Fault Diagnosis of Wind Turbine Blade Imbalance Using a Multiscale Convolutional Neural Network Based on CBAM

Wind turbine blade imbalance generates significant centrifugal forces during turbine rotation, leading to stress concentration in critical components such as blades, shafts, and drive systems. This reduces equipment lifespan and power generation efficiency. This paper proposes a wind turbine blade fault diagnosis method based on a multiscale convolutional neural network (CNN) incorporating the Convolutional Block Attention Module (CBAM). First, the Gramian Angular Summation Field (GASF) transformation is used to convert time-domain vibration signals collected by the SCADA system into 2D time-frequency images, effectively preserving the temporal correlations and frequency-domain energy distribution characteristics of the vibration signals. Then, a multiscale feature fusion module is constructed in the CNN to capture local details, medium-range associations, and global patterns of vibration features using different convolution kernels in parallel, generating a multiscale feature matrix with hierarchical representation capability. Finally, CBAM attention mechanisms are introduced during the feature fusion stage, using channel attention submodules to dynamically learn feature channel weights, enhancing frequency band features strongly related to faults, while spatial attention submodules localize sensitive regions of vibration anomalies, forming a "channel-space" dual-dimensional feature optimization mechanism. Experimental results show that the method achieves an average diagnostic accuracy of 95.56% for wind turbine blade imbalance faults.