A novel real data-driven springback prediction method for roll forming based on digital twin

Digital twins (DT) have emerged as a key technology for quality control in manufacturing, particularly in roll forming-a lightweight, energy-efficient metal sheet forming process. However, springback during roll forming poses significant challenges to product quality, as it is difficult to predict accurately using static parameters and simulations. This study develops a DT model for roll forming, proposing a method to collect real-time production data, including static, dynamic, and streaming data. Using cleaned data, predictive algorithms such as Multi-Layer Perceptron (MLP), Support Vector Regression (SVR), and Incremental Learning with Stochastic Gradient Descent (IL-SGD) are employed. IL-SGD outperforms others with a determination coefficient of 0.91 (the 0.86 of MLP and the 0.83 of SVR) and a mean absolute error of 0.1344, effectively addressing material property fluctuations and improving regression analysis. The research demonstrates the successful application of DT and predictive algorithms in roll forming, offering practical solutions for data classification and management in production. .