SCNN Solver for Finite Element Analysis with Neural Network Elements

This paper presents a structure controllable multi-layer forward neural network (SCNN) to solve liner equations and eigenvalues/eigenvectors of matrix in element stiffness simulation (ESS) approach. The stiffness matrix in ESS approach contains elements of neural network, and is usually not positive and symmetry. Therefore the conjunctive gradient method, which we applied before, is not always applicable. By using the SCNN solver, precision of present results increases dramatically comparing to the previous and is almost the same as traditional FEM analysis. Additionally, eigenvalues and eigenvectors of structure can be solved by the SCNN in ESS approach. Efficiency of the SCNN solver is demonstrated in the example. We believe the SCNN is a significant solver for ESS approach.