Abstract: With the growing adoption of electric vehicles, demand for charging infrastructure has increased significantly, highlighting the need for timely maintenance and fault diagnosis of charging piles. To effectively leverage multi-scale features in charging pile fault signals, this paper proposes a fault information fusion diagnosis method for vehicle-to-grid (V2G) charging piles with open-circuit switching tubes, based on a multi-scale convolutional neural network (CNN) and dual-attention mechanism. The approach builds upon CNNs by integrating a self-attention mechanism to emphasize critical fault signal features. Simultaneously, max pooling and average pooling layers process fault signals to extract complementary multi-scale information. Additionally, a channel attention mechanism is incorporated to enhance model performance by weighting different channel features. Fault classification is performed using a Softmax classifier. Simulation results demonstrate the method's superiority over other algorithms in convergence speed, overfitting suppression, and diagnostic accuracy, while exhibiting strong noise robustness—effectively handling noise interference in fault signals. Experimental tests show the method achieves 96.67% accuracy in locating open-circuit faults in switching tubes, providing an effective solution for diagnosing such faults in charging piles.