Electronic Nose for Predictive Maintenance

• Autor(a) principal: Giesteira, André Filipe Oliveira
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/11110/2549
• topic: Manutenção Preditiva; Sistema Embebido; Tiny Machine Learning; Identificação de Odores

Nowadays, industries are constantly evolving, striving to become as advanced as possible compared to their rivals in the same field of production. Therefore, in terms of maintenance, they cannot afford to wait for some equipment to fail and only after do the necessary maintenance, due to the fact that it takes longer and this causes production breaks[1]. By these reasons, predictive maintenance emerged, which aims, through various sensory elements inserted in industrial equipment, to monitor as well as predict when failures will occur and thus schedule in time the necessary intervention. Currently, there are several predictive maintenance methods already developed, such as fluid analysis, vibration detection, among others[2]. The proposed system aims to create a predictive maintenance method capable of detecting gases / odors through tinyML techniques and, in this way, based on odor classification, proceed to classify the existing state/problem. In order to achieve this purpose, it is necessary to the study which sensors best fit the proposed objective. In the context of a research carried out and consequent evaluation of the data collected, the sensors BME688 and MP901 were selected. Thus, the data from the sensors will be processed in an algorithm based on tinyML techniques and inserted into a microcontroller, this being the ESP32. Through the technology developed, it is possible to identify lubricant oil in different stages of its life through the odor, being able to monitor the oil CNC machines in the future. In this way, it is possible to expand, together with the most varied existing methods, the possible areas to monitor.