Selective laser melting (SLM) has been widely adopted in advanced manufacturing, yet accurate three-dimensional generation of anisotropic grain structures remains challenging due to complex thermal histories and solidification dynamics during fabrication. Existing grain generation methods suffer from limited accuracy, inadequate anisotropic feature preservation, and poor adaptability to SLM-specific microstructural characteristics. To address these limitations, this study presents SSEGAN (Statistical Structure Enhanced Generative Adversarial Network), an enhanced generative adversarial network framework specifically designed for 3D grain structure generation in SLM-processed materials. The framework generates statistically-equivalent 3D grain structures from two orthogonal cross-sectional images. Comprehensive evaluation on SLM-processed 316L stainless steel demonstrates that SSEGAN achieves 54.6% and 20.7% reductions in Fréchet Inception Distance compared to 2D to 3D GAN and CEM3DMG, respectively, while achieving TPCF errors of 0.18% and 0.25% in the XY and XZ planes, respectively. Cross-material evaluation on AlSi10Mg alloy demonstrates the applicability of the same SSEGAN architecture and training strategy to distinct SLM microstructures, providing an effective framework for microstructure generation in additive manufacturing applications.
