Feature-Based Prognostics of Rolling Element Bearings Using PCA and Exponential Trend Modeling on the IMS Dataset

Bearing failures remain one of the leading causes of unplanned downtime in rotating machinery, making accurate fault prediction essential for ensuring industrial reliability and cost-effective maintenance. This paper presents a lightweight and interpretable framework for bearing health monitoring and Remaining Useful Life (RUL) estimation using time-domain features and unsupervised statistical modeling. Leveraging the IMS bearing dataset, a suite of statistical features is first extracted from vibration signals, followed by dimensionality reduction using Principal Component Analysis (PCA) to derive a univariate health indicator. The degradation trend of this indicator is modeled using exponential curve fitting, enabling cycle-level RUL predictions based on a defined failure threshold. The method is evaluated on a test-to-failure case involving an inner race fault in Bearing 3, demonstrating a strong correlation between the modeled degradation curve and the actual failure timeline. Compared to complex machine learning models, the proposed approach offers high interpretability, low computational overhead, and reliable early warning capability, making it suitable for real-time industrial deployment. This work contributes a practical and scalable solution to predictive maintenance, with potential for extension to hybrid frameworks involving machine learning or Digital Twin systems.