An Edge-Enabled Distributed Fault Detection and Diagnosis Framework for Smart City IoT Sensor Networks Using a Hybrid KNN-Based Machine Learning Approach

Smart cities are becoming more and more dependent on extensive networks of IoT sensors to track the state of their infrastructure, environmental conditions and even public services in real time. But sensor networks face the possibility of hardware failures, communication failures, and abnormal data generation, all of which can have a negative impact on system reliability, and on the decisions made during operation. To tackle these challenges, this study aims at proposing an edge-enabled distributed self-fault detection and diagnosis framework for smart city sensor networks based on a hybrid K-Nearest Neighbours (KNN) based machine learning approach. Each sensor node autonomously derives statistical features from local measurements and verifies detected anomalies by information passed with neighbouring sensor nodes. Distance-based KNN analysis, adaptive thresholding and neighbour-assisted validation aids in reliable fault classification in the face of varying network conditions. The target of the proposed distributed architecture is to minimize the communication overhead and energy consumption, and make decentralized decisions about fault-related matters. Simulations with permanent failures, intermittent failures, and abnormal observations of data were performed to assess the framework. The proposed hybrid KNN approach performed fault detection with 94.2% accuracy, 3.9% false alarm rate, 0.94 J average energy consumption and 1785 rounds network lifetime which outperformed the centralized detection and conventional KNN approaches as reported in the comparison. These findings illustrate the potentialities of this framework to enable reliable sensor-network monitoring, energy-efficient operation and smart decision making with respect to fault.