Artificial Artificial Neural Network–Genetic Algorithm Integrated Approach for Optimizing Residual Stress and Crystallite Size in Incremental Forming of Ti–6Al–4V Alloy

Abstract

This study develops an integrated Artificial Neural Network–Genetic Algorithm (ANN–GA) approach to optimize process parameters in incremental sheet forming (ISF) of Ti–6Al–4V alloy, aiming to minimize residual stress (RS) and maximize crystallite size (D) to improve product quality. Three parameters tool radius (R), incremental step depth (S), and feed rate (F) were arranged using a Taguchi L9 orthogonal array. An ANN model (3–5–2 architecture), trained with the Levenberg–Marquardt algorithm, predicted RS and D, while GA was employed to determine optimal parameter combinations for simultaneous multi-response optimization. Experimental results showed RS between −157.11 MPa and −86.99 MPa and D from 19.67 to 21.87 nm. The ANN–GA method achieved superior prediction accuracy. The ANN model achieved a training RMSE of 0.0301 MPa for RS and 0.1394 nm for D, whereas validation RMSE values were 1.842 MPa and 0.229 nm, respectively, confirming good generalization performance. The optimal settings (R = 8.725 mm, S = 0.2588 mm, F = 1 mm·min⁻¹) reduced the magnitude of residual stress by 9.18% and increased D by 5.27% compared with the best Taguchi results. This integrated framework enhances process reliability, enables precise control of surface integrity, and provides practical guidelines for manufacturing high-performance titanium components for aerospace and biomedical applications.