Application of advanced signal processing techniques in vibration analysis and fault diagnosis of multi-dimensional anthropomorphic wind turbines

Wind turbine operating conditions are complex. To ensure the turbine's safe operation, it is essential to carry out condition monitoring and fault diagnosis of its vibration. In this paper, from the structure of wind turbines, fault types, and fault formation mechanisms, a wind turbine vibration condition monitoring system is established by designing different vibration condition monitoring sensors and combining them with the Internet of Things technology. The discrete Fourier transform is employed to preprocess the time-frequency data before extracting the specific features of the vibration signal by combining the Hilbert-Huang transform after obtaining the wind turbine vibration signal. The SC-TSFN model with spatio-temporal deep fusion is established to realize the fault diagnosis of wind turbines by combining the replaceable null convolution module, BiLSTM module and the self-attention mechanism. It has been found that when the tertiary meshing frequency fluctuates around 506.98 Hz at a fault characteristic frequency of 16.14 Hz, it indicates a fault in the tertiary high-speed shaft gear. The SC-TSFN model has a fault identification time of approximately 52 days before the actual fault downtime, and the model has a 92.05% accuracy rate for wind turbine fault identification. Relying on the signal processing technology to carry out the wind turbine vibration signal analysis and then input it into the fault identification model can realize the accurate identification of the fault state of the unit and provide technical support for the stable operation of wind turbines. © 2024 Yingwu Qin et al., published by Sciendo.