机电工程技术2024，Vol.53Issue(12)：226-230,5.DOI:10.3969/j.issn.1009-9492.2024.12.045
DenseNet多元并行网络机械故障诊断研究
Research on Mechanical Fault Diagnosis of Multivariate Parallel Network Based on DenseNet
陈洋 1张晓光 2韩小棒 3雷振兴 3陆凡凡3
作者信息
- 1. 上海智质科技有限公司,上海 201801
- 2. 上海智质科技有限公司,上海 201801||中国科学技术大学计算机科学与技术学院,合肥 230026
- 3. 安徽智质工程技术有限公司,安徽 芜湖 241000
- 折叠
摘要
Abstract
In the mechanical fault diagnosis,the accuracy of the traditional neural network model is difficult to guarantee because of the complexity and variability of the actual working conditions.In order to improve the fault recognition rate of network model,a multi-component parallel network mechanical fault diagnosis method based on DenseNet is proposed.First,the time domain signal is converted into the image containing signal features by short time Fourier transform.Then,different types of feature images are paired and input into each branch of the parallel network.Each branch is calculated in parallel to obtain different feature parameters,and the similarity between different types of fault features is compared to obtain a similarity score,and the model is optimized.The degree of model optimization is controlled by the set threshold.Finally,the mechanical fault is identified by comparing the similarity of different features,and the feature parameters with higher similarity are judged to be of the same type.The model is verified by the UConn gear fault open data set,the recognition accuracy of MBDN network can reach 97.6%,which is 8.4%higher than that of 2DCNN network.The other three evaluation indicators score 95.25%,97.43%,and 95.46%respectively,and the results show that the proposed multicomponent parallel network has better fault discrimination and accuracy than other network models.关键词
故障诊断/稠密网络/短时傅里叶变换/曼哈顿距离Key words
fault diagnosis/dense-network/shorttime Fourier transform/manhattan distance分类
信息技术与安全科学引用本文复制引用
陈洋,张晓光,韩小棒,雷振兴,陆凡凡..DenseNet多元并行网络机械故障诊断研究[J].机电工程技术,2024,53(12):226-230,5.
