Machine Learning (ML)-Based Prediction and Compensation of Springback for Tube Bending

Abstract

Bent tubes are extensively used in the manufacturing industry to meet demands for lightweight and high performance. As one of the most significant behaviors affecting the dimensional accuracy in tube bending, springback causes problems in tube assembly and service, making the manufacturing process complex, time-consuming, and difficult to control. This paper attempts to present an accurate, efficient, and flexible strategy to control springback based on Machine Learning (ML) modeling. An enhanced PSO-BP network-based ML model is established, providing a strong ability to account for the influences of material, geometry, and process parameters on springback. For supervised learning, training sample data can be collected from the historical production process or, alternatively, finite element simulation and laboratory-type experiments. Using the cold bending of aluminum tubes as the application case, the ML model is evaluated with high reliability and efficiency in springback prediction and compensation strategy of springback.