A Novel Wind Turbine Pitch Bearing Fault Diagnosis Using Signal Encoding and Attention Mechanism Dual-Channel Parallel Fusion Network

As the core component of wind turbine (WT) pitch system, pitch bearing plays an important role in the continuous safe and stable operation of WTs. However, pitch bearing fault diagnosis (PBFD) has problems such as poor diagnostic performance and low accuracy caused by strong random noise interference, small samples and variable operating conditions. For this reason, this article presents a wind turbine pitch bearing fault diagnosis approach continuous wavelet transform (CWT)-position matrix transform (PMT)-attention mechanism dual-channel parallel fusion network (ADCPFN) using signal encoding and attention mechanism dual-channel parallel fusion network. First, the CWT and PMT are utilized to encode the signals caused by vibration into time–frequency diagrams and position matrix diagrams, respectively, so as to boost the signal’s feature information by expanding the data’s visualization dimension. Second, the ADCPFN is constructed using both global and sliding window attention mechanisms to capture both global and local features in the signal-coded image data. The dual-channel parallel network structure enables the model to adequately learn various levels of feature information to suppress the interference of strong random noise on the identification of the health state of the pitch bearing. Finally, in order to verify the advantages and efficacy of the suggested approach, case studies of bearing fault diagnosis toward small samples and variable operating conditions under strong random noise were carried out, and application tests were carried out with real pitch bearing vibration data. The findings demonstrate that the presented approach performs better and has greater diagnostic accuracy when compared to the current mainstream approaches.