基於孔內壓力儀的土壤參數大數據分析

Abstract

過去，大地工程設計參數往往會使用簡單且便宜的試驗搭配轉換模型來來取得，而這些轉換模型常常具有較大的不確定性。近年來，隨著大數據的應用的發展，在各個領域皆能帶來更精確的分析與預測，因此，本研究希望藉由孔內壓力儀試驗的大數據分析來預測土壤模數(EM)及土壤的靜止土壓力係數(K0)，藉此降低推估中的不確定性。 首先，先藉由過去的文獻來蒐集由孔內壓力儀試驗所得的土壤模數(EM)及靜止土壓力係數(K0)，並同時一併蒐集其他土壤參數，以建立分析所需要之資料庫。建立資料庫完成之後，依前述的幾個經驗公式進行初步的檢核，藉此觀察資料的趨勢，再來篩選出我們想要探討相關性的參數。對於凝聚性土壤，我們關心的有: (1)液性限度(LL)；(2)塑性指數(PI)；(3)液性指數(LI)；(4)垂直有效應力(σv’)；(5)預壓密應力(σP’)；(6)標準化SPTN值(N60)；(7)PMT所得之靜止土壓力係數(K0)；(8) PMT所得之變形模數(EM)；(9)孔隙水壓常數(Bq)；(10)錐尖阻抗應力(qc)。對於非凝聚性土壤，我們關心的有: (1)孔隙比(void ratio, e0)；(2)相對密度(Dr)；(3)正規化垂直有效應力(σv’/Pa)；(4)靜止土壓力係數(K0)；(5)PMT所得之變形模數(EM)；(6)孔隙水壓常數(Bq)；(7)錐尖阻抗(qC)；(8)標準化SPTN值(N60)等。 接下來的分析將運用層級貝氏模型(HBM)來探索參數之間的相關性，並有效地捕捉資料庫中不同場址的特性。藉由結合目標場址有限的調查數據，模型可以推測出尚未掌握的資訊分佈。整個過程將使用Johnson分布系統、吉布斯取樣法以及貝氏分析中的共軛條件進行推估。結果顯示，層級貝氏模型能夠準確預測出目標場址的變形模數(EM)與靜止土壓力係數(K0)。透過可靠度觀念，這種方法能提升設計大地結構物的精準度，從而減少材料成本並提高工程的經濟效益。

In the past, geotechnical engineering design parameters were often obtained using simple and inexpensive tests combined with empirical models, which frequently involved high uncertainty. Recently, with the development of big data applications, more precise analysis and predictions have been achieved across various fields. This study aims to leverage big data analysis from pressuremeter tests to predict the soil modulus (EM) and the coefficient of earth pressure at rest (K₀), thereby reducing estimation uncertainties. First, data on soil modulus (EM) and the coefficient of earth pressure at rest (K₀) obtained from pressuremeter tests were collected from previous studies, along with other soil parameters, to establish a comprehensive database for analysis. Once the database was completed, initial checks were conducted using several empirical formulas to observe data trends and select parameters relevant to the study. The subsequent analysis employs a hierarchical Bayesian model (HBM) to explore correlations among parameters and capture the distinct characteristics of different sites within the database. By combining limited survey data from the target site, the model can infer the distribution of unknown information. This process utilizes the Johnson distribution system, Gibbs sampling, and conjugate priors in Bayesian analysis. The results demonstrate that the hierarchical Bayesian model can accurately predict the deformation modulus (EM) and the coefficient of earth pressure at rest (K₀) for the target site.By integrating reliability concepts, this method enhances the precision of geotechnical structure design, reducing material costs and increasing the economic efficiency of engineering projects.