铁谱磨粒图像的计算机图像处理

本文从计算机图像处理的系统开始，系统地介绍了在铁谱磨粒图像分析过程中，进行图像处理的各个环节。 应用分布直方图法去除磨粒图像的背景，运用阈值分割法区分了两种不同颜色的磨粒，并且介绍了提高图像质量的三种滤波方法。通过边界的跟踪搜索方法获得的能完全代表磨粒几何形状特性的边界链码，为进一步的研究奠定了基础。     

磨损磨粒的BP神经网络识别

介绍了铁谱分析技术对设备状态监测与故障诊断的方法;通过机械润滑油或液压油中微观磨损颗粒的分析来判断机器当前的工作状态。铁谱的计算机图像分析技术是近年来研究的热点。基于BP神经网络对磨损磨粒进行识别,提出了磨粒的分步识别策略,并以磨粒样本都对网络进行训练,取得了较好的识别效果。     

回归分析在铁谱技术中的应用

阐述了应用铁谱技术监测内燃机车柴油机运行状况，着重用一元线性回归制定直读铁谱磨粒浓度与机车走行公里的控制图。如果磨粒浓度位于危险线上，则预报柴油机摩擦副出现故障。     

计算机磨粒识别技术研究进展

近二十年来,铁谱图像识别技术取得了很大的发展,在图像获取、图像处理与图像特征提取等方面都有了显著的改善,随着计算机信息技术的发展,磨粒识别速度与识别正确率也有了显著提高,本文这些成绩进行了一些总结。同时指出目前还需努力的方向。     

铁谱磨粒磁化方向的自动识别方法

本文介绍了在磨粒图像中判断磨粒的磁化方向的算法，在普通的铁谱图像中，先分别计算出每一个磨粒的重心，再根据点到点的关系，求出每一个磨粒的方向，最后归纳出整个铁谱图像的磁化方向，它为判断机械部件的摩擦磨损情况提供了较好的依据。     

铁谱磨粒图像的计算机纹理分析

利用计算机图像处理技术，采用灰度共生矩阵方法提取典型磨粒图像的纹理特征参数，作为判断机械设备故障机理的判据之一。通过正常、滑动、切削、疲劳磨粒灰度共生矩阵纹理分析，得到熵、能量、惯性矩等参数的范围。     

发动机铁谱磨粒分析与磨粒识别研究

从发动机摩擦副及其摩擦磨损特点着手，提出了用于发动机磨损状态监测与诊断的主要10类磨粒，依据其识别特征提出了用于磨粒识别的14个磨粒形状、表面纹理与颜色特征；最后应用BP网络分层识别策略进行了磨粒识别。     

模糊散度理论在铁谱磨粒识别中的应用

从模糊集合的理论出发,将对称交叉熵和模糊散度理论应用于铁谱磨粒识别,分析比较了两种不同模糊隶属度函数条件下的图像分割效果,最后提出了用图像的骨架变化来提取铁谱磨粒图像形状特征的方法.     

粗糙集和证据理论在磨粒识别中的应用

为了便于对磨粒进行识别，首先利用粗糙集理论对磨粒参数信息进行约简，并形成待决策问题的经验决策表。然后，利用粗糙集理论和证据理论的关系，计算待决策信息的有关证据的基本概率指派和条件概率指派。最后，按照合成规则对上述条件概率指派进行合成，并根据决策规则对磨粒进行分类。     

基于BP神经网络的铁谱磨粒分类器设计

磨粒类型识别与分类是铁谱技术的主要内容之一,本文基于神经网络原理,探讨了磨损磨粒分类识别的神经网络模型和实现方法。     

基于磨粒分析的设备状态监测技术

对当前各种磨粒监测技术的原理、特点、适用范围进行了系统评述，并展望了磨粒监测技术发展的新思路。     

基坑变形人工神经网络预测及其网络参数优化

基坑变形人工神经网络预测受网络参数的影响较大，选取适当的网络参数才能得到较优的预测结果。本文介绍了人工神经网络原理及其网络参数的优化方法。以挡土桩桩顶水平位移预测为例，说明其具体预测步骤及网络参数优化方法。     

镍基合金涂层滑动磨损过程中磨屑的演变机理

在真空熔结镍基合金涂层(成分是0．65％C-24％Cr—3．5％Si-3．5％B-3．0％Mo-6％Fe-Ni余量)与65Mn钢组成的滑动磨损系统中，研究了磨屑的演变机理。磨损过程中磨屑不仅存在氧化、碎化等化学、机械过程，还存在因相互挤压、粘结甚至微区热压烧结等冶金过程，两方面的共同作用，使氧化物磨屑成分均匀化。应用Fe-Ni-O系统相图对高温磨损磨屑进行相分析。     

可共享的磨粒识别及磨损诊断系统

针对当前铁谱分析技术存在的不足 ,将计算机图像处理技术、数据库技术、网络通信技术相结合 ,提出了以人工智能、神经网络为核心的设计思想。介绍了基于本思想所建立的可共享的磨粒识别及磨损诊断系统WPRWDS,分析了其系统结构、实现机制、硬软件设计、功能和特点等。     

Weibull分布在基于磨屑群理论的铁谱技术中的应用

中叙述了磨屑群理论,并以其作为铁谱技术的理论依据,通过分析Ｗeibull分布的方差发现：方差的改变仅与磨屑群中大磨屑个数的改变有关,这与磨屑群理论中强调大磨屑的个数的观点是相符的,因此可以采用Ｗeibull分布作为描述润滑油中磨屑尺寸分布规律的数学工具,并据此对设备的状态进行监测。     

Applications of Artificial Intelligence to Grinding Operations via Neural Networks

In recent years, artificial intelligence played an important role in machine tool automation. Artificial neural networks, as one of the artificial intelligence algorithms, has superiority in representing the relation between the inputs and outputs of the multi‐variable system. Hence, it can be applied to sophisticated operations such as grinding operation. The aim of this research is to use artificial neural networks as the brain of grinding machine controller. The target of this controller was to achieve the desired workpiece surface roughness under grinding wheel surface topography variations. The core of the system consists of two multi‐layers feed forward artificial neural networks based on back error propagation learning algorithm. The first one was used for process design to achieve the desired surface roughness. It extracts suitable process variables such as grinding wheel speed and feed rate. The second one monitors the cutting operation using sensors' readings. It extracts the different controlling decisions; these are accept the process, redesign the process or start dressing operation under automatic control. According to these decisions, a PC master control program generates the appropriate control codes and sends them to the machine controllers to take the required actions.     

Applications of Artificial Intelligence to Grinding Operations via Neural Networks

In recent years, artificial intelligence played an important role in machine tool automation. Artificial neural networks, as one of the artificial intelligence algorithms, has superiority in representing the relation between the inputs and outputs of the multi-variable system. Hence, it can be applied to sophisticated operations such as grinding operation. The aim of this research is to use artificial neural networks as the brain of grinding machine controller. The target of this controller was to achieve the desired workpiece surface roughness under grinding wheel surface topography variations. The core of the system consists of two multi-layers feed forward artificial neural networks based on back error propagation learning algorithm. The first one was used for process design to achieve the desired surface roughness. It extracts suitable process variables such as grinding wheel speed and feed rate. The second one monitors the cutting operation using sensors' readings. It extracts the different controlling decisions; these are accept the process, redesign the process or start dressing operation under automatic control. According to these decisions, a PC master control program generates the appropriate control codes and sends them to the machine controllers to take the required actions.     

Structure and Micromorphology of Wear Debris of MC Nylon 6 under Dry Sliding: Correlation with Wear Mechanisms

The structure and micromorphology of wear debris of MC nylon 6 under dry sliding were investigated by FTIR, XRD, DSC, and FESEM, and the 3D surface topographies of friction materials before and after the friction test were observed, which will be helpful in understanding the friction and wear processes. The primary crystalline phase of both the unworn MC nylon 6 and the wear debris were α crystal, but the crystallinity of the latter was higher than that of the former. The proportion of α ₂ (002 + 202) planes increased and the reflection from the α ₁ (200) planes was suppressed in the wear debris, indicating a preferential arrangement of α ₂ (002 + 202) on the surface of the wear debris. The transition in structure of the wear debris originated from the activation of the chain segments due to the thermodynamic effects. The thermodynamic effects and high chain segment mobility resulted in the hydrogen bonding whose interchain distance is a larger rupture or even chain scission. MC nylon 6 was severely worn due to the contribution of the tearing force that resulted from the combined action of the tribo-interface adhesion and the shearing effect during friction, whereas no damage happened on the worn surface of the counterpart steel pin even if under severe sliding conditions.     

基于自组织特征映射神经网络的零件图像的识别

提出了基于小波变换提取零件图像特征和用自组织特征映射神经网络实现识别的方法，首先，对零件图像进行小波多尺度边缘检测，提取零件图像的边缘轮廓；然后将被检测的边缘轮廓图像分成若干个子区域并分别统计各子区域的边缘像素量，各子区域中的相对边缘像素系数作为零件的特征，将这些特征作为神经网络的输入样本，由自组织特征映射神经网络实现识别。实验结果表明该方法是有效的。     

基于神经网络的工程机械磨损故障诊断研究

根据工程机械结构特点、摩擦磨损规律和磨粒特征建立了标准磨粒谱 ;提出了基于神经网络的磨粒识别技术 ,设计了智能磨粒识别系统 ,诊断实例表明 ,用神经网络方法可以准确地识别工程机械磨损故障类型、程度和部位