Análise de falhas em rolamentos usando análise de vibrações e modelos de aprendizado de máquina

Abstract

Predictive maintenance is a fundamental strategy for the early identification of failures in rotating systems, particularly in bearings, which play a critical role in various industrial sectors. The premature failure of these components can lead to unplanned downtime, reduced operational efficiency, and high maintenance costs. In this context, the present study investigates the application of machine learning models for vibration analysis in bearings, aiming to detect failure-indicative patterns before severe damage occurs, enabling the proactive implementation of corrective actions. For this analysis, data were obtained from an online database provided by the Society for Machinery Failure Prevention Technology (MFPT). From these data, features in the time, frequency, and cepstrum domains were extracted and used as input for supervised classification models. The study employed two widely recognized algorithms in the literature: Random Forest and XGBoost, both of which are capable of handling large volumes of data and identifying complex patterns associated with bearing failures. The adopted approach allowed for a comparison of the robustness and predictive capabilities of each model, considering different feature selection and data processing strategies. The results highlight the effectiveness of machine learning models in failure detection, emphasizing that XGBoost demonstrated superior performance in handling raw data efficiently, whereas Random Forest, using a reduced set of variables selected based on feature importance analysis, exhibited lower computational complexity and greater interpretability. Furthermore, the analysis revealed the models’ ability to generalize failure patterns under different operating conditions, reinforcing the feasibility of artificial intelligence (AI) for predictive maintenance optimization. The proposed approach provides significant benefits to the industry, enabling cost reduction, increased equipment reliability, and mitigation of unexpected failures. Thus, the application of machine learning in bearing condition monitoring proves to be a promising solution for enhancing predictive maintenance strategies, making production processes more efficient and secure.