A novel digital twin-driven approach based on physical-virtual data fusion for gearbox fault diagnosis

The acknowledged challenge of intelligent fault diagnosis methods is that constructing a reliable diagnosis model requires numerous labeled datasets as training data, which is difficult to collect such high-quality labeled data in the practical industry. The digital twin methodology provides a brand-new and potentially powerful solution to mitigate this challenge. However, during the practical application of digital twin-driven fault diagnosis methods, an information gap can exist between the virtual and physical spaces and poses a hurdle in adopting these methods. Therefore, this paper proposes a novel digital twin-driven approach based on physical-virtual data fusion for gearbox fault diagnosis to enhance the diagnosis performance with insufficient collected fault data. One case study on a truck transmission is carried out. First, a digital twin model of the transmission is established, which can effectively mirror the vibration characteristics and generate vibration data with different health states. Second, a physical-virtual data fusion method based on the Wasserstein generative adversarial networks with gradient penalty is designed to improve the quality of the virtual fault data further. Finally, the virtual fault data through the physical-virtual fusion are used to train a fault diagnosis model. The experimental results indicate that the proposed method significantly enhances the diagnostic performance when few measured fault data from the physical space are available.