Abstract: Structured storage of fault information can be effectively implemented through knowledge graph techniques,which make up for the lack of structured fault information management ability of traditional fault diagnosis methods.However,the number of fault samples is rare in actual working conditions,and it is difficult for traditional knowledge graph techniques to complete graph construction in few-shot condition.To solve the problem,an equipment fault knowledge graph construction and inference method based on meta-learning was proposed.The method extracted fault rule chain and signal features to construct equipment fault knowledge graph.The meta-fault link prediction algorithm was proposed,which made knowledge graph have the ability of fault diagnosis in few-shot condition by using the generation strategy of negative samples in the neighborhood of the same fault cluster.The ability of fault diagnosis and similar fault query was verified by the experiments on general field dataset NELL-One and actual equipment fault datasets.

Key words: knowledge graph, fault diagnosis, meta-learning, negative sampling method

摘要： 知识图谱技术可以有效实现故障信息的结构化存储,弥补传统故障诊断方法缺乏结构化管理故障信息能力的不足。但是实际工况下故障样本数量稀少,传统知识图谱技术难以在小样本情况下完成图谱构建。针对上述问题,提出一种基于元学习的设备故障知识图谱构建及推理方法。该方法首先提取故障规则链和信号特征构建设备故障信息知识图谱;其次提出基于元学习的故障链接预测算法,通过同一故障簇邻域负样本生成策略,使知识图谱具有在小样本情况下进行故障诊断的能力;最后,分别采用通识领域NELL-One数据集和实际设备故障数据集进行实验,验证了算法的故障诊断能力和查询相似故障的能力。

关键词: 知识图谱, 故障诊断, 元学习, 负采样方法