变分模态分解与深度信念网络的双转子不对中程度识别  

Dual-rotor Misalignment State Recognition Using Variational Mode Decomposition and Deep Belief Network

作　　者：张帆宇 杨大炼[1,2] 李学军 苗晶晶[1] 张宏献 Zhang Fanyu;Yang Dalian;Li Xuejun;Miao Jingjing;Zhang Hongxian(Hunan Provincial Key Laboratory of Health Maintenance for Mechanical Equipment,Hu′nan University of Science and Technology,Hu′nan Xiangtan,411201,China;The Ministru of Education Engineering Research Center of Advanced Mining Equipment,Hu′nan University of Science and Technology,Hu′nan Xiangtan,411201,China;School of Mechatronics Engineering,Foshan University,Guangdong Foshan,528225,China)

机构地区：[1]湖南科技大学湖南省机械设备健康维护重点实验室,湖南湘潭411201 [2]湖南科技大学先进矿山装备教育部工程研究中心,湖南湘潭411201 [3]佛山科学技术学院机电工程学院,广东佛山528225

出　　处：《机械科学与技术》2020年第5期773-779,共7页Mechanical Science and Technology for Aerospace Engineering

基　　金：国家自然科学基金项目(11702091,11672106,51575178);湖南省自然科学基金项目(2018JJ3140)资助。

摘　　要：准确识别不对中严重程度是保障航空发动机双转子系统安全稳定运行的重要途径。但不对中程度信息微弱,现有方法难以对其准确识别,为此本文提出了基于变分模态分解与深度信念网络的双转子不对中程度识别方法。实验采集了3种不对中程度下的振动加速度信号,首先采用变分模态分解将振动信号分解;其次对模态函数进行分析,根据互信息理论确定VMD的分解层数,重构模态信号作为特征输入向量,并用于深度信念网络分类模型训练。通过与VMD+BP、VMD+SVM、原始信号+DBN模型的识别率进行对比分析,结果表明,本文提出的VMD+DBN模型提高了双转子不对中程度的识别准确度,验证了该方法的有效性。Accurate identification of the severity of misalignment degree is an important way to ensure the safe and stable operation of the aero-engine dual-rotor system.Because the misalignment state information often is weak,the existing methods are difficult to identify it accurately.For the above problems,this paper proposes a dual rotor misalignment degree recognition method based on variational mode decomposition and deep confidence network.The vibration acceleration signals in three misaligned degree cases are collected.Firstly,the vibration signal is decomposed by the variational mode decomposition,then the modal function is analyzed.The decomposition layer of VMD is determined according to the mutual information theory,and the modal signal is reconstructed,as a feature input vector,which is used for deep belief network classification model training.Compared with the recognition rates of VMD+BP,VMD+SVM and original signal+DBN model,the simulation results show that the proposed VMD+DBN model improves the recognition accuracy of the dual rotor misalignment degree and verifies the effectiveness of the proposed method.

关 键 词：双转子不对中 变分模态分解 深度信念网络 程度识别 

分 类 号：TH17[机械工程—机械制造及自动化]