A Predictive Crater-Overlap Model for EDM Finishing Relevant to AISI 304 Welded Joints

Electrical Discharge Machining (EDM) enables precision post-weld finishing of AISI 304 stainless steel, but stochastic spark overlaps make the fatigue-critical maximum peak-to-valley height (Rmax) difficult to predict. This study develops a validated physics-based framework quantifying how crater overlap governs Rmax evolution. Experiments on unwelded AISI 304 cylinders—proxying weld metal while excluding heat-affected zone (HAZ) effects—used Central Composite Design (20 trials, 900–9380 μJ discharge energies). Profilometry and scanning electron microscopy (SEM) correlated the crater size, overlap intensity, micro-cracking, and Rmax escalation from 18 to 85 μm. Primary and secondary crater formation under minimum and maximum overlap configurations were simulated using a 2D axisymmetric finite element model with Gaussian heat flux and temperature-dependent thermophysical properties. The predictive metric Rmax,num = (dinitial + dsecondary)/2 achieved 11–19% average error against the experimental Rmax,exp, with complementary valley depth (Rv) validation at 13% error. The Specimen 7 outlier (~50% error) reveals the limitations of deterministic modelling under stochastic debris accumulation and plasma instability at intermediate energies. Crater overlap generates secondary dimples, sharp inter-crater peaks, and rim micro-crack networks, driving the 4.7-fold Rmax increase—approaching International Institute of Welding (IIW) fatigue thresholds (<25 μm for high-cycle categories). The framework explicitly links the discharge energy, plasma channel radius (Rpc), and overlap geometry to surface topography, enabling process optimization (I·ton < 60 A·s maintains Rmax < 25 μm). Mesh independence (<2.5% convergence) and six centre-point replicates (CV = 4.2%) confirm robustness. This validated upper-bound Rmax predictor supports the digital co-optimization of welding and EDM parameters for aerospace/energy applications, with planned extensions to stochastic 3D models incorporating adaptive remeshing and real weld topographies.