AttCWKAN: A Novel Convolution Weighted Kolmogorov–Arnold Networks With Attention Mechanism for Wind Turbine Gearbox Fault Diagnosis

Accurate fault diagnosis of wind turbine (WT) gearbox is important for developing effective maintenance strategies. Multilayer perceptron (MLP)-based fault diagnosis models have achieved success but typically have a large number of model parameters that require a high volume of data and computation costs. This article proposes a novel gearbox fault diagnosis method using the Kolmogorov–Arnold theory and deep learning method, named convolution weighted Kolmogorov–Arnold network with an attention mechanism (AttCWKAN). This model uses learnable activation functions at the edges of the network instead of the point-fixed activation functions of the MLP-based model, which significantly reduces the number of parameters. The proposed model employs the nonuniform rational B-spline (NURBS) instead of B-spline in the classical Kolmogorov–Arnold network (KAN) as the activation function, enhancing the influence of key control points using weights for modeling complex curves. In addition, the convolutional attention mechanism in AttCWKAN can extract important hidden features. The experiment results from two datasets show that the average values of the $F1$ -score and accuracy of the proposed model reach {0.9913, 0.9914} and {0.9994, 0.9994}, respectively, which validates the superior performance of the proposed model. Compared with other state-of-the-art methods, the proposed model can use fewer parameters but obtain higher performance in fault diagnosis.