Collaborative Response Updating And Probabilistic Monitoring Design Methods For Slope Engineering

Landslide caused by slope failure is one of the major geoharzards in China.Proper evaluation of slope stability and deformation characteristics is vital to landslide prevention and risk control.Slope stability and deformation are affected by various uncertainties,including the inherent spatial variability of geomaterials and load uncertainty.Evaluating slope stability and predicting the deformation depend on geotechnical site investigation data,field monitoring data and calculation models.How to accurately and efficiently evaluate slope responses(e.g.,stability,deformation)considering different data and calculation models under uncertainties is a challenging task.The key challenges are:(1)Geotechnical site investigation data and monitoring data have spatial-temporal variability,and their measurements rely on the measurement location and the observation stage.As measurement data changes,it requires to constantly update the slope stability and deformation characteristics.Traditional deterministic analyses cannot quantify the effects of the changing data and the uncertainties involved;(2)The site investigation and monitoring data directly obtained are usually not identical to input parameters of the slope response evaluation model,and inverse analyses are thus required.Updating the slope response considering different data necessitates repeated inverse analyses,leading to the extremely low computational efficiency;(3)Many calculation models exist for slope response analysis.Simple models(e.g.,response surface model,RSM)are computationally efficient,but less accurate,while sophisticated models(e.g.,finite element model,FEM)are more accurate,but the induced computational burden is intensive.How to combine these two types of models to carry out model parameter inversion and update slope response in an efficient and accurate manner deserves further investigation.In addition,slope stability analysis and deformation prediction provide references for slope monitoring design.Currently,selecting and combining monitoring variables for slope monitoring design highly depend on engineering experience and subjective judgment.There is a lack of quantitative monitoring design methods,and it is difficult to consider the effects of various uncertainties on the monitoring design.Based on the abovementioned issues,the thesis focuses on slope response updating and monitoring design considering uncertainties,and takes quantitative analysis of slope uncertainty as the basis,aiming at proposing methods for efficient slope response updating and quantitative monitoring design.Research contents include efficient reliability updating of slope stability considering different site investigation data,fast slope settlement updating based on monitoring data from multiple stages and different calculation models,single monitoring variable selection and multivariable combinatorial optimization for slope monitoring design.The main work and conclusions are as follows:(1)A collaborative reliability updating approach of slope stability considering different site investigation data is proposed.The proposed method first uses Bayesian updating framework with structural reliability methods(BUS)to generate conditional random fields to characterize the spatial variability of geomaterials given site investigation data,and then employs the rejection sampling principle and collaborative analysis method(CAM)to efficiently update random fields considering different site investigation data,avoiding regenerating conditional random fields with respect to data and performing the corresponding slope stability analyses.A single-layered soil slope considering a non-stationary random field and a double-layered slope with stationary random fields are adopted to demonstrate the effectiveness and rationality of the proposed method.It shows that the proposed method can reasonably characterize the spatial variability of soil parameters considering different site investigation data(e.g.,those from different borehole locations and depths),and significantly improve the computational efficiency for slope reliability updating,which makes it possible to perform real-time reliability updating of slope stability considering soil spatial variability.(2)A collaborative updating approach of slope settlement considering monitoring data from multiple stages is developed.The proposed approach first performs driving Bayesian analysis with monitoring data obtained at early stages,and as data are sequentially obtained,it employs CAM to update slope settlement considering monitoring data obtained at later stages(or datasets from different monitoring stages).The proposed method can not only sequentially update the slope settlement given the dataset from different monitoring stages,but also provide slope settlement updating considering datasets from different locations and/or testing types.A filled slope is employed to illustrate the proposed approach.Compared with traditional Bayesian updating approach,this method can update the slope settlement efficiently and accurately considering different monitoring datasets,and provides an effective tool for exploring effects of different updating strategies(or datasets)on slope settlement updating,which enriches the data analysis method of slope monitoring.(3)A collaborative updating approach of slope settlement based on RSM and FEM is developed.The proposed method first performs driving Bayesian updating based on BUS and RSM,and explores the posterior sample space of uncertain parameters of settlement prediction model.Then,based on FEM and CAM,it performs target Bayesian updating to correct the posterior distribution of model parameters and the updated slope settlement.The unified equations are derived for driving Bayesian updating with RSM and target Bayesian updating using FEM.The method makes full use of the advantages of both models and the correlations between settlement predictions.A filled slope is adopted to demonstrate the proposed approach.Results show that the proposed approach reduces the calculation error caused by using RSM as the approximate function of FEM for Bayesian updating,and improves the computational efficiency for updating slope settlement using FEM,which opens up a new pathway for slope response updating by employing sophisticated numerical models.(4)A collaborative updating-based approach for reliability sensitivity analysis of slope stability for single monitoring variable design is developed.The proposed method conducts prior reliability analysis to quantitatively convert the prior information(including engineering experience and calculation model)into pre-monitoring information(i.e.,possible observational values(POVs)of monitoring variables),and then uses CAM to efficiently calculate the conditional probability of slope instability given different POVs of monitoring variables.The variation function of slope failure probability corresponding to different POVs of monitoring variables quantitatively reflects the sensitivity of monitoring variable with respect to slope reliability.In monitoring design,sensors can be located according to the sensitivity of monitoring variable.A reinforced slope example is used to illustrate the proposed method.Results show that the proposed method can quantitatively analyze the reliability sensitivity of monitoring variables according to prior information,and reasonably identify the most sensitive variable based on the reliability sensitivity of different monitoring variables,which provides a reference for selecting monitoring variables for slope monitoring design.(5)A collaborative updating-based approach for value of information(Vo I)analysis of monitoring variables for multivariable combinatorial optimization design is proposed.The proposed method decomposes posterior reliability analysis problem given pre-monitoring information into three probabilistic analysis problems,then employs CAM to respectively calculate the occurrence probability and conditional probability of POVs from different monitoring variable combinations,based on which the conditional probability of slope instability is efficiently updated and expected Vo I of different combinations of monitoring variables is then quantified.Maximizing the expectation of Vo I provides multivariable combinatorial optimization design.A reinforced slope example is used to illustrate the proposed method.Slope monitoring design based on Vo I analysis considering different monitoring variable combinations involves millions of reliability updating of slope stability,and traditional methods usually cannot complete the monitoring combination design.The proposed method effectively quantifies the expected Vo I of different monitoring variable combinations,and provides a reference for combination design of slope monitoring variables.