Bearing Fault Diagnosis: Datasets, Signal Processing Paradigms, and Intelligent Diagnostic Frameworks

Abstract

Rolling element bearing faults are among the leading causes of failure in rotating machinery, leading to unexpected downtime, increased maintenance costs, and reduced system reliability. Consequently, significant research efforts have been devoted to developing effective bearing fault diagnosis techniques. This paper presents a comprehensive and structured review of bearing fault diagnosis studies, focusing on three key aspects: experimental data sources, signal analysis techniques, and expert-system-based diagnostic approaches. Both publicly available benchmark datasets and customized experimental setups reported by individual researchers are systematically examined to highlight their characteristics, fault types, operating conditions, and suitability for fault diagnosis and prognostics. The review further categorizes commonly adopted signal processing techniques applied to vibration and related signals, including time-domain, frequency-domain, and time–frequency-domain methods, with particular emphasis on feature extraction strategies used to characterize bearing defects. In addition, expert systems ranging from traditional machine learning algorithms to advanced deep learning models are reviewed and compared based on their reported usage trends and diagnostic capabilities. By consolidating fragmented research outcomes, this review provides a unified perspective on current practices in bearing fault diagnosis and identifies key directions for future research toward more reliable, generalizable, and industry-oriented diagnostic frameworks.