WU Tonghai , WANG Weigang WU Jiaoyi MAO Junhong and XIE Youbai

1 Theory of Lubrication and Bearing Institute, Xi’an Jiaotong University, Xi’an 710049, China

2 Post-doctoral Research Center of Material Science and Engineering, Xi’an Jiaotong University,Xi’an 710049, China

3 School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200030, China

1 Introduction

Information extraction was always the focus of on-line ferrography. To this end, efforts were made to extract the features on the concentration, size, and material of the wear debris in lubricant[1–4]. Profound progress was achieved in wear analysis by ferrograph image investigation in off-line studies[5–6]. However, only the concentration and little morphology information of wear debris were reported by the on-line ferrography based on various signal analyses[7–10]. A newly developed on-line visual ferrograph(OLVF) sensor gives a new solution to this problem[11–12]. An on-line wear debris image provides more wear information of the running condition than an off-line one. Therefore, on-line image processing for the feature identification and wear description is of great meaning for practical applications.

The information extraction from on-line ferrograph images was preliminarily investigated in a previous study[12]. An index of particle coverage area(IPCA)reflecting relative wear debris concentration was extracted for wear degree reporting. However, the method exhibited many defects in the applications of the engines of ship craft and Caterpillar bench test. The main problems are outlined as follows.

(1) Oil color varies during engine running, which confuses the boundary identification of the objective zone in ferrograph images.

(2) Self-segmentation algorithm shows poor antiinterference ability in the case of oil darkening.

(3) IPCA gives wear debris information but no oil usage information which is found important in engine bench test.

In this research, efforts were made to improve information extraction from on-line ferrograph images for engineering application. In section 2, by using some engineering on-line images, image graying and de-noising methods are investigated for improving the accuracy and reliability of identification. In section 3, more detailed features of wear debris are studied for wear description. To facilitate engineering application, quantitative descriptors are constructed for statistically depicting the wear status and oil usage condition.

2 Identification of On-line Ferrograph Image

2.1 Image segmentation

The on-line ferrograph image from OLVF sensor is stored as a colorful image. So graying is first needed to transfer the colorful image into a binary one. Then wear debris is segmented from the background in the binary image. The two main steps of image segmentation are the objective zone identification and the self-adapting wear debris segmentation.

In a previous study[13], the gray value of each pixel was extracted from an on-line colorful image. Binary was carried out in the gray image with a global self-adapting threshold optimization. Then line scanning of pixels was carried out in the binary image, and the boundary of the objective zone was determined by the rule of gray values jumping abruptly. As a final result, an IPCA was computed within the objective zone as the relative wear debris concentration index.

Two problems were encountered in the on-line monitoring of ship craft engines. The first can be illustrated in Fig. 1, which shows an on-line ferrograph image sampled during the primary running stage. Large amounts of small wear debris were produced as shown in Fig. 1(a).The corresponding binary image is shown in Fig. 1(b).Comparison between the initial and processed image shows that much information of the small wear debris was lost during the binary processing. The second problem appeared after the engine ran for a period of time.

Fig. 1. On-line ferrograph image of ship craft engine

Fig. 2 shows the on-line ferrograph image sampled from the engine of ship craft after running 100 h. Oil color changed with rising temperature, as shown in Fig. 2(a).Also the brightness of the magnetic pole zone in the image changed because of the light scattering in the oil. The corresponding binary image, in Fig. 2(b), shows many white pixels included in the black background of the magnetic pole zones.

Fig. 2. On-line ferrograph image of ship craft engine and its binary result

A very large IPCA (45%) was computed although no wear debris was found explicitly in Fig. 2. Because of the variation of the oil color, large numbers of white points in the magnetic pole zones were produced by gray and binary processing. According to the rule of the objective zone identification, the first white point was judged as the boundary of the objective zone during line scanning.Therefore, a mistaken boundary was located in the magnetic pole zone. Such a mistaken identification led to the fact that the large black area of the magnetic pole zones was identified as wear debris, thus a big error was inevitably introduced into the IPCA computation result.

To ameliorate these problems, an improved image segmentation method was proposed with column pixel scanning and secondary binary processing. Global graying was performed in an on-line ferrograph image to get a gray image. Then the binary processing was performed also with a global self-adapting threshold to get a binary image.

Column pixel scanning was carried out in the binary image from up to down and from left to right. The number of the white pixels was accumulated during scanning. The boundary of the objective zone was determined when the ratio of white pixel number to the column pixel number was over 50%. Then the objective zone confined by the two boundaries was extracted from the initial on-line image. A binary processing was performed only in the gray space of the objective zone. Fig. 3 shows the processed result of the on-line image in Fig. 1(a). The IPCA and binary threshold of the image in Fig. 1(a) with the improved and previous method are compared in Table. The two methods produce vastly different results. Also the image in Fig. 2 was re-computed and the IPCA was 0% which agreed with the visual result.

Fig. 3. Processed result of the on-line image in Fig. 1(a)

Table. Binary threshold and IPCA of ferrograph images

2.2 Image de-noising

Image noise is unavoidably introduced during on-line image sampling. In practice, carbon build-up always darkens the engine lubricant, which is the main noise source of the on-line image.

Fig. 4 shows the on-line ferrograph image sampled from Caterpillar engine bench test and the corresponding binary result. The oil became darkening after running about 60 h,which caused the close color of the background and the wear debris in the on-line ferrograph image. Fig. 4(b) gives the evidence of background noise introduced by darkening oil flow.

De-noising of the on-line ferrograph image seems to be necessary for the engine monitoring. Morphology and de-noising methods were used in this work.

As a typical morphology method, corrosion-expansion was reported with a good depression effect on the background noise in off-line ferrograph images[13]. The binary image in Fig. 4(b) was processed with corrosionexpansion, and the result is shown in Fig. 5(a). The background noise was well depressed. However, some small wear debris and the boundary details of large wear debris were mistakenly eliminated as noise.

Linear filtering was adopted with a 5×5 filter template as shown in Fig. 6(a). The purpose of filtering is to determine with a criterion whether the current point marked by “0” is wear debris or not. In fact, the pixel points in a binary image are divided into two groups by their gray values: the white points of the background and the black points of wear debris or noise. The background points can be distinguished from the binary image by their color.

Fig. 4. Binary results of on-line ferrograph image

Therefore, the emphasis of the filtering is to distinguish the wear debris from the noise point. With the template in Fig. 6(a), the pixels in a binary image were grouped by column and line directions. Set the current pixel point as the point zero in the template, and make a judgment with other pixels corresponding with the template points. If at least one wear debris point for each point group of 1 and 2,3 and 4, 5 and 6, 7 and 8, can be found, the point zero is taken as a noise point and then set as a background point. If not, the point zero is taken as a wear debris point. In another report[13], the grouped pixels in column or line were considered, which is suitable for large wear debris.

The filtering result of Fig. 4(b) is shown in Fig. 5(b). The background noise was depressed to some degree and the boundary information was well preserved. However, some traces of background noise were still remained in the image after linear filtering. Therefore, a low-pass filter with a 5×5 filter template, as shown in Fig. 6(b), was utilized. Pixel scanning with the template was performed to determine whether the current pixel point marked by zero is wear debris or not. The determinate criterion was whether the number of the wear debris pixels surrounding the current pixel point accumulates over 18. The result is shown in Fig. 5(c). Comparatively Fig. 5(c) shows a better de-noising result and gives more boundary information of wear debris.

Fig. 5. Filtering results of the image in Fig. 4(b)

Fig. 6. Template of linear and low-pass filtering

3 Wear Feature of On-line Ferrograph Image

Ferrograph image can provide comprehensive information of wear status by feature extraction. On-line wear debris contains more information of tribosystem.Therefore, the feature extraction of on-line wear debris was studied in relation to wear debris and oil usage.

3.1 Wear debris area and number counting

Boundary chain tracking and area integration are traditional methods for single wear debris identification[14–15].However, some difficulties were encountered for such methods in processing an on-line ferrograph image containing large amounts of wear debris. Therefore, a new method, named gray stack, is designed for on-line identification of wear debris group. The principle of gray stack is depicted in Fig. 7. For a digital ferrograph image,each wear debris pixel is composed by three construct components of R, G, and B with their values ranging from 0 to 255. Thus a wear debris pixel can theoretically have as many as 2563unique values. If each wear debris corresponds with a numbered layer of the gray stack, a large stack can be established for wear debris storage.

Fig. 7. Principle of gray stack used in wear debris segmentation for on-line ferrograph image

Linear scanning was first performed for a binary image with principle from left to right and from up to down.When the first black pixel was encountered, a seeking was performed for all pixels of the wear debris. The wear debris was numbered and the corresponding gray value for all component pixels of was computed with corresponding RGB values in the image. Each component pixel of the wear debris was put into the same stack layer of the wear debris number until all pixels of the wear debris were stored. In the same way, all wear debris was segmented from the binary image with different gray values. A gray stack for all the numbered wear debris composed of all numbered component pixels was constructed.

The advantage of gray stack is that no repeated segmentation is required for every feature computation,which is very useful for on-line monitoring. The area and circumstance of each wear debris can be extracted with the pixels in its gray layer.

The roundness of wear debris is defined as

where C is the roundness of wear debris, A is the area of wear debris, and P is the circumstance of wear debris.

The equivalent diameter of wear debris is defined as

where D is equivalent diameter.

3.2 Wear index of on-line ferrograph images

The features of wear debris are widely used in wear reporting in analytical ferrograph. However, the image of characteristic wear debris is hardly captured by the on-line ferrograph. The statistic description for wear debris group is practical in engineering application[15].

Two statistic descriptors were constructed for the wear debris group in an on-line ferrograph image as follows:

where WRWRand WRWSare relative wear rate and relative wear severity respectively, Aland Asare the areas of large and small wear debris respectively, L is the image length along the oil flow, and V is sampling volume.

Oil transparency is also a possible parameter for on-line ferrograph images, which gives the information of lubricant physicochemical prosperities. The quality of on-line ferrograph images is greatly affected by oil transparency.Therefore, the gray value of binary image, defined as the index of transparency, was proposed.

Fig. 8 shows the variation of oil transparency during the Caterpillar diesel bench test. Correspondingly, the oil darkened with engine run time and became bright at every oil replacement, which indicated that the transparency agreed well with the oil usage degree.

4 Conclusions

(1) Column pixel scanning shows higher identification accuracy of objective zone boundary of a gray image than line scanning. And a secondary binary processing in an identified objective zone can improve the segmentation of small debris by eliminating the interference of the magnetic zone.

Fig. 8. Transparency variation of the oil in Caterpillar engine bench test based on on-line ferrograph images

(2) De-noising methods for on-line ferrograph images were studied. Linear filtering with definite template can depress most introduced noise points in a binary image and a low-pass filtering can well eliminate residual noise points.

(3) A gray stack was proposed to facilitate the feature extraction of wear debris of on-line ferrograph images. And two morphologic parameters were extracted for wear description.

(4) To facilitate engineering application, quantitative descriptors of WRWRand WRWSwere constructed for the statistic description of wear status. On the other hand, the oil transparency characterized by the image gray value was proposed for oil usage degree description.

[1] XIAO Hanliang. The development of ferrography in China—some personal reflections[J]. Tribology International, 2005, 38(10):904–907.

[2] ROYLANCE B J. Ferrography—then and now[J]. Tribology International, 2005, 38(10): 857–862.

[3] JONES M H. Tribology-a key element in condition monitoring[C]//Proceeding of International Conference on Condition Monitoring, Oxford, UK, 2001: 29–40.

[4] PENG Zhongxiao, KIRK T B. Computer image analysis of wear particles in three dimensions for machine condition monitoring[J].Wear, 1998, 223(1–2): 157–166.

[5] YUAN Changing, PENG Z, YAN Xinping, et al. Surface roughness evolutions in sliding wear process[J]. Wear, 2008, 265(3–4):341–348.

[6] MYSHKIN N K, MARKOVA L V, SEMENYUK M S, et al. Wear monitoring based on the analysis of lubricant contamination by optical ferroanalyzer[J]. Wear, 2003, 255(7–12): 1 270–1 275.

[7] RAADNUI S, KLEESUWAN S. Low-cost condition monitoring sensor for used oil analysis[J]. Wear, 2005, 259(7–12): 1 502–1 506.

[8] LIU Yan, XIE Youbai, YUAN Chongjun, et al. Research on an on-line ferrograph[J]. Wear, 1992, 153(2): 323–330.

[9] MILLER J L, KITALJEVICH D. In-line oil debris monitor for aircraft engine condition assessment[C]//Aerospace Conference Proceedings 2000 IEEE, Manhattan, USA, 2000, 6: 49–56.

[10] LIU Yan, LIU Zhong; XIE Youbai, et al. Research on an on-line wear condition monitoring system for marine diesel engine[J].Tribology International, 2000, 33(12): 829–835.

[11] WU Tonghai, MAO Junhong, DONG Guangneng, et al. Journal bearing wear monitoring via on-line visual ferrography[J].Advanced Materials Research, 2008, 44–46: 189–194.

[12] WU Tonghai, MAO Junhong, WANG Jingtao, et al. A new on-line visual ferrograph[J]. Tribology Transactions, 2009, 52(5): 623–631.

[13] WU Tonghai, QIU Hhuipeng, WU Jiaoyi, et al. Image digital processing technology for visual on-line ferrograph sensor[J].Chinese Journal of Mechanical Engineering, 2008, 44(9): 83–87.(in Chinese)

[14] KONG Xianmei, WANG Jiadao. The application of computer image processing in ferrogpraphy analysis[J]. Lubrication Engineering,2002 (2): 23–26. (in Chinese)

[15] LIU Jianghui. Research on analysis and recognition of particle image on digital image processing[D]. Beijing, Beijing Jiaotong University, 2007. (in Chinese)