Deep Learning Techniques for Rotating Machinery Fault Diagnosis: A Systematic Review and Comparative Study

Abstract

The reliable and safe operation of rotating machinery plays a vital role in modern industrial systems, especially in the era of Industry 4.0 and intelligent manufacturing. Among various condition monitoring techniques, vibration-based fault diagnosis has gained widespread attention because of its strong ability to detect early-stage mechanical defects. In recent years, deep learning (DL) methods have brought significant improvements to fault diagnosis by automatically learning meaningful features from raw signals and providing accurate fault classification even under complex and varying operating conditions. However, existing research is widely distributed across different machine components, fault types, datasets, and neural network architectures, making it difficult to obtain a clear and unified understanding of current advancements and remaining challenges. To address this issue, this paper presents a comprehensive review and critical analysis of eight representative deep learning–based fault diagnosis studies for rotating machinery. The selected studies are systematically examined in terms of signal characteristics, learning models, datasets, performance metrics, and application areas. A structured comparative framework is used to highlight the similarities, differences, strengths, and limitations of existing approaches. In addition, key research gaps related to data availability, model generalization, interpretability, and real-time deployment are identified. Finally, potential future research directions are discussed to support the development of scalable, explainable, and industry-ready intelligent fault diagnosis systems.