A Novel Method Using UGW-Based FAA-ELM for Wind Turbine Gearbox Health Monitoring and Fault Diagnosis Under Variable Working Conditions

Wind energy is a widely utilized renewable energy source. Enhancing the reliability of wind turbines (WTs), especially the gearboxes, is crucial for efficient power generation. Real-time structural health monitoring (SHM) of the gearbox, including pre-start self-testing and ongoing checks, is essential for achieving optimal performance and safety. In this study, we developed a laboratory gear transmission platform simulating the WT spindle gearbox and integrated it with ultrasonic guided wave (UGW) technology for active monitoring. Experiments were conducted to monitor the gearbox in three health states under five different working conditions. A modified health indicator (HI), MWF-PeEn, was extracted from UGW signals. However, significant spatial mapping variations of the HI matrix across different conditions hindered diagnostic accuracy. To address this issue, transfer learning (TL) techniques were applied to optimize the extreme learning machine (ELM), improving the spatial distribution consistency of HI, and enhancing diagnostic accuracy. The precision and recall rate of the proposed feature adaptive adjustment-extreme learning machine (FAA-ELM) model are all above 90%. Compared to recent models, the FAA-ELM showed nearly a 10% improvement in precision, demonstrating its effectiveness and robustness in real-world scenarios. This underscores the practical advantages of SHM and the FAA-ELM for reliable WT maintenance and operation.