Bayesian Model Calibration of a Triaxial Specimen Using Surface Displacement and Reaction Force Data

Precise measurement of soil deformation across a wide range of strains is vital for geotechnical engineering applications. Traditional triaxial tests are limited in their ability to accurately capture small strain behaviour, end effects, and local nonuniformities, presenting a significant challenge. Digital Image Correlation (DIC) offers a non-contact and cost-effective solution for detailed surface displacement measurements, but the challenge lies in converting this data into actionable soil parameters. This study addresses this challenge by employing Bayesian model calibration in a synthetic case study of a triaxial specimen modelled using a Finite Element Model (FEM). A sequential Bayesian calibration approach is shown to be able to quantify uncertainties and refines parameter estimates dynamically, enhancing the prediction of soil behaviour under changing stress conditions. The results show significant reductions in parameter uncertainty and convergence towards ground truth values. This methodology represents a transformative step in soil mechanics, utilising advanced statistical techniques to improve the reliability and accuracy of soil behaviour models.