计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (24): 309-318.DOI: 10.3778/j.issn.1002-8331.2208-0186

• 工程与应用 • 上一篇    下一篇

工业设备故障处置知识图谱构建与应用研究

瞿智豪,胡建鹏,黄子麒,张庚   

  1. 上海工程技术大学 电子电气工程学院,上海 201620
  • 出版日期:2023-12-15 发布日期:2023-12-15

Research on Construction and Application of Knowledge Graph for Industrial Equipment Fault Disposal

QU Zhihao, HU Jianpeng, HUANG Ziqi, ZHANG Geng   

  1. School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China
  • Online:2023-12-15 Published:2023-12-15

摘要: 利用知识图谱辅助工业设备故障处置能够有效提升故障处置效率。针对工业设备故障领域实体标注主要依靠人力,耗时耗力的问题,提出一种基于外部知识库的设备故障处置实体半自动标注方法,利用爬取的设备资料和义原等外部知识实现了领域内实体半自动标注,节省了近一半的人工标注成本。针对现有实体抽取方法应用在工业设备故障处置领域实体类型和实体标签识别错误的问题,提出一种融入词性和词边界信息的设备故障处置实体抽取方法,该方法在字嵌入时,在BERT预训练字向量的基础上融入了字所在词的词性信息和词边界信息等多源信息,获得比其他字嵌入方法更多的语义信息,结合BiLSTM和CRF构成实体抽取模型,在自建数据集上的实验结果表明该模型识别性能相较于BERT-BiLSTM-CRF,F1值提升了3.8个百分点,且在更少的迭代次数就获得较好的效果;在知识图谱应用上,提出一种多模态信息融合的设备故障处置方案推荐方法,该方法利用深度学习模型和传感器信息判断故障的发生,并基于知识图谱推荐维修人员和维修方式。

关键词: 设备故障处置, 知识图谱, 半自动标注, 实体抽取, 智能推荐

Abstract: The use of knowledge graph to assist industrial equipment fault disposal can effectively improve the fault disposal efficiency. Addressing the problem that the annotation of entities in the field of industrial equipment fault mainly relies on human resources, which is time-consuming and labor-intensive, a semi-automatic annotation method of entities for equipment fault disposal based on external knowledge base is proposed, which achieves semi-automatic annotation of entities in the field using crawled equipment information and external knowledge such as sememe, saving nearly half of the manual annotation cost. Aiming at the problem that the entity types and entity labels are incorrectly identified by using the existing entity extraction methods. The method incorporates the lexical information and word boundary information of the word in the word embedding based on the BERT pre-trained word vector to obtain more semantic information than other word embedding methods, and combines BiLSTM and CRF to form the entity extraction model in this paper. The experimental results show that the recognition performance of the proposed model has been improved by 3.8 percentage points compared with BERT-BiLSTM-CRF. At the same time, better results can be obtained with fewer iterations. On the application of knowledge graph, a multi-modal information fusion method for equipment fault disposal solution recommendation is proposed, which uses deep learning models and sensor information to determine the occurrence of faults, and recommends maintenance personnel and maintenance methods based on the knowledge graph.

Key words: equipment fault handling, knowledge graph, semi-automatic marking, entity extraction, intelligent recommendation