复合多尺度注意熵在旋转机械多工况损伤识别中的应用  

Application of composite multi-scale attention entropy in damage identification of rotating machinery under multiple working conditions

作　　者：张伟 卞其翀[3] 叶丹茜 ZHANG Wei;BIAN Qichong;YE Danqian(Department of Mathematics and Computer Science,Shanxi Normal University,Linfen 041004,China;Department of Information Engineering,Changzhi Vocational and Technical College,Changzhi 046000,China;School of Management and Economics,Fuzhou University of International Studies and Trade,Fuzhou 350200,China;School of Mechatronics and Mold Engineering,Taizhou Vocational College of Science&Technology,Taizhou 318020,China)

机构地区：[1]山西师范大学数学与计算机科学学院,山西临汾041004 [2]长治职业技术学院信息工程系,山西长治046000 [3]福州外语外贸学院经管学院,福建福州350200 [4]台州科技职业学院机电与模具工程学院,浙江台州318020

出　　处：《机电工程》2024年第3期418-429,共12页Journal of Mechanical & Electrical Engineering

基　　金：福建省中青年教师教育科研项目(JAS22192);福建省科技厅引导性项目(2021H0030)。

摘　　要：针对传统旋转机械损伤识别方法存在模型精度低和抗噪性差的问题,提出了一种基于复合多尺度注意熵(CMATE)和随机森林(RF)的旋转机械多工况损伤识别方法。首先,提出了一种新的测量时间序列复杂度的非线性动力学工具——复合多尺度注意熵;然后,利用CMATE提取旋转机械振动信号的损伤特征,其表征了旋转机械不同工况下的损伤特性;接着,将损伤特征输入至基于随机森林构造的多类别分类器中,进行了损伤识别;最后,采用滚动轴承-齿轮箱、齿轮箱和离心泵3种旋转机械数据集,并分别构造了9种工况和20种工况的多工况损伤数据集,对该损伤识别方法进行了实验研究。研究结果表明:该旋转机械损伤识别方法分别取得95%、97%和100%的识别准确率,在准确率和特征提取效率两方面优于其他的非线性动力学工具;并且在0 dB、1 dB、2 dB和3 dB这4种不同信噪比的噪声干扰下,依然取得了不错的损伤识别结果,证明了该模型具有可观的抗噪性;同时,该损伤识别方法能够稳定地识别旋转机械的不同负载和转速下的损伤,平均识别准确率分别达到了97.2%和93.5%,具有较好的实际应用潜力。Aiming at the problems of low model accuracy and poor noise resistance in the traditional damage identification methods for rotating machinery,a multi working condition damage identification method for rotating machinery based on composite multi-scale attention entropy(CMATE)and random forest(RF)was proposed.Firstly,a new nonlinear dynamic tool—composite multi-scale attention entropy,was proposed to measure the complexity of time series.Then,the damage characteristics of vibration signals of rotating machinery were extracted by CMATE to characterize the dynamic characteristics of rotating machinery under different working conditions.Then,the damage features were input into a multi category classifier based on random forest to identify the damage.Finally,three types of rotating machinery datasets,rolling bearings-gearboxes,gearboxes and centrifugal pumps,were used to construct multi-scale condition damage datasets for 9 working conditions and 20 working conditions respectively,and experimental research was conducted on the damage identification method.The results show that the method achieves the recognition accuracy of 95%,97%and 100%respectively,which is superior to other nonlinear dynamics tools in accuracy and feature extraction efficiency.Moreover,it still achieves good damage recognition results under the noise interference of 0 dB,1 dB,2 dB and 3 dB with different signal-to-noise ratios,which proves that the model has remarkable anti-noise property.At the same time,the damage identification method can stably identify the damage of rotating machinery under different loads and speeds,with an average recognition accuracy of 97.2%and 93.5%,respectively,which has application potential.

关 键 词：复合多尺度注意熵 随机森林 旋转机械 齿轮箱 滚动轴承 离心泵 多工况损伤识别 

分 类 号：TH133.3[机械工程—机械制造及自动化]