Machine learning methods for springback control in roll forming

Abstract

Springback is a critical factor that significantly influences the quality of roll forming. Accurate prediction and control of springback are crucial for the design of process parameters. This paper proposes a technique based on Support Vector Regression (SVR) and Bat Algorithm (BA) to reduce springback. Firstly, based on roll forming experiments, the SVR model optimized by algorithm based on the Simulated Annealing Particle Swarm Optimization algorithm (SAPSO) is used to predict springback and investigate the influence of forming parameters. The considered forming parameters include the mechanical properties of material (e.g. yield strength, Young’s modulus), geometries of metal sheet (e.g. sheet width), and process parameters, such as uphill value, roll gap. Then, using the Bat Algorithm based on Lévy flight disturbance, the process parameters are optimized with the predicted springback as the fitness function. The experimental results show that the springback in roll forming has been reduced by 94.47% after optimizing the process parameters. Therefore, the feasibility of the proposed springback control method is confirmed.