A spatial–semantic hierarchical knowledge graph (SSH-KG) modeling method was proposed in order to address the problem of dispersed design knowledge and difficulty in fusing with geometric models in body-in-white datum design, which leads to low automation. A four-layer ontology model, “component–feature–point cloud–datum”, was constructed. A unified representation framework integrating spatial geometry and semantics information was established. Automated mapping and reasoning from semantic rule to spatial entity were achieved through formal axiomatic definition. Experimental studies on a B-pillar reinforcement, an A-pillar reinforcement and a sill beam were conducted. Results showed that the average topological relation coverage (evaluating knowledge representation completeness) was increased from 49% to 97.4%, and the average spatial-semantic alignment rate (measuring compliance of geometric entities with semantic rules) was improved from 48.2% to 98.3% compared with conventional manual design. A closed-loop, automated workflow that transformed rule into scheme was realized. The average number of design iterations per scheme decreased from 3.5 to 1, with the total design time reduced by approximately 83%. Ablation experiments confirmed that the point cloud discretization mechanism was critical for enhancing reasoning robustness and computational efficiency.