Machine Learning-Based Prediction and Multi-Objective Optimization of Machining Parameters in Electrical Discharge Machining (EDM)

Abstract

Electrical Discharge Machining (EDM) is a non-traditional machining process used for producing complex geometries in conductive and difficult-to-machine materials. Its process responses depend on interacting electrical and thermal parameters, making prediction and parameter selection suitable for data-driven modelling. This study presents a computational machine-learning workflow for predicting material removal rate (MRR), surface roughness (Ra), and tool wear rate (TWR) from pulse-on time, pulse-off time, and peak current. A reproducible synthetic dataset containing 135 operating conditions was generated within the ranges T_on = 3–7 µs, T_off = 2–6 µs, and peak current = 10–30 A. A linear regression baseline, random forest regressor, and multilayer perceptron artificial neural network (ANN) were trained using a fixed 75:25 train-test split and evaluated using R², mean absolute error (MAE), and root mean squared error (RMSE). Across the three responses, the ANN achieved a mean test-set R² of 0.7743, compared with 0.6210 for random forest and 0.4840 for the linear baseline. The ANN was subsequently used as a surrogate model for a weighted multi-response optimization that favoured high MRR and low Ra and TWR. The resulting model-derived operating point was T_on = 7 µs, T_off = 2 µs, and peak current = 10 A, with predicted MRR = 1.0314 mm³/min, Ra = 3.7049 µm, and TWR = 0.1001 mm³/min. Because all observations are synthetic, these values are computational proof-of-concept results and must not be interpreted as experimental measurements. The workflow is intended as a reproducible template that can be retrained and experimentally validated using laboratory EDM data.