Sensitivity analysis of a three-invariant plasticity model with different sampling algorithms

ABSTRACT A Sobol sensitivity analysis is employed to identify the importance of parameters in the three-invariant soil model (GeoModel) for finite element simulation. Three commonly used sampling techniques, namely, pseudo-random, quasi-random (QR), and Latin-hypercube sampling (LHS), are investigated to determine the efficiency of each sampling technique for sensitivity analysis and identify the critical parameters for the soil model. Among the sampling methods, LHS shows the best sampling uniformity and offers more stable results and faster convergence than other methods. The analysis results also show that Young’s modulus and the initial value for the shear-yield surface are the most critical parameters for the designed outputs, which quantitatively measure the maximum total deformation (MD), plastic deformation (PV), and plastic zone size (VY). The proposed sensitivity analysis framework in this study can be applied to finite element simulations of similar engineering and scientific problems for assessing model parameters’ influence.