Identificação de Falhas em Sistemas UPS (Uninterruptible Power Supply) por meio de Deep Learning em Sistema Embarcado

Patrick R. P. Andrade; Gustavo Pessin

doi:10.20906/SBAI-SBSE-2023/3795

Patrick R. P. Andrade Programa de Pós-graduação em Instrumentação, Controle e Automação de Processos de Mineração, Universidade Federal de Ouro Preto (UFOP) e Instituto Tecnológico Vale (ITV), Ouro Preto, MG / CEMIG Distribuição S.A., Belo Horizonte, MG
Gustavo Pessin Instituto Tecnológico Vale, Ouro Preto, MG

DOI: https://doi.org/10.20906/SBAI-SBSE-2023/3795

Keywords: Failure Diagnoses in Industry, Electric Power System ( EPS ), Machine Learning, Deep Learning, Convolutionary Neural Networks (CNN), Tensor Flow Lite Micro, Sound Analysis, Spectogram

Abstract

The UPS system is extremely important for the Electric Power System (EPS), as it is responsible for ensuring monitoring and command in the case of lack of primary voltage. The three -phase rectifier is the most sensitive part of the UPS system and is the most susceptible to failures. Current rectifiers have an alarm system to indicate failures, but these alarms are most often coming late when the equipment has stopped working. This work proposes the development of an embedded solution using the Arduino Nano 33 BLE Sense and Deep Learning for Fault Identification in UPS systems through sound processing emitted by this equipment on a convolutionary neural network (CNN) using Google's Tensor Flow Lite Micro library. Results were obtained with 97.8% accuracy for identification of defective rectifiers.