基于HRFDE和GSA-PNN的旋转机械故障识别模型  

Fault identification model of rotating machinery based on HRFDE and GSA-PNN

作　　者：赫大雨 王强 HE Dayu;WANG Qiang(School of Railway Locomotive,Jilin Railway Technology College,Jilin 132299,China;School of Mechanical Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]吉林铁道职业技术学院铁道机车学院,吉林吉林132299 [2]东南大学机械工程学院,江苏南京210096

出　　处：《机电工程》2023年第12期1869-1879,共11页Journal of Mechanical & Electrical Engineering

基　　金：吉林省职业教育科研课题项目(2022XHZ009)。

摘　　要：采用波动散布熵只能提取故障振动信号的单一尺度特征,而多尺度反向波动散布熵(MRFDE)无法分析信号的高频特性信息,导致提取的故障特征不够全面,进而影响旋转机械故障识别准确率,针对这一问题,提出了一种基于层次反向波动散布熵(HRFDE)和引力搜索算法优化概率神经网络(GSA-PNN)的旋转机械故障诊断模型(方法)。首先,利用层次分割处理代替MRFDE中的粗粒化处理,提出了可以同时提取信号中低频段信息和高频段信息的HRFDE方法,并用于全面表征旋转机械故障特征中的低频和高频信息,从而生成了故障特征样本;然后,采用引力搜索算法(GSA)方法对概率神经网络(PNN)分类器的平滑因子进行了快速优化,建立了GSA-PNN多故障分类模型,对旋转机械的故障类型进行了识别和检测;最后,利用滚动轴承和齿轮箱两种典型的故障数据集,对基于HRFDE方法和GSA-PNN分类器的故障诊断方法的有效性和稳定性进行了实验分析,并将其与现有基于MRFDE、多尺度波动散布熵(MFDE)和层次散布熵(HDE)的故障特征提取方法进行了对比分析。研究结果表明:基于HRFDE方法和GSA-PNN分类器的故障诊断方法可以精准地识别旋转机械的不同故障类型,对两种数据集的识别准确率均达到了98%;而在牺牲部分故障识别效率的基础上,能够获得优于其他对比方法的故障识别准确率,其具有更好的综合性能。Aiming at the fluctuation dispersion entropy only extracted a single scale feature,the multi-scale reverse fluctuation dispersion entropy(MRFDE)couldn t extract the high-frequency characteristic signal information,resulting in the defect of the incomplete extracted fault features with the fault identification accuracy of rotating machinery.A rotating machinery fault identification model(diagnosis method)based on hierarchical reverse fluctuation dispersion entropy(HRFDE)and gravity search algorithm optimized probability neural network(GSA-PNN)was proposed accordingly.Firstly,hierarchical segmentation was used to replace the coarse-grained process in MRFDE,and an HRFDE method was proposed to simultaneously extract the low frequency and high frequency information in the vibration signal,thus the low frequency and high frequency information in the fault characteristics of rotating machinery were comprehensively characterized while the fault feature samples were generated.Then,GSA method was used to optimize the smoothness factor of PNN method rapidly,and a GSA-PNN multi-fault classification model was established to recognize and detect the fault types of rotating machinery.Finally,the effectiveness and stability of fault diagnosis method based on HRFDE method and GSA-PNN classifier were experimentally analyzed by using two typical fault data sets of rolling bearing and gearbox,and were compared with existing fault feature extraction methods based on MRFDE,multiscale fluctuation dispersion entropy(MFDE)and hierarchical dispersion entropy(HDE).The results show that the fault diagnosis method based on HRFDE method and GSA-PNN classifier can accurately identify different fault types of rotating machinery,and the recognition accuracy of two data sets reaches 98%.On the basis of sacrificing part of the efficiency,the fault identification accuracy is better than that of the other comparison methods,and the comprehensive performance is better.

关 键 词：旋转机械 反向波动散布熵 层次反向波动散布熵 故障分类器 引力搜索算法 概率神经网络 

分 类 号：TH133.3[机械工程—机械制造及自动化]