A unique statistical framework to predict the health of a machine by utilizing the vibration features of rolling element bearing data

Abstract

Rolling element bearings are critical components in machinery that rotates, supporting radial and axial loads while facilitating smooth motion. Bearing malfunctions rank among the most frequent mechanical failures in rotating machinery, resulting in significant downtime and repair costs if not identified promptly. The Root Mean Square (RMS) value is a widely used statistical feature in Condition Monitoring (CM), providing a reliable and quantitative technique for detecting early-stage bearing faults. This study introduces a method for predicting machine health by utilizing RMS values derived from baseline vibration data of Healthy-Bearing (HB), Inner-Race-Faulty-Bearing (IRFB), and Outer-Race-Faulty-Bearing (ORFB) to explore the deterioration of roller bearings under various conditions using a statistical approach. The AutoRegressive Moving Average (ARMA) (p,q) model was employed to fit the extracted vibration features from bearing data to create a stochastic relationship between the healthy data and the inner race faulty data, as well as between the healthy data and the outer race faulty data. The model's accuracy of the predictive model is evaluated using an error matrix. This study adds to advancing data-driven prognostics in industrial machinery by merging statistical modeling with vibration-based condition monitoring.