Constitutive modeling of geomaterials from non-uniform material tests

Abstract As the name implies, a non-uniform material test has a non uniform distribution of stresses and strains within the specimen. An example is a triaxial test with end friction. Unlike the conventional material tests, points in the specimen in a non-uniform material test are subjected to widely varying stress paths. In this paper a method to extract the material constitutive behavior from a non-uniform material test is presented. This method employs the use of a neural network material model in a finite element analysis. The measured boundary forces and displacements are applied in the finite element analysis to generate the input and output vectors for training the neural network material model. This method has been applied to a series of triaxial compression tests with end friction on sand.