基于图数据深度挖掘的旋转机械故障诊断

Fault diagnosis of rotating machinery based on graph data deep mining

针对旋转机械故障诊断过程面临的小样本问题,提出了一种基于图数据深度挖掘的旋转机械故障诊断方法.首先,将利用归一化处理后的监测信号重塑为汉克尔矩阵;然后,将奇异值分解得到的特征向量作为图数据的节点表示,进一步地运用边连接方式构建基于奇异值特征向量的图数据;在此基础上,利用构建的图卷积神经网络充分提取图数据中的高层次故障特征敏感信息;最后,利用softmax分类器辨识监测信号故障类别.实验结果表明:该方法能够以30%的小样本训练集实现99.28%的准确率,具备了良好的故障识别能力.

Aiming at the problem of small samples in the fault diagnosis of rotating machinery,a fault diagnosis method based on graph data deep mining was proposed.Firstly,the normalized monitoring signals were remolded into Hankel matrix.Then,the eigenvectors obtained by singular value decomposition were represented as nodes of the graph data.Furthermore,the edge connection method was used to construct the graph based on singular value eigenvectors.On this basis,the constructed graph convolutional neural network was used to fully extract the high-level fault feature sensitive information from the graph data.Finally,softmax classifier was used to identify the fault category of monitoring signals.The experimental results show that the method can achieve 99.28%accuracy with 30%small sample training set and has good fault recognition ability.