HARDENING PLASTICITY APPROXIMATED VIA ANISOTROPIC ELASTICITY - THE FOKKER-PLANCK EQUATION IN A NEURAL-NETWORK ENVIRONMENT

In the present paper a method is developed which replaces an anisotrop ic hardening plasticity problem through on appropriate sequence of ani sotropic elasticity problems. We have to solve a parameter identificat ion problem the solution of which is approximated in an appropriate ne ural network environment via supervised and unsupervised learning algo rithms. Numerical examples concerning the prediction and/or the correc tion of interpolated or extrapolated yield surfaces illustrate the the ory. The elliptic paraboloid failure surface, proposed by the first au thor [1], serves as filter in the whole procedure. The proposed algori thms give rise to a Fokker-Planck equation characterizing the elastopl astic behavior. Necessary and sufficient conditions for the convergenc e of the proposed algorithms are given.