Reliability Analysis of Rolling Bearings Using a Weighted Nonlinear Mixed-Effects Degradation Model

Abstract

Reliability assessment of rolling element bearings is critical for the predictive maintenance of industrial rotary machinery.This study proposes a Quadratic-Exponential Weighted Model (QEWM) based on Nonlinear Mixed-Effects (NLME) to characterize the degradation process of bearings. Utilizing the IMS Bearing Dataset (Set No. 2), we define the failure threshold based on the latest ISO 20816-3:2022 vibration severity standards, setting the critical RMS limit at 0.4 mm/s for Zone D. Unlike traditional models, the proposed QEWM incorporates a weight function to address heteroscedasticity, which typically intensifies during the rapid degradation phase. Model comparison based on the Akaike Information Criterion (AIC)demonstrates that QEWM significantly outperforms linear and unweighted quadratic models. To quantify the uncertainty ofthe estimation, a parametric bootstrap method with 5,000 replications was employed. The results identify a B10 life (t0.1) of165.3 hours, supported by a precise 95% confidence interval of [162.7, 168.6] hours. This research provides a robust statistical framework for bearing life prediction that aligns with international industrial standards, ensuring high precision inprognostic assessments.