引用本文:崔更申,李书漪,黄春跃,梁颖,张怀权,曹知勤.一种基于知识图谱的SMT智能故障诊断模型设计与实现[J].控制理论与应用,2026,43(2):305~315.[点击复制]
CUI Geng-shen,LI Shu-yi,HUANG Chun-yue,LIANG Ying,ZHANG Huai-quan,CAO Zhi-qin.Design and implementation of an SMT intelligent fault diagnosis model based on knowledge graph[J].Control Theory & Applications,2026,43(2):305~315.[点击复制]
一种基于知识图谱的SMT智能故障诊断模型设计与实现
Design and implementation of an SMT intelligent fault diagnosis model based on knowledge graph
摘要点击 113  全文点击 12  投稿时间：2023-09-15  修订日期：2025-09-17
查看全文  查看/发表评论  下载PDF阅读器   HTML
DOI编号  10.7641/CTA.2024.30625
  2026,43(2):305-315
中文关键词  表面组装生产  智能故障诊断  知识图谱  故障实体抽取
英文关键词  surface assembly production  intelligent fault diagnosis  knowledge mapping  fault entity extraction
基金项目  国家自然科学基金项目(61164002), 广西重点研发计划项目(桂科AB23075076), 四川省钒钛材料工程技术研究中心开放基金项目(2023FTGC08), 桂林电子科技大学研究生教育创新计划项目(2023YCXS011, 2023YCXS014, 2023YCXS019)资助.
作者单位E-mail
崔更申* 桂林电子科技大学 计算机与信息安全学院 cgs@guet.edu.cn 
李书漪 桂林电子科技大学 机电工程学院  
黄春跃 桂林电子科技大学 机电工程学院  
梁颖 成都航空职业技术学院 信息工程学院  
张怀权 桂林电子科技大学 机电工程学院  
曹知勤 四川省钒钛材料工程技术研究中心  
中文摘要
      针对表面组装生产工艺复杂、生产过程易出现设备故障和工艺缺陷的特点, 本文设计了一种基于故障知识 图谱的面向表面组装生产的智能故障诊断模型. 同时针对知识图谱构建过程的关键技术–故障实体抽取进行研究, 设计并实现了一种基于BERT-Residual-BiLSTM-CRF的面向表面组装生产故障日志的故障实体抽取模型. 首先根据 表面组装技术(SMT)故障日志文本构建故障实体抽取模型的训练、测试数据集, 其次采用TensorFlow框架搭建SMT 故障实体抽取模型, 最后利用训练好的模型进行对照试验. 结果表明, 所设计的故障实体抽取模型的故障实体识别 精确率、召回率和F值平均值较基础模型BERT-BiLSTM-CRF分别提高了约0.26, 0.28和0.24.
英文摘要
      Aiming at the complexity of the surface assembly production process, the production process is prone to equipment failures and process defects, this paper designs an intelligent fault diagnosis model based on fault knowledge graph for surface assembly production. At the same time, the key technology of knowledge graph construction process – fault entity extraction is studied, and a fault entity extraction model based on BERT-Residual-BiLSTM-CRF for surface assembly production fault logs is designed and implemented. Firstly, the training and testing datasets of the fault entity extraction model are constructed based on the text of surface mount technology (SMT) fault logs, secondly, the TensorFlow framework is used to build the SMT fault entity extraction model, and finally, the trained model is used to conduct controlled experiments. The results show that the fault entity recognition accuracy, recall and mean F-value of the designed fault entity extraction model are improved by about 0.26, 0.28 and 0.24 compared with the base model BERT-BiLSTM-CRF, respectively.