New Method of Experimentation and Material Modeling Using Neural Networks
J. Ghaboussi,* D. E. Sidarta
National Science Foundation, CMS 95-03462

Development of material models or determination of the parameters of a material model are usually based on material tests, which represent a material point. Material tests are usually designed such that the state of the stresses and strains within the sample are uniform. In this research, we are developing and testing new material experiments which induce a nonuniform state of stress within the sample. Using neural networks based on modeling techniques, we are extracting the information on the constitutive behavior of the material from the measured results of these tests. In this way, from a single test, we can extract far more information than is possible from material tests.