Deep Learning-Based Early Warning Method For Temperature-Induced Deflection Anomalies In Cable-Stayed Bridges

Cable-stayed bridges are an important infrastructure in China and occupy an important position in the development of bridges in China.As the proportion of cable-stayed bridges increases,the focus of engineering research has been transformed from bridge construction to health monitoring and post maintenance of bridges.Deflection,as an important index to measure the deformation performance of cable-stayed bridges,has been the focus of health monitoring of cable-stayed bridges.Among the many reasons affecting the deflection variation,temperature variation is one of the most important reasons and has a long-term effect on the deflection of cable-stayed bridges.Therefore,the temperature-induced deflection is closely related to the long-term service performance of cable-stayed bridges.The temperature deflection is caused by the complex temperature field composed of the main beam,towers and cables.When the cables are damaged,the temperature deflection will change abnormally,which will affect the deformation capacity and service performance of the cable-stayed bridge.At present,it is difficult to identify the abnormal changes of temperature-induced deflection caused by cable damage by conventional monitoring methods.Based on the above background,the main objective of this paper is to design a deep learning model-based method for early warning of cable-stayed bridge temperature-induced deflection,taking the Qingzhou channel bridge of the Hong Kong-Zhuhai-Macao Bridge as an engineering example and making full use of the superiority of deep learning in solving nonlinear modeling problems.The main work and contributions of this paper are as follows.(1)Based on the daily time course data of temperature and deflection of the HZMB,the long-term variation of temperature and temperature-induced deflection of the HZMB is explored,and then the correlation between temperature and deflection at various locations on the main girder is analyzed.It is concluded that the correlation between temperature and main span deflection is better than that with side span deflection;within the same span,the closer the location to the middle of the span,the more the deflection is affected by temperature changes and the stronger the correlation with temperature;for main girder deflection,the correlation with main girder temperature is better than that with cable tower temperature.(2)The multiple linear regression model of temperature-deflection and the principal component regression model of temperature-deflection were established at the mid-span position of the main span as an example,and the prediction results of the models were compared.The results prove that the principal component analysis can improve the training effect of the models.After that,using principal component analysis,two methods,support vector machine and deep learning,are used to build temperature-deflection regression models,and compare the modeling results.The results prove that: the model based on deep learning has better results.(3)The effect of data quality on the effect of deep learning is analyzed,and the deep learning model is built with distorted data and compared with the modeling results in the normal state of data.The results demonstrate that there is a big difference between the model effect in the distorted data state and the model effect in the healthy data state,which shows that the data quality has a big influence on deep learning.Finally,the regression models of temperature and full bridge deflection were established using the data in the normal state.(4)The finite element model of the Qingzhou channel bridge was established,and the change of the main girder temperature-deflection at different locations when the tension cables break was studied,and the relevant change law of the main girder temperature-deflection when the tension cables break was summarized.Finally,combining with the knowledge of hypothesis testing,a deep learning model-based method is proposed to identify the abnormal temperature-deflection of cable-stayed bridges under abnormal conditions,and reliability analysis is performed.The results show that the method can identify the abnormal deflection changes at all positions of the main span,which indicates that the method can better identify the abnormal temperature-induced deflection of cable-stayed bridges.