Bayesian updating of LRFD resistance factors of driven PPC piles from dynamic pile load tests

The current design codes in the load and resistance factor design (LRFD) framework for piles provide limited guidance on how to update the resistance factors (RFs) when dynamic pile load testing (DPLT) is conducted. Meanwhile, the calibration process for these RFs, particularly under varying site conditions and testing levels (%), remains underexplored. This study presents a novel Bayesian interface framework (BIF) for updating the LRFD RFs (Φ) using DPLT data while accounting for site variability. Four Bayesian interface-based methods (BIBMs), the Bayes’ theorem (BT), assumed density filtering, expectation propagation (EP), and conjugate priors, are employed to combine prior resistance statistics (from cone penetration test (CPT)-based pile design methods and static pile load tests (SPLT) data) with likelihood distributions derived from DPLT data. The prior data are formulated using the top-performed CPT-based “Laboratoire Central des Ponts et Chaussées (LCPC)” Pile Design Method and site-specific coefficient of variation (COV) refined through a pooled estimator approach. The likelihood distributions are adjusted for test sample size using corrected population variance. The methodology is applied to square precast prestressed concrete piles driven into Louisiana soils using a database of 80 SPLT piles. The site variability is categorized into ranges of low, medium, and high COVs. The impact of varying testing percentages of DPLTs (from 10% to 100%) and site variability (COV from 0.1 to 0.5) on RF calibration is investigated through Monte Carlo simulation. The results indicate that significant increases in RFs can be achieved even with limited DPLT; for example, testing just 10% of piles yields noticeable improvement, while approximately 45% testing achieves most of the benefit. The study demonstrates the effectiveness of BIFs, particularly BT and conjugate prior, in improving design reliability under uncertainty.