Neural Network Computing Time Analysis on Finite Element of Elastic Mechanics

Abstract

Neural network computation of the structure finite element analysis is a new parallel calculating method. Because the dynamics equation of the network corresponds to the integrated circuit, so we can obtain the solution of the finite element system equations in the circuit time constant. Many investigators have presented qualitative explanation to the method. But so far it is not completely proofed in theory. In view of the above question, on the basis of the dynamic circuit of the neural network of the finite element system equations, we derive the stabilization time of the neural network dynamic circuit and other influencing factors from matrix theory in this paper. Theoretical analysis and computer emulation show that the stabilization time of the circuit is influenced by the bias capacitance in the dynamic circuit, the minimum eigenvalue of the finite element stiffness matrix and the predefined threshold value of system steady state error.