Knowledge graphs have been recognized as a useful technique for representing knowledge as a labeled directed graph. A novel method for fault diagnosis is proposed in which a customized knowledge graph model is built for rotating machinery fault diagnosis. The proposed specialized knowledge graph is a double-layer structure with a data layer and a pattern layer for data collecting and defect pattern recognition. The creation, update, and inference methods of FDKG are proposed in this study. The proposed update method ensures that the FDKG collects information comprehensively while filtering out effective features for inference diagnosis.Experimented result shows that the FDKG can integrate multiple features perform better than existing classification methods.