Performance analysis of steel W18CR4V grinding using RSM, DNN-GA, KNN, LM, DT, SVM models, and optimization via desirability function and MOGWO.

Abstract

This study presents an innovative approach to optimizing the grinding process of W18CR4V steel, a high-performance material used in reamer manufacturing, using advanced machine learning models and multi-objective optimization techniques. The novel combination of Deep Neural Networks with Genetic Algorithm (DNN-GA), K-Nearest Neighbors (KNN), Levenberg-Marquardt (LM), Decision Trees (DT), and Support Vector Machines (SVM) was employed to predict key process outcomes, such as surface roughness (Ra), maximum roughness height (Rz), and production time. The results reveal significant improvements, with Ra values ranging from 0.231 μm to 1.250 μm (up to 81.5 % reduction) and Rz from 1.519 μm to 6.833 μm (up to 77.7 % reduction). The hybrid DNN-GA model achieved R² > 0.99, reducing prediction errors by 23-45 % compared to traditional models. Optimization via the Desirability Function achieved Ra values around 0.341 μm and Rz around 2.3 μm, with production times ranging from 1181 to 1426 s. The innovative Multi-Objective Grey Wolf Optimization (MOGWO) provided Pareto-optimal solutions, minimizing Ra to 0.3 μm, Rz to 1.5 μm, and production times between 2000 and 3000 s, offering better balance between surface quality and machining efficiency. This work highlights the unique integration of machine learning models with optimization techniques to significantly enhance grinding performance and manufacturing efficiency in high-precision industries.