Prediction Of Flutter Derivatives For ? Type Sections By Neural Networks Based On CFD

Based on the review of the basic theory of bridge flutter derivatives and the development of flutter derivative identification methods,this paper summarized the research status of flutter derivative identification.According to the insufficiency of current work,based on the two-dimensional rigid segment model wind tunnel test,the effect of three kinds of widely used turbulence models on the accuracy of identification of flutter derivatives were compared,and the turbulence model with highest accuracy for ? type sections was recommended.Based on the turbulence model,the database of flutter derivatives for ? type sections was established,and the influence of geometric parameters of ? type sections on the flutter derivatives was studied meanwhile.Furthermore,based on this database,the method of predicting flutter derivatives of ? type sections using neural networks was illustrated for the first time,which provides a new approach for the fast acquisition of flutter derivatives in the preliminary design of bridges.The main contents of this research are as follows:(1)Based on the flutter derivatives acquired from segment model wind tunnel test,accuracy of flutter derivatives acquired from numerical calculation exploiting RNG 6)- model?SST6)- model and RSM were compared,and with the comparison of streamlines calculated by the three models,RSM was proved to be more appropriate for numerical calculation of the flutter derivatives for ? type sections.(2)RSM was used in numerical calculations to obtain the flutter derivatives of ? type sections with side ratio =6~13 and size coefficient =1.1~1.3,and then a database of flutter derivatives was illustrated.Meanwhile,the effect of this two geometric parameters on the flutter derivatives were studied,with which the variation of dominate vortices in the streamline patterns was studied.(3)The neural network based on this database successfully predicted the flutter derivatives of a typical ? type section.Based on several error evaluation methods,the prediction accuracy was proved to be satisfying.This method is beneficial to improve the efficiency of acquiring flutter derivatives and calculating the critical wind speed.Finally,the BP neural network was proved to be more applicable for this nonlinear problem than RBF neural network.