Efficient Automatic Estimation of Soil Constitutive Model Parameters by Particle Swarm Optimization

Abstract Performance design is important for soil structures, and detailed analyses are performed using the finite element method and other methods. Parameters for such analysis are often estimated from soil test results. In this study, we proposed a method to estimate the parameters of the modified Cam-Clay model using dynamic multiswarm PSO (DMS-PSO). To examine the validity of the proposed method, we examined whether constitutive model parameters can be correctly estimated from the results computed by the model (consideration (1)) and whether the proposed methods can always obtain the same parameters when reproducing the experiment (consideration (2)). In the consideration (1), the search success rate was clearly increased by using DMS-PSO compared to ordinary PSO. In addition, the search was successfully conducted if the number of ‘particles’ was more than 400 and the number of ‘islands’ was more than 40. When two experiments were conducted, the search was quicker and more stable than when targeting a single experiment. In the consideration (2), parameters were able to be estimated from the experimental results automatically and reproduce the experimental results well. Since the coefficient of variation of the parameters obtained through 100 times estimation was at most 1%, this method was able to estimate almost the same parameters each time. Narrowing the solution search range of the physical properties reduced the variation in the parameters obtained. Additionally, the parameters can be estimated by at least 2 mechanical experiments.