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.     

基于BP神经网络的铁谱磨粒图像识别方法研究

在磨粒识别过程中,铁谱磨粒图像预处理和特征参数提取是关键。应用图像形态学的处理方法对磨粒图像进行预处理,结果表明,利用开运算、闭运算的图像形态学处理方法对铁谱磨粒图像进行预处理,可以消除图像二值化后留下的孤立小碎点、孔洞以及边界断点。通过磨粒图像的统计特征参数和傅里叶特征参数建立BP神经网络,并对磨粒进行识别,结果表明:采用该方法能正确识别磨粒图像,辨别磨损机制。     

一种基于小波理论的铁谱图像反锐化掩模增强法

介绍了当前常用的图像增强技术,结合铁谱图像预处理研究了基于小波域的反锐化掩模法在图像增强中的运用,并给出了具体算法。实验结果表明,基于小波分析的图像反锐化掩模增强算法能够在控制噪声的同时有效地突出图像细节,改善图像质量。小波变换使得原始图像中不同分辨率的细节特征随尺度的不同而分离开来,避免了传统算法中通过不断调整滤波器窗口大小来选择增强效果的繁琐工作;运用不同尺度下的小波分量分别进行增强,原始图像中无论较粗还是较细的边缘都能同时得到增强,从而既能有效地检测出边缘又能突出图像的表面纹理细节。     

多层次信息融合在铁谱图像磨粒识别中的应用

针对铁谱图像磨粒识别中异类信息综合利用率较低的问题,提出多层次信息融合的铁谱图像磨粒识别方法。首先,在铁谱图像二值化分割的基础上进行二值滤波,结合彩色铁谱图的R、G、B三分量,实现铁谱图像的彩色滤波。其次,以实际采集的磨粒图像样本为例,提取滤波后二值图像的形态特征,以及滤波后彩色图像的颜色特征;在特征层利用PCA对异类特征进行维数约简,并结合SVM和k-fold交叉验证,实现形态特征和颜色特征的特征层融合;在决策层将异类特征的SVM概率输出结果作为D-S证据理论的基本概率分配函数,实现形态特征和颜色特征的决策层融合。通过与形态学滤波结果对比,验证了本文提出滤波方法的优越性;其次,不同层次的信息融合结果表明,与单独使用颜色特征和形态特征相比,异类信息融合后可实现优势互补,有效提高故障磨粒的识别准确率。     

图像颜色特征提取在铁谱图像分类及磨粒识别中的应用研究

铁谱图像所包含的彩色信息对磨粒识别、磨损形式分析非常重要。对图像分类技术进行研究，主要对铁谱图像的颜色特征进行研究，提出了聚类树分析、模糊聚类技术与统计分析相结合的定量研究方法，对铁谱图像进行背景、磨粒区域分割，以获得可进行定量分析的磨粒。计算的颜色特征为铁谱图像的进一步处理和识别以及磨粒的机器自动识别、磨损形式分析奠定了基础。     

铁谱磨粒多重分形特征研究

为有效描述铁谱磨粒特征,提出用多重分形谱参数表达磨粒形态特征的新方法。选择盒计数法计算磨粒图像的多重分形谱,研究磨粒多重分形谱的有效性,分析磨粒多重分形谱参数的不变性和鲁棒性;确定磨粒图像预处理方法,并对4类典型磨粒的多重分形谱参数进行统计分析。结果表明:将多重分形谱参数应用于磨粒识别,总识别率为82.5%。磨粒具有明显的多重分形特性,可用多重分形谱参数来描述磨粒的形态特征;多重分形谱参数具有平移不变性,但对灰度变化和噪声干扰的鲁棒性较差,在提取多重分形谱参数时,需要对磨粒图像做严格的预处理。     

A New On-Line Visual Ferrograph

An on-line visual ferrograph (OLVF) characterized by direct reading and on-line analysis was developed based on magnetic deposition and image analysis. A digital sensor was integrated with a CMOS image sensor to obtain images of deposited wear debris under illumination conditions. An electromagnetic instrument was designed to deposit the wear debris flowing through an oil flow channel. The oil flow channel, fixed on the electromagnet, was arranged parallel to the magnetic flux in the air gap between two electromagnet poles. The deposition effect on wear debris was analyzed theoretically. The result shows that the wear debris in different sizes can be deposited in the same zone by controlling the oil flow rate and magnet field intensity. Corresponding application software for image sampling and processing was developed. An index of relative wear debris concentration, IPCA (Index of Particle Coverage Area), is given as an output in addition to wear debris images. Finally, two kinds of experiments were specified to assess the effect and validity of the OLVF. The results show that the OLVF has effective deposition and identification for both relatively large and small wear debris with rational control parameters. The validity examinations with the commercial particle quantifier (PQ) and direct reading ferrograph (DR) show that the OLVF has an approaching trend to the reference instruments in both heavily and lightly contaminated oil.     

旋转式铁谱仪生成谱片观察规范的建立

本文以磨屑的群体分析理论为依据,以旋转式铁谱仪生成的谱片为研究对象。对于任意一个油样,为了反映油样中磨屑群的群体特征,运用数理统计的有关原理和方法,以谱片上大磨屑对全体磨屑的面积比和大磨屑的数量两个群体特征参数为例,论证了所需谱片的最小数目以及每一张谱片上所需观察的最小视场的数目,据此建立了旋转式铁谱仪生成谱片的观察规范。     

炼化设备油液分析诊断软件开发

介绍炼化设备油液分析诊断软件,该软件由铁谱定量分析模块、自动铁谱显微图像分析模块、油液分析数据库模块、磨损状态评判模块和诊断报告输出模块组成,通过铁谱分析仪、铁谱显微镜获取油样磨损信息,监测大型炼化设备润滑油品质量,并对磨损状况智能分级,为设备维护和管理提供依据。     

机械设备磨损故障的分类与铁谱诊断的探讨

根据智能化铁谱诊断的要求,本文给出了机械设备磨损故障的定义并对机械设备磨损故障进行了系统的分类。提出将铁谱诊断水平划分成三个级别,综合分析了铁谱诊断方法,给出了一个智能化铁谱诊断模型。     

基于差商的油液监测铁谱图像自适应分割

针对油液监测中铁谱磨粒图像分割阈值难以选取的问题,本文提出一种基于差商的自适应铁谱图像分割算法。首先,将铁谱磨粒灰度图像转换成三维灰度直方图,并对其进行切片分析;然后,引入Newton插值多项式,将不同切片所得的频数作为切片灰度-频数曲线的插值点,基于差商构造第一类可接受函数和第二类可接受函数,结合实验数据确定两类误差,选取同时满足两类误差的最小灰度值作为分割阈值;最后,用本文方法对不同类型的磨粒图像以及添加高斯噪声和椒盐噪声后图像分别进行分割实验,并与经典的迭代阈值法、Otsu算法、最大熵法进行了比较。实验结果表明,本文方法受噪声干扰较小,误检率和漏检率整体优于其他3种算法。对分割所得的磨粒图像进行特征提取,并利用支持向量机进行识别,本文方法对3种故障磨粒识别准确率最高,达到82.86%,虽在运行时间上无明显优势,但综合性能最优,能满足油液监测过程中铁谱图像自适应分割的需求。     

基于Mask R-CNN的铁谱磨粒智能分割与识别

针对铁谱图像因背景复杂、尺寸分布广、颗粒重叠等导致难以精确分割与识别的问题,以相似度高的疲劳剥块、严重滑动磨粒、层状磨粒共3种异常磨粒作为研究对象,提出基于深度神经网络模型Mask R-CNN的对多目标铁谱磨粒进行智能分割与识别的方法,并对特征提取层分别选用深度不同的残差网络ResNet50和ResNet101进行对比试验。实验结果表明,基于迁移学习方法的Mask R-CNN+ResNet101模型能够在复杂背景下对多目标、多类型、多尺寸的相似磨粒进行有效分割与识别,测试集的平均精度高达76.2%,模型具有较好的泛化能力。     

基于图像分析和野点检测的航空发动机磨损状态识别

研究了基于发动机滑油滤磨屑图像的磨损状态自动识别技术。首先采用最大熵法和数学形态学方法,提取滑油滤磨屑图像中反映磨损状态的特征量;然后采集反映正常状态的航空发动机滑油滤图像,通过图像分析与特征提取,构造出仅包含正常样本的训练样本集,最后用野点检测方法对训练样本进行学习,并使用遗传算法对野点检测参数进行优化,得到了滑油滤磨屑图像的正常域,并以此来识别航空发动机磨损状态的严重程度。开发了发动机滑油滤监控系统(engine oil filter monitoring system,EOFMS),实现了基于野点检测的磨屑图像识别功能,并利用实际航空发动机滑油滤磨屑图像进行了实验分析,结果验证了该方法的有效性。     

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。     

Ferrographic Analysis of Wear Debris from Boundary Lubrication Experiments with a Five Ring Polyphenyl Ether

The types of wear particles generated by a five-ring polyphenyl ether in boundary lubrication experiments in various atmospheres were determined by ferrographic analysis. The types of wear particles observed included: cylindrical or rock-like organo-metallic debris, adhesive and cutting wear particles, and some spherical debris. Interpretations as to the mechanism of generation of the various types of particles are presented.     

图像可视在线铁谱传感器的图像数字化处理技术

为实现图像可视在线铁谱传感器的磨粒图像自动辨识,建立数字图像获取系统,探讨铁谱图像数字化处理方法。研究了铁谱图像的预处理方法,对比在RGB和YUV颜色空间对铁谱图像的灰度化处理效果,采用不同的微分模板对平滑后图像进行锐化处理;探讨减背景法和自动阈值法在铁谱图像磨粒分割中的应用效果;给出适用于在线铁谱图像的定量描述方法。研究表明,采用YUV颜色空间的明视度分量可以得到平滑的灰度图像,合理的模板选择可以使微分法在锐化磨粒边缘的同时保持整体图像的平滑;铁谱图像的磨粒分割结果表明,减背景法由于采用人工选取门限值而难以适用于在线铁谱图像的处理,而自动阈值法可以根据铁谱图像自动选取合适的阈值以达到良好的分割效果;采用磨粒百分覆盖面积作为定量指标可反应良好分割的铁谱图像中的磨粒统计质量分数。     

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.     

Decision of lubricated friction conditions for materials of automobile transmission gear using neural network

It is hard to inspect the state of lubrication of an automobile transmission visually. Thus, it is necessary to develop a new inspection method. Wear debris can be collected from the lubricants of an operating transmission of an automobile, and its morphology will be directly related to the friction condition of the interacting materials from which the wear debris originated in the lubricated transmission. In this study, wear debris in lubricating oil are extracted by membrane filter (0.45 μm), and the quantitative values of shape parameters of wear debris are calculated by digital image processing. These shape parameters are studied and identified by an artificial neural network algorithm. The results of the study may be applicable to the prediction and diagnosis of the operating condition of transmission gear.     

BP神经网络在基于铁谱技术的磨损状态评价体系中的应用

根据智能化铁谱状态监测的要求，将对设备磨损状态反应比较敏感的三个定量铁谱参数作为输入参数，应用BP神经网络对设备的磨损状态进行了自动识别。经过现场验证，此方法结合定性铁谱分析可以较准确地评判设备的磨损状态级别。