Research On Vortex-Induced Vibrations Of Long-Span Bridges Based On Prototype Monitoring And Machine Learning

In order to meet the needs of transportation development,China has built many longspan bridges.The increase in span makes the bridge more flexible and more sensitive to wind.Therefore,wind resistance design becomes one of the control factors of bridge design.Although divergence flutter has been avoided by increasing the critical wind speed of the flutter in the bridge design,it is difficult to avoid the occurrence of selfexcited vortex-induced vibration.In addition,the increase of span reduces the natural frequency of the bridge structure,thus reducing the critical wind speed of vortex-induced vibration and increasing the occurrence frequency of vortex-induced vibration at low wind speed.In recent years,vortex-induced vibrations have been observed many times on many long-span bridges.Large-amplitude vortex-induced vibrations cause fatigue damage to bridge structures and threaten driving safety.Therefore,it is of great significance to study the vortex-induced vibration of long-span bridges.Although a lot of research on the mechanism and modeling of vortex-induced vibration have been done based on wind tunnel test,it is difficult to simulate the spatio-temporal characteristics of real wind environment and the high Reynolds number effect of full-scale structure,leading to some discrepancy between wind tunnel test results and real wind effects on prototype bridges.Therefore,this paper studies the vortex-induced vibration of a prototype bridge in a real complex wind environment based on prototype monitoring big data and machine learning algorithms.Firstly,an automatic identification method for bridge vortex-induced vibration based on clustering algorithm is proposed.A vortex-induced vibration identification feature space characterized by root mean square of acceleration and vibration single-frequency characteristics is proposed.An automatic identification method for bridge vortex-induced vibration based on clustering algorithm is established and applied in the long-term vibration monitoring data of a long-span bridge.166 vortex-induced vibration events are identified,including six modes of vortex-induced vibrations.Further,the cluster analysis of wind speed fields of vortex-induced vibrations has discovered 6 clusters of wind speed fields and their correspondence with vortex-induced vibration modes,revealing the influence mechanism of the wind speed field on the vortex-induced vibration mode of the bridge and the influence of the non-uniformity of the wind field on the vortex-induced vibration mode of the bridge.Secondly,a decision tree algorithm-based bridge vortex-induced vibration mode identification method and a support vector regression-based bridge vortex-induced vibration statistical response time series prediction method are proposed.A bridge vortexinduced vibration mode identification decision tree model with 1-minute mean wind speed and mean wind direction as inputs and bridge vortex-induced vibration mode label as output is proposed.The vortex-induced vibration mode of a long-span bridge is accurately predicted by the model using the wind field monitoring data.A support vector regression model for prediction of bridge vortex-induced vibration statistical response time series with 1-minute mean wind speed and mean wind direction as external inputs and 1-minute bridge deck displacement rms value as output is proposed.The 1-minute displacement rms amplitude time series of the vortex-induced vibration under action of space-time-varying wind is accurately predicted for a long-span bridge by the model using the wind field monitoring data.The model identifies the wind speed range and the wind direction range of the bridge vortex-induced vibration.It can also reveal the influence of the span-wise non-uniformity of wind field on the response of vortex-induced vibration.Thirdly,a neural network-based modeling method for the differential equation of bridge vortex-induced vibration displacement amplitude is proposed.The feedforward neural network model and recurrent neural network model of instantaneous displacement amplitude of bridge vortex-induced vibration with instantaneous mean cross wind speed as external input and time derivative of instantaneous displacement amplitude as output are proposed.The vortex-induced vibration displacement amplitude time series of a longspan bridge under space-time-varying wind is accurately predicted by the recurrent neural network model using the wind field monitoring data.The advantage of recurrent neural network on this dynamical system modeling is revealed by the comparison of structures of feedforward neural network and recurrent neural network.Fourthly,a sparse identification method for time-varying dynamics of bridge vortexinduced vibration is proposed.A time-varying dynamics identification method for bridge vortex-induced vibration based on sparse identification of nonlinear dynamical system is proposed.The time-dependent ordinary differential equation of displacement amplitude is identified for each vortex-induced vibration event of a long-span bridge using the vibration and wind monitoring data.The time-varying dynamics of the bridge vortexinduced vibration its mechanism are revealed by the identified time-dependent ordinary differential equations.Further,the cluster analysis of the function term coefficients of differential equations has revealed 7 clusters of function term coefficients,revealing the different dynamics of bridge vortex-induced vibration and the relationship between the dynamics and the self-excited effect in vortex-induced vibration.