Bayesian inversion for parameter estimation of advanced soil constitutive model and associated uncertainty quantification: Transitional Markov Chain Monte Carlo versus Particle Filtering

Precise estimation and uncertainty quantification of parameters of advanced soil models become crucial for reliable constitutive predictions and risk-informed design of geo-structures. Bayesian updating offers a robust approach for addressing model and measurement uncertainties while yielding probabilistic parameter estimates. Moreover, it circumvents the need of iterative calibration commonly adopted in laboratory test-based model fitting, thereby reducing subjectivity and enhancing reproducibility. Nevertheless, such approaches for soil constitutive parameter estimation from single-element triaxial tests across multiple confining and density states have not been explored extensively. In the present study, inverse estimation approaches powered by the Bayesian inference technique, i.e. Transitional Markov Chain Monte Carlo (TMCMC) and Particle Filtering (PF), have been employed in this context. An exhaustive comparative study has been conducted with a set of synthetic data, reflecting their relative performance in terms of computational efficiency, convergence rate, prediction accuracy, and model precision. The methods have further been employed with the drained triaxial test data of Hostun sand to ascertain their real-life applicability. It has been perceived that PF outperforms TMCMC in accuracy, precision, as well as computational cost. As a spin-off, this study also identified the inherent correlations between the critical state parameters, which are not readily evident through deterministic approaches.