To address the limitations of traditional motor fault diagnosis methods, which rely on specialized knowledge and experience and exhibit low accuracy in handling complex data, a motor fault diagnosis method that combines the Symmetric Dot Pattern (SDP) with the Multichannel Residual Fusion Graph Capsule Network (MCFRes-CapsGNN) was proposed. Snowflake diagrams were constructed using symmetric polar coordinate transformation techniques, and multi-sensor signals were fused through color multi-channel fusion technology to generate color fused images, optimizing the visual representation of signal features. Superpixel segmentation was applied to divide the images into superpixel blocks, and graph-structured data was constructed based on their texture, color, and distance features. The MCFRes-CapsGNN model was used to process the graph-structured data, which effectively captured hierarchical structures and complex spatial relationships within the data and enhanced feature learning capabilities. Experimental results demonstrate that the proposed method achieves an accuracy of 98.12% in diagnosing common fault types, outperforming other intelligent fault diagnosis algorithms.