Machine Learning-Driven Condition Monitoring and Performance Optimization of Thrust Bearings: State-Of-The-Art Review, Challenges, And Future Perspectives

Abstract

Thrust bearings are critical components in high-capacity rotating machinery such as hydroelectric generators, turbines, compressors, marine propulsion systems, and vertical pumps, where they support axial loads and ensure operational stability. Their performance is governed by complex thermo-hydrodynamic interactions involving lubrication behavior, temperature distribution, structural deformation, and rotor dynamics. Degradation mechanisms including lubrication starvation, thermal instability, wear, fatigue, cavitation, and misalignment can significantly impair efficiency and reliability, making advanced condition monitoring essential for safe and economical operation. Recent advances in Machine Learning (ML) and Deep Learning (DL) have transformed machinery health monitoring by enabling automated fault diagnosis, anomaly detection, health assessment, and Remaining Useful Life (RUL) prediction from multi-sensor data. However, the unique fluid-film lubrication characteristics and coupled thermo-hydrodynamic behavior of thrust bearings present challenges that are not adequately addressed by conventional approaches developed primarily for rolling-element bearings. This review critically examines recent developments in ML-driven condition monitoring and performance optimization of thrust bearings. The analysis encompasses sensing technologies, signal-processing techniques, supervised and unsupervised learning methods, deep learning architectures, transfer learning, prognostic frameworks, and uncertainty-aware RUL prediction. Particular emphasis is placed on Physics-Informed Machine Learning (PIML), which integrates physical laws with data-driven models to improve robustness, interpretability, and predictive accuracy. Emerging technologies including Digital Twins, Explainable Artificial Intelligence, Federated Learning, and Edge Intelligence are also discussed. By identifying key challenges related to data scarcity, model generalization, uncertainty quantification, and industrial deployment, this review provides a structured research roadmap and highlights future directions toward autonomous, reliable, and self-optimizing thrust-bearing management systems.