CycleGAN With Momentum Control for Intelligent Bearing Fault Diagnosis Driven by Dynamic Models

Rolling bearings play a crucial role in ensuring reliability and safety of mechanical systems and their fault diagnosis has received much attention in the industrial field. However, since rolling bearings mostly operate under normal conditions, it is difficult to obtain a large amount of data on different types of fault conditions in reality. As a result, the problem of imbalanced data limits the application of various intelligent diagnostic methods. To address this problem, this article proposes a novel bearing dynamic model-driven 1-D cycle consistency generative adversarial network (Cycle-GAN). Furthermore, in order to avoid the problem of falling into local minima that is often encountered during network training, a momentum control strategy is introduced. This study proposes a dynamic model-driven 1-D Cycle-GAN (CycleGAN-MC) for rolling bearing fault diagnosis. Experimental results demonstrate that the proposed method can address the data imbalance issue, reducing the percent root mean square difference (PRD) between generated and experimental signals up to 11.78%, a 15% improvement over traditional methods. Moreover, the accuracy of fault classification using data generated by CycleGAN-MC reached 98.7%, representing a 4.4% increase compared to the original Cycle-GAN. These results demonstrate the effectiveness of CycleGAN-MC in enhancing diagnostic accuracy and stability.