Building a framework for probabilistic assessment accounting for soil, spatial, operational and model uncertainty, applied to pile driveability

Abstract

Monopile driveability is associated with a large number of soil, hammer and model uncertainties. In this paper, a probabilistic framework is provided to understand what role variability plays in this problem. The framework is applied to a project where piles were driven into London Clay. Uncertainties on different variables are estimated, such as pile weight, water level during installation, hammer efficiency, soil variability and spatial variability and uncertainty of empirical parameters as well as uncertainty of the model itself. A Bayesian mixture model and conditional random field theory is applied to consider the spatial variability between the cone penetration tests (CPT). The combination of the spatial variability and parameter uncertainty provides a prior estimation of the self-penetration of the piles by carrying out Monte Carlo simulations with the model.It is possible to calibrate the model using pile self-weight penetration, with an improved estimate of the probabilistic predictions. This is done with Markov Chain Monte Carlo simulations, to quantify the posterior probability of the parameters and finally to create a posterior predictive contour of the self-penetration model. This approach is used to repeat the back-analysis of pile driving into London Clay, with prior accounting of the inherent variability.