【作者】 张龙；

【导师】 王静秋；

【作者基本信息】 南京航空航天大学 ， 机械设计及理论， 2013， 硕士

【摘要】 磨损是造成机械设备故障和失效的主要原因之一，对机械设备进行磨损工况监测和故障诊断可以实现设备的预防性维修，减少停机损失并保证人员与设备安全，避免突发故障和灾难的发生，具有重要意义。铁谱分析是上世纪70年代出现的一种磨损颗粒分析和机器状态监测技术，是目前最经济且有效的磨损检测方法之一。铁谱图像处理与分析是现代铁谱分析技术的关键环节。实现铁谱图像中磨粒的准确分割，对后续的磨粒分析和识别，以及磨损状态监测和故障诊断有重要的意义。本文主要对铁谱图像磨粒分割和磨粒边缘检测进行了研究，具体如下：(1)针对铁谱图像中较难分割的磨粒沉积链和异常大磨粒，提出融合改进分水岭变换和蚁群算法(Combining Watershed andAnt ColonyAlgorithm,CWACA)的铁谱图像分割方法。首先，采用彩色梯度算子获得铁谱图像的梯度图，采用标记分水岭变换对开闭重建后的铁谱梯度图像进行处理，实现磨粒的初步分割；其次，将颜色差异项和动态搜索半径引入到蚁群聚类中，通过改进蚁群聚类准则，合并相似区域，实现异常大磨粒的分割。但是，因为磨粒沉积链上许多细小的磨粒彼此连接、性质相近甚至相同，因此也有可能被错误地合并到一起。因此在进行蚁群聚类后，还需要利用形状参数——长宽比，对聚类结果进行判别和修正，从而实现不同类型磨粒的准确分割。最后，将本文算法与大津阈值法、K均值聚类和模糊C均值聚类算法进行比较，实验结果证明，本方法是一种对磨粒沉积链和异常大磨粒进行分割的有效方法。(2)本文利用蚁群算法对磨粒边缘进行检测。首先利用阈值法提取包含磨粒边缘信息的目标点，将蚂蚁分布在目标点上，可以将蚂蚁快速引导到磨粒边缘周围；然后利用像素梯度值和颜色差异大小引导蚂蚁移动，从而实现磨粒边缘检测；最后探讨了蚁群算法中不同参数对检测结果的影响。本课题以Visual C++6.0为平台，利用OpenCV函数库实现各算法。实验结果证明了本文提出的图像分割和边缘检测方法的可行性和有效性。更多还原

【Abstract】 Wear is one of the main reasons caused mechanical equipment failure. Taking wear conditionmonitoring and fault diagnosis of mechanical equipment is significant, because it can achieve thepreventive maintenance of equipment, reduce the cose of downtime, ensure the safety of personneland equipment and avoid the occurrence of sudden failure and disaster. Ferrography is a wear particleanalysis and machine condition monitoring technology which appeared in the1970s. Currently, it hasbecome one of the most economic and effective methods of wear detection. The processing andanalysis of ferrograph image is the key of modern ferrograph technology. Achieving the accuratesegmentation of wear debris has an important impact on the subsequent particle analysis andrecognition, even the wear condition detection and fault diagnosis of a mechanical system.In this paper, the researches of wear debris segmentation and edge detection have been studied,as follows:(1) Aims at the accurate segmentation of wear debris chains and the abnormal large wear debriswhich are difficult to segment in ferrograph image, this paper presented a ferrograph imagesegmentation method combining watershed and ant colony algorithm. Firstly, the gradient image iscomputed by the presented colorful gradient operator, and then it is segmented by the labelingwatershed algorithm, this step can achieve the pre-segmentation of wear debris. Secondly, thedynamic searching radius and regional color difference have been introduced to the ant colonyclustering algorithm, the similar regions are merged by the improved ant colony clustering criteria,and in this step the segmentation of abnormal large wear debris is obtained. However, because thereare so many small homogeneous or adherent wear particles that some of them may be mistakenlymerged together. In order to achieve the accurate segmentation of the different type of wear particles,the discrimination and correction of clustering results should be taken by using the shape parameterslike aspect ratio after the ant colony clustering. Finally, the proposed algorithm is compared with Otsuthreshold method, K-means clustering and Fuzzy C-means clustering, the experiments show that thismethod is an effective way in the segmentation of wear debris chain and abnormal large wear debris.(2) In this paper, the ant colony algorithm is adopted in wear particle edge detection. Firstly, thetarget pixels which contain the edge information of wear particles can be extracted by thresholdmethod, and then the ants are distributed in these pixels. Secondly, the ants move in the guide of pixelgradient values and color difference to achieve the edge detection of wear debris. Finally, this paper discusses the influence of different parameters to the detection results.In this paper, Visual C++6.0is used as a software platform and OpenCV library used to fulfilledthe algorithms. The experimental results demonstrate that the methods of image segmentation andedge detection presented in this paper are feasible and effective.更多还原