Integration of Hybrid Machine Learning and Multi-Objective Optimization for Enhanced Turning Parameters of EN-GJL-250 Cast Iron

Abstract

This study aims to optimize the turning parameters for EN-GJL-250 grey cast iron using hybrid machine learning techniques integrated with multi-objective optimization algorithms. The experimental design focused on evaluating the impact of cutting tool type, testing three tools: uncoated and coated silicon nitride (Si3N4) ceramic inserts and coated cubic boron nitride (CBN). Key cutting parameters such as depth of cut (ap), feed rate (f), and cutting speed (Vc) were varied to examine their effects on surface roughness (Ra), cutting force (Fr), and power consumption (Pc). The results showed that the coated Si3N4 tool achieved the best surface finish, with minimal cutting force and power consumption, while the uncoated Si3N4 and CBN tools performed slightly worse. Advanced optimization models including improved grey wolf optimizer–deep neural networks (DNN-IGWOs), genetic algorithm–deep neural networks (DNN-GAs), and deep neural network–extended Kalman filters (DNN-EKF) were compared with traditional methods like Support Vector Machines (SVMs), Decision Trees (DTs), and Levenberg–Marquardt (LM). The DNN-EKF model demonstrated exceptional predictive accuracy with an R2 value of 0.99. The desirability function (DF) method identified the optimal machining parameters for the coated Si3N4 tool: ap = 0.25 mm, f = 0.08 mm/rev, and Vc = 437.76 m/min. At these settings, Fr ranged between 46.424 and 47.405 N, Ra remained around 0.520 µm, and Pc varied between 386.518 W and 392.412 W. The multi-objective grey wolf optimization (MOGWO) further refined these parameters to minimize Fr, Ra, and Pc. This study demonstrates the potential of integrating machine learning and optimization techniques to significantly enhance manufacturing efficiency.