基于ICEEMDAN-MPE和GWO-SVM的滚动轴承故障诊断方法

针对滚动轴承故障状态难以准确且快速的识别,提出了一种基于改进自适应噪声完备集成经验模态分解(Improved Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise, ICEEMDAN)-多尺度排列熵(Multi-Scale Permutation Entropy, MPE)和灰狼算法优化支持向量机(Grey Wolf Optimization Algorithm-Support Vector Machine, GWO-SVM)结合的故障诊断方法。首先将轴承信号进行ICEEMDAN分解,然后选取其中相关性较大的IMF(Intrinsic Mode Function)分量计算多尺度排列熵构成特征集合,最后通过GWO-SVM算法进行故障状态识别。通过滚动轴承数据集和不同算法的对比实验,验证了ICEEMDAN-MPE-GWO-SVM方法的有效性,表明该方法可以准确且快速的诊断滚动轴承的故障情况。     

基于倒阶比谱的滚动轴承故障诊断

为从变速滚动轴承振动信号中提取轴承故障特征,利用了基于倒阶比分析的诊断方法.该方法利用计算阶比跟踪算法,采集转速信号和振动信号,并对振动信号进行角度域重采样,获得角度增量恒定的角度域周期平稳信号,再进行倒谱分析,最后依据倒阶比谱分析结果进行故障识别和诊断.由于阶比分析能将非平稳时域信号转换为平稳的角度域信号,而倒谱分析能很好的抑制频率模糊,所以将阶比分析和倒谱分析相结合,充分利用二者优点,能从干扰中准确识别故障特征,适合于变速滚动轴承信号的故障特征提取.实验证明该方法有效,能够提取变转速滚动轴承故障特征,具有一定的工程应用价值.     

基于MR-DCA的滚动轴承微弱故障诊断

【目的】针对滚动轴承微弱故障难以识别的问题,提出了一种基于MR-DCA的滚动轴承故障诊断方法。【方法】利用最大相关峭度解卷积以及共振稀疏分解的方法对输入样本进行预处理,可以有效地滤除原信号中的噪声,突出故障冲击成分。将所获得的故障分量的二维时频图以及原始信号作为网络的训练样本,经两个特征学习模块后,使用注意力机制对输入特征进行筛选,通过权值重分配可以有效地提高模型计算效率和识别精度。为了验证模型性能,使用某大学的滚动轴承微弱故障数据进行故障诊断分析,同时开展消融实验,对诊断模型各个模块的有效性进行验证。【结果】结果表明,所提出的方法识别准确率更高,且具有更快的训练速度和迭代速度。【结论】所提模型在进行滚动轴承微弱故障诊断时可以实现良好的诊断性能。     

基于深度卷积自编码器和多尺度残差收缩网络的滚动轴承寿命状态识别

针对滚动轴承早期故障识别困难、退化性能难以准确评估的问题,提出了基于深度卷积自编码器(DCAE)和多尺度残差收缩网络(MSRSN)的滚动轴承寿命状态识别方法。首先,为获得清晰的故障特征频率及倍频,将原始数据样本转换为包络谱输入深度卷积自编码器中,实现轴承寿命状态特征的自动提取与表达,并基于多维尺度分析(MDS)算法约简寿命状态特征获得低维特征,然后计算低维特征空间内样本间的欧几里得距离(ED),即为轴承性能衰退评估指标;其次,为全面提取轴承性能衰退特征,提出了改进的多尺度残差收缩网络识别模型,并开发了ReLU与DropBlock正则化相结合的新激活策略增强模型的抗噪性;最后,将所提方法及对比方法应用于轴承全寿命实验数据。实验结果表明：笔者提出的性能衰退评估指标能够精准地识别轴承性能退化起始点以及刻画轴承的退化趋势,所提出的改进的多尺度残差收缩网络识别模型在S_SNR=-4～6 dB环境中平均识别正确率为91.75%,能够准确识别轴承寿命状态,验证了方法的实用性以及有效性。     

结合CEEMDAN和灰度关联分析方法的滚动轴承性能退化评估

针对大型机械设备中滚动轴承容易发生故障的问题,提出一种将自适应噪声的完备经验模态分解(CEEMDAN)和灰度关联分析相结合的滚动轴承性能退化评估方法。首先利用CEEMDAN对轴承全寿命周期的振动信号进行分解,得到能量熵特征,其次以正常状态下的特征矢量作为灰度关联分析的参考序列,然后计算轴承全寿命周期内的特征矢量与正常特征矢量的关联度,作为性能退化过程的定量评估指标,结果表明该方法能及时发现早期故障,并能很好的描述轴承退化的各个阶段。最后利用基于CEEMDAN和Hilbert包络解调的方法对评估结果的正确性进行了验证。     

一种基于EMD的共振解调改进方法

针对传统共振解调方法中共振频带需要人为确定和故障轴承振动信号信噪比低的缺陷,提出了一种基于EMD算法的共振解调改进方法。该方法首先对轴承振动信号进行EMD分解,然后自适应地筛选出高频固有振动频率附近的基本模式分量（IMF）,并对单分量进行滤波处理,最后重构选取滤波后的基本模式分量,并对重构信号进行包络解调分析,得到故障特征频率和故障类型。滚动轴承故障诊断表明,改进方法不仅能够自适应地确定共振频带,而且可以有效地提取故障特征,识别故障类型。     

基于隐马尔科夫模型的滚动轴承性能衰退评估

针对滚动轴承全寿命周期内健康状态的变迁和性能衰退的识别和评估问题,引入隐马尔科夫模型(hidden Markov model,HMM),利用Baum-Welch算法对滚动轴承全寿命周期振动信号数据进行建模,利用Viterbi算法解算和检验模型,最后通过Forward-Backward算法计算测试数据的概率分布;通过校验数据概率分布的隐含序列和观测序列分别表示状态变迁概率和性能衰退评估结果.研究结果表明:该方法能够快速解算滚动轴承状态迁移的概率分布,有效识别性能衰退状态,为预防性维修提供参考.     

VW-MSPC方法在变压器状态评估中的应用研究

针对变压器状态评估变量指标之间具有相关性以及重要性不同的特点,在基于马氏距离的传统多元统计控制图中引入加权马氏距离,设计一种基于变量加权的多元统计过程控制的变压器评估方法。该方法结合关联规则方法给出的变压器关键指标变量体系利用多元控制图的波动检出能力,将变压器各类故障对应的多元实时监测数据转化为一个检验统计量,进而对变压器运行状态进行评估。电流回路过热故障的实例分析表明,相对于现行的变压器状态评估标准,该方法具有更快的故障检出速度;对于已知故障的样本数据的回顾检测表明该方法具有较高的故障检出率。     

融合PHOG和MSLBP特征的铁路扣件检测算法

为了提高铁路扣件检测的识别率和鲁棒性,以及扣件图像PHOG特征的有效性,提出了简单有效的枕肩定位算法,该算法首先在提取PHOG特征前,根据枕肩、扣件和背景间的位置关系去除冗余背景信息;然后,模拟人眼视觉注意机制,设计MSLBP特征采样方式,提取扣件图像的宏观纹理特征;最后,采用分层次加权融合的方法联立两类特征,并采用SVM分类器进行扣件分类识别,提出一种基于计算机视觉和PHOG-MSLBP融合特征的缺陷识别算法.将该算法应用于实验,结果表明:与使用PHOG、MSLBP单一特征相比,基于PHOG-MSLBP融合特征检测算法的平均识别率分别提高了6.3%、4.5%,且鲁棒性更强,可满足扣件缺陷自动化检测的需要.     

基于虚拟仪器的机车滚动轴承故障诊断研究

介绍了机车滚动轴承故障诊断的机理。利用LabVIEW软件和相关硬件,设计了一种机车滚动轴承故障诊断系统。该系统结合某一型号机车轮对,利用CompactDAQ和NI9233加速度采集模块采集滚动轴承振动信号。根据滚动轴承机理,振动信号包含滚动轴承故障信息。因此,运用小波变换函数,对振动信号进行时频域分析处理,获取有用的故障特征信息,进而确定故障类型。从实验结果可以看出,该方法能准确地找到故障频率,对于机车滚动轴承故障诊断有很好的指导意义。     

Monotonicity Evaluation Method of Monitoring Feature Series Based on Ranking Mutual Information

As a prerequisite for effective prognostics, the goodness of the features affects the complexity of the prognostic methods. Comparing to features quality evaluation in diagnostics, features evaluation for prognostics is a new problem. Normally, the monotonic tendency of feature series can be used as the visual representation of equipment damage cumulation so that forecasting its future health states is easy to implement. Through introducing the concept of ranking mutual information in ordinal case, a monotonicity evaluation method of monitoring feature series is proposed. Finally, this method is verified by the simulating feature series and the results verify its effectivity. For the specific application in industry, the evaluation results can be used as the standard for selecting prognostic feature.     

Graph and hint based algorithm for machining feature automation recognition and mapping

Feature recognition is a process of extracting machining features which has engineering meaning from solid model, and it is a key technology of CAD/CAPP/CAM integration. This paper presents an effective and efficient methodology of recognizing machining feature. In this approach, features are classified into two categories: pocket feature and predefined feature. Different feature type adopts its special hint and heuristic rule, and is helpful to recognize intersection feature. Feature classification optimizes search algorithm and shortens search scope dramatically. Meanwhile, extension and split algorithm is used to handle intersecting feature. Moreover, feature mapping based on machining knowledge is introduced to support downstream application better. Finally, case studies with complex intersecting features prove that the developed approach has stronger recognizing ability.     

智能化滚珠丝杠副退化状态的评估方法

现有滚珠丝杠副退化状态评估方法通常假设已有充足且带标签的数据集,但实际工程应用中故障成本过高、获取标签难度过大,难以在特定工况下获得大量带标签数据集. 针对上述问题,提出一种基于多尺度对抗域对抗学习的智能化状态评估方法,结合注意力卷积神经网络模块和域对抗学习模块,利用不同工况下采集的传感器信号建立深度学习模型,从而自适应地学习域不变特征并实现高效的知识复用和特征迁移;利用多工况下采集的滚珠丝杠副退化信号构建试验数据集来验证方法的有效性. 研究结果表明:本文方法在6个标签缺失跨工况条件下的滚珠丝杠副退化状态识别子任务中均取得了高于89.02%的识别准确率;能够充分迁移带标签数据的关键特征,实现了标签样本缺失条件下目标工况退化状态识别.      

分层CS-LBP和HOG特征融合的人脸识别研究

针对单独的纹理特征只能提取图像的纹理信息而不能得到图像轮廓边缘信息的问题,在人脸识别的特征提取研究中提出了分层CS-LBP和分层HOG特征的融合方法.对图像分别进行多次CS-LBP和HOG特征的提取,得到分层CS-LBP特征提取图像和分层HOG特征提取图像,对分层的特征提取图像再次提取分层CS-LBP和分层HOG特征,并将两种分层特征进行融合,得到更有效的人脸的纹理及边缘轮廓特征.在ORL和GT人脸库上的实验结果验证了所提出的分层特征融合方法的有效性.     

基于弹性网格和方向线素特征的脱机手写汉字识别

弹性网格和方向线素特征已广泛应用于许多手写汉字识别系统中,并认为是目前较成熟和有效的手写体汉字特征.本文将传统弹性网格特征和方向线素特征相结合然后运用BP神经网络对其进行识别.实验证明,本文特征提取方法在识别率上优于传统的弹性网格方法.     

Latent Supportive Utility of Irrelevant Attributes in Feature Selection

This paper proposed a novel feature selection method LUIFS (latent utility of irrelevant feature selection) that not only selects the relevant features, but also targets at discovering the latent useful irrelevant attributes by measuring their supportive importance to other attributes. The method minimizes the information lost and simultaneously maximizes the final classification accuracy. The classification error rates of the LUIFS method on 16 real-life datasets from UCI machine learning repository were evaluated using the ID3, Nave-Bayes, and IB (instance-based classifier) learning algorithms, respectively; and compared with those of the same algorithms with no feature selection (NoFS), feature subset selection (FSS), and correlation-based feature selection (CFS). The empirical results demonstrate that the LUIFS can improve the performance of learning algorithms by taking the latent relevance for irrelevant attributes into consideration, and hence including those potentially important attributes into the optimal feature subset for classification.     

Cyclostationary property based spectrum sensing algorithms for primary detection in cognitive radio systems

To implement the primary signal without interference in cognitive radio systems, cognitive radios can detect the presence of the primary user in low SNR. Currently, energy detector is the most common way of spectrum sensing because of its low computational complexity. However, performance of the method will be possibly degraded due to the uncertainty noise. This paper illustrates the benefits of one-order and two-order cyclostationary properties of primary user’s signals in time domain. These feature detection techniques in time domain possess the advantages of simple structure and low computational complexity comparing with spectral feature detection methods. Furthermore, performance of the one-order and two-order feature detection is studied and the analytical results are given. Our analysis and numerical results show that the sensing performance of the one-order feature detection is improved significantly comparing with conventional energy detector since it is robust to noise. Meanwhile, numerical results show that the two-order feature detection technique is better than the one-order feature detection. However, this benefit comes at the cost of hardware burdens and power consumption due to the additional multiplying algorithm.     

A Novel Method for Character Segmentation in Natural Scenes

Introduction Nowadays the applications of scene text recog-nition are rapidly expanding with the developmentof portable digital imaging devices[1]. However,character extraction from scene images has alwaysbeen a challenging problem due to complex back-ground, uneven illumination, shadows and noise ofimages[2]. Besides, languages also impose anotherlevel of variation in text. Characters in such lan-guages as Chinese, Japanese and Korean are usual-ly composed of several strokes, which do not n…     

邮件过滤系统中预处理方法的研究与实现

完成了邮件过滤系统中的预处理工作。实现了信息增益特征选择算法,通过实验对比,得出了PU系列语料库合适的特征维数。使用词频反文档频率公式计算了特征词的权重,通过算法把标准邮件集处理成了支持向量机算法可以直接处理的向量空间模型的形式。     

基于时频感知加权的车辆路面冲击声品质评价

为表征与量化人对路面冲击声的主观感受，首先，对减速带工况冲击非平稳噪声信号进行声时感知时长定义，同时根据人耳听声可辨性将声时历程分为冲击段、峰值段及衰减段；进而，以小波变换提取冲击噪声中的主冲击与多重微冲击特征信息，组成冲击声品质评价的基础特征阵；然后，类比峰值因子法定义频域滤波因子，并基于序关系分析法确定时变感知加权系数，组建时频滤波网络对基础特征阵加权且建立冲击声品质时频感知评价指标；最后，基于实车过减速带冲击噪声测试数据计算声品质指标，并进行对比验证.研究结果表明：所提时频感知加权评价指标与主观评价的相关系数在车速20 km/h时为0.927，在车速30 km/h时为0.922；在考虑路面冲击声声时历程全程评价时，经典的声品质评价指标（特征频带时变响度）与主观评价的相关系数在车速20 km/h时为0.933，在车速30 km/h时为0.649；所提时频感知加权评价方法对于车速为20 km/h与30 km/h的情况具有较好的适用性.