GTN poroplastic damage model construction and forming limit prediction of magnesium alloy based on BP-GA neural network

Abstract

During plastic deformation processes, the ductile solids damage and fracture at the microscopic level originate from the evolution of micro -voids, including nucleation, growth and coalescence of voids. In this study, the fracture caused by damage of the AZ31 magnesium alloy sheet is analyzed using the Gurson-Tvergaard-Needleman (GTN) model. The damage parameters of the GTN model are determined by statistical void volume fractions (VVF) in three damage stages during uniaxial tensile experiments with scanning electron microscopy (SEM). The damage parameters are then optimized by the Back Propagation-Genetic Algorithms (BP-GA) neural network. According to the result, the main GTN damage parameters: the initial void volume, the critical volume fraction, the void volume fraction of the nucleation part, and the void volume fraction when the material finally fails are obtained, respectively. Comparing the simulation with the test, the results of the optimized parameters fit better. The void growth reflects the damage during the deformation process, and the GTN model can accurately predict the ductile damage. Furthermore, the experimental forming limit diagrams (FLDs) of the AZ31 sheet at 200℃ are accurately predicted using the parameters obtained by the GTN model. Good agreement has been observed between the experimental and predicted FLDs.