Sistema ciberfísico IoT para detecção de anomalias em ambientes internos utilizando autoencoder embarcado e alerta móvel

Abstract

This monograph presents the development of a system for anomaly detection based on IoT sensor data, seeking to automatically identify atypical behaviors, such as fire hazards, in residential or industrial environments. The system integrates sensors for continuous acquisition of temperature and flammable gas concentration, which are subsequently processed by an Autoencoder model responsible for learning the normal operating pattern. The implemented architecture integrates collection, preprocessing, inference, and communication of results in a compact and efficient manner, without the need for heavy computational infrastructure. The main results obtained include quantitative mean squared error (MSE) values ​​within the expected limits to distinguish normal and anomalous conditions, as well as good model performance in both simulated tests and real experiments. The system demonstrated consistent capacity for reconstruction and separation between normal patterns and significant deviations. In addition, the project implements a real-time alert mechanism, allowing the user to be notified immediately through connected devices. The central objective of this work is to present a lightweight, embedded, and IoT-oriented solution capable of performing real-time anomaly detection, with lower computational cost and greater portability compared to conventional approaches.