基于元学习的小样本多模态TE过程故障诊断  

Fault Diagnosis of a Small Sample Multimodal TE Process Based on Meta Learning

作　　者：杨青 华春丽 朱美臣 吴东升 王笑臣 YANG Qing;HUA Chunli;ZHU Meichen;WU Dongsheng;WANG Xiaochen(Shenyang Ligong University,Shenyang 110159,China;Zhejiang Linix Motor Co,Ltd,Dongyang 322118,China)

机构地区：[1]沈阳理工大学自动化与电气工程学院,沈阳110159 [2]浙江联宜电机有限公司,浙江东阳322118

出　　处：《沈阳理工大学学报》2024年第2期52-57,共6页Journal of Shenyang Ligong University

基　　金：辽宁省教育厅高等学校基本科研项目(LJKMZ20220618)。

摘　　要：为解决多模态TE过程在小样本条件下故障诊断精度低的问题,提出了一种基于深度最近邻神经网络(DN4)与压缩和激励(SE)模块结合的集合型故障诊断方法(SEDN4)。首先,使用小波包变换将多模态过程数据转换成二维图像,划分元学习任务;然后,由嵌入网络进行局部特征提取,获得局部特征描述符;最后,使用k近邻搜索得到预测值。当新模态产生时,基于已有模型设计经验,在小样本条件下可快速得到新模态故障诊断模型。实验结果表明,本文方法在小样本条件下能够较好地实现多模态故障诊断,提高了故障诊断准确率,诊断效果较好。To address the issue of low accuracy of fault diagnosis of the multimodal TE process with limited samples.An ensemble fault diagnosis method SEDN4 based on deep nearest neighbor neural network(DN4)combined with squeeze and excitation(SE)blocks is proposed.First,the multimodal process data is converted into a two-dimensional image using the wavelet packet transform to delineate the meta-learning task.Local feature extraction is then performed by the embedding network to obtain local feature descriptors.Finally,a k-neighborhood search is used to find the expected values.With the new modes,based on the experience of existing model designs,the new fault diagnosis model can be obtained quickly under small sample conditions.According to the experimental findings,the method can be used to realize multimodal fault diagnosis well under small sample conditions.The improved fault diagnosis accuracy leads to better diagnostic results.

关 键 词：多模态TE过程 故障诊断 小样本学习 元学习 

分 类 号：TP277.3[自动化与计算机技术—检测技术与自动化装置]