In this study, an artificial neural network (ANN) model was constructed to investigate the relationship between the roller path parameters to form a cylindrical cup in multi-pass conventional spinning and the thickness distribution throughout the height of a workpiece. Furthermore, the path parameters that simultaneously realize multiple target values of the workpiece dimensions were calculated instantly by the iterative solution based on the constructed model. A systematic design of the path parameters for a constant thickness distribution was established as follows. First, the roller path was expressed using 12 parameters. Second, the workpieces were spun under various experimental conditions, which were determined by partial randomization of the orthogonal array based on the Taguchi method. Third, an ANN model was trained by considering seven path parameters as inputs and five forming result values as outputs (cup height, wall thickness at 25%, 50%, and 75% of the cup height, and residual path length). Finally, the path parameters required for realizing a constant thickness were determined using an ANN model with an iterative solution. Although several samples of the training dataset exhibited non-uniform thickness distributions, the workpieces that were spun under the parameters obtained via iteration exhibited a constant thickness distribution. The parameters responsible for stretching the material in the radial direction significantly affected the thickness distribution. The most influential parameter was the increment in the axial start position for each curved pass.
Graphical abstract