Using Bayesian Analysis to Implement the Specific Site Variability into LRFD Design of Piles

Abstract

The study developed a two-level Bayesian framework to account for site specific variability in bias estimates for pile capacity evaluations using cone penetration test (CPT) data. The framework updated a weak prior for the bias factor with regional data in level 1 and with site specific data in level 2. A confidence bias site parameter was introduced to give more weight to site specific data. The framework improved existing methods by combining regional data, site specific data, and engineering judgement. The proposed approach was applied to assess the bias factors for pile capacity at three sites in Louisiana: Houma Bridge, Gibson Highway and Causeway Boulevard. The resulting bias factors were used to estimate the site-specific resistance factors for LRFD based design, which are typically calibrated using statewide or nationwide data. The results highlight that the selection of prior data in level 1 Bayesian analysis has little effect on the updated posterior data of specific site. In general, the updated posterior parameters for the specific new site lie between the prior₂ parameters and the likelihood₂ parameters, taking into consideration the specific site variability. Posterior₂ data can be used to determine the LRFD resistance factor (${\upphi }_{R}$) for the design of piles based on pile-CPT design methods for the specific site. More weight should be given to new pile load test data using the confidence bias site parameter, which depends on the site condition and extent of testing.