Experimental optimization and machine learning-based prediction of warm incremental sheet forming of AZ31 magnesium alloy

This study evaluates the warm incremental sheet forming (WISF) behavior of AZ31 magnesium alloy through variable wall angle experiments, with the aim of improving formability and reducing forming force. Four process parameters forming temperature, step-down, spindle speed, and tool diameter were investigated at three levels using a Taguchi L27 orthogonal array. Grey Relational Analysis (GRA) was employed to evaluate the experimental results and identify the optimal combination of parameters for multi-response performance, while Analysis of Variance (ANOVA) was used to determine the contribution of each process parameter. The findings indicate that a high forming temperature and low step-down have the greatest influence on increasing the achievable wall angle and reducing forming force. Random Forest-based machine learning models were employed as supplementary predictive tools for both responses. The Random Forest models showed promising preliminary predictive performance for maximum wall angle and moderate performance for forming force. Feature-importance analysis provided supplementary insights into parameter influence within the investigated experimental domain. However, given the limited dataset of 27 experimental runs, further validation using additional experimental data is required to establish the models’ generalizability. The optimum condition identified by Grey Relational Analysis was a forming temperature of 250 °C, a step-down of 0.2 mm, a spindle speed of 250 rpm, and a tool diameter of 8 mm, under which a maximum wall angle of 66.5° and a forming force of approximately 0.639 kN were obtained. These results demonstrate that the combined experimental, statistical, and data-driven framework can effectively identify favorable WISF conditions for AZ31 magnesium alloy within the investigated process window.