-
Gabriel Vinicios M. Fernandes
Programa de Pós-Graduação em Instrumentação, Controle e Automação de Processos de Mineração, Universidade Federal de Ouro Preto e Instituto Tecnológico Vale e Vale AS, Ouro preto, MG, Brasil Vale S.A
-
Agnaldo Rocha Reis
Depto. de Engenharia de Controle e Automação da Escola de Minas da Universidade Federal de Ouro Preto, MG
-
Frederico Gadelha Guimarães
Machine Intelligence and Data Science (MINDS) Laboratory,Universidade Federal de Minas Gerais, Belo Horizonte, MG
Keywords:
Fuzzy Time Series, machine learning, mining, fault detection
Abstract
Brazil is one of the most important producers of iron ore in the world and with the potential to grow over the years. The mining sector demands the adoption of new technologies for the development of specialized systems that increase the efficiency in the production process. Mining must embrace innovation for automation of repetitive tasks, systems integration, continuous process improvement, disaster risk reduction and adaptation to the global context. In this sense, the article proposes the use of Fuzzy Time Series(FTS) to learn a representation of the data that is more effective for the classification stage, using data originating from Plant A of the Ferro S11D Project, from Vale S.A., located in Canaã dos Carajás, Pará, Northern region of Brazil. The result demonstrated a significant improvement in failure prediction from the addition of PWFTS (Probabilistic Weighted Fuzzy Time Series) techniques to the XGBoost (Extreme Gradient Boosting) algorithm. From the proposed methodology, there was an increase in accuracy from 79,3% to 98,9%.
