In industry, especially aircraft fields, the hydroforming process is widely used for sheet metal forming, because of its advantage, such as uniform thickness forming. Since the hydroforming operation is performed in the high-pressure chamber controlled by the pressure control valve, the determination of back-up fluid pressure in the chamber is one of the most essential tasks.
Recently, to determine the back-up fluid pressure, many investigators have studied the relationships between the forming pressure and material characteristics of the product and geometrical parameters in the forming process by analytic approach of the plastic deformation. However, it's very difficult to obtain such relationships by an analytical method. In this thesis, we present an estimation system which estimates the back-up pressure of hydroforming process utilizing a multi-layered neural network. The neural network learns the nonlinear relationship between the back-up pressure and the geometric state variables of hydroforming process. The proposed method does not necessitate sophisticated analysis on hydroforming process but only needs qualitative understanding of physical phenomena associated with the process. In addition, for the estimation of the forming pressure, we can minimize the error by simplify the modelling of the hydroforming process.
A series of experiments are performed for a number of different forming conditions by using various punch shapes, sizes, and drawing ratios. To minimize the number of the experimental data set needed for neural network training, we utilize Taguchi experiment method. From the experiments suggested by the Taguchi method we collected sets of data which yield uniform thickness product and utilized as a set of data for the multi-layered neural network.
The experimental results show that the nueral network well approximates the nonlinear relationship between the back-up pressure and the geometric state variables of hydroforming process, thus giving the good estimation of back-up pressure vs punch stroke curve.