Federated Learning Framework for Privacy-Preserving Smart Infrastructure Monitoring

Abstract

Smart cities, intelligent transportation systems, and industrial infrastructures increasingly rely on IoT, edge computing, and AI to enable real-time monitoring and predictive maintenance. However, centralized machine learning raises concerns regarding data privacy, communication overhead, security, and data ownership. This paper proposes a Federated Learning Framework for Privacy-Preserving Smart Infrastructure Monitoring (FL-PSIM), which enables decentralized model training without transferring raw infrastructure data. Edge devices collaboratively share encrypted model updates using secure aggregation, differential privacy, and adaptive encryption techniques to preserve confidentiality. The framework also incorporates edge-cloud collaboration to balance computational efficiency, model accuracy, and network resource utilization. Designed to support heterogeneous sensor environments across transportation, energy, industrial, and urban systems, FL-PSIM optimizes global learning while maintaining local data privacy. Experimental results demonstrate improved monitoring accuracy, anomaly detection, communication efficiency, scalability, and resilience against cyber threats compared with centralized AI approaches. The proposed framework provides a secure, privacy-preserving, and scalable foundation for next-generation smart infrastructure, supporting sustainable digital transformation, smart cities, and Industry 5.0 applications.