Springback, the elastic recovery of sheet metal after unloading, is a critical bottleneck to dimensional accuracy in precision forming. This study develops an interpretable, data-driven framework for predicting springback in V-bending by integrating controlled experimentation, machine learning (ML), and explainable artificial intelligence (XAI). A comprehensive dataset of 192 trials was generated using AA3105 aluminum, AISI 1020 mild steel, and C11000 copper under varied bending forces, holding times, and heat-treatment conditions. Nineteen regression models across linear, ensemble, kernel-based, and neural-network architectures were evaluated. Polynomial Ridge Regression yielded the superior predictive performance ( R 2 ≈ 0.8432), demonstrating robust generalization and low error. To transcend the black-box nature of conventional ML, SHAP-based analysis was employed to quantify feature importance. Results identified bending force as the primary driver of springback, followed by holding time and material properties, consistent with physical forming theory. By capturing nonlinear thermo-mechanical interactions without the computational overhead of finite-element analysis, this framework enables rapid, trustworthy prediction. This integration of experimentally validated machine learning with explainable analytics facilitates precise die compensation and process optimization, ultimately streamlining data-driven decision-making in sheet-metal forming. • Accurate springback prediction using Polynomial Ridge Regression • SHAP-based feature analysis enhances model interpretability • Integrates material and process parameters for precision forming • Experimental dataset validates data-driven prediction approach • Supports Industry 4.0 adoption in sheet metal forming
Karim et al. (Wed,) studied this question.