静力触探识别场地土层分布的贝叶斯学习方法研究

Cone penetration test (CPT) is usually performed vertically to identify subsurface soil stratification.  However, due to time and budget constraints, the number of CPT soundings performed in a site is often limited, leading to a great challenge in properly interpreting CPT data and identifying stratification in unsounded area along horizontal direction.  A Bayesian learning method is presented in this paper to address this difficulty.  The method can predict soil classification and stratification in a two-dimensional (2D) vertical cross-section using a limited number of CPT soundings.  The method consists of three components: (1) 2D interpolation of CPT data using Bayesian learning; (2) determination of soil behavior type (SBT) using Robertson chart at every location in the 2D cross-section, including locations with and without CPT soundings; (3) and soil layer/zone delineation using an edge detection method.  High-resolution CPT data and SBT information in the 2D vertical cross-section can be obtained.  Soil layer/zone boundaries are delineated automatically.  The method is illustrated using a simulated example.  The results suggest that the method performs well even when only five sets of CPT soundings are available.