Research on a multi-sensor information fusion method for wind turbine gearbox fault diagnosis based on the BLSCFN model

Abstract

The internal structure of wind turbines is complex, and their operating conditions vary widely. When a gearbox failure occurs, complex coupling effects emerge among different parts of the turbine, causing collected signals to be easily interfered with by other components. Traditional methods relying on a single signal for wind turbine gearbox fault diagnosis often result in low accuracy. This paper proposes a wind turbine gearbox fault diagnosis method based on the BLSCFN model with multi-sensor information fusion to address this issue. First, the collected gearbox vibration signals are processed using Fast Fourier Transformation (FFT), and the frequency spectrum of the vibration signals is used as the training input for a convolutional neural network (CNN). Simultaneously, the stator current signals are input into a Bi-directional Long Short-Term Memory network (BiLSTM) to capture the temporal relationships of the stator current from both forward and backward directions. Then, a cross-attention mechanism is introduced to calculate the attention scores between the stator current features and the gearbox vibration features. Using a designed linear weighted fusion strategy, information interaction and fusion of the two different source feature signals are conducted to obtain the relevant parts of the input features. Experimental results show that compared to other commonly used methods and single-sensor gearbox fault diagnosis techniques; the proposed method achieves superior diagnostic performance. This multi-sensor fusion approach effectively improves accuracy and reliability of wind turbine gearbox fault diagnosis.