Poincaré Features for Estimation of Remaining Useful Life in Roller Bearings

Prognostics and Health Management (PHM) are essential for optimizing maintenance strategies for rotating machinery. By enabling predictive maintenance, PHM facilitates targeted interventions, minimizing unnecessary maintenance, extending component lifespan, and optimizing resource allocation. This research investigates using Poincaré Plot (PP) and statistical Health Indicators (HIs) derived from vibration signals for prognostics in roller bearings. Utilizing the IMS bearing dataset, PP-based HIs demonstrated high value of prognostics metrics and were used to train a Long Short-Term Memory (LSTM) model for Remaining Useful Life (RUL) estimation. The LSTM model achieved high precision in RUL prediction and exhibited good generalization capabilities when validated with an independent test dataset. Both statistical HIs and PP-based HIs proved very accurate for roller-bearing prognostics.