SMART SENSORS FOR ANOMALY DETECTION DRIVEN BY TINYML IN IOT ARCHITECTURE
Eduardo Pereira, Leonardo S. Marcondes, Josemar M. Silva · 2023
In general, the architecture of Internet of Things (IoT) systems is organized in layers, starting with the perception layer, or physical computing, which involves the sensing process, followed by communication, processing, or middleware layer, application, and business layers. In this paper, we propose a new middleware layer positioned just above the perception layer, by integrating more powerful microcontrollers capable of running artificial intelligence algorithms, known as TinyML. We have developed an algorithm based on extreme value theory for real-time anomaly detection. The results indicate the feasibility of implementing the proposed architecture for real-time monitoring and automated decision-making in industrial systems.