Prediction Of Flutter Critical Wind Speed Of Flat Steel Box Girder Based On Neural Network

It is widely used in long-span suspension bridge for the flat steel box girder which has good mechanical performance and convenient construction.With the soft features,the long-span bridge is easy to produce flutter,which has great damage and serious consequences.Therefore,it is of great significance to study the flutter of flat steel box girder.Traditional wind tunnel flutter test taking a lot more time,aerodynamic measures are blind selecting,numerical simulation take more time and the accuracy is poor.In order to avoid the severe dependence of the pre-selection of aerodynamic measures on the wind tunnel test and reduce the not good enough trial,in view of the accuracy and efficiency of static coefficient calculated by CFD,this paper studies the flutter by using the static coefficient instead of the flutter derivative,an artificial neural network model is established based on the wind tunnel test data to predict the flutter critical wind speed of flat steel box girder.The main contents of this paper are as follows:(1)Three flat steel box girder models with different ratios of width to depth are designed,each model has three groups ratios of torsional to bending frequency,five groups of aerodynamic measures are used to carry out wind tunnel force and vibration tests under the same torsional bending frequency ratio.Take the test results verify that the parameter F,which is composed of lifting moment coefficient(C_M),lifting force coefficient(C_L)and their slope product(C′_MC′_L),represents the aerodynamic damping of flat steel box girder and whether is feasible to judge the relative value of flutter critical wind speed.Compared and analyzed the effects of ratio of torsion to bending frequency and central stabilizer measures on flutter stability.(2)Take 135 flutter critical wind speed data and 12 groups of real bridge data which including the influence factor of flutter are unified.Take section width to height,ratio of torsion to bending frequency,mass per linear meter,mass inertia moment per linear meter,and C_M、C_L、C′_MC′_L as the final factor affecting flutter stability,BP and RBF neural network are established.Compared the prediction accuracy,recognition rate and generalization performance of the two neural networks.The appropriate BP neural network model is obtained,which proves that the selected neural network has certain accuracy and reliability.(3)Taking a real bridge as an example,take the static three component coefficients calculated by CFD and the dynamic characteristics calculated by ANSYS into the neural network model to verify the feasibility of the neural network in practical engineering application.The results show that the BP neural network based on the above data samples can accurately and quickly predict the flutter critical wind speed of flat steel box girder with different aerodynamic measures,and can be applied to the pre-selection of flutter suppression measures.