter identification and correction of the static finite element model of the bridge structure respectively.1.The application,research significance and present situation of the finite element model correction technique and neural network in this paper are expounded.This paper introduces the theory and method of finite element model modification,introduces the different classification methods of the finite element model modification,the selection of parameters to be modified,the evaluation of model correction effect and the construction of objective function.2.Neural networks are introduced.Firstly,the mechanism of BP neural network based on two kinds of neural networks is introduced,secondly,based on the advantages of genetic algorithm and neural network,an integrated neural network is established,and a generation counter neural network is constructed,which is suitable for this paper.3.Using the Machine learning algorithm library and the depth Learning algorithm library in the Python language,the Integrated neural network and the generation counter neural network are built.Based on the design data,the finite element model is built and the numerical model test is carried out.In this paper,the uniform design method is introduced to generate the training data for the neural network,and the perturbation of the sample data is staged and the neural network is fine-tuned.The rationality and application value of the method are proved by the numerical model correction.4.According to the static and dynamic load test data and strengthening scheme of the strengthened rigid-frame arch bridge,the finite element calculation model is established by using MIDAS/FEA finite element software to extract the model response data under the relevant load.Combined with the data of static load test,build integrated neural network,generate-type countermeasure neural network,construct training data,debug network model,and realize the parameter identification and correction of FEM model according to the measured data and output forecast result.On this basis,the structure is segmented reasonably and the parameter refinement is identified and corrected.The results show that the accuracy of the calculation results of the finite element model parameter segmentation recognition is improved,and the finite element model of the strengthened rigid frame arch bridge,which is identified and modified by different neural networks,can accurately reflect its stress condition under specific load conditions.A precise finite element model is provided for the evaluation of the technical condition of the structure,which provides a reference for the evaluation of the structure reinforcement effect and the damage identification.