Stretch bending process design by machine learning

Stretch bending is a popular forming technique aiming at forming the long product’s axis into a desired curvature with the cross-section remaining constant. The most efficient and convenient method to prevent the defects in stretch bending such as axial springback and cross-sectional distortion is the optimization of forming path. In this work, a function of stretch length and bending angle is utilized to represent the forming path, and two new variables, total stretch length and distribution ratio, are proposed to better describe the effect of forming path on forming quality. Based on the data obtained from finite element simulation, a machine learning model is established to rapidly predict forming quality of the rail with a hat-shaped section. The forming path is optimized by multi-objective optimization method based on NSGA-II algorithm. The results show that the stretch bending quality is mainly determined by the total stretch, while the stretch distribution in three forming steps (e.g., pre-stretch, bend stretch and post-stretch) is a minor factor in controlling stretch bending quality with a few exceptions. With total stretch’s increment, the forming angle firstly increases dramatically and tends to a steady state at last, while the cross-sectional distortion increases constantly. The optimal forming path obtained indicates that the stretch of optimal forming process is an equal-ratio combination of the post-stretch and bending stretch.