Increasing the Production Accuracy of Profile Bending with Methods of Computational Intelligence

Important quality criteria for profile bending are an accurate profile contour and an accurate cross section. During the bending process, torsion of the profile, deformation of the cross section, and deviations at the profile contour can occur. If these undesired effects are too large, the bent profile is not usable. Critical causes of profile cross section deformation are thin wall thicknesses with hollow sections. The profile torsion is favored by asymmetrical profile cross sections. These effects can be minimized by a production-correct profile design, whereby a trade-off between a production-correct design and the boundary conditions exists. Furthermore, undesired variations in the profile material properties and the profile cross section lead to deviations in the profile contour. These deviations cannot be reduced by design but by usage of a closed-loop control during bending. In this article, a software system for three-roll bending is presented that minimizes undesirable effects during bending by structure optimization of the profile cross section and application of closed-loop control. The structure optimization is based on an evolutionary algorithm and the process control uses a neuro-fuzzy controller. The structure of the software system and results of experiments are presented and discussed.