Generative Domain Generalization Network based on Local Maximum Mean Discrepancy for Bearing Fault Diagnosis of Wind Turbine

The wind turbines (WTs) usually work in the variable environment. It leads to a frequent impact force act on WT bearings in different working conditions. It would take a look as a typical cross-domain fault diagnosis (FD) issue. Nevertheless, Intelligent FD models may face challenges in generalization when confronted with novel working conditions and managing low-resource datasets. To address the aforementioned challenges, we present a Generative Domain Generalization Network based on Local Maximum Mean Discrepancy (GDGN-LMMD). Within the framework of GDGN-LMMD, we introduce a potential domain generation network (PDGN). The network aims to achieve domain generalization under low-resource datasets by enhancing the variety of sample distributions under known working condition. Furthermore, we introduce a new domain generalization network based on LMMD (DGN-LMMD). This network is tailored to extract domain-invariant fault information from the source domain and potential domains, enabling generalization to previously unseen working conditions. Ultimately, the comprehensive case study demonstrates that GDGN-LMMD enhances diagnostic accuracy in unseen working conditions.