To examine stress redistribution phenomena in bridges subjected to varying operational conditions,this study conducts a comprehensive analysis of three years of monitoring data from a 153-m double-deck road–rail stee...To examine stress redistribution phenomena in bridges subjected to varying operational conditions,this study conducts a comprehensive analysis of three years of monitoring data from a 153-m double-deck road–rail steel arch bridge.An initial statistical comparison of sensor data distributions reveals clear temporal variations in stress redistribution patterns.XGBoost(eXtreme Gradient Boosting),a gradient-boosting machine learning(ML)algorithm,was employed not only for predictive modeling but also to uncover the underlying mechanisms of stress evolution.Unlike traditional numerical models that rely on extensive assumptions and idealizations,XGBoost effectively captures nonlinear and time-varying relationships between stress states and operational/environmental factors,such as temperature,traffic load,and structural geometry.This approach allows for the identification of critical periods and conditions under which stress redistribution becomes significant.Results indicate a clear shift of stress concentrations frombeamends toward mid-span regions following the commencement of metro operations,reflecting both structural adaptation and localized overstress near arch ribs.Furthermore,the model generates robust predictions of stress evolution,demonstrating potential applications in early warning systems and fatigue risk assessment.This work represents the first application of interpretable gradient-boosting techniques to stress redistribution modeling in double-deck bridges.In addition,a Stress Redistribution Index(SRI)is proposed,derived from this monitoring study and finite-element-based transverse load distributions,to quantify temporal stress shifts between midspan and edge beams.The results provide both theoretical contributions and practical guidance for the design,inspection,and maintenance of complex bridge structures.展开更多
基金supported by the Key Technologies Research and Development Program under Grant 2021YFB1600300.
文摘To examine stress redistribution phenomena in bridges subjected to varying operational conditions,this study conducts a comprehensive analysis of three years of monitoring data from a 153-m double-deck road–rail steel arch bridge.An initial statistical comparison of sensor data distributions reveals clear temporal variations in stress redistribution patterns.XGBoost(eXtreme Gradient Boosting),a gradient-boosting machine learning(ML)algorithm,was employed not only for predictive modeling but also to uncover the underlying mechanisms of stress evolution.Unlike traditional numerical models that rely on extensive assumptions and idealizations,XGBoost effectively captures nonlinear and time-varying relationships between stress states and operational/environmental factors,such as temperature,traffic load,and structural geometry.This approach allows for the identification of critical periods and conditions under which stress redistribution becomes significant.Results indicate a clear shift of stress concentrations frombeamends toward mid-span regions following the commencement of metro operations,reflecting both structural adaptation and localized overstress near arch ribs.Furthermore,the model generates robust predictions of stress evolution,demonstrating potential applications in early warning systems and fatigue risk assessment.This work represents the first application of interpretable gradient-boosting techniques to stress redistribution modeling in double-deck bridges.In addition,a Stress Redistribution Index(SRI)is proposed,derived from this monitoring study and finite-element-based transverse load distributions,to quantify temporal stress shifts between midspan and edge beams.The results provide both theoretical contributions and practical guidance for the design,inspection,and maintenance of complex bridge structures.
