Open-Set Fault Diagnosis of Wind Turbine Gearbox Based on Inverse Lumped Parameter Thermal Network and Open-Set Kolmogorov–Arnold Network

The fault of the wind turbine gearbox will cause long-term downtime loss. At present, most fault diagnosis methods are based on a closed-set strategy, leading to misdiagnose when facing unknown faults. Therefore, a fault diagnosis method for wind turbine gearbox based on inverse lumped parameter thermal network (ILPTN) and open-set Kolmogorov-Arnold network (OSKAN) is proposed. First, the lumped parameter thermal network (LPTN) is established based on the gearbox structure. Second, a fault feature construction method based on the inverse heat transfer problem of the gearbox is proposed. The inverse heat transfer problem is solved by the double-distributed fuzzy inference (DDFI) method, and the fault features are constructed based on the inverse thermal conductance and heat source intensity. Third, we proposed OSKAN to realize open-set fault diagnosis of wind turbine gearbox. Finally, the method is validated by the 2 MW wind turbine data of a wind farm in northern Hebei, China. The results show that the weighted $F1$ Score of OSKAN can reach 0.9929 in cases with different openness, which is superior to the other five comparison methods and can effectively identify unknown faults. The fault features are well interpretable. This method is of great significance to improve the maintenance efficiency and reduce the downtime.