Prognóstico de falhas em rolamentos baseado em modelos granulares evolutivos

Abstract

Predictive maintenance is essential to ensure operational continuity in the industry, being one of the pillars of Industry 4.0, where sensors and cyber-physical systems enable continuous equipment monitoring. In this context, Prognostics and Health Management (PHM) emerges as an approach for failure prediction through the estimation of the Remaining Useful Life (RUL) of machines using algorithms based on physical models, statistical methods, and computational intelligence techniques capable of analyzing operational parameters and calculating the time to failure. However, traditional prediction methods face limitations such as dependency on large volumes of training data, fixed structures and parameters, high computational cost, and low interpretability. These constraints make such methods ineffective in real-world scenarios, where there is a lack of historical data, abrupt operational changes, and the need for computationally efficient and explainable models. To overcome these limitations, this research proposes the AutoCloud Predictor, a granular and evolutionary model that learns continuously and autonomously, adjusting its parameters and expanding or reducing its structure according to the characteristics of the data. Additionally, its granularity-based approach enables the generation of more interpretable and adaptable predictions to system variations. Preliminary tests of the AutoCloud Predictor demonstrate its ability to address the challenges presented, offering more adaptable and interpretable predictions. However, its accuracy has not yet reached the expected level, highlighting the need to explore potential improvements addressed throughout this study.