An Adaptive Federated Learning Model for Fault Detection and Node Reuse Using Deep Learning Mechanism in Wireless Sensor Networks

The growing complexity of Wireless Sensor Networks (WSNs) within Internet of Things (IoT) enabled applications presents significant challenges related to fault detection, energy efficiency, and network resilience. Node failures can cause severe disruptions, leading to increased energy consumption, data loss, and reduced network lifespan. Conventional fault detection methods lack flexibility when dealing with node relocations, while machine learning-based techniques often overlook energy efficiency considerations. This research proposes the Adaptive Federated Learning for Fault Detection and Node Reuse (AFL-FDNR) framework, which combines federated learning with Deep Q-Network (DQN) to provide an adaptive, energy-efficient solution for fault detection and node recovery. Federated learning ensures scalable, privacy-preserving model training across sensor nodes, while DQN optimizes node recovery decisions based on real-time network conditions. Simulation results demonstrate that AFL-FDNR achieves 96.01% fault detection accuracy, reduces energy consumption to 1910 J, and extends network lifetime to 11,452 hours. These metrics demonstrate the framework’s ability to enhance the reliability, efficiency, and sustainability of large-scale WSNs, making it a robust solution for modern IoT applications.