ResNet-LSTM并行网络转子故障迁移诊断方法  

Rotor Fault Transferable Diagnosis Method for ResNet-LSTM Parallel Networks

作　　者：向玲[1] 张兴宇 胡爱军[1] 邴汉昆 杨鑫 XIANG Ling;ZHANG Xing-yu;HU Ai-jun;BING Han-kun;YANG Xin(Department of Mechanical Engineering,North China Electric Power University,Baoding 071003,Hebei Province,China;Huadian Electric Power Research Institute Co.,Ltd.,Hangzhou 310030,China)

机构地区：[1]华北电力大学机械工程系,河北保定071003 [2]华电电力科学研究院有限公司,杭州310030

出　　处：《动力工程学报》2023年第1期41-47,共7页Journal of Chinese Society of Power Engineering

基　　金：国家自然科学基金资助项目(52075170,52175092)。

摘　　要：为提高小样本下的转子故障识别精度,提出了基于残差网络(ResNet)和长短期记忆网络(LSTM)的并行神经网络(RLPN)转子故障迁移诊断方法。首先,使用卷积层和池化层作为模型的前置特征提取器,提取信号的浅层特征;然后,利用ResNet模块提取转子信号的空间特征,利用LSTM模块提取转子信号的时间特征;最后将提取的时间和空间特征融合,对转子的不同工况开展迁移学习,以实现故障诊断。结果表明:该方法能够提升故障的分类性能,有效识别转子故障,诊断结果优于已有的智能故障迁移诊断方法。In order to improve the rotor fault identification accuracy under small samples,a rotor fault transferable diagnosis method based on residual network(ResNet)and long short-term memory network(LSTM)was proposed.Firstly,the convolutional layer and pooling layer were used as pre-feature extractors to acquire shallow features of signals.Then the ResNet module was used to extract the spatial features of rotor signals,and the LSTM module was used to extract the temporal features of rotor signals.Finally,the time and space features extracted were fused,and the rotor fault diagnosis was achieved by transferring learning between different working conditions.Results show that this method can improve the performance of fault classification while effectively identify rotor faults,and the diagnosis results are better than the existing intelligent fault transferable diagnosis methods.

关 键 词：故障诊断 转子故障 残差网络 长短期记忆网络 并行神经网络 

分 类 号：TH165.3[机械工程—机械制造及自动化]