Autonomous Decision-Making in Additive Manufacturing via Integration of Machine Learning with Digital Twin Architectures

Abstract

Additive manufacturing has emerged as a revolutionary paradigm in modern industrial production systems, enabling unprecedented geometric complexity and functional integration. This research investigates the symbiotic integration of machine learning algorithms with digital twin architectures to facilitate autonomous decision-making in metal-based additive manufacturing processes. We propose a novel framework that synthesizes real-time sensor data acquisition, multi-physics simulation, and reinforcement learning to optimize process parameters dynamically during fabrication. The methodology employs tensor-based representation learning coupled with graph neural networks to capture the complex spatial-temporal correlations inherent in melt pool dynamics. Empirical validation on Ti-6Al-4V and Inconel 718 specimens demonstrates that our approach reduces geometric deviations by 37.4\% and porosity defects by 42.8\% compared to conventional feedback control systems. Furthermore, the computational overhead of the proposed system adds only 1.3\% to total fabrication time while enhancing mechanical properties by approximately 18.5\% across multiple metrics. This research establishes a foundational architecture for self-regulating additive manufacturing systems capable of autonomous adaptation to material and process variability without human intervention.