Integration of laboratory testing and constitutive modeling of soils

Abstract

A soil constitutive model that correctly captures soil behavior under general loading modes is requisite to solving complex boundary value geotechnical engineering problems. Available laboratory tests provide information on material behavior within a very limited range of stress–strain paths and do not cover the full range of loading paths experienced in a boundary value problem. Soil behavior information developed from most laboratory tests are often limited and insufficient to validate constitutive model performance under general loading conditions. This paper explores a new approach for linking soil laboratory testing and constitutive model development through the use of a novel computational framework: self-learning simulations (SelfSim). SelfSim is an inverse analysis approach that extracts underlying material constitutive behavior from boundary measurements of load and displacement. SelfSim is applied to two simulated laboratory tests, a triaxial compression shear test with no-slip frictional ends (loading base and cap), and a triaxial torsional shear test with no-slip frictional ends. The frictional ends result in non-uniform states of stress and strain throughout the tested specimen. SelfSim successfully extracts the diverse stress–strain response experienced throughout the specimens. A neural network-based constitutive model is developed using extracted soil behavior from both laboratory tests and used successfully in the forward prediction of the load-settlement behavior of a simulated strip footing. The results demonstrate that SelfSim establishes a direct link between laboratory testing and soil constitutive modeling to extract soil behavior under complex loading modes and to readily develop corresponding constitutive models.