Study On Nonlinear Aerodynamic Force And Flutter Of Long-span Suspension Bridge Based On Deep Learning

Under the guidance of new era strategies such as "The Belt and Road Initiative","Transportation Power Strategy" and "Maritime Power Strategy",the long-span suspension bridges in China have taken a new step forward.With the development of long-span and light flexibility of suspension bridges,the structures become more and more sensitive to wind load,and the nonlinear influences become more and more prominent.This makes nonlinear flutter analysis a difficult problem that must be overcome in wind resistant design of long-span bridges,and the nonlinear aerodynamic force is its foundation and premise.Combined with the deep learning technology and relying on the Ma’anshan Yangtze River Highway Bridge project(a three-tower suspension bridge,main span: 2 × 1080m),a series of research work,from twodimensional section numerical simulation to three-dimensional fine analysis of the whole bridge,from nonlinear aerodynamic force to nonlinear flutter response,has been carried out step by step:(1)The nonlinear aerodynamic numerical simulation of bridge section is carried out.Taking the two-dimensional scale deck section as the object,the CFD numerical model is established.Based on this model,the numerical simulations of static state,forced vibration and free vibration are realized successively.In the process,the aerodynamic parameters such as aerostatic three-component coefficients and flutter derivatives are identified systematically,the influence of railings and other auxiliary facilities on aerodynamic parameters is explored,the aerodynamic nonlinear characteristics are displayed,and the flutter critical wind speed is calculated.The accuracy of aerodynamic CFD numerical calculation method is verified by comparing wind tunnel test results of the segmental model.(2)The nonlinear aerodynamic reduced order models(ROMs)of bridge deck based on deep learning are constructed.A large amount of nonlinear aerodynamic force data is obtained by CFD forced vibration numerical simulation,and the displacement signals are synthesized by harmonic superposition.Two types of deep learning frameworks such as feedforward neural network(FNN)and long and short-term memory(LSTM)network are introduced.According to the characteristics of the two frameworks,the training,verification and test sets are constructed,and then two kinds of nonlinear aerodynamic ROMs are established.The performances of the two models are compared under various working conditions such as forced vibration and free vibration,and the importance of fluid memory effect on aerodynamic timedomain modeling is evaluated.(3)A nonlinear flutter analysis method of the three-dimensional full-bridge based on deep learning is proposed.Based on LSTM network,the aerodynamic ROM of full-scale deck section is constructed.Combined with the full-bridge finite element model,the restart technique was implemented to achieve the cross platform dynamic iterative calculation of the structural response and aerodynamic force in time-domain.The influence of additional excitation is eliminated by adjusting structural damping.This method comprehensively considers multiple effects on the full-bridge scale,including the three-dimensional effect,static wind effect,as well as aerodynamic and geometric nonlinearities,thus the nonlinear flutter phenomenons(i.e.soft flutter and flutter mode transition)observed in the wind tunnel test of the full-bridge aeroelastic model are reproduced.(4)The nonlinear characteristics and influencing factors of the full-bridge flutter response are explored.Based on the full-bridge scale,the nonlinear flutter responses of the bridge deck,the middle tower,the main cable and suspenders are analyzed successively.On these basis,the important soft flutter characteristics are discussed including the amplitude dependence of aerodynamic damping,the morphological characteristics of the middle tower participating in flutter and the instantaneous torsional center of the bridge deck,and the stress of suspenders in soft flutter is also evaluated.Besides,the flutter time-domain calculation based on linear selfexcited force is carried out.Finally,the influence of nonlinear factors on flutter calculation is summarized by integrating three kinds of flutter calculation methods involved in the full paper and two kinds of wind tunnel tests.