Abstract:

The wind turbine gearbox cannot effectively collect vibration signals under complex faults, which leads to the decline of fault early warning accuracy of wind turbine gearbox. To address this issue, this study investigated the twin modeling of gearbox fault early warning system based on spatio-temporal characteristics. Through the information acquisition subsystem and optical fiber sensing technology, the time sequence and spatial position data of the wind turbine gearbox during operation were collected in real time to obtain spatio-temporal characteristic data. By using the twin space, the collected spatio-temporal characteristic data of the gearbox were transmitted to the virtual space. By adopting the involute tooth-profile drawing method, a wind turbine gearbox twin model was established by integrating geometric, physical, and behavioral models. Simulation data were acquired, and the composite fault early warning results were obtained via probabilistic neural networks. Experiments show that the system can effectively collect the spatio-temporal characteristic data of the wind turbine gearbox and establish the twin model of gearbox. Under the composite fault condition, the mean square error is less than 0.05 m/s², and the correlation coefficient is greater than 0.95, indicating that the system has high data acquisition accuracy and strong correlation. It can effectively complete the composite fault early warning of gearbox and significantly improve the accuracy and efficiency of fault early warning.

Key words: spatio-temporal characteristic, probabilistic neural network, wind turbine gearbox, composite fault, early warning system