Deep learning in industrial machinery: A critical review of bearing fault classification methods

• Published in: Applied soft computing Vol. 171; p. 112785
• Main Authors: Rehman, Attiq Ur, Jiao, Weidong, Jiang, Yonghua, Wei, Jianan, Sohaib, Muhammad, Sun, Jianfeng, E, Shiju, Rehman, Khalil Ur, Chi, Yongwei
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2025
• Subjects: Deep learning algorithms; Fault classification methods; Industrial machinery; Rolling Bearing; Fault classification methods; Rolling Bearing; Deep learning algorithms; Industrial machinery
• ISSN: 1568-4946
• DOI: 10.1016/j.asoc.2025.112785

The review provides an overview of the state-of-the-art in Deep Learning (DL) algorithms for rolling bearing fault classification which remains vital in industrial sectors including transportation, energy, manufacturing, and so forth. Even though they experience a variety of faults, rolling bearings are very crucial in ensuring machine efficiency. This prompts the review of the DL application, which is continuously growing, for intelligently detecting these faults. It comprehensively analyses DL models including Convolutional Neural Networks (CNNs), Auto-Encoders (AEs), Deep Belief Neural Networks (DBNs), Recurrent Neural Networks (RNNs), Generative Adversarial Networks (GANs), and some advanced networks i.e., Transfer Learning (TL), Transformer Neural Network (TNN), Self-Supervised Learning (SSL), Federated Learning (FL), Meta-learning and Interpreting Neural Networks assessing their effectiveness and limitations in fault classification. Thus, the current review is unique among available literature at present since it bridges this crucial gap by including all forms of advanced networks and gives an insight into the potential and challenges. Besides, it emphasizes the importance of different sensing techniques and key datasets in the field to show the contribution towards advancements of DL applications. Finally, referring to current challenges and recommendations for future research directions encompassing environmental adaption, sensor deployment, data preprocessing, model training enhancements, algorithm selection, classifier development, and systematic documentation frame the conclusive part of the paper. This review will serve as an important source for diligent researchers in legitimacy approaches of machinery reliability improvement by means of DL-based techniques for rolling bearing fault classification. •This review analyzes deep learning models and their roles in feature extraction, managing complex datasets, and enhancing fault diagnosis accuracy.•It evaluates the effectiveness and limitations of advanced networks in fault classification, bridging a key gap in the literature and providing insights into their potential and challenges.•The review highlights the importance of sensing techniques and key datasets in advancing deep learning for rolling bearing fault classification, concluding with challenges and future research directions to guide machinery reliability improvement.•This manuscript reviews deep learning methods for rolling bearing fault classification, comparing models based on data requirements, adaptability, interpretability, and complexity. It highlights strengths, limitations, scalability in industrial settings, and challenges with large datasets.