Dynamic machinery health prognostics using nonlinear stochastic modeling

Over time, wear and tear are caused by the continuous working of equipment components. Enhanced prognosis framework implementation is imperative to improve equipment reliability and prevent catastrophic breakdowns. Reliable fault identification under imbalanced data conditions remains a critical challenge in predictive maintenance. This study presents an innovative hybrid prognostic framework that combines principal component analysis (PCA) with a nonlinear stochastic Markov Auto-Switch Regression (MASR) model for enhanced bearing fault prediction. Unlike conventional approaches, PCA is employed not only for dimensionality reduction but also for addressing class imbalance and feature redundancy in vibration data by isolating dominant fault-sensitive components. This enhances the discriminative capability of the subsequent stochastic model under skewed data distributions. The MASR model is then applied to forecast the transition of bearings from a healthy to a faulty state, capturing nonlinear dynamics and abrupt shifts in system behavior. The results are confirmed using residual diagnostics and variable gradient analysis techniques for assurance of the prediction of bearing degradation. The model effectively identified regime changes due to loading on the bearing and its transitions from normal to faulty states. The proposed model achieved a transition probability of 85.9% for the defective state and 15.1% for the normal state, indicating the persistence of fault regimes once initiated. During the normal state, small residual variations (0.01–0.05) were observed, while during fault progression, higher residual fluctuations (−0.218 to 0.144) confirmed the onset of defects. The proposed PCA-MASR framework provides a novel and computationally efficient approach for early detection of bearing degradation under imbalanced operating conditions, enabling proactive maintenance, reduced downtime, and improved reliability of mechanical systems.