用知识图谱和大模型提升制造故障原因跨线复用能力
Fault Cause Identification across Manufacturing Lines through Ontology-Guided and Process-Aware FMEA Graph Learning with LLMs
- 通过大模型将FMEA文档转为统一知识图谱，融合领域概念与流程顺序
- 采用过程感知的图神经网络，使故障原因检索准确率提升至0.719
- 适合需要跨产线复用诊断经验的智能制造场景
复杂工程系统中的故障原因识别因系统复杂性、频繁重构及诊断知识复用性差而面临挑战，自动化制造产线是典型应用场景。尽管故障模式与影响分析（FMEA）包含宝贵专家经验，但其在异构系统配置间的复用受自然语言差异、术语不一致和流程不同阻碍。为此，提出OGPAL框架，结合制造领域概念化与图神经网络推理，增强FMEA复用性。首先，利用大语言模型（LLM）支持的本体引导信息抽取，将多条产线的FMEA文档转化为统一知识图谱，捕捉动作、状态、部件、参数等域概念；其次，采用带过程感知评分函数的相对图卷积网络（RGCN），学习兼顾语义关系与流程顺序的嵌入表示；最后，通过链接预测检索并排序与目标产线流程一致的候选故障原因。在汽车压力传感器装配线案例中，OGPAL优于SOTA检索增强生成基线（nDCG@20=0.450）和RGCN方法（0.559），达到最佳性能（0.719）。消融实验验证了LLM驱动的领域概念化与过程感知学习的贡献。结果表明，该框架有效支持对异构诊断知识的推理，提升了FMEA知识在产线间的迁移能力。
原文摘要 · Abstract (English)
Fault cause identification in complex engineered systems remains challenging due to system complexity, frequent reconfigurations, and the limited reusability of accumulated diagnostic knowledge, with automated manufacturing lines representing a prominent application domain. Although Failure Mode and Effects Analysis (FMEA) worksheets contain valuable expert insights, their reuse across heterogeneous system configurations is hindered by natural language variability, inconsistent terminology, and process differences. To address these limitations, we propose OGPAL (Ontology-Guided and Process-Aware Learning), a framework that enhances FMEA reusability by combining manufacturing-domain conceptualization with graph neural network reasoning. First, FMEA worksheets from multiple manufacturing lines are transformed into a unified knowledge graph through ontology-guided information extraction supported by a large language model (LLM), capturing domain concepts such as actions, states, components, and parameters. Second, a Relational Graph Convolutional Network (RGCN) with the process-aware scoring function learns embeddings that respect both semantic relationships and sequential process flows. Finally, link prediction is employed to retrieve and rank candidate fault causes consistent with the target line's process flow. A case study on automotive pressure sensor assembly lines demonstrates that OGPAL outperforms a state-of-the-art retrieval-augmented generation baseline (nDCG@20 = 0.450) and an RGCN approach (0.559), achieving the best performance (0.719) in fault cause identification. Ablation studies confirm the contributions of both LLM-driven domain conceptualization and process-aware learning. These results indicate that the framework effectively supports reasoning over heterogeneous diagnostic knowledge and improves the transferability of FMEA knowledge across manufacturing lines.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。