Métodos de inteligência artificial para detecção de falhas industriais aplicados em um sistema de manufatura: uma análise comparativa de desempenho

Abstract

This work addresses the functioning of industrial processes and the detection of faults, aiming to contribute in an instructive way to the modern industry, especially in the context of paint manufacturing. Using a software-simulated environment that fundamentally replicates industrial operational scenarios, the research justifies its choice in this sector due to its economic relevance and the growing need for improvements. The central objective is to carry out a comparative analysis of certain machine learning techniques, such as Support Vector Machines, Kth Nearest Neighbor, Multilayer Perceptron and Random Forest, to detect failures in the production process. Simultaneously, a database was developed to represent the functioning of the system during the manufacturing process, containing variables that reproduce the typical operation of this process based on the software explored. The implementation of these techniques demonstrates the behavior of fault detection in the industrial manufacturing system, proposing improvements in the quality of the production process and reduction of operational costs. The insertion of faults in the process, focusing especially on the paint production and mixing system, exposed the fault detection models to various correct or unexpected situations in their operation, providing comparative analyzes for the development of an instructive and applicable approach, contributing for the advancement and improvement of methods in industrial processes.