Aplicações de machine learning na engenharia mecânica: um estudo de caso para diagnóstico da operabilidade de sistemas de abastecimento de água

• Main Author: Carvalho, Matheus Vinícius de
• Publication Date: 2021
• Format: Bachelor thesis
• Language: por
• Source: Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT))
• Download full: http://repositorio.utfpr.edu.br/jspui/handle/1/28098
• topic: Aprendizado do computador; Método de estudo de casos; Software gratuito; Bombas hidráulicas elétricas; Manutenção; Machine learning; Case method; Open source software; Pumping machinery, Electric; Maintenance; CNPQ::ENGENHARIAS::ENGENHARIA MECANICA

This paper aims to apply concepts and techniques of machine learning in the area of predictive maintenance in mechanical engineering. Several software has been developed to assist in extracting information from the large amount of data made available by Big Data. The choice of Orange Data Mining was made in order to take advantage of the characteristics contained in it, mainly by using visual programming. The paper uses real data from the Tanzanian government. Ensuring water for tanzania's population makes it important to make a prediction of pump operability at water stations. Among the techniques of accuracy assessment of the machine learning algorithm available, the one used was the classification accuracy because it has easy understanding and good results. Both the cleaning and classification of variables were done using tools present in Orange Data Mining. The evaluation models used in software, decision trees, random forests and gradient boosting, flexible models that work with different types of variables and support nonlinearity. After adjusting a history of 47520 training data and 11880 test data, an accuracy of 78.8% was obtained using random forests, 77.6% using the boosting gradient and 75.7% using the decision trees method. The final result shows that it is possible to forecast the operability of pumps in filling stations with accuracy close to 77% in a first analysis, being able to achieve better results with a data update and optimization of inconsistent variables.