Machine learning in additive manufacturing of aluminum alloys: a review of fundamentals, challenges, and evolution of intelligent pathways

The complex process-structure-property relationships in aluminum alloy additive manufacturing (AM) necessitate advanced methodologies beyond conventional approaches. This review charts the systematic evolution of machine learning (ML) as a transformative tool within this specific field. It details a progression from foundational data-physics integrated modeling, which enhances the prediction of thermal-stress fields and microstructural outcomes for aluminum systems, to the implementation of intelligent control frameworks. Key technical pathways are examined, including reinforcement learning (RL) for autonomous process optimization, multimodal in-situ monitoring for comprehensive process awareness, graph neural networks (GNNs) and Transformers for capturing non-Euclidean and long-range dependencies, explainable AI for mechanistic interpretation, as well as physics-informed neural network (PINN) and ML-driven design of novel aluminum alloys. This review also examines the practical challenges and technical demands in industrial applications. To address the critical challenge of data scarcity, advanced paradigms such as meta-learning, few-shot learning, and cross-domain transfer learning are analyzed for their role in efficient knowledge utilization. These interconnected advancements are facilitating a paradigm shift in aluminum alloy AM, from experience-based methods toward a knowledge-driven, intelligent manufacturing future. The realization of this future is contingent upon the development of integrated multi-scale modeling platforms, open data ecosystems, and truly self-adaptive production systems.