Hybrid Ensemble Learning with Uncertainty Quantification for Multi-objective EDM Parameter Optimization

Abstract

Electrical Discharge Machining (EDM) remains critical for precision machining of hard-to-cut materials, yet parameter optimization continues to challenge manufacturers due to complex non- linear interactions and conflicting objectives. This research presents a novel Hybrid Ensemble Learning (HEL) framework integrating adaptive feature selection, dynamic weight optimization, and uncertainty quantification for multi-objective EDM optimization. The framework was validated on a comprehensive dataset of 150 experimental runs with five input parameters (pulse-on time: 10-200 µs, pulse-off time: 10-100 µs, peak current: 1-30 A, gap voltage: 20-120 V, electrode diameter: 0.5-10 mm) and four response variables. The HEL framework demonstrated exceptional predictive accuracy with R² values of 0.961, 0.966, 0.890, and 0.753 for Material Removal Rate (MRR), Tool Wear Rate (TWR), Surface Roughness (Ra), and Overcut (OC) respectively. Multi- objective optimization using modified NSGA-II identified optimal parameters achieving MRR of 51.56 mm³/min, TWR of 10.66 mm³/min, Ra of 5.84 µm, with estimated OC below 18 µm, representing 23.5% improvement in productivity and 18.2% reduction in tool wear compared to conventional approaches.