Automated Identification Of Momdal Parameters Of The Long-Span Bridge Considering Uncertainties Quantification

Long-span bridges continue to show the characteristics of super-long span and flexibility,and their sensitivity to vibration caused by ambient excitation is becoming increasingly significant.After a long-term operation,the in-service performance of the components such as the main girder and cables of the bridge will deteriorate to varying degrees,resulting in the frequent occurrence of various abnormal vibrations and structural damage.Operational modal analysis is vital for structural health monitoring of long-span bridges,which provides crucial parameters for structural vibration mitigation and in-service condition evaluation.It is of great significance to promote the automation of operational modal analysis and realize the online monitoring of structural dynamic property for ensuring the healthy operation of bridges.However,influenced by test noise and model error,the operational modal analysis inevitably contains uncertainties,which weakens the reliability of structural dynamic performance analysis.In order to improve the refinement level of online monitoring of operational modal parameters of long-span bridges,this study proposes an automated modal identification method considering uncertainties quantification utilizing Bayesian inference and machine learning algorithms.Based on this,the Bayesian operational modal on-line analysis software is developed.A series of research and validation for the proposed method are carried out relying on the monitored data from the structural health monitoring system(SHMS)of the Sutong Yangtze River Bridge and the Nanjing Qixiashan Yangtze River Bridge.The main research contents are as follows:(1)Operational modal analysis for separated modes of the long-span bridge based on Bayesian inference.The genetic algorithm is combined with the fast Bayesian fast Fourier transform(FBFFT)approach to search for the most probable values of modal parameters.Moreover,the asymptotic estimation interval is established to constrain the search space of parameters.Thus,a modal identification method for long-span bridges with separated modes is developed.The performance of the proposed method is verified by analyzing the numerical simulation data.Furthermore,the range of asymptotic estimation interval and the data duration in modal identification is discussed to determine their proper values.Based on the monitored acceleration data of the main girder of the Sutong Yangtze River Bridge,the operational modal analysis of the bridge is carried out to verify the effectiveness of the proposed method in the modal identification of the long-span bridge with separated modes.(2)Operational modal analysis for close modes of the long-span bridge based on Bayesian inference.The generalized FBFFT approach is adopted to estimate the close modal parameters of long-span bridges.The effects of data duration and frequency bandwidth on the identification uncertainty of close modes are separately discussed.Furthermore,the uncertainty properties of separated and close modes are compared.Based on the monitored acceleration data of the stiffening girder of the Nanjing Qixiashan Yangtze River Bridge,the modal parameters of the bridge are identified,which verifies the effectiveness of the generalized FBFFT approach in the operational modal analysis of the long-span bridge with close modes.(3)Automated Bayesian modal parameters identification approach based on the cross-modal assurance criterion(CMAC)matrix.In order to extract the frequency response interval of target modes,the singular value(SV)decomposition of structural vibration data is carried out to establish the CMAC matrix,which is then reconstructed by utilizing the convolution autoencoder.The Kohonen network is used to cluster the normalized SV sequence established by utilizing the SV data in the extracted modal response interval to recognize the type of target modes.Based on the above method,the automation of Bayesian modal identification is realized.The performance of the machine learning models in the proposed method is analyzed by numerical simulation.Furthermore,the identification accuracy and computational efficiency of the proposed method are verified.(4)Development and application of the Bayesian operational modal online analysis software for long-span bridges.The approaches such as the triple standard deviation method and the generalized regression neural network are applied to detect and recover the abnormal monitoring data.The reliability of the estimations is evaluated according to their variation coefficient.The Bayesian operational modal online analysis framework for the structural health monitoring of long-span bridges is established based on the above method.Then the Bayesian operational modal online analysis software is developed accordingly.The application of the software is implemented by utilizing the long-term acceleration data recorded by the SHMS of the Sutong Yangtze River Bridge and the Qixiashan Yangtze Yangtze River Bridge.Moreover,the evolution characteristics of modal parameters of relative bridges are analyzed.