Kalman Filter-based Monitoring And Prediction Of Building Deformation

Structural Deformation is an important indicator for the structural safety assessment and damage diagnosis.It is a complicated and time-varying system with multi-factors coupling.Accurate measurement and prediction of structural deformation are still challenges in civil engineering.Aiming at the errors brought by the multiple time-varying effects and uncertain factors in the data acquisition,abnormal data detection,missing data recovery,dynamic displacement estimation and deformation prediction,this paper systematically studies the method of structure deformation monitoring and prediction based on multi-model/multi-rate Kalman filtering.Based on the state space model of the structural deformation and displacement,the high-frequency and low-frequency errors in the displacement estimation are corrected by fusing different types of data with different sampling frequencies,and uncertainty errors in the deformation prediction are reduced by interactive fusion of multiple deformation state models.Consequently,an optimal estimation method for the multivariate coupled time-varying system of structural deformation is established to improve the accuracy of structural deformation measurement and prediction.The main contents of this research are as follows:(1)In order to investigate the deformation pattern of the structure under multiple timevarying effects such as temperature,concrete shrinkage,concrete creep,and construction load during the construction period,data on structural deformation,temperature and strain during the construction period are collected via a structural health monitoring system on a practical structure.The uneven temperature distribution inside and outside the main structure as well as in different directions was investigated.The strain development of the vertical and horizontal members of the structure under the uneven temperature distribution was revealed.In addition,the development of vertical shortening and differential shortening during the construction period of the main body of the structure was investigated.Plenty of monitoring data were provided for the subsequent research works of this paper.(2)To address the problem of low efficiency in detecting multiple data anomalies in field monitoring,a data anomaly identification method based on feature extraction and pattern recognition neural network(PRNN)was proposed.Feature sets were established based on the characteristics of different data anomalies.By doing so,the lengthy original data samples were transformed into short feature vector samples.Only a small number of labeled feature vector samples were used to train the PRNN,which significantly improved the efficiency of data processing and network training while ensuring accuracy.The polarized AUCs metric was introduced to accurately describe the detection performance,which improves the optimization efficiency of feature selection and network parameters.As a result,highly efficient and accurate detection of multi-type data anomalies were achieved.This method accuracy was verified using the monitoring data of an actual structure,and the influences of different factors on the method accuracy were analyzed.(3)To address the problem of inaccurate recovery of long-term missing deformation data,an interactive multi-model Kalman filter(IMM-KF)based method for recovering missing deformation data was proposed.The state model of the structural strain under multiple effects such as upper load,temperature,concrete shrinkage,and creep was established.The state model was extended into multiple interactive state models(IMM)by considering the perturbation of structural deformation coefficients caused by uncertainties.Using the calculated elastic deformation,concrete shrinkage and creep,and the measured temperature as inputs.The optimal estimation of structural strain was obtained via the fusion of the multiple state models,reducing the estimation errors caused by the parameter perturbations.As a result,the missing strain data was accurately recovered.This method was verified by the data from concrete shrinkage creep tests.Moreover,field monitoring data during the construction period were used to verify the performance of this method in recovering long-term missing strain data.(4)To address the problems of difficulty and low accuracy in the measurement of structural dynamic displacement,a multi-rate Kalman filtering-based method was proposed to measure dynamic displacement by fusing acceleration and strain data.Firstly,a strain-displacement transformation formula was derived based on the geometric conversion relationship between the overall structural displacement and local strain,which is not dependent on the modal shapes.For the case of different sampling rates of strain and acceleration in practical monitoring,the high-frequency acceleration and low-frequency strain-displacement are optimally fused by multi-rate Kalman filtering.The fusion of strain and acceleration data not only makes up for the lack of high-frequency information of the low-frequency strain but also modifies the low-frequency error of acceleration,which significantly improves the accuracy of dynamic displacement.The accuracy and effectiveness of this method were verified by a steel cantilever beam vibration test and field monitoring of dynamic displacement of two supertall structures.(5)To address the problem of inaccurate prediction due to the perturbation of structural deformation prediction model parameters,a structural deformation prediction method based on an interactive multi-model adaptive Kalman filter(IMM-AKF)was proposed.The uncertain errors generated by environmental changes and boundary constraints are equated to fluctuations of the structural deformation coefficients.Therefore,interactive multi-state models of structural strain were established,and the structural strain was predicted by Kalman filtering.The convergence performance of the filter was judged according to the residuals between the predicted and measured strains.The filter error parameters were adaptively corrected by using the forgetting factor to improve the prediction accuracy of structural strain.The accuracy of the method was verified by the data of concrete shrinkage and creep test data,and the strain measurements during the construction period of a real supertall structure.