Machine Learning-Based Modeling and Multi-Criteria Optimization of Wire-EDM Parameters

Wire-EDM (Wire Electrical Discharge Machining) plays a crucial role in the precision manufacturing of complex geometries, particularly for materials that are difficult to machine using conventional methods. Intelligent modeling and optimization of process parameters are required to improve performance, especially to maximize Material Removal Rate (MRR) and minimize kerf width. This study presents a machine learning (ML)-driven approach for predictive modeling and multi-objective optimization of Wire-EDM input settings. Using 31 experimental data points from existing literature, the influence of four important parameters i.e. V g (Average gap-voltage), W f (wire feed-rate), T on (pulse-on time) and T off (pulse-off time) on MRR and kerf width was systematically investigated. Three advanced regression techniques i.e. RFR (random forest regression), SVR (support vector regression), and GBR (gradient boosting regression) are utilized to capture the complex nonlinear relationships between inputs and outputs. Among these, the SVR model demonstrated superior performance, achieving a high R 2 value of approximately 0.92, indicating strong predictive accuracy. To interpret the model and assess feature significance, SHAP analysis is conducted, identifying gap voltage as the most influential factor affecting both MRR and kerf width. The optimized parameters generated by the SVR model led to enhanced machining performance, aligning well with the experimental results. This research underscores the potential of tree-based ensemble learning techniques like gradient boosting and random forest regression in advancing Wire-EDM process optimization, offering a robust, intelligent framework compatible with the goals of smart manufacturing and Industry 4.0. The methodology sets a strong precedent for data-driven decision-making in precision machining.