基于信息熵的滚动轴承声发射信号故障诊断

利用声发射信号进行低速滚动轴承故障检测.在状态特征提取的基础上,从信息融合的思想出发,建立一种基于时域的奇异谱熵、频域的功率谱熵、时-频域的小波能谱熵和空间特征谱熵的故障诊断方法,并作为综合评价滚动轴承运行状态的定量特征指标.     

基于声发射技术的滚动轴承故障检测

利用声发射传感器、PCI-2声发射数据采集卡和AEwin软件系统建立滚动轴承故障检测试验台,进行轴承故障诊断。利用能量分析法对采集的声发射(AE)信号进行初步故障诊断,对自相关处理后的声发射信号进行FFT变换得到频域信号,并采用模糊识别法中的"择近原则",通过建立隶属函数和贴近度函数来识别轴承的故障类型。     

声发射法在滚动轴承故障识别中的应用

本文研究了用声发射法（简称AE）诊断滚动轴承故障的新技术，首次提出用谐振能量比判别法对滚动轴承故障进行定量诊断，在自制的实验装置上应用简单而有效的声发射信号的提取与处理系统，成功地诊断了轴承的故障，实验证明，本文提出的声发射检测方法简单，易行、是一种很有前途的方法。     

基于振动和声发射的滚动轴承故障检测

滚动轴承故障早期微弱特征信息的提取对于保障机械系统的正常运行具有十分重要的意义。鉴于单一类型传感器采集到的信息局限性,有时会造成诊断准确率较低。提出了一种将振动加速度与声发射两种检测信息融合的方法,并将其应用于滚动轴承故障的诊断。首先构建滚动轴承多传感器故障信息处理与融合的算法模型,随后基于滚动轴承故障实验平台与测试系统,获得滚动轴承典型状态的振动加速度和声发射信号,并对实验数据进行分析处理;最后,在此基础上,将振动加速度和声发射两种信号数据特征进行融合,完成对滚动轴承故障的诊断。研究结果表明,该方法对于滚动轴承故障模式的识别较为有效。     

基于双谱分析的滚动轴承故障模式识别

当滚动轴承发生故障时,其产生的振动信号一般是包含较强噪声的非高斯和非线性信号。本文对高阶统计量方法用于滚动轴承故障特征提取进行了研究,提出了基于双谱估计的滚动轴承故障诊断方法。利用这种方法可以同时获得包含滚动轴承故障信号幅值和相位信息的双谱特征图谱。研究表明,双谱图谱可以有效地监测滚动轴承工作状态的模式,因而可以快速地识别滚动轴承不同的故障特征。     

基于声发射技术的铁路货车滚动轴承故障诊断研究

采用声发射技术对铁路货车滚动轴承的故障诊断是一种新型的无损检测方法。对采集到的声发射信号进行基于小波变换的带通滤波处理后,故障特征信息明显,容易区分出铁路货车滚动轴承的不同状态。将声发射技术用于铁路货车滚动轴承的状态监测与诊断中,经大量试验证明,该法能准确区分轴承状态的好坏与故障类型。     

基于LSTM-RNN的滚动轴承故障多标签分类方法

为了提高长短时记忆神经网络模型(long short-term memory recurrent neural network, 简称LSTM-RNN)对滚动轴承故障分类的正确率并减少训练样本量,提出一种基于多标签LSTM-RNN的滚动轴承故障分类方法。首先,建立滚动轴承故障信号仿真模型,分析滚动轴承故障仿真信号频谱特征及其故障分类特点;其次,结合多标签LSTM-RNN模型结构特点,对滚动轴承频谱特征向量进行编码,并利用仿真故障信号验证多标签LSTM-RNN分类方法的有效性;最后,搭建滚动轴承故障模拟试验台,采集3类转速不同故障类型滚动轴承故障振动信号,并采用3种特征提取方法得到共9组试验数据,基于该数据对多标签LSTM-RNN分类方法和单标签LSTM-RNN分类方法进行对比试验。试验结果表明：多标签LSTM-RNN分类方法相比于单标签LSTM-RNN分类方法,平均分类正确率从69.07%提高到99.21%;在保证两种分类方法正确率相近情况下,多标签LSTM-RNN分类方法训练所需样本量比单标签LSTM-RNN分类方法平均减少69.55%。多标签LSTM-RNN分类方法适用于复杂振动信号分类,对于实现快速准确的旋转机械故障诊断具有应用价值。     

铁路货车轴承故障声发射信号的小波分析处理

将声发射技术用于铁路货车轴承故障信号的采集，用小波分析进行带通滤波，从而减少了背景噪声和其他工况的干扰，成功地将正常轴承与故障轴承区分出来。     

借助振动和声发射诊断滚动轴承的失效

借助振动和声发射(ＡｃｏｕｓｔｉｃＥｍｉｓｓｉｏｎ,简称ＡＥ)来检测和诊断轴承的滚动接触疲劳失效,已经做了大量的研究工作(ＤａｒｌｏｗａｎｄＢａｄｇｌｅｙ,1975;Ｂｅｒｒｙ,1991;Ｃａｒｔｅｒ,1996;Ｂａｒｋｏｖ等,1996;Ｊａｍｅｓ等,1973;Ｅｎｓｏｒ等,1975及Ｉｎｏｕｅ等,1987)。在通过振动和ＡＥ检测轴承的失效中,快速傅里叶变换(ＦＦＴ)分析被广泛采用。然而ＦＦＴ分析需要许多步骤才能得到轴承中有缺陷的零件的频谱。为了在轴承应用中推广现代诊断技术(ＣＤＴ),有必要研究更简便的分析方法。本文所述方法是通过ＡＥ声源定位和测…     

基于混合域特征集与加权KNN的滚动轴承故障诊断

针对滚动轴承的早期故障特征微弱难以有效辨识的问题,提出基于混合域特征集与加权K-近邻分类器的滚动轴承早期故障诊断方法。首先,基于时域、频域、时频域信号处理方法计算滚动轴承早期故障的特征指标量,构造混合域特征集,再将混合域特征集输入给KNN实现滚动轴承的早期故障诊断。实验结果表明,基于混合域特征集与加权K-近邻分类器的滚动轴承早期故障诊断方法能够有效地提取滚动轴承早期故障的低维敏感特征,而且结构稳定,诊断精度高,可以推广应用于滚动轴承的实时在线监测。     

基于包络谱分析的滚动轴承故障诊断分析

介绍了包络谱分析方法的基本原理,它是一种基于滤波检波的振动信号处理方法,也是诊断设备零件损伤故障的一种有效的手段,尤其对初期故障和信噪比比较第的故障信号,识别能力很强。重点分析了包络谱分析方法在轴承故障诊断中的应用。通过对滚动轴承故障诊断的实例分析,验证了包络谱分析运用于诊断设备零件损伤故障所取得的效果。     

基于角域级联最大相关峭度反褶积的滚动轴承早期故障诊断

变转速工况是某些启制动工作制设备常用的工作方式,针对启制动工作制下滚动轴承故障的振动信号呈现非平稳特性,加之现场环境噪声的干扰,难以从原始故障信号中提取特征频率。提出基于角域级联最大相关峭度(CMCKD)的滚动轴承故障诊断方法。首先将时域非平稳故障信号进行角域重采样转换为角域内的平稳信号;然后用级联最大相关峭度反褶积对故障信号进行处理,抑制信号中的噪声,提取信号中的周期冲击成分。通过对仿真和实验数据的分析,验证了角域级联最大相关峭度反褶积方法的有效性。     

基于Teager-Huang变换的滚动轴承故障诊断方法

针对滚动轴承故障振动信号的非平稳特征,介绍了一种基于Teager-Huang时频谱和边际谱的滚动轴承故障诊断方法。详细阐述了Teager-Huang时频谱和边际谱的计算方法及物理意义。给出了该故障诊断方法的步骤,并对仿真和实际轴承的滚动体故障、内圈故障和外圈故障信号进行了分析和故障诊断。结果表明,基于Teager-Huang变换的故障诊断方法具有计算速度快,估计准确稳定的特点,是准确判断滚动轴承故障状态的一种有效新方法。     

基于GRU与迁移学习的滚动轴承故障诊断

现阶段基于深度学习的故障诊断需要大量的数据,而制作数据集是一项耗时耗力的工作。针对这一缺点,提出一种基于门控循环单元(Gate Recurrent Unit,GRU)与迁移学习的滚动轴承故障诊断方法。该方法利用与目标域特征相似且易获得源域数据的特点训练网络,确定网络结构和参数,冻结经过训练的卷积神经网络(Convolutional Neural Networks,CNN)和GRU,用小样本目标域数据训练该网络,微调全连接层和分类层,达到迁移的目的。实验对比分析表明,基于GRU与迁移学习的滚动轴承故障诊断方法明显优于基于BP神经网络和基于概率神经网络(Probabilistic Neural Network,PNN)方法的故障诊断,能够更加准确地进行故障分类,为小样本数据集下的故障诊断提出了新思路。     

基于KPCA的轴承故障状态分类研究

提出基于核函数主元分析的轴承故障分类方法。该方法通过计算轴承振动信号原始特征空间的核函数来实现原始特征空间到高维特征空间的非线性映射。通过振动测试仪获取轴承在正常、外圈破损和保持架损坏状态下的实验数据,比较主元分析与核函数主元分析的故障分类效果。实验表明,核函数主元分析更适合提取故障信号的非线性特征,对故障特征状态有更好的分类效果,并对分类器有较强的鲁棒性。     

Rolling bearing composite fault diagnosis method based on EEMD fusion feature

Aiming at the problem that the composite fault vibration signal of rolling bearing is complex and it is difficult to effectively extract the impact characteristics of the composite fault, a composite fault diagnosis method of rolling bearing based on multi-scale fuzzy entropy feature fusion is proposed. Compared with traditional fault feature extraction methods that can only extract single fault feature information, this method can increase the discrimination of composite fault features, effectively separate multiple composite fault features, and more comprehensively characterize composite fault feature information. First, the signal is processed by EEMD, getting a series of IMF components. Secondly, the energy and kurtosis index of the IMF component are calculated, the appropriate IMF component is selected through the correlation coefficient to obtain a new time series, the multi-scale fuzzy entropy is calculated, and feature fusion performed. Finally, the least square support vector machine is used to diagnose the fault of the fusion feature. The method is verified by a mechanical failure simulation test bench. The experimental results show that this method can quantitatively characterize the data information of fault signal, improve the anti-interference ability, have good feature extraction ability of composite fault of rolling bearings, and can effectively identify the type of composite fault. Compared with the method using multi-scale fuzzy entropy, energy and kurtosis index alone, the accuracy of fault diagnosis increases by 8.12 % and 11.65 %, respectively.

     

Rolling bearing fault convolutional neural network diagnosis method based on casing signal

Affected by the transmission path, it is very difficult to diagnose the vibration signal of the rolling bearing on the aircraft engine casing. A fault diag     

Rolling element bearing fault detection based on optimal antisymmetric real Laplace wavelet

The presence of periodical impulses in vibration signals usually indicates the occurrence of rolling element bearing faults. Unfortunately, detecting the impulses of incipient faults is a difficult job because they are rather weak and often interfered by heavy noise and higher-level macro-structural vibrations. Therefore, a proper signal processing method is necessary. We proposed a differential evolution (DE) optimization and antisymmetric real Laplace wavelet (ARLW) filter-based method to extract the impulsive features buried in noisy vibration signals. The wavelet used in paper is developed from the fault characteristic signal model based on the idea of sparse representation in time-frequency domain. We first filter the original vibration signal using DE-optimized ARLW filter to eliminate the interferential vibrations and suppress random noise, then, demodulate the filtered signal and calculate its envelope spectrum. The analysis results of the simulation signals and real fault bearing vibration signals showed that the proposed method can effectively extract weak fault features.     

基于SK-NLM包络的滚动轴承故障冲击特征增强

非局部均值算法(NLM)是活跃于图像信号处理领域的一种新方法,因其良好的去噪特性,近几年来在滚动轴承故障诊断领域也开始获得应用。NLM利用样本点邻域窗口包含的局部结构为基本单元,通过对相似成分加权运算后取其平均值以达到抑制噪声干扰、突出故障冲击特征的目的。但对于强噪声条件下的低信噪比信号而言,NLM滤波效果并不理想。提出一种结合谱峭度(SK)和NLM权重包络谱的故障诊断方法,首先对原始信号进行SK分析得到最优中心频率及带宽构成最优滤波器,初步消除环境干扰及测量噪声;其次对NLM算法进行改进,不再以滤波信号为分析对象,而是直接利用NLM加权运算得到的信号样本点权值分布曲线作为预处理信号的包络信号,从权重角度使故障冲击得到二次增强,消除SK带通滤波器的带内噪声;最后对权值分布曲线进行包络谱分析,进而得到诊断结果。通过仿真信号、实验室信号及工程实际信号分析对所提方法进行了验证,并与最小熵解卷积(MED)进行了对比。     

Rolling element bearing fault diagnosis based on the combination of genetic algorithms and fast kurtogram

The rolling element bearing is a key part in many mechanical facilities and the diagnosis of its faults is very important in the field of predictive maintenance. Till date, the resonant demodulation technique (envelope analysis) has been widely exploited in practice. However, much practical diagnostic equipment for carrying out the analysis gives little flexibility to change the analysis parameters for different working conditions, such as variation in rotating speed and different fault types. Because the signals from a flawed bearing have features of non-stationarity, wide frequency range and weak strength, it can be very difficult to choose the best analysis parameters for diagnosis. However, the kurtosis of the vibration signals of a bearing is different from normal to bad condition, and is robust in varying conditions. The fast kurtogram gives rough analysis parameters very efficiently, but filter centre frequency and bandwidth cannot be chosen entirely independently. Genetic algorithms have a strong ability for optimization, but are slow unless initial parameters are close to optimal. Therefore, the authors present a model and algorithm to design the parameters for optimal resonance demodulation using the combination of fast kurtogram for initial estimates, and a genetic algorithm for final optimization. The feasibility and the effectiveness of the proposed method are demonstrated by experiment and give better results than the classical method of arbitrarily choosing a resonance to demodulate. The method gives more flexibility in choosing optimal parameters than the fast kurtogram alone.