A mixing-locally polynomial Regression method and Global Sensitivity analysis – (MRGS) procedure is used in assessing foundation and abutment uncertainties of concrete arch dams. Latin hypercube sampling is employed for generating samples in the sensitivity analysis. The model consists of a typical 80 m high arch dam along with its foundation and abutments. A potentially movable rock wedge confined by three discontinuities located in the Karun IV dam’s abutment is incorporated in the model. The random variables include (a) variables related to the geometry of foundation discontinuities, (b) variables associated with the strength of discontinuities, and (c) rock mass properties. The response parameter (Damage Parameter (DP)) is the maximum displacement of the foundation in contact with the dam body. Sobol indices as a measure of sensitivity to response change are computed for each random variable, and the most important random variables affecting the dam's stability are identified. Results show that compared with the first-order polynomial, the second-order functions lead to a more efficient sample size. The proposed methodology can be used practically in the reliability analysis of not only any existing concrete arch dam but also for similar civil engineering systems.

混合局部多项式回归方法和全局敏感性分析 - (MRGS) 程序用于评估混凝土拱坝的地基和桥台不确定性。拉丁超立方采样用于在灵敏度分析中生成样本。该模型由一座典型的 80 m 高拱坝及其基础和桥台组成。由位于 Karun IV 大坝桥台的三个不连续点限制的潜在可移动岩石楔被纳入模型。随机变量包括 (a) 与基础不连续性几何相关的变量，(b) 与不连续性强度相关的变量，以及 (c) 岩体特性。响应参数（Damage Parameter (DP)）是基础与坝体接触的最大位移。计算每个随机变量的 Sobol 指数作为响应变化敏感性的度量，并确定影响大坝稳定性的最重要的随机变量。结果表明，与一阶多项式相比，二阶函数导致更有效的样本量。所提出的方法不仅可用于任何现有混凝土拱坝的可靠性分析，也可用于类似的土木工程系统的可靠性分析。