支持向量机在油液磨粒在线监测系统中的应用

在发动机油液在线监测系统中,运用支持向量机的方法对油液中的磨粒进行分类识别,并结合运用最近邻法对分类器的训练过程进行优化;其中基于支持向量机的磨粒分类器的输入为磨粒的主轴长度、孔隙率、圆度、角二阶矩、梯度熵和纹理相关性等参数,输出为滑动磨粒、切削磨粒、球状磨粒和疲劳剥块4种磨粒种类;搭建油液在线监测实验平台进行磨粒分类识别实验,结果表明,基于支持向量机的磨粒分类器的分类准确率高达94.7%,并且由于最近邻法的使用分类器的处理速度也提高了30%.     

基于支持向量机的铁谱磨粒模式识别

将支持向量机方法用于铁谱磨粒模式识别，以磨粒样本的圆形度、细长度、散射度和凹度4个形态特征量作为支持向量机分类器的输入，以滑动磨损、切削磨损、正常磨损和疲劳点蚀4种磨损形式作为分类器的输出，建立基于支持向量机的磨粒分类器；研究支持向量机中误差惩罚系数和核参数对磨粒分类器的性能影响；通过实验比较了基于支持向量机与基于BP神经网络的磨粒分类器的性能，结果表明，基于支持向量机的磨粒分类器分类准确率为96％，基于BP神经网络的磨粒分类器分类准确率为90％。     

油液在线监测系统中的磨粒识别

针对机械设备磨损状态监测要求,构建了基于显微图像分析的油液在线监测系统.根据系统的光路特点,对磨粒图像进行了基于彩色特征的转换,并通过与背景图像的差值处理来快速提取磨粒目标.基于最小二乘支持向量机设计了两类磨粒分类器,并利用粒子群优化算法对最小二乘支持向量机模型中的参数进行了优化选取.在此基础上,根据磨粒识别体系,设计了磨粒综合分类器.最后,利用铁谱分析技术对系统性能和识别效果进行了检验,结果表明,系统的识别精度达到95%以上,满足磨粒在线监测要求.     

基于支持向量机的产品编号识别

为了提高仓库管理系统的性能,将支持向量机用于产品编号的模式识别。采用基于投影法的图像处理算法提取编号数字;对倾斜的数字进行矫正,并对提取的数字进行归一化;构造支持向量机分类器对归一化的数字进行识别;通过对一组数字样本的测试,分析了支持向量机参数与分类器的识别率的关系。测试结果表明,支持向量机分类器可以在小样本的情况下获得较高的识别率。     

油液在线监测系统中磨粒识别技术研究

针对磨损状态监测要求，构建了基于显微图像分析的油液在线监测系统。根据系统光路特点，对磨粒图像进行了基于彩色特征的转换，并通过与背景图像的差值处理来快速提取磨粒目标。基于最小二乘支持向量机设计了磨粒两类分类器，并利用粒子群优化算法对最小二乘支持向量机模型中的参数进行了优化选取；根据磨粒识别体系，设计了基于最小二乘支持向量机的磨粒综合分类器。最后，利用铁谱分析技术对系统性能和识别效果进行了检验，结果表明本系统具有较高的检测精度和识别效果。     

基于分层模糊支持向量机的油液磨粒自动识别

提出基于模糊支持向量机的机械设备在用油液磨粒自动识别方法。首先利用K-均值聚类算法对磨粒图像进行分割,提取磨粒的形状尺寸特征参数、边缘细节特征参数、表面纹理特征参数作为其量化表征,分别选择最能反映待识别磨粒特征的参数作为各个二分类器的输入向量;然后结合二叉树法和一对多法间接构造磨粒的分层多类别分类器模型,在训练过程中同时利用粒子群算法优化分类器的参数,建立一种参数自适应的模糊支持向量机分层多类别分类模型。将该模型应用到旋挖钻机在用油液的磨损颗粒识别中,识别率最高达90%。该模型结构简单、分类精度好,在磨粒识别领域较大的工程应用价值。     

基于支持向量机与最近邻分类器的模拟电路故障诊断新策略

针对模拟电路的故障诊断和定位问题,提出了一种基于支持向量机分类器(SVC)和最近邻分类器(NNs)的模拟电路故障诊断新策略,利用SVC解决高维故障样本的分类问题,而采用NNs解决故障样本间的重叠问题。首先建立"1-v-r"结构的SVC对电路故障样本进行训练,并根据训练参数构建故障字典;其次,在测试阶段,根据算法决定利用SVC或NNs对未知样本进行测试。本文设计的故障分类器方法简单,结构确定,通过对两个模拟电路的实验表明,所提出的方法性能要优于常规的"1-v-r"支持向量机分类器;与"1-v-1"支持向量机分类器的诊断性能较为接近,但测试时间较后者显著减少,较为适合模拟电路的故障诊断。     

基于遗传算法的多尺度支持向量机及其在机械故障诊断中的应用

通过对支持向量机核函数的分析发现,当对样本的各个特征赋予不同大小的尺度参数时,可以避免冗余特征干扰分类,增强关键特征在分类中的作用,提高支持向量机分类器的学习和泛化能力。在此基础上,提出一种具有不同特征尺度参数的支持向量机(简称多尺度支持向量机),并通过遗传算法最小化LOO(leave-one-out)泛化错误上限估计,根据各个特征的识别能力赋予其不同大小的尺度参数。将多尺度支持向量机用于轴承故障诊断,实验结果表明,与传统的单尺度参数支持向量机相比,多尺度支持向量机具有更好的泛化能力。对压缩机气阀的故障识别表明,尺度参数的大小直接反映了对应特征识别能力的大小,因此可以依据尺度参数的大小进行特征选择,保留关键特征,剔除冗余特征。     

基于多主元特征与支持向量机的动态过程质量异常监控模型

为简化多支持向量机识别模型的计算复杂度、提高动态过程质量异常模式的识别精度,提出一种基于多主元特征与支持向量机相结合的动态过程异常监控模型。利用主元分析方法对动态数据进行特征提取,将所提取的不同主元特征作为支持向量机分类器的输入对模型进行训练。将识别效率高的主元特征对应的转换矩阵与多支持向量机相结合,构建了基于多主元特征的多支持向量机识别模型,对质量异常模式进行识别。仿真实验表明,所提基于多主元分析支持向量机识别模型的识别精度比传统基于主元特征或其他特征提取方法的识别模型有显著提高,且训练所需时间大大减少。     

一种用于磨粒识别的基于改进PSO 算法的支持向量机模型

为提高磨粒智能识别的准确率,以传统支持向量机和粒子群优化(PSO)算法为基础,提出一种基于改进PSO算法的支持向量机(SVM)识别模型。该识别模型的惩罚参数和核函数参数可同时得到最佳优化,从而可建立模型参数最优的自适应SVM识别模型。采用该识别模型对油液中的磨粒进行智能识别,结果表明该模型识别准确率高达98%,明显优于BP神经网络模型。     

铁谱磨粒形态分形特征参数提取方法研究

阐述了分形与分形维数的定义，给出了铁谱磨粒分形维数和无标度区间的计算原理。编制了磨粒图像处理及磨粒分维计算的计算程序，并以片状磨粒为实例，提取了该磨粒的轮廓分维特征、磨粒表面纹理特征、表面纹理方向特征与磨粒表面间隙度特征四个特征值，为有效识别磨粒提供科学依据。     

Intelligent identification of wear mechanism via on-line ferrograph images

Condition based maintenance（CBM） issues a new challenge of real-time monitoring for machine health maintenance. Wear state monitoring becomes the bottle-neck of CBM due to the lack of on-line information acquiring means. The wear mechanism judgment with characteristic wear debris has been widely adopted in off-line wear analysis; however, on-line wear mechanism characterization remains a big problem. In this paper, the wear mechanism identification via on-line ferrograph images is studied. To obtain isolated wear debris in an on-line ferrograph image, the deposition mechanism of wear debris in on-line ferrograph sensor is studied. The study result shows wear debris chain is the main morphology due to local magnetic field around the deposited wear debris. Accordingly, an improved sampling route for on-line wear debris deposition is designed with focus on the self-adjustment deposition time. As a result, isolated wear debris can be obtained in an on-line image, which facilitates the feature extraction of characteristic wear debris. By referring to the knowledge of analytical ferrograph, four dimensionless morphological features, including equivalent dimension, length-width ratio, shape factor, and contour fractal dimension of characteristic wear debris are extracted for distinguishing four typical wear mechanisms including normal, cutting, fatigue, and severe sliding wear. Furthermore, a feed-forward neural network is adopted to construct an automatic wear mechanism identification model. By training with the samples from analytical ferrograph, the model might identify some typical characteristic wear debris in an on-line ferrograph image. This paper performs a meaningful exploratory for on-line wear mechanism analysis, and the obtained results will provide a feasible way for on-line wear state monitoring.     

Collection of solid and gaseous samples to diagnose inertial confinement fusion implosions

Collection of representative samples of debris following inertial confinement fusion implosions in order to diagnose implosion conditions and efficacy is a challenging endeavor because of the unique conditions within the target chamber such as unconverted laser light, intense pulse of x-rays, physical chunks of debris, and other ablative effects. We present collection of gas samples following an implosion for the first time. High collection fractions for noble gases were achieved. We also present collection of solid debris samples on flat plate collectors. Geometrical collection efficiencies for Au hohlraum material were achieved and collection of capsule debris (Be and Cu) was also observed. Asymmetric debris distributions were observed for Au and Be samples. Collection of Be capsule debris was higher for solid collectors viewing the capsule through the laser entrance hole in the hohlraum than for solid collectors viewing the capsule around the waist of the hohlraum. Collection of Au hohlraum material showed the opposite pattern: more Au debris was collected around the waist than through the laser entrance hole. The solid debris collectors were not optimized for minimal Cu backgrounds, which limited the conclusions about the symmetry of the Cu debris. The quality of the data limited conclusions on chemical fractionation effects within the burning, expanding, and then cooling plasma.     

Enhancement of oil debris sensor capability by reliable debris signature extraction via wavelet domain target and interference signal tracking

On-line oil debris monitoring is an effective approach to detecting machine component wear through estimating the size and the quantity of metallic debris in the lubricating oil. However, oil debris (particle) signatures are often contaminated by background noise and vibration interference during the operation of the oil debris sensor. As such, the accuracy of debris measurement and counting depends largely on the authenticity of the extracted debris signature. Considering characteristics of both target and interference signals obtained by the oil debris sensor, we propose a novel debris signature extraction technique to improve the oil debris measurement capability based on the wavelet domain information. In each wavelet scale of the oil debris sensor output signal, the debris coefficients are detected based on the singularity of the debris signal. The interference coefficients are estimated by adaptive linear prediction. The overlapped debris and interference coefficients are separated by a new prediction strategy involving alternating applications of forward and backward predictors. The differences between the mixture and the estimated interference coefficients are employed to reconstruct the debris signature. The proposed technique is evaluated using both uni- and bi-excitation experimental data and compared with a recently reported method. The experimental results and comparisons indicate that the proposed new method can extract the debris signature more truthfully, and thus improve the oil debris monitoring accuracy in real applications.     

Ferrography—then and now

Since 1971, when ferrography was first introduced, there have been many developments in the techniques used to process samples and also the way in which debris is analysed and interpreted in relation to the underlying causes that produced them. The science and application of ferrography embraces both the engineering and medical fields and this has led to new and exciting ways of understanding the associated phenomena and how to extract the best information for the advancement of the subject matter.The purpose of this special issue is to provide an overview of some of the principal developments that have taken place over the past three decades, thereby placing on record what is widely acknowledged to be one of the most significant developments in the field of wear debris technology.     

Microscopic changes in surface topography and elemental concentration on the track formed by the intermittent motion of friction

A preliminary investigation of the molecular weight of polytetrafluoroethylene (PTFE) wear debris was undertaken. It was found that the molecular weight of PTFE wear debris generated from thrust washers was drastically reduced compared with that of the thrust washer. A consistent difference in the molecular weight of the wear debris was observed in samples obtained at high and at low pressure velocities. A fluorocarbon composite material yielded wear debris of molecular weight even lower than that of unfilled PTFE wearing below its limiting pressure-velocity.     

A comparative evaluation of RPD and ferrographic diagnostic methods for turbine engine lubricant samples

A rotary particle depositor (RPD) was evaluated and compared with analytical (AF) and direct reading (DR) ferrographs to ascertain if the RPD is better suited for analysis of wear debris in turbine engine lubricant samples than the AF and DR. Sieved iron particles were added to synthetic turbine engine lubricants at various concentrations and were analysed by RPD and ferrographic methods and compared. Lubricant samples containing a fine test dust were also analysed by RPD and AF to determine the effectiveness of the RPD's centripetal acceleration and the AF's gravity flow in eliminating non-ferrous contamination from the sample. Finally, comparisons were made between actual lubricant samples from a turbine engine simulator analysed by RPD and ferrographic methods. For lubricant samples with relatively small amounts of non-ferrous contamination or samples where non-ferrous particles are important in machine health condition assessment, the AF and DR ferrographs are equal or superior to RPD for lubricant analyses. For samples with high levels of non-ferrous contamination, the RPD is superior in eliminating unwanted non-magnetic particles and permitting a less obstructed view of the important ferrous wear debris particles.     

Surface roughness evolutions in sliding wear process

Wear debris analysis is a technique for machine condition monitoring and fault diagnosis. One key issue that affects the application of wear debris analysis for machine condition monitoring is whether the morphology of the wear particles accurately depicts their original states and the surface morphology of the components from which the particles separate. This study aimed to investigate the evolution of the surface morphology of wear debris in relation to change in the surface morphology of wear components in sliding wear process. Sliding wear tests were conducted using a ball-on-disc tester under proper lubrication and improper lubrication conditions. The study of the particle size distribution and the surfaces of both the wear debris and the tested samples in relation to the wear condition and the wear rates of the wear components were carried out in this study. The evolutions of the surface topographies of both the wear debris and the wear components as wear progressed were investigated. This study has provided insight to the progress of material degradation through the study of wear debris. The results of this research have clearly demonstrated that: (a) there is a good correlation of the surface morphology of wear debris and that of the wear components, and (b) the surface morphology of wear debris contains valuable information for machine condition monitoring.     

On the Friction of Carbon Black- and Silica-Reinforced BR and S-SBR Elastomers

Friction of carbon black- and silica-reinforced elastomers is studied experimentally and theoretically, using Persson’s model. The effect of reinforcement fillers on elasticity was determined by dynamical mechanical analysis. Carbon black-filled samples have a higher Young’s modulus than the silica-filled compounds. Silica-filled rubbers have a higher tan (δ) at lower temperatures and a lower loss tangent at higher temperatures, which is a rough indication for higher wet grip and lower rolling resistance, respectively. Friction tests on a ball-on-disk test rig were performed to study the effect of the reinforcement fillers and their amount on the friction between rubber samples (disks) and relatively smooth or rough granite surfaces (balls). The results were discussed and compared with the friction model presented by Persson. It was shown theoretically and experimentally that hysteresis does not play a significant role in the friction of rubber samples in contact with smooth granite and that it plays a minor role in contact with a rough granite sphere. Therefore, the hysteresis contribution of friction can be neglected in the contact of rubbers with just smooth spheres. Moreover, a higher friction coefficient is seen in samples with a higher content of fillers. Silica-filled compounds show a slightly higher coefficient of friction compared with the carbon black-filled compounds. The effect of attached wear debris to the granite surfaces on the friction level has been studied. The results are supported by SEM and confocal microscopic images of the wear debris itself and wear debris attached to the granite spheres, respectively.