Advanced constitutive model calibration using genetic algorithms: some aspects

Abstract

The complex nature of soil behaviour can only be modelled by advanced constitutive models. This type of models has the disadvantage of requiring a great number of parameters (material model constants and internal variables initial values) that must be defined. The application of Genetic Algorithms, GAs, appears to be a useful technique to solve this type of problem. Previous applications of GAs to the determination of soil models’ parameters have shown some efficiency and flexibility. However, some questions are not completely clarified when the methodology is applied to advanced constitutive models, such as: the amount of information required to obtain a “unique” solution in terms of parameters, the evolution of the parameters that have no influence on the particular type of test being simulated and the definition of a criterion to evaluate the similarity between polygonal curves. This paper intends to clarify some of these topics.