BEM-NN computation of generalised Newtonian flows (Q1604021)

Using boundary element method, the non-Newtonian fluid flows such as generalized Newtonian flows (GNF) are studied in the paper. The constitutive equation for GNF is given by using general viscosity function including power-law model, Carreau-Yasuda model, Carreau model and Cross model. The governing equations are presented in boundary element-neural network (BEM-NN) formulation. In oder to avoid the direct computation of volume integrals for global approximation of field variables, the approach in the paper adopts the multiplayer perceptron networks and radial basis function networks. As numerical examples, circular Couette flow and plane Poiseuille flow are calculated for power-law, Carreau-Yasuda and Cross models. A new linear boundary element techniques is proposed where a general treatment of corner problem and exact calculations of singular integrals are used. The prensent approach is able to achieve convergence for power-law fluids at small law indices, which is difficult in general.