Integrated FEM and neural network-based optimization of the drawing process for variable cross-section forgings

Abstract

This study built a quality prediction model for the drawing process by integrating finite element simulation and neural network technology. It meticulously examines the influence of pivotal process parameters, including temperature, friction, speed, anvil width ratio (b/h), width-height ratio (w/h), and reduction ratio (ξ), on the overall drawing process. To facilitate a precise characterization of the workpiece's shape post-drawing, four key parameters - fullerring coefficient (F), barrelling degree (B), elongation ratio (E), and spread ratio (S) are introduced.The findings reveal that while friction and speed exert minimal influence on the drawing process, factors such as temperature, b/h, w/h, and especially the ξ, hold significant sway. Notably, a single-step reduction of 40% emerges as an optimal choice, with caution advised when exceeding a 50% reduction due to the potential implications on the internal stress state of the workpiece. A neural network framework is employed to analyze collected training data, establishing a high-precision predictive model with a post-training correlation coefficient of 0.973. Model robustness is validated through rigorous testing across three independent datasets. Building on deformation mechanisms of single-step drawing processes, we propose a customized process scheme for variable cross-section forgings. Optimized parameters—temperature (900°C), pressing speed (5 mm/s), friction factor (0.3), ξ=40%, and w/h=1.2—achieve higher efficiency with critical metrics demonstrating exceptional performance: maximum forming force (F=2.84) and minimal barreling distortion (B=5.85%). A single-pass strategy is recommended for production simplification, contingent upon maintaining anvil width exceeding workpiece elongation. These findings provide a robust technical foundation for enhancing production quality and efficiency in variable cross-section forging.