CPT-based reliability analysis of offshore monopile foundations using geology-informed hierarchical Bayesian spatial modelling

Uncertainty in ground conditions strongly influences the reliability of offshore wind monopiles, yet design practice still treats spatial variability, soil stratification and property-transformation uncertainty in isolation. This paper presents an end-to-end Bayesian framework that carries these ground-related uncertainties from cone penetration test (CPT) data through to a monopile serviceability failure probability. A geology-informed hierarchical Gaussian-process regression (HGPR) model produces full posterior predictive distributions of cone resistance in three dimensions, with the soil-unit classification uncertainty entering in the prediction stage through a mixture of per-soil-unit Gaussian Processes. A Bayesian area-wide model then maps cone resistance to the effective friction angle, capturing also the transformation uncertainty, and a tailored conditional Monte Carlo scheme integrates these sources efficiently into the failure probability. Applied to a North Sea offshore wind farm with complex glacial stratigraphy, the framework demonstrates that transformation uncertainty has a large influence on the reliability while vertical spatial variability is largely averaged out along the pile. Loads, pile geometry and material properties are treated deterministically, so the workflow propagates the ground-related uncertainties to structural performance and provides a transparent, practical basis for reliability-based monopile design.