An Artificial Intelligence Application for In-Process Springback Control of Sheet Metal Bending

Abstract Bending is the fastest and most efficient process commonly used in the industry for processing thin metal sheets into three-dimensional shapes by localized deformation using only a single geometrical die. However, suppliers provide metal sheets with variations in dimension and mechanical properties, which causes inconsistencies in the final angle after bending. This requires manual checking and correction of each angle, resulting in inefficiency. The problem can be resolved by considering the variations in the sheets and adjusting the bending stroke accordingly. This study used neural network technology to create a model that predicts the final stroke required based on load measurements during the bending process. The model was implemented and validated using a laboratory press. With a root-mean-square error of less than 0.27 deg, the model demonstrates its feasibility for practical industrial applications within the range of its training data.