On Bridge Structure Health Management Based On Big Data Analysis

Bridge is an important part of road infrastructure,and the related research on characteristic of bridge structure has been widely concerned.Nanjing Dashengguan Bridge is an important railway channel across the Yangtze River.On regard of its size and security maintenance needs,the bridge is equipped with a well-defined sensor monitoring system.Every day these sensors produce a lot of data.Research on monitoring data can further understand the bridge structure characteristic and help to design bridge warning systems for real-time state monitoring.Traditional data processing methods are difficult to analyze huge amount of monitoring data and processing time is unacceptable.This thesis applies big data processing technology to improve performance of data analysis and analyse the bridge monitoring data comprehensively.Furthermore,with the combination of big data realtime processing technology and anomaly detection algorithm,a bridge real-time prewarning system is designed and implemented,which is able to generate on-line warns in small batch interval based on sensor monitoring data.The main contributions of this thesis are summarized as following:(1)Based on the monitoring data analysis requirements,a bridge big data platform is proposed.It covers two main modules,namely off-line data analysis module and online data analysis module.Off-line data analysis module is used to store data.What's more,correlation analysis is applied to the data and a warning model is designed.Online data analysis module is put forward for real-time analysis of sensor monitoring data using the obtained warning model.It monitors the bridge real-time running state.(2)Due to the strong correlation between temperature and displacement data,regression algorithm is used to construct the displacement prediction model.It combines 3-Sigma interval estimation theory with the obtained prediction model to construct point estimation warning model.The experimental results show that both anomaly precision ratio and recall ratio are about 90%.(3)In terms of the stationarity of vibration acceleration time series data,vibration acceleration time series warning mode is designed using time series cutting method and isolation forest anomaly detection algorithm.The experimental results show that F1 value of iForest+Kmeans algorithm outperforms the original iForest algorithm with about 4%.(4)Both the off-line and on-line analysis performance of the bridge big data platform are evaluated.The experimental results show that off-line analysis performance is from 50% to 90% higher than that of the tradition methods.On-line warning analysis works well with the interval of five seconds.