115 / 2023-09-19 22:27:12
An Improved Graph Sampling-based Machine Fault Diagnosis under Edge Computing Architecture
edge computing, fault diagnosis, graph sampling, graph convolutional network.
Final Paper
Jiongyi Chen / Huazhong University of Science and Technology
Jie Liu / Huazhong University of Science and Technology
Chaoying Yang / Huazhong University of Science and Technology
Shiyu Liu / Huazhong University of Science and Technology
Xiaozhen Qin / Huazhong University of Science and Technology
Zhemei Fang / Huazhong University of Science and Technology
基于边缘计算架构,旋转机械的实时故障诊断能够及时发现故障,从而增强工业生产的安全性。近年来,图论在特征提取的性能改进方面得到了广泛的关注。然而,将基于图论的故障诊断方法迁移到边缘侧应用存在一定的局限性:1)使用原始特征作为节点属性会导致计算复杂度高。2)在诊断过程中大量使用训练样本会导致诊断延迟延长和性能低效。该文提出一种边缘计算架构下基于图采样的改进机器故障诊断方法。具体而言,建立同时 (FC-EP) 的特征压缩边预测器在构造节点连接时将压缩特征输出为节点属性。在此基础上,采用图采样方法来减少模型训练和推理的每个实例中使用的节点数量。然后,利用采样图来训练切比雪夫图卷积网络(GCN)。完成模型训练后,将诊断节点作为新节点添加到采样图中,模拟实时故障诊断场景。随后,将这种方法部署到树莓派平台上进行验证。对比实验结果表明,所提方法具有较高的诊断精度和较低的计算延迟。
Important Date
  • Conference Date

    Nov 02

    2023

    to

    Nov 04

    2023

  • Dec 15 2023

    Draft paper submission deadline

  • Dec 20 2023

    Registration deadline

Sponsored By
IEEE Instrumentation and Measurement Society
Xidian University
Contact Information