Recent advances in journal bearings: wear fault diagnostics, condition monitoring and fault diagnosis methodologies

This review comprehensively encompasses a range of recent studies on journal bearings, emphasizing wear fault diagnostics, condition monitoring, and fault diagnosis methodologies. A significant finding reveals a shift back to the utilization of journal bearings in various rotating machinery such as compressors, motors, turbines, and pumps. Various methodologies employed in these recent studies include vibration analysis, machine learning, deep learning, and both numerical and experimental simulations. Key findings indicate that ensemble models, such as the CNN and deep neural network (CNNEPDNN) model, significantly improve convergence speed, test accuracy, and F-Score in bearing fault diagnosis by 15-20% compared to individual models. Additionally, convolutional autoencoders have demonstrated impressive performance, achieving an average Pearson coefficient of 91% in wear estimation, underscoring the critical importance of predictive maintenance. Despite these remarkable advancements, challenges persist due to the lack of uniform evaluation criteria and the focus on specific error types under particular operating conditions. Collaborative efforts among researchers are essential for developing robust and broadly applicable diagnostic models. Addressing these ongoing issues will further enhance condition monitoring and defect detection, leading to more reliable and academically rigorous diagnostic methods applicable in diverse real-world scenarios.