Prediction of Wear in Roll Forming Using Data-Based Analysis and Modeling

Knowing exactly when a tool needs to be replaced helps to avoid unscheduled downtime due to failure. In roll forming, as in many other forming processes, especially abrasive wear is critical for tool life. Since numerical modeling of wear phenomena is complex and time-consuming due to required microscopic accuracy, a practical data-based approach is investigated to predict wear states. In roll forming, driven tool rolls are responsible for both the incremental profile bending at each forming step and the overall profile transport induced by all forming steps. The driving diameter of a tool roll is defined as being located at the position with rolling friction, i.e., without relative sliding velocity between the tool and the workpiece. The effective direction of the tool torque corresponds to the transport direction of the frictional force between tool and workpiece. Sensor data shows, that the torque of a driven tool roll is sensitive to changes in the tool gap between the rolls, to changes in transport velocity and to changes in geometrical bending parameters. Therefore, it is considered to be an important inline measured process variable. In this approach, idealized wear states in terms of enlarged tool radii are experimentally investigated using four discrete levels. As a result, the average torque level show correlations to the wear states for all the four investigated forming steps individually and simultaneously allowing to determine data trends. The trends are explained by a geometric displacement of the driving diameter at larger tool radii. Furthermore, stationary and non-stationary forming conditions are identified from data, allowing the forming process to be analyzed over time. Using the presented approach, four discrete wear states are classified for one of the forming stages by an artificial neural network. Using these results, torque monitoring can be applied to predict wear and estimate the end of the tool life.