Probabilistic characterization of inherent and epistemic geotechnical uncertainty in soil constitutive models using polynomial chaos expansion and monotonic drained triaxial tests

This study proposes a probabilistic, uncertainty-informed framework for calibrating advanced soil constitutive models (SCMs), particularly, advanced critical state-based models, to accurately capture uncertainty in soil behavior in geotechnical applications. The proposed framework incorporates Polynomial Chaos Expansion (PCE) metamodels to optimize sensitivity analysis (SA) and enable Bayesian updating of SCM parameters, ensuring precise calibration that addresses inherent and epistemic uncertainties. Additionally, Random Forest (RF) analysis is employed to validate initial statistical assumptions and parameter trends during SA and enhance the robustness in the calibration process. Monotonic drained triaxial tests are used within this framework to calibrate the SANISAND model, an advanced critical state-based SCM for sand, with a focus on Nevada Sand soil due to its significance in geotechnical engineering. The framework estimates parameters’ joint probability density functions (PDFs) from experimental data, providing probabilistic insights into model responses under varying confining pressures and relative densities. By reducing computational demands and integrating uncertainty quantification, this approach offers an efficient and accurate calibration process, improving SCM predictive capability and reliability for use in finite element (FE) analyses. This study demonstrates the framework's application and validation to Nevada Sand and proposes PDFs with correlation coefficients for the SANISAND model, accelerating its integration in posterior stochastic geotechnical system-level modeling.