Fault diagnosis of wind turbine structures with a triaxial vibration dual-branch feature fusion network

The structural safety and fault diagnosis of wind turbines have emerged as key requirements for maintaining the power output performance and reliability of the large-scale wind power industry. Due to the unpredictable operating conditions and the diversity of fault varieties of wind turbines, accurate fault diagnosis poses significant challenges. This paper proposes a novel triaxial vibration-based dual-branch feature fusion network (TriVib-DBFFN) for structural health monitoring of wind turbines. The network is developed with a learnable Fast Fourier transform (FFT) layer by combining conventional signal processing methods with the adaptability of neural networks. Especially, it innovatively includes a dual-branch feature fusion network that is able to adaptively integrate meaningful features in both the time and frequency domains. This fusion method significantly improves diagnostic performance under diverse operating conditions. In addition, this study can reveal the specific signal directions and frequency components prioritized during feature extraction by analyzing the weighting outcomes obtained from the model training. Therefore, common faults in a wind turbine system including foundation looseness, tower tilt, and blade asymmetricity at different degrees can be diagnosed with high accuracy.