FE-based NN for finite element analysis

An architecture of Finite Elements based Neural Networks (FE-based NN) and its algorithms for modeling dynamic problems is proposed in this paper. An energy-function-based methodology is applied to combining Finite Element Method (FEM) with an ANN topology. The FE-based NN consists mainly of two node-variable layers and one element-subnet layer. Unknown inputs of the network are updated during the iterations using an inversion technique with imposed boundary conditions. A modified Newmark method is adopted for solving the dynamic problems. Results of the simulation on dynamics show promising results.