Cross-Domain Intelligent Diagnosis of Mechanical Equipment Using Gauss-Wasserstein Distance

Transfer learning technology has found widespread application in the intelligent diagnosis of rotating machinery. However, the commonly used distance metrics, struggle to accurately capture divergence between distributions and are sensitive to kernel functions, resulting in limitations when faced with significant operating condition discrepancies. To overcome the problem, this study introduces a novel cross-domain diagnosis algorithm that utilizes the Gauss-Wasserstein distance. Firstly, we provide an empirical estimation of this distance and successfully incorporate it into the deep diagnosis model. Secondly, this distance metric comprehensively considers both first-order and second-order fault information across two domains, facilitating a more effective cross-domain transfer process by minimizing loss. The efficacy of the proposed method is illustrated using a bearing dataset.