面向研发全链路的智能审图系统研究与应用OA
Research and Application of an Intelligent Drawing Review System for the Entire R&D Process
针对复杂装备研发全链路中多源异构数据难统一、规范规则难落地、跨环节审查协同不足及人工审查效率与一致性受限等问题,文章基于知识工程、规则驱动审查和检索增强生成方法,提出并实现一套面向研发全链路的智能审图系统.该系统采用 4 层架构:通过多模态解析与数据归一化实现文档、PLM(产品生命周期管理)数据与二维/三维模型的统一表达;基于文档结构恢复与工程知识图谱完成规范知识建模,并以图谱关系驱动生成可执行审查规则;通过多智能体工作流编排与可控 RAG(检索增强生成)推理实现自动化协同审查;在输出层固化证据链路与追溯机制,确保结论可解释、可复核.文档审查、PLM 集成审查与产业链协同轻量化审查的工程验证结果表明,该系统具有良好的通用性、扩展性与工程适配性,系统辅助审查效率较传统人工方式提升20倍以上,人工审签工作量减少约80%,可支撑研发质量闭环.
To address challenges in unifying multi-source heterogeneous data,enforcing engineering standards and rules,enabling collaborative review across R&D stages,and improving the efficiency and consistency of manual review for complex equipment development,this paper presents the development and implementation of an intelligent drawing review system for the entire R&D process based on knowledge engineering,rule-driven review,and retrieval-augmented generation(RAG).The system is designed with a four-layer architecture.Multimodal parsing and data normalization are employed to establish a unified representation for documents,product lifecycle management(PLM)data,and 2D/3D models.Document structure reconstruction and engineering knowledge graphs are used to model standards and specifications,and executable review rules are then derived from graph relationships.Multi-agent workflow orchestration and controllable RAG-based inference are incorporated to enable automated and collaborative review.Evidence chains and traceability mechanisms are embedded at the output layer to ensure that review results are explainable and verifiable.Engineering validation in three representative scenarios—general document review,PLM-integrated review,and lightweight review for industrial-chain collaboration—demonstrates that the proposed system offers good generality,scalability,and adaptability to practical engineering environments.Compared with conventional manual review,the system improves review efficiency by more than 20 times and reduces review and approval workload by approximately 80%,thereby supporting closed-loop quality management throughout the entire R&D process.
刘振波;冯立杰;王添;周叶红;王锐;涂本荣
中车工业研究院有限公司,北京 100070中车工业研究院有限公司,北京 100070||中车科技创新(北京)有限公司,北京 100192中车工业研究院有限公司,北京 100070中车长江车辆有限公司,湖北 武汉 430212中车长春轨道客车股份有限公司,吉林 长春 130062中车南京浦镇车辆有限公司,江苏 南京 210031
交通工程
智能审图研发全链路多源异构数据知识图谱多智能体协同
intelligent drawing review systementire R&D processmulti-source heterogeneous dataknowledge graphmulti-agent collaboration
《控制与信息技术》 2026 (4)
68-78,11
中国中车股份有限公司重大专项科技研究开发计划(2026CKA673、2026CKA674)
评论