Efficient automatic estimation of soil constitutive model parameters via particle swarm optimization

In this study, we automatically estimated the parameters of the modified Cam-Clay model, a representative constitutive model for soil. The estimation was carried out by minimizing the objective function using the dynamic multiswarm particle swarm optimization (DMS-PSO) algorithm, which is an improvement over the original PSO. The objective function was newly defined by quantifying the discrepancy between the targeted results and the model calculations in q-p′-v space. DMS-PSO divides particles into several islands to search globally and prevent local solutions, and even particles that fall into a local solution can be relocated. To evaluate the automatic estimation performance of DMS-PSO, we examined whether model parameters could be correctly estimated from the calculation results (Consideration (1)) and whether the DMS-PSO algorithm could consistently obtain the same parameter values when reproducing the experimental results (Consideration (2)). Regarding Consideration (1), the objective function was consistently smaller than 1.0 × 10^–6 when the number of particles was greater than 400 and the number of islands was greater than 40. At this time, the parameter values could be estimated to the fifth decimal place. When two experiments were conducted, the estimation was obtained approximately 1.5 times faster than when only one was conducted. Regarding Consideration (2), the coefficient of variation of the parameters obtained from 100 estimations was at most 1%, and the parameter values were estimated to be approximately the same each time. In addition, narrowing the solution search range based on soil physical properties could reduce the variation in parameters by approximately 10%. Additionally, the parameters could be accurately estimated by data from at least two mechanical experiments.