Automating site characterization from pile field data

Metamodeling techniques are increasingly applied to soil–structure interaction and geotechnical design due to their capacity to achieve high predictive performance and adaptability with significantly reduced computational demand. Focus is given on updating site-specific geotechnical parameters from monitoring information. However, depending on project site heterogeneity, anisotropy and constitutive model selected, the dimensionality of the calibration problem can increase substantially. For such scenarios, the quality of surrogate prediction might be prohibitive in terms of performance. In addition, an often-neglected aspect is metamodel application to simplified geological-based ground models over mechanical soil behavioral profiles computed from available in-situ testing. This study introduces two Bayesian practice-oriented methodologies to update geotechnical constitutive model parameters from limited pile field data and eliciting error propagations within heterogenous CPT-based ground models: a surrogate-based approach and a hybrid-adaptive workflow. An adaptive Differential Evolutional Transitional Markov Chain Montecarlo is proposed to assess the asymptotical convergence toward high dimensional posterior distribution. These methodologies are applied on a drilled-and-grouted pile, embedded in a natural soil deposit, subjected to static and cyclic tensile loading. The proposed workflows are designed to leverage standard features of commercial geotechnical software and established constitutive models, enabling practical implementation. The two methods propose optimization frameworks capable of jointly updating high dimensional geotechnical parameters using limited site-specific pile field test data.