Calibration of Soil Model Parameters Using Particle Swarm Optimization

In this paper, a neuro-fuzzy model in conjunction with particle swarm optimization (PSO) are used for calibration of soil parameters used within a linear elastic-hardening plastic constitutive model with the Drucker-Prager yield criterion. The neuro-fuzzy system is used to provide a nonlinear regression between the deviatoric stress and axial strain (sigma(d) - epsilon) obtained from a consolidated drained triaxial test on samples of poorly graded sand. The soil model parameters are determined in an iterative optimization loop with PSO and an adaptive network based on a fuzzy inference system such that the equations of the linear elastic model and (where appropriate) the hardening Drucker-Prager yield criterion are simultaneously satisfied. It is shown that the model parameters can be determined with relatively high accuracy in spite of the limited insight gained by a single set of data. To verify the robustness of the technique, a second set of data obtained under different confining pressures is then used in a separate run. The outcome shows a close match with the same order of accuracy. (c) 2012 American Society of Civil Engineers.