As industrial equipment becomes more complex, accurate bearing fault diagnosis is critical for ensuring production safety and cost-effectiveness. Traditional single-channel diagnostic methods, however, struggle to achieve high accuracy. To address this, this paper proposes a bearing fault diagnosis model based on an Improved Black Kite Algorithm (IBKA) and feature fusion. First, vibration signals are transformed into time–frequency images using Synchronized Wavelet Transform (SWT), and spatial features are extracted via a 2D Convolutional Neural Network (2D-CNN). Then, temporal dynamics are captured using Gated Recurrent Units (GRU), enhanced by a Multi-Head Self-Attention (MSA) mechanism to emphasize key features. Finally, the IBKA optimizes model hyperparameters, incorporating strategies like elite inverse learning to enhance global search capabilities. Experiments on the Case Western Reserve University bearing dataset show that the proposed model achieves a test accuracy of 99.6%, demonstrating its superiority in fault classification compared to other models.