Probabilistic Characterization Of Underground Stratigraphy And Its Uncertainty Based On Cone Penetration Test

Cone penetration test(CPT)is a widely used in-situ geotechnical test which is fast,economical,and can provide nearly continuous measurements of soil properties.Meanwhile,stratigraphy identification based on CPT data is an important aspect of geotechnical investigation,which affects geotechnical engineering analysis and design.However,CPT can not provide soil samples so that sectional drawing of soil layer can not be obtained.Traditional schemes of stratigraphy identification based on CPT rely on engineering experience and subjective jugdements.Thus,there inevitably exists uncertainty,and the reliability of stratigraphy is unknown.Therefore,how to identify soil layers objectively and quantify the uncertainty of the stratigraphy profile rationally to reflect its reliability is still a crucial problem to be solved.To address the aforementioned problems,this thesis develops a Bayesian framework,which consists of two parts:(1)determination of the number of soil layers(N)and(2)identification of the most probable soil boundaries.Within this framework,the Bayesian function of underground stratigraphy is derived based on CPT data and a given N considering spatial variability and statistical property of soil mechanical parameters.The uncertainty of stratigraphy profile is quantified to provide guidelines for further investigation and geotechnical design.The most probable N is determined by comparing model evidences of different models.To solve the complex high-dimensional Bayesian updating problem,a Subset Simulation-based Bayesian updating algorithm is adopted to generate equivalent samples following posterior distribution and obtain model evidences.The proposed method is applied to soil behavior type index and specific penetration resistance,and standard deviations of soil boundaries are used to quantify the uncertainty of boundaries.The effectiveness and correctness of the method are verified.Results manifest that the identified soil layers reflect the statistical property of soil mechanical parameters,and the uncertainty of boundaries is affected by the difference of adjacent two soil layers.This thesis is organized as follows.First,the research background and significance are introduced.Then,the Bayesian model identification and selection methods are discussed,and a method based on Bayesian theory is proposed for the probabilistic characterization of stratigraphy and its uncertainty.Finally,the proposed method is applied to soil behavior type index data and specific penetration resistance data.