To address the challenges posed by the complex and variable operating conditions, exacerbated nonlinear fault characteristics, and insufficient feature extraction capability of existing models in the fault diagnosis of offshore wind turbine bearings, a novel fault diagnosis method based on a Kolmogorov-Arnold Convolutional Neural Network with Global Attention Mechanism (KACN-GAM) is proposed. The raw vibration signals are first transformed into two-dimensional images using Gramian Angular Difference Fields (GADF) and used as model inputs. The nonlinear activation functions of Kolmogorov-Arnold Networks (KAN) are integrated into the convolutional layers of Convolutional Neural Network (CNN) to extract fault features in a more flexible manner. A global attention mechanism is employed to strengthen the model’s ability to extract global contextual information. The diagnostic results are obtained through a combination of global average pooling and fully connected layers. Experiments conducted on a real-world offshore wind turbine bearing dataset demonstrate that the proposed method can effectively identify various fault types, exhibiting good generalization capability and high accuracy.