Parameter Estimation of Particle Flow Model for Soils Using Neural Networks

Abstract

A calibration process is developed to determine the parameter values. Three-axial compressions tests in laboratory and neural network are used to determine the material internal friction angle and stiffness, respectively. These tests are repeated numerically using PFC models with different sets of particle friction coefficients and particle stiffness values. Three-axial compressions tests are found to be dependent on both the particle friction coefficient and the particle stiffness. The compression test results can be used to determine a unique set of particle friction and particle stiffness values. The calibration process is validated by modelling filling process of head chamber of shield machine. It is shown that the parameter estimation procedure proposed in the paper can accurately predict the deformation characteristics and flow patterns of conditioned soils.