Metamodelling for geotechnical reliability analysis with noisy and incomplete models

A kriging-based metamodelling approach for analysing the structural reliability of a sheetpile wall in dyke is formulated. This specific problem characterised by high target reliabilities (Pf∼10−7) combination with noisy and incomplete numerical model response. Starting from original formulation active learning Monte Carlo simulation (AK-MCS), robust two-stage metamodel framework formulated adaptive multiple importance sampling, Gaussian process classification kernel enhancements. Learning functions convergence criteria are revised to maintain consistency The developed applied analysis soil-structure interaction involving body, which representative class problems encountered engineering practice. Low dimensional example studies demonstrate workings give insight into Full probabilistic analyses then performed estimate probabilities failure updating context. results show that after several necessary enhancements classical formulations, approaches can be used successfully computational models as often geotechnical