Data-driven Stochastic Subspace Identification Of Bridge Health Monitoring Method Based On Time-series Data

Bridge is an integral part of public transport infrastructure.To understand the health status of its service life is the key to ensure the safe passage of pedestrians and vehicles.Because of the great advantages of non-contact measurements,their wide frequency range of response,their sub-millimetric displacement sensitivity,and their quick set-up,ground-based microwave interferometry has been used for non-contact vibration and displacement monitoring of various bridges.In Structural health monitoring(SHM),parameter identification is one of the most crucial parts to detect the damage position and damage degree of the structure,and accurately evaluate the residual life,including the natural frequency,mode shape and damping ratio of the stochastic subspace system identification has many advantages,such as robustness,numerical stability,high identification accuracy and the ability to distinguish close modes.In recent years,it has become a standard tool to modal identification of civil and mechanical engineering structures.However,the accuracy of SSI is limited by the noise content of the monitoring data.Considering that there are many uncontrollable factors in the acquisition data environment,preprocessing the data before the algorithm runs is the premise to ensure the accuracy of the algorithm.At the same time,the main problem of SSI method is modal order estimation.In order to avoid missing modes,the singular value average method will lead to overestimation of the modal order,so as to produce the false mode problem.Modal order estimation using the stabilization diagram has a complex construction process and a large amount of calculation.In order to solve the problems of traditional SSI method in the process of modal identification,this paper develops in four steps: method comparison selection,relevant parameter selection,calculation efficiency and modal order estimation:(1)This paper systematically introduces the basic state space model of random subspace,and introduces the linear algebra tools(matrix projection,least square method,Kalman filter,etc.)in data-driven random subspace recognition.The performance of the three SSI methods(N4SID,COV-SSI,DATA-SSI)is compared through simulation experiments.The characteristics of the three methods are analyzed according to the identified modal parameters(natural frequency,damping ratio,mode shape)and calculation efficiency.The results show that N4 SID has the highest calculation efficiency and the lowest accuracy.Due to over estimation operation,it is easy to have repeated false modes.The calculation efficiency of COVSSI and data SSI method is equivalent,and COV-SSI speed is slightly superior.The recognition frequency of the two methods is highly consistent with the theoretical value,while data SSI performance is better than COV-SSI in the damping identification part.In conclusion,DATASSI with better comprehensive performance is selected as the main identification method.(2)In the process of modal parameter identification,some parameters of DATA-SSI need to be specified manually.They affect the effectiveness of the algorithm.Improper selection will lead to unreliable identification results.Specific parameters: Hankel matrix row number;the number of rows in the past output subpartition;the number of rows in the future output subpartition.This paper mainly discusses the function of these parameters in the process of data SSI identification,and analyzes how to select the most suitable and minimum values of these parameters.The influence of these parameters on the identification results is illustrated by a simulation model.Then select the appropriate values for these parameters according to the given value specification.Finally,the results of identification are compared with those of other modal parameter identification methods.(3)Aiming at the calculation efficiency and modal order estimation of DATA-SSI,an improved method based on the basic principle of DATA-SSI is proposed in this paper.The first is the calculation efficiency of the algorithm.The first step is the detection and removal of abnormal data.The detection data of the boxplot can show the abnormal data and the degree of data dispersion intuitively,which provides a reference for the selection of response signals of system modal identification.The second step "compresses" Hankel matrix,because the small column vector of matrix will lead to the low resolution projection of subspace,the norm of column vector is taken as the standard to remove the "ill conditioned vector" in matrix.The second improvement is to introduce an exact mode estimation method,which can effectively avoid spurious modes.The main content is to enhance the differentiation of the power spectral density of the measured signal by building the autocorrelation matrix,so that the connection point between the signal and the noise subspace will change significantly.In calculating the relative difference between the continuous eigenvalues of the matrix,the distinction between useful information and useless information will be further expanded.In this paper,the timeseries data of Beishatan bridge obtained by GB-SAR is used to verify the optimization algorithm.The results show that the algorithm proposed in this paper can effectively identify the modal parameters of the bridge and provide data basis for the health assessment of the bridge.