一种基于改进DeepID2网络的转子碰摩声发射信号识别方法  

A Rotor Rub-Impact AE Signal Recognition Method Based on Improved DeepID2 Network

作　　者：杨伟博 李晶 赵杰[1] 夏昌炜 徐虹飞 YANG Weibo;LI Jing;ZHAO Jie;XIA Changwei;XU Hongfei(School of Information and Communications Engineering,Nanjing Institute of Technology,Nanjing Jiangsu 211167 China;School of Information Engineering,Nanjing Audit University,Nanjing Jiangsu 211815 China)

机构地区：[1]南京工程学院信息与通信工程学院,江苏南京211167 [2]南京审计大学信息工程学院,江苏南京211815

出　　处：《电子器件》2022年第1期128-132,共5页Chinese Journal of Electron Devices

基　　金：国家自然科学基金项目(51908285,62001215);南京工程学院校级科研基金项目(YKJ201975)。

摘　　要：提出了一种基于改进CNN的转子碰摩故障信号识别方法。针对传统CNN经常出现的梯度消失问题,把网络层的各层特征信息连接到一起,可有效避免边缘信息的损失,保留各层的信息特征;同时提取转子碰摩故障信号的声谱图以及差分特征等多通道图像的输入特征与优化后的CNN网络模型相适应;再利用包含各层信息的融合特征输入网络的全连接层,对转子碰摩故障信号进行分类识别。实验结果表明,提出的基于改进CNN的转子碰摩故障信号识别算法,与传统深度神经网络模型相比,识别率有较大提高。A rotor rub-impact fault signal recognition method based on improved CNN is proposed.Firstly,to solve the gradient disappearance problem that often occurs in the traditional CNN,an improved CNN model is proposed.To avoid the loss of edge information,the fully connected layer was used to connect the feature information of each layer in the network layer,which maximum retention of information features at each layer.Then,combining the acoustic spectrum of rotor rub-impact fault signal with its differential characteristics to construct the multi-channel image input characteristics suitable for CNN network.Finally,the full connection layer of the network is input by using the fusion feature containing the information of each layer to classify and identify the rotor rub-impact fault signal.The experimental results show that the recognition rate of the improved CNN recognition algorithm is improved compared with the traditional CNN and other network models.

关 键 词：卷积神经网络 声发射 碰摩故障识别 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]