| A large-scale simulation model is needed for the uncertainty stochastic sensitivity analysis of the mechanical properties of beam-column connection.The finite element method often exceeds the computing power of the computer and cannot meet the analysis requirements and computational efficiency of stochastic sensitivity analysis.With the development of artificial intelligence,the calculation efficiency problem can be solved by a combination of the finite element model and a mathematical model,but the accuracy of the analysis results depends on the accuracy of the mathematical model.However,using artificial neural networks to construct mathematical models also has numerous problems,such as falling into local extrema.The mathematical model was established with the end-plate joint of steel frame as the main research object in this thesis,the PSO-CNN model,which stands for Particle Swarm Optimization Convolutional Neural Network is presented.To address the computational efficiency and calculation accuracy problems of neural network models,which need to be solved urgently to conduct the large number of simulations required for stochastic sensitivity analysis.This model was used to improve the training and prediction accuracy of CNN,where the initial learning rate of the CNN was optimized by PSO.Additionally,the PSO was combined with the Adam optimizer of CNN to automatically screen the optimal network structure.The accuracy of the PSO-CNN model and the CNN model was verified by comparing the high-dimensional function examples and the structural models.The results indicated that the training and prediction accuracy of the PSO-CNN model was greatly improved compared with that of the CNN model.It has good generalization ability.Python and ANSYS are integrated interactively to achieve batch automatic modeling and post-processing of the extended end-plate connection joints and the middle column end-plate connection joints.The reliability of the modeling method used in this thesis is verified by comparing with the test results.Additionally,the influence of a single design variable on the initial rotational stiffness of the end-plate connection,the bending moment bearing capacity of the linear elastic section and the ultimate bending moment bearing capacity is obtained by single parameter change of the joint model.For practical engineering applications,if only considering the change of a single variable and ignoring the coupling between variables,there may be a large deviation in the analysis results.Therefore,this thesis proposes a global stochastic sensitivity analysis method based on Latin hypercube method and Sobol sensitivity,which considers the interaction between multiple variables and quantifies the influence of input parameters on model output.At the same time,Spearman rank correlation coefficient was used to analyze the positive and negative impacts of parameter changes on structural response trends,in order to obtain more accurate structural response predictions.Through the global stochastic sensitivity analysis of the extended end-plate connection joint,the influence degree of each parameter on the initial rotational stiffness of the joint is determined,and the corresponding design suggestions are given.These suggestions are consistent with the design specifications and the research results of other scholars,which verifies the reliability of the method in this thesis.It has a reference effect on the structural design optimization of end-plate connection joints,and can be applied to other types of structural sensitivity analysis. |
