Deformation error compensation of industrial robots in single point incremental forming by means of data-driven stiffness model

Abstract

Hole flanging by single point incremental forming offers great potential to reduce equipment costs in the forming of small batches and flanges. This potential is further increased by using industrial robots as manipulators. However, industrial robots are characterized by a low stiffness, which results - due to the acting process forces - in a low dimensional accuracy of the formed flanges. In this paper, the pose dependent stiffness of an industrial robot is modeled by data-driven regression models based on a gaussian process and artificial neural network. Both models are capable to predict the deformation error accurately. The gaussian process model shows a slightly higher generalizability compared to the artificial neural network and is hence used to predict the deformation error in single point incremental forming. An offline compensation method is proposed to generate an adjusted toolpath. Validation experiments with a heavy-duty industrial robot show the effectiveness of the proposed model. The dimensional error of the formed flange diameters could be reduced by about 50 %, resulting in more accurate and circular flanges.