This research focuses on the challenge of assessing the health status and fault-type diagnosis in rolling element bearings (REBs). A key obstacle in this field pertains to feature extraction to exhibit generalizability across a wide array of machines and operating conditions. To address this challenge, new dimensionless features extracted from wavelet transform (WT) and fast Fourier transform (FFT) are proposed in this study. Mamdani fuzzy inference systems are developed based on the authors' extensive years of expertise in condition monitoring. One fuzzy system is dedicated to REB health status detection, while three other fuzzy systems are designed for fault-type diagnosis. The generalizability of this approach is validated through testing on four distinct REB vibration datasets. These datasets include two laboratory datasets, XJTU and PRONOSTIA, a dataset from the authors' university, and an industrial data collected from three industrial sites. The health status detector demonstrates strong performance across all datasets, achieving an average accuracy of 96.8% when an acceptable discrepancy of 0.2 is considered, and 90.4% with a stricter discrepancy of 0.1. Meanwhile, the fault-type diagnosis system is assessed on datasets with available fault-type labels, achieving an average accuracy of 98.7%. Results underscore the robust generalization of the proposed dimensionless features which are insensitive to operating conditions.
Keywords: Feature extraction; Fuzzy logic; Generalization; Intelligent diagnosis; Rolling element bearing.
Copyright © 2026. Published by Elsevier Ltd.