Aplicação de técnicas de machine learning na análise preditiva de falhas para otimização da manutenção industrial

Abstract

With the diffusion of Industry 4.0, the operational mode of industries has been significantly impacted by digital transformation. The use of advanced technologies, such as Machine Learning, emerges as a powerful tool to drive greater efficiency and competitiveness among companies. This optimizes the scheduling of predictive maintenance, allowing for the anticipation of machine and equipment failures through historical data, thus avoiding unplanned downtime and maximizing asset availability. In this context, the utilization of Machine Learning techniques is justified, enabling the automated and intelligent analysis of large volumes of data, facilitating the identification of patterns and correlations that are difficult to detect. The main objective of this work is to identify the incidence of failure mode parameters in machinery, aided by artificial intelligence, within the scope of Mechanical Engineering. The synthetic data used for this purpose was obtained from the "Machine Predictive Maintenance Classification" dataset on the Kaggle website, which simulates the parameters of a real machine in both its operational and failure states. The Machine Learning models employed for failure prediction were Random Forest, Support Vector Machine and Multilayer Perceptron, and their accuracies were compared to select the best model.