Comparative study of statistical and soft computing approaches for forecasting material removal rate and temperature in the EDM process of Inconel 718

This study develops an integrated finite element method (FEM)–regression–artificial neural network (ANN) framework for forecasting the material removal rate (MRR) and peak temperature in the Electrical Discharge Machining (EDM) of Inconel 718, a nickel-based superalloy widely employed in aerospace and tooling industries. A transient thermal FEM model was implemented in ANSYS Workbench, incorporating Gaussian-distributed heat flux, temperature-dependent thermos-physical properties, and a dynamically expanding spark radius to realistically simulate single-spark energy dissipation. Sensitivity analyses were performed for heat transfer fraction (F = 0.2–0.4), spark radius partitioning, and flushing efficiency (ηf = 0.1–1.0) to ensure numerical stability and model reliability. The predicted thermal fields and crater geometries were benchmarked against published EDM studies, exhibiting < 10% deviation in MRR and peak temperature trends, thereby validating model fidelity. Two predictive approaches were developed using the combined FEM and experimental dataset (36 trials): (i) second-order regression models analyzed through Analysis of Variance (ANOVA), and (ii) feedforward backpropagation ANNs trained and tested with k-fold cross-validation to avoid overfitting. Regression models revealed dominant linear and interaction effects of discharge current, pulse-on time, and heat transfer fraction on both responses, while ANN models achieved superior predictive accuracy (R2 > 0.99) with narrow prediction intervals, demonstrating strong generalization capability. Beyond MRR and temperature, the proposed framework is designed for extension to additional EDM responses such as surface roughness, tool wear, and recast layer thickness, thereby offering a comprehensive predictive environment for process optimization. A comparative discussion of computational efficiency versus prediction accuracy establishes the practical utility of the FEM–ANN hybrid strategy for high-performance EDM applications. The presented methodology bridges the gap between physics-based simulation and data-driven learning, providing an experimentally consistent and computationally efficient tool for intelligent machining of Inconel 718 and other high-temperature alloys.