Machine learning-assisted prediction and Bayesian optimisation of wire EDM process parameters for surface roughness of AISI S7 tool steel

Abstract

AISI S7 tool steel finds application in aerospace, automotive, and tooling industries for its high toughness and impact resistance. Machining this material with acceptable quality remains challenging. This study investigates the influence of Wire Electrical Discharge Machining (WEDM) process parameters, including pulse–on time (T on ), pulse–off time (T off ), discharge voltage (V), and wire tension (WT), on the surface roughness of AISI S7 tool steel using a machine learning–assisted predictive framework. Experimental WEDM data were used to develop and evaluate regression models, namely Linear Regression, Lasso, Elastic Net, Polynomial Regression, Random Forest, XGBoost, LightGBM, Support Vector Regression, and Voting Regressor. Model performance was assessed using the determination coefficient ( R ²), root mean square error, and absolute error. The results revealed that pulse–on time was the most influential parameter affecting surface roughness, followed by pulse–off time and discharge voltage, whereas wire tension exhibited a comparatively lower effect. Among the evaluated models, LightGBM achieved the highest prediction accuracy, with a mean R ² of 0.968 ± 0.011, RMSE of 0.100, and MAE of 0.081 under 5–fold cross–validation. Bayesian optimization was employed to identify optimal conditions, and experimental validation confirmed close agreement between predicted and measured surface roughness values. Scanning electron microscopy analysis revealed WEDM surface features, including craters and re–solidified debris, supporting the machining trends. The study demonstrates that integrating machine learning and Bayesian optimization provides an effective approach for predicting and optimizing surface roughness in WEDM of AISI S7 tool steel. It is recommended that pulse–on time be carefully controlled to achieve improved surface quality and machining performance.