Material-Aware Cross-Alloy Transfer Learning for Universal Relative Density Prediction in Laser Powder Bed Fusion of 316L Stainless Steel, AlSi10Mg, and Inconel 718

Achieving near-full density remains one of the primary objectives in laser powder bed fusion (LPBF), as relative density strongly influences the mechanical performance, fatigue resistance, and structural reliability of additively manufactured components. Recent advances in machine learning have enabled accurate density prediction; however, most existing models are developed for a single-alloy system and require extensive experimental datasets, limiting their applicability to new materials. Consequently, developing data-efficient predictive frameworks that can generalize across multiple alloy classes remains a significant challenge. In this study, a novel material-aware cross-alloy transfer learning framework is proposed for universal relative density prediction in LPBF-manufactured metals. The framework integrates experimental datasets from three fundamentally different alloy families—316L stainless steel, AlSi10Mg aluminum alloy, and Inconel 718 nickel-based superalloy—within a unified machine learning architecture. Unlike conventional alloy-specific models, the proposed approach incorporates both process parameters and intrinsic thermophysical material descriptors, including thermal conductivity, melting temperature, density, and specific heat capacity, enabling the model to learn transferable process–material relationships across distinct alloy systems. The key novelty of this work lies in transferring knowledge acquired from data-rich alloys to data-scarce alloys through a material-aware learning strategy. Rather than treating each alloy as an independent problem, the framework identifies common densification mechanisms shared across different materials and leverages these relationships to improve prediction accuracy while reducing data requirements. Explainable artificial intelligence techniques are further employed to reveal the influence of process and material parameters on densification behavior, providing valuable insight into LPBF process optimization. The proposed methodology establishes a foundation for universal predictive models in metal additive manufacturing and demonstrates how transfer learning can accelerate process qualification for emerging alloys with limited experimental data. By combining cross-alloy knowledge transfer, material-aware feature engineering, and explainable AI, this work offers a scalable pathway toward intelligent, data-efficient manufacturing and next-generation digital process design.