Intelligent optimization of cold radial forging process for 20CrMnTiH alloy based on GA-BP and performance analysis

Cold radial forging (CRF) is recognized as one of the most effective manufacturing processes for the production of hollow components. Nonetheless, even minor alterations in the alloy composition during the cold deformation processing phase can significantly influence the material's manufacturability. This study addresses the challenges associated with high optimization costs, complex data acquisition, and the prediction of forming quality when integrating multiple processes, such as heat treatment and material composition. We propose the development of an intelligent predictive model aimed at forecasting the forming quality of the 20CrMnTiH alloy during CRF, utilizing a back-propagation neural network optimized by genetic algorithm (GA-BP). The assessment of forming quality is based on parameters such as damage, residual stress, and equivalent strain. The study investigates the impact of alloy composition and spheroidal annealing (SA) process parameters on forming quality. The performance of the three GA-BP prediction models is superior in terms of the coefficient of determination (R2) and mean square error (MSE), when compared to a conventional BP neural network and four other machine learning techniques, including gradient-boosted decision trees, random forests, support vector regression, and logistic regression. A comprehensive comparative analysis of the evaluation metrics across all three prediction models, alongside multi-objective optimization, indicates that the Pareto solution set generated by NSGA-II exhibits optimal distribution uniformity. The optimized process parameters (0.17%C-0.24%Si-0.81%Mn-0.03%P-0.03%S-1.15%Cr-0.05%Ni-0.06%Ti, Te: 60 degrees C, ATi: 4.57 h) resulted in a reduction of residual stresses by 12%, equivalent strain by 15%, and component damage by 30%. The results demonstrate that the methodology proposed in this paper not only enhances the quality of CRF molding but also significantly improves the accuracy of the predictive model for forming quality.