Fault diagnosis method of wind turbine bearing based on time graph fusion

The introduction of the dual-carbon target has heightened interest in wind power generation as a key clean energy source. Bearing fault diagnosis is essential for ensuring the efficient and stable operation of wind turbines. However, bearing vibration signals are often affected by external interference, making feature extraction and fault classification challenging. This paper proposes a novel fault diagnosis method for wind turbine bearings, integrating optimized variational mode decomposition (VMD), Gram angle difference field coding, and a Swin Transformer-One-dimensional Convolutional Neural Network Efficient Channel Attention Network (1DCNN ECANet) parallel network. The methodology consists of two main stages: first, an improved dung beetle optimization algorithm is used to optimize VMD parameters; second, a parallel network is constructed using the Swin Transformer-1DCNN ECANet model. Finally, the effectiveness of the proposed approach is validated using the public dataset from Case Western Reserve University.