Parameter Estimation of a Critical-State Based Strain Hardening Soil Constitutive Model Using Particle Filtering Framework

In order to capture the underlying deformation mechanism of soil, advanced soil constitutive models often involve several model parameters, which further enable to mimic the complex behavior of soil accurately. However, the intricacies involved in calibrating the associated large number of model parameters often deter the use of such advanced constitutive models. The sampling-based particle filtering (PF) approach, often used in conjunction with other numerical techniques, emerges as an efficient tool to solve related boundary value problems. The present study aims to explore the potential application of PF for estimating the parameters of a critical state-based soil constitutive model using triaxial test data. The employed soil constitutive model is a non-associative one and incorporates shear-strain hardening/softeningresponse following a hyperbolic relationgoverned by the state variables. A stress-based single elementalgorithm has been implemented to replicate the triaxial shearing response of soil following the constitutive model, where the model parameters are considered to evolve in the parameter space. The PF maximizes the Bayesian likelihood of these model parameters by updating their posterior probability density functions, where the density functions are approximated as particles. As a result, it enhances the predictive accuracy and reduces the disparity between the numerical simulation and experimental observation. To validate the proposed approach, it has beenfirst tested against a synthetic triaxial shearing dataset under different confining pressure and density states. Finally, the proposed method has been applied to estimate the model parameters pertaining to thetriaxial test dataavailable for Hostun sand, and the model predictions obtained by employing the mean estimates of these parameters have been further presentedfor elucidating the efficacy of the proposed approach.

Keywords: Soil constitutive model, Particle filtering, Parameter estimation, Inverse analysis.