CEEMD和小波半软阈值相结合的滚动轴承降噪

针对滚动轴承振动信号降噪处理时如何保证信号边缘信息完整性的问题,提出将互补集合经验模态分解(complementary ensemble empirical mode decomposition,简称CEEMD)与小波半软阈值相结合的信号降噪方法,对滚动轴承故障高频振动信号进行降噪处理。首先,采用CEEMD方法对故障振动信号进行分解,针对信号特点自适应获取不同频段模态分量;其次,将对包含噪声污染的高频信号模态分量进行相关性分析,得到含噪成分较高的高频模态分量,进一步采用小波半软阈值进行降噪处理;最后,将降噪后的模态分量同残余分量进行信号重构,完成降噪过程。分析结果表明,相对于传统小波阈值降噪和CEEMD强制降噪方法,提出的方法能够有效去除高频信号的噪声,且最大程度地保证了原始信号的完整性,降噪效果更好。     

改进的ESMD用于公共场所异常声音特征提取

由于公共场所异常声音的特殊性及背景噪声的复杂性,极点对称模态分解(ESMD)用于异常声音分解时,存在一些理论和技术上的缺陷。经分析认为公共场所异常声音为非线性、非平稳信号,背景噪声服从T分布。为此,提出改进的ESMD用于公共场所异常声音分解,得到有利于识别的特征。所提出方法的特点是将T分布噪声序列添加到具有背景噪声的异常声音信号中,以减小背景噪声对特征提取的影响;将模态分量的排列熵作为判定异常声音与背景噪声的准则,自适应筛选有效的模态分量;用对称中点插值法替代极值中点奇偶插值法,以缓解ESMD插值端点不明确带来的模态失真。在公共场所异常声音数据库上进行了相关实验。实验结果表明,所提出的方法与目前典型的时频信号处理方法相比,在提高公共场所异常声音分类识别率的同时,缩短异常声音的分解时间,是一种有效的公共场所异常声音特征提取方法。     

Target location method for pipeline pre-warning system based on HHT and time difference of arrival

The increasing serious pipeline leakage accidents are caused by third part damage. The third party damage activity around pipeline can generate seismic wave, and the seismic wave can be used for target identification and damage activity location. This paper presents a novel passive location method based on arrival time difference of specific seismic wave characteristic frequencies. The seismic signals are typically non-stationary and the conventional methods cannot analyze them well. Hilbert–Huang Transform (HHT), including empirical mode decomposition (EMD) and Hilbert transform, is a new time–frequency analysis method and can be used for seismic signals analyzing. Firstly, EMD is applied to process the signals and obtain the intrinsic mode functions (IMFs) features. The kurtosis features are used to identify targets and characteristic frequencies are selected as principal components according to IMFs energy features. These principal components are processed by windowed harmonic wavelet transform and then instantaneous features of seismic signals can be extracted. TDOA can be deduced from the arrival time difference of principal frequency components. Finally, target location can be achieved by the time difference analyzing, sensors layout and the relative position between sensors and targets. The seismic signals acquired from field experiment are analyzed and the results are discussed.     

Application of Hilbert-Huang Decomposition to Reduce Noise and Characterize for NMR FID Signal of Proton Precession Magnetometer

The parameters in a nuclear magnetic resonance free induction decay (FID) signal contain information that is useful in magnetic field measurement, magnetic resonance sounding and other related applications. A real time sampled FID signal is well modeled as a finite mixture of exponential sequences plus noise. We propose to use the Hilbert-Huang transform (HHT) for noise reduction and characterization, where the generalized Hilbert-Huang represents a way to decompose a signal into so-called intrinsic mode function (IMF) along with a trend, and obtain instantaneous frequency data. First, acquiring the actual untuned FID signal by a developed prototype of proton magnetometer, and then the empirical mode decomposition is performed to decompose the noise and original FID. Finally, the HHT is applied to the obtained IMFs to extract the Hilbert energy spectrum of the signal on the frequency axis. By theory analysis and the testing of an actual FID signal, the results show that, compared with general noise reduction methods such as auto correlation and singular value decomposition, combined with the proposed method can further suppress the interfered signals effectively, and can obtain different components of FID signal, which can be used to identify the magnetic anomaly, the existence of groundwater etc. This is a very important property since it can be exploited to separate the FID signal from noise and to estimate exponential sequence parameters of FID signal.     

IVMD融合RobustICA的内燃机噪声源分离

为了准确分离识别内燃机的主要噪声源,提出了一种改进变分模态分解融合鲁棒独立分量分析的方法。首先,针对变分模态分解方法的分解数选择问题进行了算法优化,提出了基于重构信号能量比和中心频率的改进变分模态分解方法,并利用仿真信号进行了验证;其次,进行了内燃机噪声试验,利用改进变分模态分解将单通道信号分解成多个信号分量,根据信号分量与源信号的互信息主要分量识别,克服了主要噪声分量选择客观依据不足的问题;最后,通过鲁棒独立分量分析提取主要噪声分量的独立成分,并结合相干分析和时频分析进行噪声源识别。结果显示,所提出的方法能够有效进行噪声源分离,可成功识别出燃烧噪声、活塞敲击噪声和空压机噪声等内燃机主要噪声源。     

基于小波包分解与主流形识别的非线性降噪

为解决工程实际中强噪声、非线性且频率成分复杂的振动信号降噪问题,提出了基于小波包分解和主流形识别的非线性降噪方法。采用小波包分解将原始振动信号正交无遗漏地分解到各频带范围内,根据各子频带中信噪空间分布,分别采用相应参数对小波包分解系数进行相空间重构;采用局部切空间排列(local tangent space alignment,LTSA)主流形识别方法在高维相空间中实现信号与噪音的分离,并重构出降噪后的一维小波包分解系数,最后进行小波包分解重构得到降噪后的振动信号。通过仿真实验和实例应用对本文所提方法的有效性进行了验证,试验结果表明本文方法具有良好的非线性降噪能力。     

基于自适应CEEMD的非平稳信号分析方法

由于标准的互补集总经验模态分解(complementary ensemble empirical mode decomposition,简称CEEMD)在处理模态混叠问题时缺乏自适应性,其本质是分解信号获得的本征模态函数(intrinsic mode function,简称IMF)之间产生了一定的信息耦合现象,使IMF分量不能正确地反映信号的真实成分。因此,提出了在使用CEEMD分解信号的过程中嵌入网格搜索算法(grid search algorithm,简称GSA),以最小二乘互信息(least squares mutual information,简称LSMI)为网格搜索算法的适应度函数,构造一个自适应CEEMD方法。该算法通过自适应地搜索最佳的白噪声幅值,修正信号分解过程中产生的少量的耦合频率成分,确保每个IMF分量之间信息的正交性,以进一步抑制模态混叠问题。最后,通过仿真实验验证了该方法的有效性,并将该方法用于提取滚动轴承微故障的特征频率。实验结果表明,该算法在滚动轴承的微故障特征提取应用中具有更少的迭代数、IMF分量以及相对更小的计算量。     

基于变模式分解的爆震特征识别方法

基于经验模态分解(empirical mode decomposition,简称EMD)算法因递归分解模式所造成的固有缺陷,将使用变分原理进行分解的变模式分解(variational mode decomposition,简称VMD)算法引入到爆震识别领域,发现VMD算法对比EMD算法有较高的计算效率与准确性,而且表现出了较好的鲁棒性,更加适合于在混有强烈背景噪声的缸盖振动信号中提取爆震特征。在此基础上,针对VMD算法分解层数需要手动选择的缺点,利用各阶分量的中心频率之差,提出了一种可以自适应选择VMD分解层数的方法。这种方法的思路为利用VMD算法对信号从一个较小的层数开始进行分解,逐个增加分解层数,直至各阶分量中心频率差值满足预先设定的阈值为止,即可得到最佳分解结果。经实验数据验证与对比,结果显示了这种方法的优越性。     

采用EEMD和WPT的结构损伤特征提取方法

为了解决传统小波或小波包变换方法对结构损伤振动信号频率分辨率不高、易受邻近谐波交叠影响的问题,提出了一种基于聚类经验模式分解(EEMD)和小波包变换(WPT)的结构损伤特征提取方法.首先对原始信号进行EEMD分解,提取包含结构损伤信息的固有模式分量(IMF),再对其进行正交小波包分解,并计算小波包相对能量分布.该方法用于美国土木工程师学会(ASCE)提出的钢结构框架的损伤特征提取,结果表明:EEMD方法具有白噪声的剔除特性,可避免模式混叠的发生;不同检测节点处不同损伤工况的IMF小波包相对能量分布有显著的差异,可以作为一种理想指标表征结构损伤特征.     

基于CEEMDAN与自适应阈值降噪的滚动轴承故障诊断

针对滚动轴承故障信息受到噪声污染而难以识别的问题,提出一种基于自适应噪声完备集合经验模态分解和自适应阈值降噪(CEEMDAN-ATD)的滚动轴承故障诊断方法。首先对原始振动信号进行CEEMDAN分解;其次利用灰色关联分析法(GRA)筛选出噪声主导和信号主导的分量;然后对噪声主导分量分别进行自适应阈值降噪(ATD)处理,并与信号主导分量进行重构;最后通过分析重构信号的Teager能量谱实现滚动轴承故障的识别。采用凯斯西储大学轴承数据对所提方法进行验证,并与完全总体经验模态分解-自适应阈值降噪(CEEMD-ATD)和CEEMDAN-小波阈值降噪(CEEMDAN-WTD)2种方法作比较,结果表明,所提方法表现出较好的自适应性和去噪效果,能够较好地服务于滚动轴承故障诊断。     

变结构SVD算法及其在信号分离中的应用

利用奇异值分解（Singular value decomposition,SVD）进行信号处理的关键在于矩阵的构造,为利用SVD分离信号中的不同频率成分,提出一种变矩阵结构递推SVD算法,其思想是在SVD递推分解过程中逐次改变矩阵的结构,每进行一次SVD分解,矩阵的结构就规律性地变化一次,由此形成对信号中不同频率成分的适应性,从而达到将其分离出来的目的。推导出这种变结构SVD的信号分解算法,证明了这种算法可以将原始信号分解为一系列分量信号的线性组合。进一步从理论上分析了这种算法的信号分离机理,证明了对于一些特定的频率结构,这种变结构SVD算法可以实现对原信号中单个频率分量的逐次分离。最后通过对模拟信号和工程实际信号的分离实例证实了变结构SVD算法良好的信号分离效果,并与小波分析和多分辨SVD方法进行了比较,结果表明变结构SVD的信号分离结果优于这两种方法。     

液压泵故障的小波变换诊断方法

分析了小波变换的时 -频局部化特性及基于多分辨分析的信号小波分解重构算法 ,研究了信号局部奇异性在小波变换下的特性。根据故障信号和噪声的局部奇异性在小波变换下的模极大值在不同尺度上的传播特性不同的特点 ,并利用小波分解重构算法 ,对泵壳振动加速度信号进行了分解、去噪和重构。大大改善了监测信号的信噪比 ,对故障特征信号进行了时域定位 ,提取了故障特征频率。     

Duffing振子与响应灵敏度结合的结构损伤检测方法

阐述了Holmes型Duffing方程的混沌特性用于强噪声背景下弱信号检测的机理。针对强噪声背景下的结构损伤定位问题,提出了基于Duffing振子和响应灵敏度结合的结构损伤定位方法,仅通过一组特定的参数来实现对于噪声背景下响应信号的提取,避开了传统的Duffing振子参数选择的繁琐性,适用于更高的噪声水平,同等噪声水平下较传统的时域响应灵敏度方法计算量更小。将该方法用于强噪声背景下三维桥梁结构模型的损伤检测,结果表明,该方法在强噪声背景环境下能较好地实现损伤定位,为强噪声背景环境下的实际工程结构损伤识别提供了良好的思路。     

基于模态分量优化重构和稀疏表达的 静电信号联合降噪方法

航空发动机静电监测技术表现出了较高的故障预警能力,但原始静电信号常包含较多噪声,为提高故障信息提取的准确性,必须对静电信号进行降噪处理。本研究首先介绍了静电监测技术的原理,分析了信号的噪声的来源和主要构成;针对静电信号耦合噪声滤除问题,引入了信号稀疏表达和经验模态分解理论,研究了模态分量的筛选依据和相关准则,并提出了一种基于模态分量优化重构和稀疏表达的联合降噪算法和具体流程;利用所提方法对涡扇发动机试车实验中采集的实际静电信号进行了降噪效果验证,并与其它方法进行了对比。结果表明本文方法在滤除随机噪声以及工频干扰的同时能更高程度的保留有用异常颗粒信号,稀疏迭代次数在设置为20～50时均能够较好提取异常信号。     

基于自适应谐波分量提取的航空发动机附件传动系统变速故障诊断方法

针对当前基于无键相阶次跟踪(TLOT)的故障诊断所面临的转速谐波分量提取存在误差累积效应、瞬时相位难以准确估计等问题,本文提出了一种基于自适应谐波分量提取的航空发动机附件传动系统变速故障诊断方法。首先,通过低通滤波和降采样优化搜索空间并提升计算速度,在此基础上利用自相关平均周期进行自适应辛几何模态分解;其次,采用基于替代数据检验的伪谐波分量识别方法,完成转速谐波分量自适应分离结果的稀疏化表征。最后,基于转速谐波分量瞬时相位计算结果,对原始非平稳信号进行等角度重采样,利用傅里叶变换获取阶次谱以实现旋转机械装备的变速故障诊断。通过与典型信号分解方法对比,验证了所提方法的有效性;此外,对法国Safran某型航空发动机扫频试车过程中附件传动系统实测数据进行分析,所得阶次相对误差为0.059%,优于同类方法计算结果,进一步显示了其工程应用价值。     

基于判别字典学习的电能质量扰动识别方法

电能质量扰动识别方法通常是先通过数字信号处理工具对信号进行检测和特征提取,再采用人工智能方法对特征进行分类识别,增加了识别过程的复杂性和冗余性。提出一种基于判别字典学习(DDL)的稀疏表示电能质量扰动识别方法,可有效减少识别步骤、降低复杂性,并提高识别率。该方法首先采用主成分分析方法将K类扰动训练样本集降维为扰动降维特征训练样本集,由各类样本分别训练出冗余子字典,然后级联成判别字典。接着基于l0范数算法求解出降维测试信号在该判别字典下的稀疏表示矩阵,最后利用不同的冗余子字典重构测试样本,由冗余残差最小值确定目标归属类,实现对电能质量扰动信号的识别。仿真实验结果表明该方法能有效地对不同电能质量扰动进行识别,过程简单、数据量少、抗噪声鲁棒性好,在信噪比20 d B以上的噪声环境中电能质量扰动识别准确率达到95%以上。     

Vibration component separation by iteratively using stochastic resonance with different frequency-scale ratios

It is well-known that stochastic resonance (SR) is mainly used for signal denoising and weak signal detection. In this paper, we firstly find the frequency range selection characteristic (filtering characteristic) of re-scaling frequency SR (RFSR) caused by the driving frequency limitation of bistable SR. It then follows that a novel approach to separate vibration components with different frequencies by iteratively using SR is explored. The frequencies of most vibration signals exceed the driving frequency limitation, thus by use of different frequency-scale ratios, the vibration signals with different frequency range can be extracted by RFSR. Firstly, a small frequency-scale ratio is used to obtain the vibration signal with a narrow frequency range, i.e. low frequency vibration. As the output of SR may have a phase lag, a simple phase-shift correction method is proposed to improve the accuracy of signal component separation. The phase-shift corrected signal of RFSR output is separated from the original vibration signal and the residue is treated as the new vibration signal. Then, increasing the frequency-scale ratio according to a searching algorithm, the vibration signal with higher frequency can be obtained by RFSR. Through this iterative process, several harmonic vibration components can be separated from the original noisy vibration signal. The proposed method, empirical mode decomposition (EMD) and Hilbert vibration decomposition (HVD) are respectively applied to analyzing a simulated vibration signal and extracting the fault feature of a rotor system. The contrastive results show that this proposed method has good frequency resolution and can successfully separate monocomponent harmonic signals from a strongly noisy multicomponent harmonic vibration signal while EMD and HVD cannot.     

SWT based separation method for periodic signal with non-stationary noise and its application in EMF

Periodic signal superimposed with strong non-stationary noise that follows an approximate 1/f distribution cannot be easily separated with traditional signal processing methods. Using the stationary wavelet transform, the noisy signal is decomposed into wavelet coefficients including both the detail and approximate coefficients. According to its periodic feature, detail coefficients on each scale are extracted to form the same-phase sequences which consist of coefficients with the same phase values in each cycle. The amplitude probability distribution functions of same-phase sequences follow approximate Gaussian distribution. Therefore, noise in the same-phase sequences can be removed with the non-linear median filter and moving average filter. Since non-stationary noise follows approximate 1/f distribution, the approximate coefficients on the lowest frequency level have strong non-stationary property. Due to spectrum leakage of different frequency sections, the leakage signal components are superimposed on the approximate coefficients. Three different filtering methods are proposed to process the approximate coefficients in order to extract the useful signal components and to reconstruct the periodic signal accurately. Finally the proposed method is used to process the output signal of electromagnetic flowmeter during slurry flow measurement under different slurry concentrations and different flow rates. Results show that the proposed method is effective in the separation of periodic signal and strong non-stationary noise which follows the approximate 1/f distribution.     

变分模态分解在转子故障诊断中的应用

针对转子故障诊断问题,提出一种基于变分模态分解(variational mode decomposition,简称VMD)的信号处理方法。该方法在获取分解分量的过程中通过迭代搜寻变分模型最优解来确定每个分量的频率中心及带宽,从而能够自适应地实现信号的频域剖分及各分量的有效分离,对各单分量信号进行希尔伯特变换,即可得到瞬时的频率和幅值信息。对仿真信号和典型转子故障信号进行VMD方法和经验模态分解(empirical mode decomposition,简称EMD)方法的分析比较,以验证所提方法的有效性。仿真信号的分解结果表明,变分模态能够准确分离出信号中的固有模态分量且不存在模态混叠;转子故障实验信号的分析结果表明,所提方法能够有效提取出明显的故障特征,从而准确诊断出转子存在的故障。     

基于ITD-形态滤波和Teager能量谱的轴承故障诊断

针对强背景噪声下滚动轴承振动信号故障特征信息难以提取的问题,提出了结合固有时间尺度分解(ITD)-形态滤波和Teager能量谱的滚动轴承故障特征提取与诊断方法。首先对滚动轴承振动信号采用ITD方法分解,得到若干个固有旋转分量;考虑到噪声主要分布在高频段,取前2个高频的固有旋转分量进行形态滤波,并将滤波后的信号与剩余固有旋转分量重构;对重构信号计算Teager能量算子并绘制Teager能量谱,从Teager能量谱中可以识别出故障特征。将本方法应用于滚动轴承的内圈故障和外圈故障诊断,结果表明ITD-形态滤波可以有效去除振动信号中的背景噪声并保留冲击特征,Teager能量谱可以直观并准确显示出故障特征。