在大地工程中,許多學者提出經驗式來預測目標設計參數,然而這些經驗式在使用上會受到特定土壤參數限制,當缺少特定土壤參數時就無法使用這些經驗式來進行預測,並且在使用上會受到土壤種類與地區的影響,導致預測結果存在許多不確定性。因此本研究希望藉由數據驅動法以及階段式貝氏模型來量化場址特性,並建立一個多變數機率模型,針對淺基礎的極限承載力與沉陷量進行預測。 透過前人文獻回顧,蒐集並了解與淺基礎在垂直載重下之行為相關的土壤參數來建立全球資料庫,接著篩選出較有探討性的參數,包含:(1)判釋之極限承載力(q_um);(2)標準貫入試驗(N_60);(3)圓錐貫入試驗之椎間阻抗(q_c);(4)正規化之(N_1 )_60;(5)正規化之Q_tn;(6)基礎寬度(B);(7)Vesic(1973)公式計算之極限承載力(q_uc);(8)雙曲線擬合參數a;(9)雙曲線擬合參數b。 在建置多變數機率模型時,先藉由Johnson分布系統(Johnson distribution system)將參數轉換至標準常態空間,再利用吉普斯取樣法(Gibbs sampler, GS)、共軛條件(conjugate conditions)與階段式貝氏模型 (hierarchical Bayesian model, HBM)來學習參數間的相關性和資料庫中的場址特性,同時將資料庫中空缺的資料填補,接著加入目標場址的已知資訊,便能預測預求得之目標參數,隨著已知資訊的增加,不確定性便隨之降低,於可靠度的觀念下能更加準確地去進行淺基礎設計並節省設計成本。
In geotechnical engineering, many scholars have proposed empirical formulas to predict design parameters. However, they are limited by specific soil parameters. And these empirical methods will be affected by soil types and regions, resulting in many uncertainties in the prediction results. Therefore, this study uses data-driven methods and Hierarchical Bayesian model(HBM) to quantify site characteristics, and establishes a multivariable probability model to predict the ultimate bearing capacity and settlement of shallow foundations. In this study, the parameters in the database are:(1) measured ultimate bearing capacity (q_um) ;(2) standard penetration test N_60 value;(3) cone penetration resistance (q_c);(4) normalized (N_1 )_60;(5) normalized Q_tn;(6) foundation width (B);(7) calculated ultimate bearing capacity by Vesic (1973) (q_uc);(8) hyperbolic fitting parameter a;(9) hyperbolic fitting parameter b. To build a multivariate probability model, using Johnson distribution system to convert the parameters into standard normal distributions, then applying Gibbs sampler method, conjugate conditions and the hierarchical Bayesian model (HBM) to capture the correlation behavior in site-specific data and fill the missing data. With the amount of the known information increase, the uncertainty of prediction will decrease. Based on reliability-based design, it can be more economical and accurate in the design of shallow foundations.