Borehole and CPTU Integrated Probabilistic Site Characterization with Noisy Data Filtering

The presence of noisy data or outliers in borehole drilling and piezocone penetration test (CPTU) may lead to biased quantification of uncertainties for soil classification and soil property evaluation. Several statistical methods have been proposed in the literature to identify and filter such noisy data; nevertheless, they may not properly interpret the physical meaning of noisy data. This study incorporates a widely used CPTU-based soil classification chart into a coupled Bayesian machine learning method to achieve noise filtering in the integration of borehole and CPTU data. CPTU data are converted to soil behavior types or soil zones based on the soil classification chart and probabilistically related to the authentic soil types by a transition probability matrix under the Bayesian framework. The proposed method is demonstrated by the probabilistic site characterization of the Hong Kong-Zhuhai-Macao Bridge project. Results indicate that noisy data far away from the dominating data set can be properly identified and separated. The proposed model successfully captures the variation of major soil strata revealed by borehole logs and simultaneously identifies the minor variations of significant thin layers recorded by the CPTU data. A sensitivity study indicates that the spatial autocorrelation of soil type should be properly considered in the noise filtering of soil classification.

Automated summary

This study combines Bayesian machine learning with CPTU data and utilizes a soil classification chart for noise filtering. It can accurately identify and separate noisy data, and effectively capture soil stratification variations.