Prediction Of Buffeting Responses Of Large-span Cable-stayed Bridges Using Gaussian Process Regression Approach

In recent years,with the development of structural materials and engineering technology,bridge engineering expanded rapidly,large numbers of long-span bridges have sprung up.To date.However,the bridge safety accidents also occurred frequently in this period.Therefore,it is necessary to apply health monitoring systems(HMS)to improve real-time damage identification and status assessment during the construction and operation of bridges,so as to reduce the probability of accidents.For now,many larges bridges have installed HMS to collect the concerned parameters such as the vibrations,forces,stresses,temperature fields and etc.However,the amount of sensors for most HMS is less than 1500.Sensors are arranged at the most interested positions for the sake of cost.Thus,the responses at overwhelming majority locations cannot be directly monitored.Even worse,some data may be deficient due to the failure of data collection,transfer,and/or storage system.Therefore,it is significant to develop efficient prediction approaches for spatial extension and data recovery.This paper uses the machine learning method: Gaussian process regression(GPR)as basis,to propose an approach for predicting the bridge vibration responses at the positions where sensors are not arranged.The main work and conclusions are as follows:(1)Based on the harmonic superposition method,the three-dimensional pulsating wind field of long-span cable-stayed bridge is simulated.Combined with the characteristics of the bridge and the correlation of natural wind,the fluctuating wind field is simplified.The fast Fourier transform technique is used to simulate the time history of the pulsating wind speed of the main beam and the main tower.The power spectrum and correlation function are used to verify the simulation results.(2)Establish the finite element model of long-span cable-stayed bridge and analyze its dynamic characteristics based on ANSYS.Establish the finite element model from four part:The main girder system,cable-stayed system,pylon system,auxiliary pier and transition pier system.Analyze the first 16 modes.Buffeting simulation of long-span cable-stayed Bridges.Simulated wind load from three parts: static wind load,buffeting load and aerodynamic self-excitation load.Analyze the characteristic of buffeting response.These buffeting response data will be use to test the method raised after.(3)Prediction of vibration response of long-span cable-stayed Bridges.A novel approach using the machine learning strategy of Gaussian process regression(GPR)is proposed to predict the bridge vibration responses at the positions where sensors are not arranged.the dynamic responses at all nodes are obtained and used as data source to demonstrate theapplicability and accuracy of the GPR approach.