Improved Hybrid DBN-HEVA Framework for Accurate Prediction and Multi-Objective Optimization of Die Sinking EDM Performance

Electrical discharge machining (EDM) has made it possible to cut conductive metals that are either uncut or difficult to cut using standard machining techniques. In this EDM process, an accurate and flexible prediction and optimization model is required to evaluate this process and increase machining efficiency and surface quality. This study develops a novel adaptation of an improved Attention-Guided Deep Belief Network integrated with an Adaptive Human Eye Vision Algorithm (AG-DBN-AHEVA) for accurate prediction of EDM performance characteristics. Using an experimental dataset of 81 runs for accurate performance prediction, this work also examines the impact of EDM input parameters on material removal rate (MRR), tool wear rate (TWR), and Surface roughness (SR). To ensure high data quality and better generalization capability, pre-processing techniques including outlier detection, normalization, correlation analysis, and SMOTE-based regression augmentation were applied. The results showed that the suggested model outperformed the existing models, achieving the lowest prediction errors (MAE [Formula: see text] 0.007, RMSE [Formula: see text] 0.010) and the greatest [Formula: see text] value of 0.994. Furthermore, to attain the most effective EDM performance, multi-objective African Vultures Optimization is used to optimize MRR while decreasing TWR and SR. The ablation experiment, which demonstrates that the proposed model achieves higher prediction accuracy when compared to its reduced equivalents, justifies the inclusion of each component in the final design. Overall, this study demonstrated that the intelligent methodology proposed is a dependable, comprehensible, and highly effective AI-based approach for modeling and optimizing intelligent EDM processes in a high-technology manufacturing scenario.