有限标记样本下基于GSSL-GraphSage的半监督故障诊断方法

Semi-supervised fault diagnosis using GSSL-GraphSage under limited labeled samples

中文关键词  故障诊断  GraphSage网络  有限标记样本  半监督学习  标签传播策略

英文关键词  fault diagnosis  GraphSage network  limited labeled samples  semi-supervised learning  label propagation strategy

中文摘要

      鉴于在实际工程中采集的齿轮箱标注监测数据是有限的, 且基于图神经网络的齿轮箱故障诊断方法研究 仍存在标签信息挖掘不充分的问题, 本文提出一种有限标记样本下基于图的半监督学习(GSSL)与图采样聚合算 法(GraphSage)的齿轮箱半监督故障诊断方法. 基于K最近邻算法和基于图的标签传播策略, 将标签信息沿边传播给 分布相似的邻域样本, 从而充分利用有限样本的标签信息, 提高模型性能. 将每个振动频谱样本视为一个节点构建 基于图的半监督学习框架, 最后将半监督学习框架输入到节点级GraphSage网络中进行故障分类, 避免新加入节点 重新训练的情况, 可有效防止训练过拟合, 增强泛化能力. 将所提方法用于分析齿轮箱故障实验数据, 结果表明所 提方法能够在6%的低标签情况下准确诊断齿轮箱的不同故障模式, 验证了对齿轮箱故障诊断的可行性和有效性.

英文摘要

      In view of the limited labelled monitoring data collected in actual engineering and insufficient label information mining problem of gearbox fault diagnosis method based on Graph neural network, a Graph-based semi-supervised learning (GSSL) under limited labelled samples is proposed. GSSL and Graph sample and aggregate (GraphSage) algorithm for gearbox semi-supervised fault diagnosis. Based on the K-nearest neighbor algorithm and the graph-based label propagation strategy, the label information is propagated to neighborhood samples with similar distribution along the edge, so as to make full use of the label information of limited samples and improve the model performance. Each vibration spectrum sample was regarded as a node to construct a graph-based semi-supervised learning framework. Finally, the semi-supervised learning framework was input into the node-level GraphSage network for fault classification, avoiding the situation of new nodes being retrained, which could effectively prevent the over-fitting of training and enhance the generalization ability. The proposed method is used to analyze the experimental data of gearbox faults. The results show that the proposed method can accurately diagnose different fault modes of gearbox under the condition of 6% low label, which verifies the feasibility and effectiveness of gearbox fault diagnosis.