Research on the neural networks used for shaping tubes by the liquid extrusion process

Liquid extrusion, as a new kind of metal forming process for shaping tube and bar products directly from liquid metal, can reduce the intermediate steps and production costs and make the materials doubly strengthened. But it has not been widely used since the process parameters are now selected by experience, which can easily result in a high reject rate. In order to analyze the contributing factors of the process, the artificial neural network method was used in this paper. The network architecture was determined by adopting 125 sets of experimental data of the shaping tubes of AlCuSiMg alloy as samples and, by contrast, one or two hidden layers and the numbers of nodes and other network parameters. The knowledge base for the process parameters of liquid extrusion has been established. The values predicted by the knowledge base are very consistent with the practical ones. The result shows that the introduced method is feasible and effective.