Prediction Approaches With Applications Of Multisource Structural Health Monitoring Data For Long-Span Bridges

Rational interpretation of massive structural health monitoring(SHM)data is critical to ensuring the safety and operation efficiency of significant infrastructures,which has attracted increasing attention in civil engineering.This research aims to develop prediction approaches of multi-source data collected from long-span bridges.The main contents of this thesis are presented as follows.(1)Modeling and forecasting the strain of a long-span bridge using an improved Bayesian dynamic linear model(BDLM).An improved BDLM,which considers an autoregressive(AR)component in addition to the trend,seasonal,and regression components,is presented in this thesis to improve the computational accuracy.The model parameters are estimated by combing the expectation-maximization(EM)with Kalman filter algorithms.The order of the BDLM is determined by the measure criteria of Akaike information criterion(AIC)and root mean squared error(RMSE).The realtime monitoring data collected from a long-span cable-stayed bridge is utilized to demonstrate the feasibility of the improved BDLM.Its performance is compared with the AR model,multiple linear regression model,and BDLM without the AR component.(2)Anomaly detection of SHM data using the improved BDLM.The subspace identification method is introduced to overcome the initialization issue of the EM algorithm.The log-likelihood difference of consecutive time steps is then used to determine thresholds without introducing extra anomaly detectors.The proposed BDLM-based approach is first verified by the simulation data and then applied to the SHM data collected from two long-span bridges.(3)BDLM with switching for performance alarm of bridge expansion joints using SHM data.An approach based on the BDLM and Markov-switching theory is presented for bridge alarm of bridge expansion joints.A weighted combination of BDLMs is employed to estimate the real state of the expansion joint displacement.The EM algorithm initialized by the subspace method is utilized to optimize the parameters of switching BDLM.Then,the EM algorithm is compared with the Newton-Raphson approach in terms of computational accuracy and efficiency.The present approach is validated through the simulated data and then is applied to the expansion joint of a longspan bridge.(4)Bayesian dynamic regression for reconstruction of missing data in SHM.To address issue that static regression model fails to accurately capture the relationship in related variables,this study presents a Bayesian dynamic regression(BDR)method to reconstruct the missing SHM data.The Kalman filter and EM algorithms are employed to estimate the state variables(regressors)and parameters.The moving window strategy is introduced to reduce the computational expense of the BDR model.Two cases,including a laboratory building model and a long-span cable-stayed bridge,are utilized to examine the reconstruction performance of the multivariate BDR method.(5)A probabilistic approach for short-term prediction of wind gust speed using ensemble learning.As an efficient strategy to improve the performance of a single model,the ensemble model including random forest(RF),long-short term memory(LSTM),and Gaussian process regression(GPR)models,is presented in this research.The outputs of RF and LSTM are set as inputs of the GPR model to perform probabilistic prediction.The performance of the present methodology is compared with the persistence model(PM),RF,LSTM,GPR,averaging,and gradient boosting regression decision tree(GBDT)model.(6)A probabilistic framework for predicting typhoon-induced dynamic responses of a long-span bridge.As opposed to traditional model-based analysis involving low efficiency,TIR prediction is performed from a data-driven perspective.The quantile random forest(QRF)with Bayesian optimization is presented for probabilistic prediction.The predictor variables are obtained from parameters related to typhoon characteristics.In particular,QRF can be utilized to rank the relative importance of predictor variables.To illustrate the superiority of the QRF with Bayesian optimization,it is compared with various optimization algorithms and models.The SCB is used as a test bed to illustrate the effectiveness of the present method.