Data-Driven Modelling and Predictive Control for the Geometry and Mechanical Properties of Freeform Bending Processes

Freeform bending is a technique used to bend different tube profiles into complex structures. In the realm of Aritificial Intelligence and Machine Learning, data-driven approaches in modeling and in optimization based control contribute to the job being executed, by increasing product quality and reducing energy consumption and material waste. In previous works, soft-sensors as well as different factors affecting the geometry and the residual stresses have been investigated and been utilized in a preliminary closed-loop control structure. Later on, this control structure has been replaced with a control strategy called Residual Strategy Algorithm , that could, based on previous deviations in the geometry, proactively react in order to reduce their effects. In this paper, the previously developed Residual Strategy Algorithm is extended to include both, the geometry as well as the residual stresses of the tube being bent. Also the modelling of both will be discussed altogether. The validation of the algorithm is done using simulation results.