卷积神经网络与知识图谱结合的轴承故障诊断
Bearing Fault Diagnosis Based on Convolution Neural Network and Knowledge Graph
李志博 1李媛媛 1蔡寅1
作者信息
- 1. 上海工程技术大学 电子电气工程学院,上海 201620
- 折叠
摘要
针对目前旋转机械故障诊断时,存在单一地利用振动数据、诊断结果模糊的问题,提出一种卷积神经网络(Convolutional Neural Network,CNN)与知识图谱结合的故障诊断方法.该方法以原始轴承数据和机理知识作为输入,然后进行实体抽取和数据标注,利用本文提出的端到端多尺度注意力机制神经网络模型进行故障诊断,最终构建知识图谱,实现故障信息的详细展示,进行辅助诊断.利用两份数据集进行实验验证,采用全新的数据处理方法,结果表明,所提出的算法在160种故障类型中加权F1值相比基准模型提高11.03%,并且利用传统故障诊断实验和其他算法对比充分证明本文提出的模型具有较强的稳定性和泛化性能.
Abstract
Aiming at the problems of using vibration data only and the fuzzy diagnosis results in the current fault diagnosis of rotating machinery,a fault diagnosis method based on convolution neural network(CNN)and knowledge graph is proposed.In this method,the original bearing data and mechanism knowledge are taken as the input,and the entity extraction and data annotation are carried out.The proposed end-to-end multi-scale attention mechanism neural network model is used to carry out the fault diagnosis of bearings,realize relationship extraction,and finally build a knowledge map to realize the detailed display of fault information for auxiliary diagnosis.Experimental verification is carried out on two datasets,and a new data processing method is adopted.The results show that the value of weighted F1 in the proposed algorithm is 11.03%higher than that of the benchmark model in 160 fault types,and the traditional fault diagnosis experiments and other comparative algorithms fully prove that the proposed model has strong stability and generalization performance.
关键词
故障诊断/卷积神经网络/知识图谱/轴承Key words
fault diagnosis/convolutional neural network/knowledge graph/bearing引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据