基于长短时记忆模型的风机故障诊断  

Fan Fault Diagnosis Based on Long-short Term Memory Network

作　　者：刘瑶[1] 徐海平 初宁[1] 郑枫 伍柯霖 吴大转[1] LIU Yao;XU Hai-Ping;CHU Ning;ZHENG Feng;WU Ke-Lin;WU Da-Zhuan(College of Energy Engineering,Zhejiang University,Hangzhou 310027,China;The 705th Research Institute Kunming,CSIC,Kunming 650118,China)

机构地区：[1]浙江大学能源工程学院,杭州310027 [2]中国船舶集团有限公司第七〇五研究所昆明分部,昆明650118

出　　处：《工程热物理学报》2020年第10期2437-2445,共9页Journal of Engineering Thermophysics

基　　金：国家自然科学基金资助项目(No.6170144);浙江省重点研发计划(No.2019C01147);压缩机技术国家重点实验室开放基金(No.SKL-YSL201812,No.SKL-YSL201903)。

摘　　要：风机运行发生故障可能造成重大财产损失和人身安全问题。本文利用集合经验模态分解对风机的振动信号进行预处理,提取出振动信号的低频分量作为输入,利用一种基于长短时记忆的循环神经网络分类器,对风机的振动时序信号数据进行深度学习,进行故障分类。其应用过程包括模型设计,模型训练和诊断实现算法,并且利用仿真信号对模型进行验证,再进行风机运行过程中的部分故障模式识别。长短时记忆模型通过增加"记忆门"、"遗忘门"等单元,有效改善循环神经网络模型的梯度爆炸现象,提高分类性能,并且直接对时序信号数据进行处理,便于工业应用。通过对JNF282B-9D型离心式风机进行常见故障的模式识别与诊断,正确率达96.9%。该系统能实现准确的故障诊断,便于生产运营。Failure in the operation of the fan may cause significant property damage and personal safety issues.This paper uses ensemble empirical mode(EEMD)decomposition to pre-process the vibration signal of the fan,extract the low-frequency component of the vibration signal as input,and use a recurrent neural network classifier based on long short time memory network(LSTM),and apply deep learning on the vibration time signal data of the fan For fault classification.The application process includes model design,model training and diagnosis implementation algorithms,and the simulation signal is used to verify the model,and the recognition of the failure mode recognition during the operation of the fan.By adding"memory gate","forget gate"and other units,the long-short-term memory network can effectively improve the gradient explosion phenomenon of the recurrent neural network model,improve the classification performance,as well as directly process the time-series signal data for industrial application.Through the pattern recognition and diagnosis of common faults of JNF282 B-9 D centrifugal fan,the detection accuracy rate is 96.9%.Therefore,the proposed approach can realize accurate fault diagnosis and facilitate production and operation.

关 键 词：长短时记忆模型 集合经验模态分解 风机 故障诊断 

分 类 号：TH17[机械工程—机械制造及自动化]