The Diagnosis of Wind Turbine Blade Imbalance Using Dual-Input Signals in Parallel CNN-Transformer

Abstract

Prolonged exposure of wind turbine blades to wind forces can lead to blade twisting and structural loosening. These defects result in uneven mass distribution, causing severe vibrations in wind turbines, which reduce energy efficiency and increase operational costs. To address the challenges of weak vibration signal feature extraction and poor diagnostic model performance caused by blade mass imbalance, this paper proposes a dual-signal parallel CNN-transformer model based on Fast Fourier Transform (FFT) and Variational Mode Decomposition (VMD). A convolutional neural network (CNN) is employed to extract spatial features from the fused time-frequency domain signals, while the time-domain signals are input into a transformer encoder to capture long-term temporal dependencies. A cross-attention mechanism integrates temporal and spatial features by computing attention weights, allowing the model to focus on critical features while reducing computational complexity. Experiments using the vibration data of the wind turbine nacelle collected through the SCADA system show that when the stacked time-frequency signals are used as input, the accuracy of the model is increased by 36.04%, 7.34% and 5.41% compared with the original signal, FFT-processed signal and VMD-processed signal, respectively. The proposed method achieves a diagnostic accuracy of 97.5% under full-sample conditions and 95% under low-sample conditions.