Implementation of ANN-based rock failure criteria in numerical simulations

摘要

In this paper, the application of Artificial Neural Networks (ANNs) as a basis for new generation of rock failure criteria is illustrated. As an example, a typical series of the results of triaxial compression tests on Indiana limestone were fitted using an ANN. In order to evaluate the relative accuracy of the trained ANN, two well-known conventional criteria of Mohr-Coulomb and Hoek-Brown were also used to fit the data. It was observed that the ANN-based criterion can give more accurate predictions of strength for both brittle and ductile failure modes. Subsequently, the explicit formulation of the ANN-based criterion and equations for instantaneous values of an equivalent Mohr-Coulomb criterion were derived. Finally, the formulas were incorporated in numerical simulations of triaxial compression tests and circular tunnels in anisotropic in-situ stress fields. The accurate results of these simulations showed that ANN-based failure criteria can be successfully implemented in numerical analyses.