[Data] Learning Composition-Sensitive Signatures in PBF-LB: A Lightweight, Modality-Aware, Explainable Graph-Attention Sensor Fusion Framework for In-Situ Monitoring of Graded 316L–CuCrZr Alloys

Real-time composition monitoring in Laser Powder Bed Fusion (LPBF) of multi-material structures remains a critical challenge due to the complex and transient nature of melt pool emissions across spatially varying material gradients. In this study, we propose a modality-aware learning framework that fuses acoustic emission (AE) and optical back-reflection signals for spatiotemporally resolved classification of local composition in graded 316L–Cu alloys. The framework leverages a hybrid architecture that seamlessly integrates learnable shapelet extraction with graph-based attention mechanisms. Shapelets serve as interpretable, data-driven descriptors of modality-specific temporal behavior, enabling the model to capture fine-grained waveform features with minimal computational overhead. These compact representations are subsequently structured into a temporal graph, where nodes correspond to dual-modality signal segments and edges encode their pairwise relationships across the process timeline. A Graph Attention Network (GAT) operates over this representation, using modality-specific attention heads to adaptively prioritize sensor streams based on their relevance to compositional variation. Applied to LPBF data spanning five copper concentrations (20%–100%), the framework achieves up to 92% classification accuracy using only ~4,000 trainableparameters—outperforming unimodal baselines while maintaining strong generalization across adjacent classes. Saliency and activation analyses further reveal a class-conditional shift in sensor relevance, with optical signals gaining dominance at higher Cuconcentrations, in line with copper’s increased reflectivity. This lightweight, interpretable, and physically grounded framework advances in-situ monitoring for additive manufacturing and offers a scalable foundation for sensor fusion in the fabrication of compositionally graded materials.