Curtain wall systems (CWS) are essential components of modern architecture, yet their metal sections exhibit high slenderness ratios, complex multi-curvature geometries, and shape specificity. These characteristics make such components prone to springback after machining, adversely affecting dimensional accuracy and construction quality. Reliable springback prediction is critical, but conventional finite element method (FEM)–based approaches are computationally inefficient when applied to large-scale and diverse components. To overcome this limitation, this study proposes a data-driven springback prediction framework for curved metal sections that integrates numerical simulation, data augmentation, and machine learning model. Experimental validation demonstrates that the proposed framework achieves robust prediction performance, with an average coefficient of determination (R2) of 0.875, a mean absolute error (MAE) of 0.047, a root mean square error (RMSE) of 0.075, and a Pearson correlation coefficient (PCC) of 0.936. Prediction stability across different datasets is further examined, and computational time comparisons confirm substantial efficiency gains over FEM. The proposed framework offers a viable pathway to improving manufacturing accuracy, accelerating quality control, and preserving the geometric fidelity of CWS metal sections.
Jiang et al. (Tue,) studied this question.