Avaliação de Desempenho de Modelos Preditivos para Estimação da Vida Útil Remanescente em Rolamentos de Máquinas Rotativas

This paper presents a study on remaining useful life estimation techniques for bearings in rotating machines. Two methods based on mathematical models (exponential growth) and two data-driven methods (autoencoder and LSTM network) were compared. To mitigate the restriction in data availability, synthetic signals generated from real signal characteristics were employed in training. Quantitative metrics were calculated for four real signals from two databases. The results highlight the potential and limitations of the evaluated strategies. It is concluded that current predictive models do not show substantial differences in performance, and therefore, robust, and universal RUL estimation still represents an open challenge.