Implementation of edge AI for early fault detection in IoT networks: evaluation of performance and scalability in complex applications

The exponential growth of Internet of Things (IoT) deployments has introduced critical demands in reliability, energy efficiency, and real-time fault detection. Traditional cloud-based solutions suffer from excessive latency and energy overhead due to continuous data transmission and centralized analysis. To address these limitations, this study introduces an edge-based artificial intelligence (AI) architecture tailored for early fault detection in heterogeneous IoT networks. The architecture leverages recurrent neural networks and autoencoders optimized for time-series anomaly detection, enabling local inference directly on edge nodes. The system was evaluated under realistic laboratory conditions using a range of IoT devices and edge computing platforms, including Raspberry Pi and Nvidia Jetson. Experimental results demonstrate a 92.0% fault detection rate with a response time consistently under 150 ms, significantly outperforming cloud-based approaches in both latency and energy metrics. Energy consumption was reduced to 50 Wh under standard conditions, and the architecture successfully scaled to support up to 500 IoT devices, maintaining stable detection accuracy above 88%. These results validate the proposed edge AI system as a scalable and energy-efficient alternative for real-time fault monitoring. Its low-latency, decentralized nature makes it suitable for deployment in industrial automation, smart city infrastructure, and mission-critical IoT applications where operational continuity and autonomous decision-making are essential.