柴油机振动信号快速稀疏分解与二维特征编码

针对柴油机故障诊断方法中的信号时频表征及特征提取问题,提出一种基于振动信号快速稀疏分解与二维时频特征编码识别的柴油机智能故障诊断方法。首先,为了获得时、频聚集性优良的时频图像,提出一种随分解残差信号自适应更新Gabor字典的改进匹配追踪(adaptive matching pursuit,简称AMP)算法,利用AMP算法将柴油机振动信号分解后叠加各原子分量的Wigner-Ville分布,获取原信号的稀疏分解时频图像;然后,为提取时频图像的特征参量,提出了双向二维非负矩阵分解(two-directional,2-dimensional non-negative matrix factorization,简称TD2DNMF)算法,用于对时频图像的幅值矩阵进行特征编码,获取蕴含在时频图像内部的低维特征,并利用最近邻分类器实现了时频图像的自动分类识别。将提出的方法应用于4种不同状态柴油机气门故障的诊断试验中,结果表明,该方法能够获得无交叉项干扰、聚集性好的时频图像,使各时频分量的物理意义更加明确,并改进了传统图像模式识别中的特征参数提取方法,是一种有效的柴油机故障诊断方法。     

基于K-奇异值分解和层次化分块正交匹配算法的滚动轴承故障诊断

利用层次化分块正交匹配算法（HBW-OOMP）的高稀疏性和运算速度快等优点,提出了一种基于K-奇异值分解（K-SVD）字典和HBW-OOMP算法的故障轴承诊断方法。首先利用K-SVD自学习训练方法得到包含冲击成分的冗余字典,克服了固定结构字典适应性不强的缺点。然后采用基于分块思想的HBW-OOMP算法进行原子的选取和稀疏系数的求解,以重构信号包络谱峭度最大为终止条件,自适应确定分解次数。最后应用所提方法对仿真信号和故障轴承实验信号进行故障特征提取,结果表明该方法能够有效提取强背景噪声下故障特征成分,具有一定的应用前景。     

基于字典学习和稀疏编码的振动信号去噪技术

针对现有机械振动信号去噪算法需要一定先验知识的问题,提出了一种基于字典学习和稀疏编码的自适应去噪滤波方法。根据信号的本质特性,应用在线字典学习方法对原始数据进行学习和训练,寻求数据驱动的最优字典空间。引入正交匹配追踪算法,确定原始信号在最优字典空间上的稀疏表示。基于稀疏编码和优化字典,重构原始信号,实现信号去噪。仿真和试验结果表明,相对于现有去噪方法,基于字典学习和稀疏编码的方法自适应能力强,去噪效果好。     

基于改进稀疏编码的微弱振动信号特征提取算法

针对强噪声环境下难以有效提取微弱振动信号特征的问题,提出了基于改进字典学习和移不变分量过滤(IDL-SICF)的稀疏编码振动信号特征提取算法。首先,将振动信号进行分段和平滑预处理以降低数据处理复杂度,接着利用改进的字典学习和高效系数求解算法构建基于移不变稀疏编码的自适应滤波器,然后过滤字典原子重构的移不变分量以获得表征信号本质特征的最优基函数,取最优基函数对应的移不变分量的特征频率强度作为评价信号特征提取效果的优劣。仿真和实测数据的试验结果表明,相比于现有微弱振动信号提取算法,提出的算法具有更强的特征提取能力,在实际应用中具有较高的可行性。     

A vibration analysis method based on hybrid techniques and its application to rotating machinery

Vibration-based condition monitoring and fault diagnosis technique is a most effective approach to maintain the safe and reliable operation of rotating machinery. Unfortunately, the vibration signal always exhibits non-linear and non-stationary characteristics, which makes vibration signal analysis and fault feature extraction very difficult. To extract the significant fault features, a vibration analysis method based on hybrid techniques is proposed in this paper. Firstly, the raw signals are decomposed into a few product functions (PFs) using local mean decomposition (LMD), and meanwhile instantaneous frequency and instantaneous amplitude also are obtained. Subsequently, Fourier transform is performed on the derived PFs, and then, according to the spectra features, the useful PFs are selected to reconstruct the purified vibration signals. Lastly, several different fault features are fused to illustrate the operating state of the machinery. The experimental results show that the proposed method can accurately extract machine fault features, which proves that the combined application of LMD and other signal processing techniques is a successful scheme for the machine vibration analysis.     

Anti-aliasing lifting scheme for mechanical vibration fault feature extraction

A troublesome problem in application of wavelet transform for mechanical vibration fault feature extraction is frequency aliasing. In this paper, an anti-aliasing lifting scheme is proposed to solve this problem. With this method, the input signal is firstly transformed by a redundant lifting scheme to avoid the aliasing caused by split and merge operations. Then the resultant coefficients and their single subband reconstructed signals are further processed to remove the aliasing caused by the unideal frequency property of lifting filters based on the fast Fourier transform (FFT) technique. Because the aliasing in each subband signal is eliminated, the ratio of signal to noise (SNR) is improved. The anti-aliasing lifting scheme is applied to analyze a practical vibration signal measured from a faulty ball bearing and testing results confirm that the proposed method is effective for extracting weak fault feature from a complex background. The proposed method is also applied to the fault diagnosis of valve trains in different working conditions on a gasoline engine. The experimental results show that using the features extracted from the anti-aliasing lifting scheme for classification can obtain a higher accuracy than using those extracted from the lifting scheme and the redundant lifting scheme.     

Fault diagnosis of rotating machinery based on the statistical parameters of wavelet packet paving and a generic support vector regressive classifier

The fault diagnosis of rotating machinery has attracted considerable research attention in recent years because such components as bearings and gears frequently suffer failure, resulting in unexpected machine breakdowns. Signal processing-based condition monitoring and fault diagnosis methods have proved effective in fault identification, but the revelation of faults from the resulting signals requires a high degree of expertise. In addition, it is difficult to extract the fault-induced signatures in complex machinery via signal processing-based methods. In this paper, a new intelligent fault diagnosis scheme based on the extraction of statistical parameters from the paving of a wavelet packet transform (WPT), a distance evaluation technique (DET) and a support vector regression (SVR)-based generic multi-class solver is proposed. The collected signals are first pre-processed by the WPT at different decomposition depths. In this paper, the wavelet packet coefficients at different decomposition depths are referred to as WPT paving. Statistical parameters are then extracted from the signals obtained via the WPT at different decomposition depths. In selecting the sensitive fault features for fault pattern expression, a DET is employed to reduce the dimensionality of the feature space. Finally, a SVR-based generic multi-class solver is proposed to identify the different fault patterns of rotating machinery. The effectiveness of the proposed intelligent fault diagnosis scheme is validated separately using datasets from bearing and gearbox test rigs. In addition, the effects of different wavelet basis functions on the performance of the proposed scheme are investigated experimentally. The results demonstrate that the proposed intelligent fault diagnosis scheme is highly accurate in differentiating the fault patterns of both bearings and gears.     

一种频域特征提取自编码器及其在故障诊断中的应用研究

提出一种可以直接从振动信号中提取频域特征的非对称自编码器方法。与传统自编码器以重构振动信号作为目标输出不同,频域自编码器使用振动信号的频谱作为目标输出,这种非对称的自编码器可以学习振动信号与其频谱之间的映射关系,使得编码器可以输出频域特征。为了说明提出的频域自编码器的特征提取效果,在轴承数据集上进行特征提取和故障诊断实验,在没有引入标签信息的情况下,频域自编码器提取到的特征表现出较好的聚类效果,能够区分轴承的不同故障类型;进一步进行了泛化实验,训练分类器时使用1%的有标签样本,可以达到90%以上的故障分类准确率。实验结果表明,频域自编码器与传统自编码器相比,可以更好地提取振动信号的故障特征信息,具有一定的实用价值。     

基于相关正交匹配追踪算法的风电机组滚动轴承稀疏故障诊断方法

针对风电机组滚动轴承故障信号的非平稳、强噪声污染等导致的有效冲击特征难以检测的问题,提出了一种基于相关正交匹配追踪（COMP）算法的稀疏故障诊断方法。基于COMP算法,在每次迭代后,首先根据内积大小依次计算原子与残差的相关系数,将相关系数最大的原子与其他符合条件的原子合并,将合并后的原子作为一个新原子;然后,利用这些新原子重新构成一个与信号相关度较强的新字典,对信号进行稀疏表示;最后,通过分析稀疏表示结果的包络谱实现滚动轴承故障的准确诊断。由于该方法重构的新原子与残差的相关性较强,因此只需较少的迭代次数就可得到较高的稀疏表示精度。仿真试验和工程应用验证了所提方法的有效性和实用性。     

Intelligent Fault Diagnosis of Rotary Machinery Based on Unsupervised Multiscale Representation Learning

The performance of traditional vibration based fault diagnosis methods greatly depends on those handcrafted features extracted using signal processing algorithms, which require significant amounts of domain knowledge and human labor, and do not generalize well to new diagnosis domains. Recently, unsupervised representation learning provides an alternative promising solution to feature extraction in traditional fault diagnosis due to its superior learning ability from unlabeled data. Given that vibration signals usually contain multiple temporal structures, this paper proposes a multiscale representation learning (MSRL) framework to learn useful features directly from raw vibration signals, with the aim to capture rich and complementary fault pattern information at different scales. In our proposed approach, a coarse-grained procedure is first employed to obtain multiple scale signals from an original vibration signal. Then, sparse filtering, a newly developed unsupervised learning algorithm, is applied to automatically learn useful features from each scale signal, respectively, and then the learned features at each scale to be concatenated one by one to obtain multiscale representations. Finally, the multiscale representations are fed into a supervised classifier to achieve diagnosis results. Our proposed approach is evaluated using two different case studies: motor bearing and wind turbine gearbox fault diagnosis. Experimental results show that the proposed MSRL approach can take full advantages of the availability of unlabeled data to learn discriminative features and achieved better performance with higher accuracy and stability compared to the traditional approaches.     

改进Alexnet的滚动轴承变工况故障诊断方法

旋转机械中的滚动轴承常工作在变负荷、强噪声的环境中,而传统的滚动轴承故障诊断方法难以在复杂工况下自适应地提取对其故障诊断有利的特征,针对此问题,提出一种改进AlexNet的滚动轴承变工况故障诊断方法。首先,将采集的一维时域信号按横向插样构建便于改进AlexNet输入的二维特征图,于现存的纵向插样和二维频谱而言,保留了特征自动提取过程中振动信号的时序性和关联性;其次,改进调整AlexNet卷积层的功能层且经过卷积和次采样等操作,从二维特征图中自动提取出利于滚动轴承状态辨识的特征;最后,以softmax的交叉熵为损失函数,利用Adam按小批量迭代优化法实现对滚动轴承故障的诊断。通过与多种方法对滚动轴承不同位置、不同损伤程度的12类状态诊断效果比较,结果表明,该方法对变负荷、强噪声条件下的滚动轴承故障诊断的精度更高,鲁棒性更强。     

基于ADMM字典学习的滚动轴承振动信号稀疏分解

在稀疏分解过程中,字典模型构建的结果会直接影响稀疏分解的效果。为获得结构更好的字典,提出了一种基于交替方向乘子法(ADMM)的字典学习方法,在字典学习过程中采用交替方向乘子法逐个更新字典中原子,得到的字典具有良好的结构。将该字典学习方法应用到滚动轴承振动信号稀疏分解中,能获得更快的字典学习速度和更好的稀疏分解效果。与K-SVD字典学习方法相比较,证明了所提方法在轴承信号稀疏分解中的优越性。     

基于DBN的故障特征提取及诊断方法研究

随着装备日趋复杂化,依靠专家经验或信号处理技术人工提取和选择故障特征变得越来越困难。此外,以BP神经网络、SVM为代表的浅层模型难以表征被测信号与装备健康状况之间复杂的映射关系,且面临维数灾难等问题。结合深度置信网络(DBN)在提取特征和处理高维、非线性数据等方面的优势,提出一种基于深度置信网络的故障特征提取及诊断方法。该方法通过深度学习利用原始时域信号训练深度置信网络并完成智能诊断,其优势在于能够摆脱对大量信号处理技术与诊断经验的依赖,完成故障特征的自适应提取与健康状况的智能诊断,该方法对时域信号没有周期性要求,具有较强的通用性和适应性。在仿真数据集和轴承数据集上进行了故障特征提取和诊断实验,实验结果表明:本文提出的方法能够有效地从原始信号中进行多种工况、多种故障位置和多种故障程度的故障特征提取和诊断,并且具有较高的故障识别精度。     

非凸罚正则化稀疏低秩矩阵的大型减速机圆锥滚子轴承微弱故障诊断

在强烈外界噪声下或轴承故障早期发展阶段,从轴承非平稳故障信号中提取微弱冲击成分是一个难点,针对这一问题,提出了一种新的基于非凸罚正则化稀疏低秩矩阵（Non-convex penalty regularization sparse low-rank matrix,NPRSLM）的轴承微弱故障特征提取方法。该方法不依赖振动信号结构的先验知识,也无需采集大量的样本信号来训练字典,避免了传统稀疏表示设计冗余字典带来的缺乏物理意义,通用性差等缺陷。该方法的核心思想是把采集的振动信号与待提取的故障脉冲看作一维矩阵（向量）,通过求解稀疏正则化的反问题得到故障脉冲信号。在建模上,通过引入非凸罚函数代替了传统最小化L1-norm融合套索算法,建立非凸罚正则化稀疏低秩矩阵模型,理论推导了所建立模型的严格凸性,并利用交替方向乘子法（Alternating direction method of multipliers,ADMM）对模型进行求解,同时讨论了模型参数对模型算法的收敛性问题、凸性与非凸性边界取值问题等。仿真算例与大型减速机圆锥滚子轴承诊断实例表明：该方法不仅能提取隐藏在强烈外界噪声中的微弱冲击特征,而且改善了传统最小化L1-norm融合套索算法在提取微弱故障冲击时产生的脉冲能量大幅衰减与脉冲数目丢失问题。     

ARTIFICIAL NEURAL NETWORK BASED FAULT DIAGNOSTICS OF ROLLING ELEMENT BEARINGS USING TIME-DOMAIN FEATURES

A procedure is presented for fault diagnosis of rolling element bearings through artificial neural network (ANN). The characteristic features of time-domain vibration signals of the rotating machinery with normal and defective bearings have been used as inputs to the ANN consisting of input, hidden and output layers. The features are obtained from direct processing of the signal segments using very simple preprocessing. The input layer consists of five nodes, one each for root mean square, variance, skewness, kurtosis and normalised sixth central moment of the time-domain vibration signals. The inputs are normalised in the range of 0.0 and 1.0 except for the skewness which is normalised between −1.0 and 1.0. The output layer consists of two binary nodes indicating the status of the machine—normal or defective bearings. Two hidden layers with different number of neurons have been used. The ANN is trained using backpropagation algorithm with a subset of the experimental data for known machine conditions. The ANN is tested using the remaining set of data. The effects of some preprocessing techniques like high-pass, band-pass filtration, envelope detection (demodulation) and wavelet transform of the vibration signals, prior to feature extraction, are also studied. The results show the effectiveness of the ANN in diagnosis of the machine condition. The proposed procedure requires only a few features extracted from the measured vibration data either directly or with simple preprocessing. The reduced number of inputs leads to faster training requiring far less iterations making the procedure suitable for on-line condition monitoring and diagnostics of machines.     

基于最优 IMF 分量和 K-SVD 的滚动轴承故障声音信号特征提取

针对滚动轴承声音信号中周期性冲击故障特征难提取的问题,提出了基于最优 IMF 分量与 K-SVD 字典学习相结合的轴承故障特征提取方法。首先,利用 VMD 分解原始信号获得一系列 IMF 分量;其次,利用 SAF 指标自适应选取最优 IMF 分量,并作为训练信号;最后,利用 K-SVD 字典学习方法训练出字典库,通过正交匹配追踪算法( OMP )对原始信号处理得到稀疏信号,并对稀疏信号进行包络谱分析。仿真及实验结果表明,对比传统 K-SVD 字典学习方法,该方法得到的稀疏信号信噪比( SNR )更高,能更准确地提取滚动轴承周期性冲击,增强了轴承故障特征。     

基于压缩感知弱匹配追踪算法的信号特征提取

针对滚动轴承故障诊断中的特征提取问题,提出一种基于压缩感知弱匹配追踪算法的特征提取方法。针对轴承故障信号特征特点构建了一个由傅里叶字典和冲击时频字典组成的联合字典,作为弱匹配追踪算法中的过完备冗余原子库。进而利用改进的简化粒子群寻优算法在联合字典原子库中寻找最能匹配轴承故障信号特征的原子,实现故障信号的快速高效稀疏分解。在信号重构阶段提出了一种改进的阈值降噪策略,解决了软阈值降噪存在恒定偏差以及硬阈值降噪的不连续问题。对CWRU（Case Western Reserve University）轴承数据中心所提供的标准轴承故障信号和某钢厂滚动轴承实测信号进行了仿真,仿真结果验证了该方法的优越性。     

基于拉普拉斯分值和模糊C均值聚类的滚动轴承故障诊断

针对滚动轴承故障振动信号的非平稳特征和故障征兆的模糊性,提出了基于拉普拉斯分值和模糊C均值（FCM）聚类的滚动轴承故障诊断方法。该方法首先在时域和频域对滚动轴承振动信号进行特征提取,组成初始特征向量;然后利用拉普拉斯分值进行特征选择,形成故障特征向量;最后以FCM聚类为故障分类器,实现滚动轴承不同故障类型的识别。应用实例和对比实验表明,该方法能有效提取滚动轴承振动信号特征,诊断滚动轴承故障。     

基于非线性模式分解的旋转机械复合故障特征提取方法

针对旋转机械复合故障振动信号的非平稳特征,提出了一种基于非线性模式分解（NMD）的故障特征提取方法。该方法首先通过NMD将振动信号分解为若干个具有实际物理意义的非线性模态(NM)分量和一个残余分量之和,然后对各NM分量采用包络谱分析提取故障特征。仿真信号的分析结果验证了NMD方法的优越性,在此基础上将NMD方法应用于旋转机械复合故障诊断中,实验数据的分析结果表明,该方法能有效提取出旋转机械复合故障的特征。     

Sparse signal decomposition method based on multi-scale chirplet and its application to the fault diagnosis of gearboxes

Based on the chirplet path pursuit and the sparse signal decomposition method, a new sparse signal decomposition method based on multi-scale chirplet is proposed and applied to the decomposition of vibration signals from gearboxes in fault diagnosis. An over-complete dictionary with multi-scale chirplets as its atoms is constructed using the method. Because of the multi-scale character, this method is superior to the traditional sparse signal decomposition method wherein only a single scale is adopted, and is more applicable to the decomposition of non-stationary signals with multi-components whose frequencies are time-varying. When there are faults in a gearbox, the vibration signals collected are usually AM-FM signals with multiple components whose frequencies vary with the rotational speed of the shaft. The meshing frequency and modulating frequency, which vary with time, can be derived by the proposed method and can be used in gearbox fault diagnosis under time-varying shaft-rotation speed conditions, where the traditional signal processing methods are always blocked. Both simulations and experiments validate the effectiveness of the proposed method.