Abstract
The qualification of structural components for high-performance applications requires integrated design strategies that couple fundamental alloy chemistry with manufacturability and performance under complex thermal cycles. This work presents MAPLE-Net (Material-Agnostic Porosity Learning Engine), a physics-informed machine learning framework for the systematic prediction of porosity across diverse alloy systems in Laser Powder Bed Fusion (LPBF). MAPLE-Net combines advanced graph-based learning architectures with physically meaningful elemental descriptors to encode candidate alloy-process combinations while explicitly enforcing machine-specific process constraints. The framework enables a material-agnostic approach to qualification, reducing the reliance on alloy-specific trial-and-error campaigns and extensive post-build characterization. Building on a foundational dataset of over 3,000 builds across 50 distinct metal alloys, MAPLE-Net represents a machine-learning-enhanced computational platform where models trained on composition-process relationships augment descriptor-based screening to predict relative density across unseen chemistries. The architecture is uniquely positioned to extend predictions to High-Entropy Alloys (HEAs), Complex Concentrated Alloys (CCAs), and Multi-Principal Element Alloys (MPEAs), with modular potential for transfer learning to predict mechanical properties such as hardness. Selected alloy candidates are validated through a tight modeling-experiment loop involving advanced density characterization techniques to calibrate performance in challenging low-density regimes. The results demonstrate the effectiveness of a chemistry-first, data-augmented approach for accelerating the discovery and qualification of advanced structural materials for aerospace, nuclear, and energy sectors.