Deep Learning Based Structural Health Monitoring And Early Warning For Long-Span High-Speed Railway Bridges

With the rapid development of China’s national economy and the improvement of comprehensive national strength,China’s high-speed railway construction has developed rapidly,and has become an important support for the national transportation artery and the sustainable development of the national economy.Bridges account for a large proportion in China’s high-speed railway,therefore,it is necessary to ensure the good service state of high-speed railway bridges to meet the traffic requirements of high-speed trains.In recent years,structural health monitoring technology has developed by leaps and bounds,especially with the rise of big data,deep learning and artificial intelligence technology,the data-driven structural health assessment method provides a technical means for damage identification and safety early warning of high-speed railway bridges.Taking a highway railway Yangtze River Bridge as the engineering background,this dissertation studies and puts forward the technical framework of long-span high-speed railway bridge monitoring and early warning system,and builds the corresponding bridge structure health monitoring platform.Combined with the measured data and based on the deep learning method,this dissertation systematically studies the health monitoring and early warning method and technical system of long-span high-speed railway bridges,and its effectiveness and reliability have been verified by practical engineering application.The relevant research results are of great significance for the long-life and safe service of long-span high-speed railway bridges and promoting the intelligent construction and advanced maintenance of high-speed railway infrastructure.The main contents of this dissertation are as follows:1.Health monitoring data cleaning method based on Bi LSTM modelThis dissertation systematically studies the typical signal anomaly types of bridge monitoring data,formulates the anomaly classification and processing standards,creatively combines the sensitive indicators of various anomalies,completes the data automatic repair by using the Bi LSTM model,and constructs a set of standardized,automatic and accurate bridge monitoring data cleaning process,Prepare data for real-time evaluation and early warning of bridges.2.Vibration performance monitoring and early warning of bridge structure based on LSTM classification networkThrough the decomposition and reconstruction of multi-layer wavelet packets,the dynamic strain response of the bridge under the action of high-speed train is extracted;By constructing LSTM classification network,the intelligent recognition and extraction of non-stationary signal of vehicle actuation strain are realized;The vehicle induced strain characteristic parameters of the key components of the bridge are extracted by adaptive filtering method;Based on probability statistical analysis,the intelligent monitoring and early warning of bridge vehicle induced vibration performance is realized.3.Monitoring and early warning of bridge thermal deformation based on LSTM modelBy extracting the temperature induced deflection of the bridge and introducing the complexity of the uneven temperature field of the main beam,the correlation model between the temperature field and the temperature induced deflection of the main beam is established by using the methods of multiple regression and machine learning,and the correlation characteristics between the temperature field and the temperature induced deflection of the main beam are explored;Furthermore,the nonlinear fuzzy regression model between temperature field and temperature induced deflection is established by using LSTM model.On this basis,based on the time-varying statistical characteristics of the regression results,a dynamic early warning method for the abnormal change of the main beam alignment of high-speed railway bridges is proposed4.Monitoring and early warning of bearing capacity of bridge structure based on LSTM modelBased on the LSTM(Long Short-Term Memory)neural network mining,the nonlinear mapping relationship model between the monitoring data of different measuring points is established.Combined with PCA method,a comprehensive index that can quantitatively analyze the degradation degree of bridge bearing capacity is proposed,an early warning evaluation method of bridge bearing performance based on structural distributed strain time-varying statistical model is further established.In conclusion,the technical framework of long-span high-speed railway bridge monitoring and early warning system,health monitoring data cleaning method based on in-depth learning,high-speed railway bridge structural vibration performance monitoring and early warning method,temperature induced deformation monitoring and early warning method and bearing performance monitoring and early warning method proposed in this dissertation are conducive to timely grasp the performance status of high-speed railway bridges and ensure the safe operation of high-speed railway bridges,which has a good application prospect.