Research On Temperature Effect Analysis And Early Warning Method Of Suspension Bridge Based On Monitoring Data

In order to ensure the normal operation and structural safety of long-span suspension bridges,it is very important to evaluate and early-warning the safety status of the bridges in real-time operation.The typical characteristics of bridge measurement information are multi-source,volume and variety.It is urgent to find suitable data analysis methods to deal with it,so as to carry out multi-source data fusion,feature engineering,pattern recognition and other work.For long-span steel suspension bridge,because of its structural characteristics as flexible system and good thermal sensitivity of steel,the steelbox suspension bridge will have a large structural response change under the action of temperature.The monitoring data of suspension bridge are separated,and the influence of temperature on the structural deformation and internal force of suspension bridge is analyzed.Considering the diversity of the environment and the multi-scale temperature effect of the background bridge,combined with a large number of monitoring data and calculation results of finite element model of suspension bridge,the multi-level threshold line is set for the suspension bridge,and appropriate machine learning algorithm is selected to predict the measured data,so as to achieve the goal of accurate and effective early warning of bridge health monitoring system.The main contents and research results of this paper are as follows:Firstly,in order to eliminate the end-point effect of Variational Mode Decomposition(VMD)in separation study of temperature effect of suspension bridge measured data,a three-section cross-signal decomposition method is proposed.Considering the influence of the number of modes(k)and penalty factor(α)on the decomposition effect of VMD,a temperature effect separation method based on optimized VMD is formed and applied to the real-time data processing and analysis of suspension bridge.The research shows that the optimized VMD algorithm can successfully separate the temperature effect in the signal.By analyzing the correlation between each mode and the measured temperature,it is proved that the temperature effect is mainly concentrated in the low frequency band,and the structural response of the suspension bridge is mainly affected by the temperature.The results show that the method is effective and can be used for the data analysis of longterm health monitoring of bridges.Secondly,in view of the high sensitivity of suspension bridge to temperature,the local steel box girder model is established by using ANSYS to analyze the temperature distribution and temperature stress changes of steel box girder under different sunshine intensity,It is concluded that under the action of sunlight,the vertical and horizontal temperature distribution of steel box girder is uneven,and the uneven temperature distribution leads to the uneven stress distribution.Then,combined with the measured temperature and deflection data,the influence of multi-scale temperature on deflection is analyzed,and the linear relationship between temperature and deflection in different seasons is obtained;Based on MIDAS / civil,the whole bridge model of suspension bridge is established,and the deflection changes under different tower beam temperature difference,cable temperature difference and steel box girder temperature difference are calculated.According to the finite element calculation results and the deflection / strain limit values of steel box girder suspension bridge specified in the code,multi-level threshold lines are set for the deflection / strain of different sections of the background bridge.Finally,to solve the problem that long short term memory network(LSTM)can not effectively predict long-term memory,an improved long-term memory accumulation model(N+LSTM)is proposed.The deflection/strain trend term extracted by the optimized VMD method is used as the input of the N+LSTM prediction model,and the trend prediction result of the specified length is obtained.According to the close degree between the prediction results and the threshold line,the alarm level is determined to achieve the goal of pre alarm before the occurrence of bridge structure damage.