• bearings
• predictive maintenance

Industrial maintenance strategies have evolved from reactive and preventive approaches to predictive maintenance (PdM), leveraging data-driven techniques for enhanced efficiency. This thesis focuses on PdM for journal bearings in rotating machinery, using temperature signals for anomaly detection. Failures are identified when bearing temperatures exceed critical thresholds, with deviations in predicted versus actual temperatures signaling potential defects. The study explores various forecasting models, including decision trees and LSTMs, but emphasizes Foundation Models (FMs) for their adaptability and generalization. A real-world dataset from a turbo-generator is used to evaluate these models, demonstrating that FMs outperform conventional approaches. The findings highlight the potential of next-generation AI models in PdM, reducing unplanned downtime and improving industrial reliability.