车辆支腿液压系统的自注意力池化图神经网络故障诊断  

Self-attention Pooled Graph Neural Network Fault Diagnosis of Vehicle Outrigger Hydraulic System

作　　者：郭媛 李孟飞[1,2] 汪胜 曾良才 GUO Yuan;LI Mengfei;WANG Sheng;ZENG Liangcai(Key Laboratory of Metallurgical Equipment and Control Technology,Ministry of Education,Wuhan,Hubei 430081;Hubei Key Laboratory of Mechanical Transmission and Manufacturing Engineering,Wuhan,Hubei 430081;Precision Manufacturing Research Institute,Wuhan University of Science and Technology,Wuhan,Hubei 430081)

机构地区：[1]冶金装备及其控制教育部重点实验室,湖北武汉430081 [2]机械传动与制造工程湖北省重点实验室,湖北武汉430081 [3]武汉科技大学精密制造研究院,湖北武汉430081

出　　处：《液压与气动》2024年第9期72-80,共9页Chinese Hydraulics & Pneumatics

摘　　要：针对特种车辆支腿液压系统故障信号的复杂性、特征混叠等问题,提出一种基于自注意力池化图神经网络的车辆支腿液压系统故障诊断方法,并介绍了车辆支腿的常见故障模式和失效机理。将故障信号转换为2D特征图表示,并提出一种改进的3D结构的特征图。以故障特征图作为输入,将图卷积与自注意池化相结合进行特征提取,通过全连接层对提取的特征进行分类识别。结果表明:与2D特征图相比,所提3D特征图提高了模型2%~3%的诊断精度;与原来的池化方法相比,加入自注意力机制的图神经网络在支腿故障数据集上准确率提高了7%~8%,表现出了较高的诊断精度和稳定性,为液压系统故障诊断提供了方法参考。Aimed at the complexity and feature aliasing of fault signals in the hydraulic system of special vehicle support legs,a fault diagnosis method for vehicle support leg hydraulic system based on self-attention pooling graph neural network is proposed,and common fault modes and failure mechanisms of vehicle support legs are introduced.Convert the fault signal into a 2D feature map representation and propose an improved 3D structured feature map.Taking the fault feature map as input,combining graph convolution with self-attention pooling for feature extraction,and then classifying and recognizing the extracted features through a fully connected layer.The results show that compared with 2D feature map,the proposed 3D feature map improves the diagnostic accuracy of the model by 2%to 3%;Compared with the original pooling method,the graph neural network with self-attention mechanism has improved the accuracy by 7%to 8%on the support leg fault dataset,demonstrating high diagnostic accuracy and stability,providing a method reference for hydraulic system fault diagnosis.

关 键 词：故障诊断 液压系统 自注意力 图神经网络 

分 类 号：TH137[机械工程—机械制造及自动化]