Finite Element Analysis System Using the Neural Network Constitutive Properties

摘要

The inelastic finite element analysis is indispensable for reliable estimation of the material behaviors, which are under such severe environments as the cyclic loading and the elevated temperature. The inelastic material properties have been expressed in terms of constitutive equations. However, it is not easy to determine parameters included in these equations from the experimental data. In our previous studies [1][2], we have found that the neural networks have a comparable ability to estimate the stress-strain curves under uniaxial loading and stationary temperature conditions to the conventional inelastic constitutive equations. When we actually make use of the neural network to model the constitutive properties in the structural analysis, it is necessary to implement the neural network modeler for a finite element method (FEM) code. In this paper, we incorporate the neural networks for modeling the inelastic material behaviors under uniaxial loading and stationary temperature conditions into an FEM code instead of the inelastic constitutive equations.