Improvement on On-line Ferrograph Image Identification
Improvement on On-line Ferrograph Image Identification
复制标题
DOI：
10.3901/cjme.2010.01.001
复制
发表时间：
2010
影响因子：
4.2
通讯作者：
Tonghai Wu
中科院分区：
文献类型：
--
作者：
Tonghai Wu
新开发的在线可视铁谱仪（OLVF）为发动机磨损状态监测提供了新的途径。然而，在线磨损碎片图像处理的可靠性在船舶发动机监测和卡特彼勒台架测试中都受到挑战，这在之前的研究中没有报道。在监控引擎和处理图像方面遇到了两个问题。首先，监测一段时间后，细小的磨损碎片就很难从图像背景中识别出来。其次，由于运行一段时间后油液变黑，背景噪声大大降低了磨屑的识别精度。因此，对图像处理中采用的方法进行了研究。磨粒识别中出现问题的两个主要原因概括如下。一般而言，二值阈值由图像全局像素决定，容易受到图像中非目标区域的影响。由于监测过程中油色变浅，二值图像中目标区域的边界被误识别。因此，进行了如下改进。通过扫描一列像素来识别全局二值图像中的目标区域，然后进行限制在目标区域内的二次二值处理以识别小磨损碎片。使用特定模板的线性滤波来抑制二值图像中的噪声，然后进行低通滤波以消除残余噪声。此外，通过灰堆分离各个磨粒，提取单个磨粒的形貌参数，并提出WRWR（相对磨损率）和WRWS（相对磨损严重度）两个指标进行磨损描述。为发动机在线监测提供了新的指标。
A newly developed on-line visual ferrograph(OLVF) gives a new way for engine wear state monitoring. However, the reliability of on-line wear debris image processing is challenged in both monitoring ship engines and the Caterpillar bench test, which weren’t reported in previous studies. Two problems were encountered in monitoring engines and processing images. First, small wear debris becomes hard to be identified from the image background after monitoring for a period of time. Second, the identification accuracy for wear debris is greatly reduced by background noise because of oil getting dark after running a period of time. Therefore, the methods adopted in image processing are examined. Two main reasons for the problems in wear debris identification are generalized as follows. Generally, the binary threshold was determined by global image pixels, and was easily affected by the non-objective zone in the image. The boundary of the objective zone in the binary image was misrecognized because of oil color becoming lighter during monitoring. Accordingly, improvements were made as follows. The objective zone in a global binary image was identified by scanning a column of pixels, and then a secondary binary process confined in the objective zone was carried out to identify small wear debris. Linear filtering with a specific template was used to depress noise in a binary image, and then a low-pass filtering was performed to eliminate the residual noise. Furthermore,the morphology parameters of single wear debris were extracted by separating each wear debris by a gray stack, and two indexes, WRWR (relative wear rate) and WRWS (relative wear severity), were proposed for wear description. New indexes were provided for on-line monitoring of engines.