Industrial data analysis faces unprecedented challenges in balancing the need for collaborative machine learning with stringent privacy requirements and regulatory compliance. Federated learning emerges as a transformative paradigm that enables distributed learning across industrial networks while preserving data sovereignty and confidentiality. This comprehensive review examines federated learning architectures specifically designed for industrial applications, addressing unique challenges including heterogeneous data distributions, communication constraints, and security requirements. We analyze various aggregation algorithms, privacy-preserving mechanisms, and deployment strategies across manufacturing, energy, and supply chain domains. Our analysis demonstrates that federated learning systems achieve comparable performance to centralized approaches while reducing data breach risks by up to 90% and enabling compliance with regulations such as GDPR and industry-specific standards. The paper identifies key research directions including robust aggregation methods, edge computing integration, and blockchain-based coordination mechanisms for next-generation industrial federated learning systems.