摘要
针对汽车发动机故障率高且种类多,故障征兆与故障之间存在多对多的复杂耦联关系,故障溯源难度大、准确率低等问题,提出了融合知识图谱和多元神经网络的发动机故障智能预测方法。将发动机运行状态、故障现象、故障原因和维修记录作为输入信息,通过知识抽取、消歧和加工形成为可表示、可推理的结构化知识网络,并进行特征向量转换;建立了包含故障记录嵌入层、卷积层、GRU门控层和注意力机制的多元神经网络通路,通过特征向量训练形成了发动机故障预测模型,实现了发动机定性故障现象到定量故障推理,再到定性故障预测输出的映射变换;通过实际维修案例验证了所提KG-CNN-GRU-Att方法的可行性和有效性。
Aiming at the problems of high incidence rate and multiple types of automobile engine faults,complex coupling relationship between fault symptom and faults,and difficulty in fault traceability and low accuracy,an intelligent prediction method of engine fault based on knowledge graph and multiple neural network is proposed.Firstly,the engine running state,fault phenomena,fault causes and maintenance records are taken as input information,and the structured knowledge network which can be represented and inferred is formed through knowledge extraction,disambiguation and processing.The eigenvectorsare carried out for the knowledge network.Secondly,a multi-neural network pathway including fault record embedding layer,convolutional layer,GRU(Gate Recurrent Unit)gating layer and attention mechanism is established,and the engine fault prediction model is formed through eigenvector training,which will realize the mapping transformation from qualitative fault phenomena to quantitative fault reasoning,and then to qualitative fault prediction output.Finally,the feasibility and effectiveness of the proposed KG-CNN-GRU-Att(Knowledge Graph-Convolutional Neural Network-Gate Recurrent Unit-Attention)method are verified by a practical maintenance case.
作者
阴艳超
王宵
徐成现
YIN Yanchao;WANG Xiao;XU Chengxian(Faculty of Mechanicaland Electrical Engineering,Kunming University of Scienceand Technology,Kunming 650500,China)
出处
《机械科学与技术》
CSCD
北大核心
2023年第12期2055-2063,共9页
Mechanical Science and Technology for Aerospace Engineering
基金
国家自然科学基金项目(52065033)。
关键词
知识图谱
多元神经网络
发动机故障
向量转换
智能预测
knowledge graph
multivariate neural network
engine fault
vector transformation
intelligent prediction
