Application of Federated Machine Learning for Cross-Platform Knowledge Sharing in Distributed Additive Manufacturing Environments

Abstract

Additive manufacturing (AM) processes have been advancing rapidly across multiple industrial sectors, generating vast amounts of process data that remain largely siloed within individual organizations. This paper introduces a novel federated machine learning (FML) architecture optimized for cross-platform knowledge sharing in distributed AM environments while preserving data privacy and sovereignty. Our approach enables collaborative learning across geographically dispersed production nodes without requiring the centralization of sensitive proprietary data. The proposed framework implements a hierarchical attention-based neural architecture with differential privacy guarantees that maintain \(\varepsilon\)-differential privacy at a threshold of \(\varepsilon = 1.35\) while achieving convergence rates 27\% faster than traditional federated averaging methods. Performance evaluation across five distinct AM platforms demonstrates significant improvements in part quality prediction (14.2\% reduction in mean absolute error), anomaly detection sensitivity (19.8\% increase), and build failure prevention (22.5\% decrease in false negatives). Furthermore, the system's communication overhead scales sublinearly with the number of participating nodes, requiring only 8.7\% additional bandwidth when doubling the participant count. This research establishes a robust foundation for inter-organizational knowledge sharing in AM contexts, potentially accelerating process optimization, material development, and quality assurance across the distributed manufacturing ecosystem while maintaining competitive boundaries between participating entities.