Aplicação de algoritmos de aprendizado de máquina na manutenção preditiva: uma análise das técnicas de IA para predição de falhas

With the significant increase in the volume of operational data collected, machine learning algorithms have stood out in identifying the need for predictive maintenance in equipment, especially in scenarios where the partial or total inoperability of a piece of equipment is extremely costly. This study analyzed the feasibility of three algorithms, training them and comparing their performance using synthetic data that simulates real-world scenarios. The results point to strong potential for the use of the Random Forest algorithm, which achieved an accuracy of 0,99 in the tests conducted, with a training time of 0,22 seconds on the test set. Furthermore, this algorithm outperformed the others in all evaluation metrics considered, establishing itself as the most effective approach among those analyzed. Thus, the results obtained can aid in the selection of more efficient algorithms for predictive maintenance in industrial environments with limited data.