Estudo da aplicação de técnicas de análise de dados e Machine Learning em uma indústria de arroz parboilizado : manutenção preditiva e estudo de falhas em uma caldeira

Abstract

The rise of Industry 4.0 and its technologies along with Big Data bring with them many opportunities for industry. Allied to this, Data Analysis and Machine Learning use these large volumes of data generated to assist in decision making. Despite the ascendancy of Industry 4.0, the Brazilian reality shows that most industries are not ready for this change and are mostly characterized as Industry 2.0. In this sense, a rice processing industry presented itself as promising for study purposes due to its several areas and data generation, therefore, the objective of this paper is to study the application of Data Analysis and Machine Learning techniques through Python programming language in a rice processing industry in Campanha Region that has characteristics of an Industry 3.0. The area selected as the object of study for the application of the mentioned techniques is a mixed boiler that presents data collection by the company itself every hour through annotations on printed spreadsheets and a high recurrence of failures. From this, the data is digitized through Microsoft Excel and studied using the Jupyter Notebook platform. The Data Analysis and Machine Learning techniques are intended to perform a study of boiler failures and then a classification of these failures, in addition to this, to perform a predictive maintenance study by predicting the remaining life of the boiler. For this, Linear Regression, XGBoost Regressor and Random Forests Classifier and Regressor are used. The results obtained in this study show the validity of working with Data Analysis in an Industry 3.0 to understand the boiler process, its failures and the behavior of process variables in the face of some of the faults presented. On the other hand, it is believed that the adversities found in the data make the Machine Learning algorithms for classification of fault conditions and prediction of remaining useful life do not obtain good results.