Parameter and model identification using the particle filter for geotechnical applications

A computational method incorporating the finite element model into data assimilation technique is presented for identifying the yield function of the elasto-plastic constitutive model, corresponding parameters and/or initial/boundary conditions based on the sequential measurements of hypothetical soil tests and an actual construction sequence to overcome some difficulties for elasto-plastic problems from which the existing inverse analysis strategies have suffered.Yield function and/or parameter identification of the elasto-plastic constitutive model should be made by considering the measurements of deformation and/or pore pressure step by step from the initial stage of construction and throughout the deformation history under the changing traction boundary conditions due to the embankment or the excavation, because the ground behavior is highly dependent on the loading history. Thus, it appears that sequential data assimilation techniques, such as the Ensemble Kalman Filter (EnKF) or the Particle Filter (PF), are the preferable tools that can provide estimates of the state variables, i.e., deformation, pore pressure, and unknown parameters, for the constitutive model in geotechnical practice. An appropriate set, consisting of the yield function of the constitutive model and the parameters of the constitutive model, can be simultaneously identified by the PF in order to describe the most suitable soil behavior. This paper discusses the priority of the PF in its application to initial/boundary value problems for elasto-plastic materials and demonstrates a couple of numerical examples.