Bayesian Optimization for CPT-Based Prediction of Impact Pile Drivability

Abstract

Pile drivability predictions require information on the pile geometry, impact hammer, and the soil resistance to driving (SRD). Current SRD prediction methods are based on databases of long slender piles from the oil and gas industry and new, robust, and adaptable methods are required to predict SRD for current offshore pile geometries. This paper describes an optimization framework to update uncertain model parameters in existing axial static design methods to calibrate SRD. The approach is demonstrated using a case study from a German offshore wind site. The optimization process is undertaken using a robust Bayesian approach to dynamically update uncertain variables during driving to improve simulations. The existing method is shown to perform well for piles with geometries that reflect the underlying database such that only minimal optimization is required. For larger diameter piles, relative to the prior best estimate, optimized results are shown to provide significant improvements in the mean calculations and associated variance of pile drivability as more data is acquired. The optimized parameters can be used to predict SRD for similar piles in analogous ground conditions. The demonstrated framework is adaptable and can be used to develop site-specific calibrations and advance new SRD methods where large pile driving data sets are available.