Publications / 2006 Proceedings of the 23rd ISARC, Tokyo, Japan
In this paper, a new methodology for identifying numerous elastic parameters of an orthotropic ground material from field measurements is presented. For a better understanding, a simpler problem, which, however, is in the same concept, is adopted for describing the strategy, which is to identify an orthotropic material from a single structural test. At the heart of the methodology is the self-learning algorithm that is to extract various stress-strain relationships from a single structural test and train a neural network with the relationships in finite element framework. The constitutive matrix resulting from the trained neural network based constitutive model (NNCM) is compared with the conventional constitutive matrix for an orthotropic material to determine the nine independent elastic constants. An example is given for better understanding of the methodology proposed.