Federated Learning and Privacy-Preserving AI for Smart Manufacturing and Industrial IoT: Frameworks, Architectures, and Industrial Applications

Abstract

The proliferation of intelligent devices, sensors, and networked systems in modern manufacturing has generated an unprecedented volume of operational, behavioral, and environmental data—data that, if harnessed effectively, could enable transformative improvements in process efficiency, product quality, and operational safety. However, the data privacy constraints imposed by competitive dynamics, regulatory requirements (including GDPR and sector-specific data protection mandates), and organizational boundaries increasingly prevent manufacturers from centralizing their data for AI model training. Federated learning (FL)—the paradigm of training machine learning models across distributed data sources without exchanging raw data—has emerged as the dominant solution to this tension between data utility and data privacy in smart manufacturing. This review provides a comprehensive and critical synthesis of federated learning and privacy-preserving AI for smart manufacturing and industrial IoT. We examine FL frameworks and architectures for industrial AI, including federated learning for human-robot collaboration, federated time-series forecasting in collaborative manufacturing, FL-based intrusion detection for industrial IoT cybersecurity, and the integration of FL with edge AI and digital twin platforms. We further demonstrate how advances in industrial sensing—precision 3D optical metrology, collaborative robotic inspection systems, and automated software testing frameworks—provide critical data streams and deployment contexts for federated industrial AI. A central contribution of this review is the articulation of a unified Federated-Edge-Physical (FEP) architecture that integrates federated learning, edge AI inference, and real-time physical process control for privacy-preserving, low-latency, and scalable smart manufacturing.