Research Article The Segmentation of Wear Particles …downloads.hindawi.com/journals/at/2016/4931502.pdf · The Segmentation of Wear Particles Images Using-Segmentation Algorithm

Research ArticleThe Segmentation of Wear Particles Images Using119869-Segmentation Algorithm

Hong Liu Haijun Wei Lidui Wei Jingming Li and Zhiyuan Yang

Laboratory of Oil Analysis Shanghai Maritime University 1550 Lingang Avenue Shanghai 201306 China

Correspondence should be addressed to Haijun Wei haijun welson163com

Received 30 October 2015 Revised 24 March 2016 Accepted 31 March 2016

Academic Editor Patrick De Baets

Copyright copy 2016 Hong Liu et alThis is an open access article distributed under theCreative CommonsAttribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

This study aims to use a JSEG algorithm to segment the wear particlersquos imageWear particles provide detailed information about thewear processes taking place between mechanical components Autosegmentation of their images is key to intelligent classificationsystemThis study examined whether this algorithm can be used in particlesrsquo image segmentation Different scales have been testedCompared with traditional thresholding along with edge detector the JSEG algorithm showed promising result It offers a relativelyhigher accuracy and can be used on color image instead of gray image with little computing complexity A conclusion can bedrawn that the JSEG method is suited for imaged wear particle segmentation and can be put into practical use in wear particlersquosidentification system

1 Introduction

With the development of complex machines the demandfor effective and economic maintenance schedules has grownrapidly these years Condition-based maintenance is an ideaof using real-time data to prioritize and optimize main-tenance resources instead of regular periodic inspectionObserving the state of the system is known as conditionmonitoring Such a system will determine the equipmentrsquoshealth and act only when maintenance is actually necessary[1 2] Sounds vibrations and temperatures are informativefeatures often measured to estimates the machine conditionyet wear particle analysis holds its own advantages [3]Wear particles (or wear debris) are generated as componentsmove relative to each other The parameters that define wearparticles such as their quantity shape and size reflect thewear modes wear mechanisms and severity associated withtheir generation [4]Wear debris contained in the lubricationoil carry detailed and important information about thecondition of the machine [5] The particle characteristics aresufficiently specific so that the operating wear modes withinthe machine may be determined allowing prediction of theimminent behavior of the machine [6 7] Wear particles areusually categorized by their wear modes such as adhesiveparticles or fatigue particles

In practice wear particles are usually gathered by fer-rography which uses a high-gradient magnetic field toattract and hold particles from a fluid sample as it flowsdown a specially prepared microscope substrate [8] Theschematic of ferrography are demonstrated in Figure 1The ferrography technology allowed wear particles to beobserved and analyzed for the first time In the early daysof condition monitoring practitioners determined the wearcondition merely by the amount of wear debris The devel-opment of ferrography provided a wide array of studiesthat helped identify the characteristics of wear particlesand the mechanisms by which they generated The problemwith currently employed ferrography techniques is that theparticle morphology assessment particle classification andmachine status evaluation rely heavily on human expertisewhich is time-consuming costly and not always reliable [1]In recent years tribologists and engineers made considerableefforts to build autoclassification system for imaged wearparticles Such a system would need both tribology researchand image processing techniques Most of the work focusedon how to identify wear particles [1] It is reported that ant-colony algorithm [9] deterministic tourist walking [5] andfuzzy 119888-means algorithm [10] can be used to identify wearparticles Also in [11] attempts had been made to identify

Hindawi Publishing CorporationAdvances in TribologyVolume 2016 Article ID 4931502 10 pageshttpdxdoiorg10115520164931502

2 Advances in Tribology

(a) Analytical ferrogram

0

10

20

30

40

50

Exit end Drain tube

Nonwettingbarrier

Wear particlesnear exit are

submicroscopic

Entry region wherefluid first touches down

on the ferrogram

Dist

ance

from

exit

end

(mill

imet

ers)

(b) Deposition pattern on a ferrogram

Figure 1 Picture of analytical ferrogram and its schematic diagram

adhesive fatigue and abrasive particles using area perimeterand elongation parameters study shows that these simpleparameters can be effective for certain types of wear particlessuch as abrasive particles and sphere like fatigue particlesThesame research group also compared the different dimensionreduction methods and texture features [12 13] In the studyby [14] fractal dimension had been used to analyze weardebris surface In the study by [15] a small online monitoringsystem for gearbox have been built and gray level andintegrated morphological features have been used for wearparticlesrsquo image segmentation

The possibility of using a pattern recognition system forwear particle analysis without the need of a human expertholds great promise in the condition monitoring industryStill there are several problems that need to be resolved Oneof the first and most important of them is autosegmentationfor wear particlesrsquo image In order to autoclassify wearparticles one must first demarcate the object appearing indigital images The accuracy of segmentation will directlyaffect the subsequent feature extraction classification andidentification of wear particles Therefore wear particlesegmentation is the vital first step for wear particlesrsquo imageanalysis However the existing segmentationmethods are notadaptable for complicated ferrograph images for two reasons

(1) Wear particles deposited on the glass substrate oftenhave complicated background Large abnormal parti-cles such as severe sliding particles or chunky fatigueparticles are main target yet they are surrounded bycountless microparticles which often share similarcolor making accurate segmentation by thresholdtechnique a hard job

(2) Since the pictures are microscopic image the edges ofthe object are frequently blurred leading to difficultyin segmentation by edge detection algorithm

As demonstrated in Figure 2(b) these particles havecomplicated texture features and their background are filled

withmicroparticles and some of the edges were blurred Evenin the area of computer vision segmentation is actually oneof the central and most difficult practical problems [16 17]Yet until recently not much work has been done In [18]gray level and integrated morphological features are used tosegment the agglomerated particles yet the background ofthe image is rather simple and clean algorithms adaptive tocomplicated background shall be tested It is reported in [9]that ant-colony algorithm can be used to complete the taskyet computing efficiency needs to be improved Li et al usedmorphological erosion and dilation operations on binaryimages the edge of a single wear particle could be detectedwith the Laplace operation [19] Yet how to get accurate binaryimage in itself is a difficult problem

In this research attempts have been made to solvethe problems mentioned above improving the result ofautomatic analysis for ferrography images We represent theimproved JSEG algorithm for wear particlesrsquo color imagesThe object of this study is to see whether this algorithmcan be used in particle image segmentation Parameters weretested according to the characteristics of wear particles Acomputer program is developed to preprocess images Theimage samples used are generated using tribology testingmachines Experiment results showed that JSEG is suitablefor wear particlersquos image segmentation

2 Experiment and Method

21 Collection of Wear Particlersquos Images Wear testing is oftencarried out to study tribological performance of certainmaterials or lubricants In this research however it is donefor collecting wear particles and building a database of theirimages

The adhesive wear testing was carried out on a pin-on-disk machine which consists of a horizontal rotating discand a deadweight-loaded pin The stationary pin is stainlesssteel (hardness 175HV) and the disk is made of cast iron

Advances in Tribology 3

OptM 500x100120583m

(a)

OptM 500x100120583m

(b)

OptM 500x100120583m

(c)

Figure 2 Wear particlesrsquo images demonstration

(300HV) Pin and disc diameters were 10mm and 60mmrespectively Disc average roughness was Ra = 25120583m Thetests were conducted in a laboratory air the temperatureis 22∘C and relative humidity is 50 The load is 10 kgso the contact pressure is around 127MPa The rotatingspeed is set to 500 rpm and the time is set to 90 minutesgiving an approximate total sliding distance of 7500m Asolvent resistant PVC ring is fitted over the circumferenceof the disk to block the wear particles In order to simulateboundary lubrication a layer of CD40 lubricant is appliedto the disk in the beginning The experiment shows that thelubrication reduced the friction coefficient from average of04 to no more than 02 After the experiment the disk waswashed with acetone and the wear particles were gatheredwith ferrography

The fatigue wear particles were generated by four-balltesting machine The material of the ball is GCr15 (hardness700HV) Maximum load and speed were set to 1800N and300 rmin In order to generate fatigue wear particles thetime is set to 6 hours Before the test the balls should beultrasonic cleaned in petroleumetherTheparticles generatedwere collected and then ferrography was used to place themon amicroscopy glass slide To expand the database andmakethe result more robust particles from real-world machinesrsquolubricant system were also collected using ferrography

22 JSEG Algorithm Introduction The JSEG algorithm [20]can be separated into two parts In the first part colorsin the image are quantized to several representative classesthat can be used to differentiate regions in the image


Peer group filteringcolor space quantization

Color class-map

Spatial segmentationJ-image calculation

J-image

Region growing

Segmentation results

Wear particlesrsquo color image

Figure 3 Schematic of the JSEG algorithm for particlesrsquo image

This quantization is performed in the color space withoutconsidering the spatial distribution of the colors Then theimage pixel values are replaced by their corresponding colorclass labels thus forming a class-map of the imageThe class-map can be viewed as a special kind of texture composition Inthe next part spatial segmentation is performed on class-mapwithout considering the corresponding pixel color similarityIt is difficult task to analyze the similarity of the colors andtheir distributions at the same time The schematic of thealgorithm is shown in Figure 3

In order to remove noise in color image filter is usedsuch as vector median filter or directional distance filterThese traditional filters are implemented uniformly across thewhole image regardless of the condition each pixelThus thesefilters always modify pixels that are not corrupted by noiseIn this research a nonlinear algorithm called peer groupfiltering is used to improve the quality of segmentation Let1199090(119899) refer to an image pixel vector (in color image it contains

three values red green and blue) which characterizes thecolor information at position 119899 centered in a 119908 times 119908 windowSort all the pixels in the window according to their distancesto 1199090(119899) in ascending order and denote them as 119909

119894(119899) The

Euclidean distance measure is used here

119889119894(119899) =

10038171003817100381710038171199090(119899) minus 119909

119894(119899)

1003817100381710038171003817 119894 = 0 119896

1198890(119899) le 119889

1(119899) le sdot sdot sdot le 119889

119896(119899)

(1)

The peer group 119875(119899) of size119898(119899) for 1199090(119899) is defined as

119875 (119899) = 119909119894(119899) 119894 = 0 119898 (119899) minus 1 (2)

The filtering then can be done to the peer group membersonly instead of the pixels of whole window and thus canavoid edge blurring The problem left is how to determinethe appropriate size 119898(119899) for each peer group based on localstatistics In this study a threshold 119879(119899) is set such that

119889119898(119899)minus1

(119899) le 119879 (119899)

119889119898(119899)

(119899) gt 119879 (119899)

(3)

Here 1D distances 119889119894(119899) for Fisherrsquos discriminant estima-

tion [21] are used The criterion to be maximized is

119867(119894) =

10038161003816100381610038161198861(119894) minus 119886

2(119894)1003816100381610038161003816

2

1199041

2

(119894) + 1199042

2

(119894)

119894 = 1 119896 (4)

where

1198861(119894) =

1

119894

119894minus1

sum

119895=0

119889119895(119899)

1198862(119894) =

1

119896 + 1 minus 119894

119896

sum

119895=119894

119889119895(119899)

1199041

2

(119894) =

119894minus1

sum

119895=0

10038161003816100381610038161003816119889119895(119899) minus 119886

1(119894)

10038161003816100381610038161003816

2

1199042

2

(119894) =

119896

sum

119895=119894

10038161003816100381610038161003816119889119895(119899) minus 119886

2(119894)

10038161003816100381610038161003816

2

(5)

Themethod calculate119867(119894) for each 119894 and finds the cut-offposition where119867(119894) is the maximum that is

119898(119899) = arg max119894

119867(119894) (6)

But before searching for the cut-off position an extracalculation is done to remove impulse noise The followingtest is performed on the first and the last 119872 points of 119909

119894(119899)

usually 119872 is set to half of 119908 first-order differences of 119889119894(119899)

119891119894(119899) are calculated

119891119894(119899) = 119889

119894+1(119899) minus 119889

119894(119899) (7)

If 119891119894(119899) ge 120572 the end points 119909

119895(119899) for 119895 le 119894 or 119895 gt 119894

are considered as impulse noise and removedThe remaining119889119895(119899) are used to compute the real peer group After removal

of noise and finding cut-off position the center pixel 1199090(119899) is

replaced by the weighted average of its peer group members

119909new (119899) =sum119898(119899)minus1

119894=1

119908119894119901119894(119899)

sum119898(119899)minus1

119894=1

119908119894

119901119894(119899) isin 119901 (119899) (8)

where119908119894are the standardGaussianweights depending on the

relative positions of 119901119894(119899) with respect to 119909

0(119899)

After peer group filtering cluster technique is used forcolor quantization 119870-means start with 119896 random clustersand then iterativelymove items between clusters minimizingvariability between clusters As demonstrated in Figure 4Figure 4(a) is original image Figure 4(b) is images afterquantization and Figure 4(c) is images after both peer groupfilter and color quantization The quantization reduces thetotal kinds of color to only 15 while the texture of theparticle retains almost the original style considering the real-time image should be no more than 400 times 400 (for largenumber of pixels canmake calculation complexity grow to anunacceptable level) the imagewill lose little information aftercolor quantization Compared to result of direct quantization


(a) (b) (c)

Figure 4 The quantized wear particlesrsquo images

the proposedmethod smooths the original picture making itmore reliable for subsequent segmentation

As mentioned above the filtering and color quantizationand the total variety of color reduce to a very small numberAll the pixel colors are replaced by their corresponding colorclass labels The newly constructed image of those labels iscalled a class-map The class-map in fact reflects the texturecomposition of the image Figure 5 gives an example ofclass-map [20] for demonstration there are three labels redyellow and blue The value of each point (pixel) in the class-map is the image pixel position a 2D vector (119909 119910) Let 119885 be

the set of all 119873 data points in a class-map Let 119911 = (119909 119910) let119911 isin 119885 and let119898 be the mean

119898 =

1

119873

sum

119911isin119885

119911 (9)

Suppose 119885 has 119862 classes 119885119894 119894 = 1 119862 let 119898

119894be the

mean of the119873119894data points of class 119885

119894

119898119894=

1

119873119894

sum

119911isin119885

119911 (10)


(a) Class-map 1 119869 = 1720 (b) Class-map 2 119869 = 0855 (c) Class-map 3 119869 = 0

Figure 5 An example of different classes distribution and their corresponding 119869

Let

119878119879= sum

119911isin119885

119911 minus 1198982

119878119882=

119862

sum

119894=1

119878119894=

119862

sum

119894=1

sum

119911isin119885119894

1003817100381710038171003817119911 minus 119898

119894

1003817100381710038171003817

2

(11)

where 119878119882is the total variance of points belonging to the same

class Define

119869 =

119878119879minus 119878119882

119878119882

(12)

For the condition of an image consisting of several homo-geneous color regions the color classes are more separatedfrom each other and the value of 119869 is large On the other handif all color classes are uniformly distributed over the imagethe value of 119869 tends to be small For instance three imagesin Figure 5 all have 3 classes and the number of each classis completely the same Yet the last 119869 is the smallest becausethe classes are uniformly distributedTherefore in a particlersquosimage inside the particle texture is unified in somewaywhileoutside the particle it is completely different Moreover local119869 value can be very large near region boundariesWe can nowthink of constructing an imagewhose pixel values correspondto these 119869 values over small windows centered at the pixelsThese images are called 119869-image The higher the local 119869 valueis the more likely the corresponding pixel is near a regionboundary

Given the characteristics of the 119869-image region-growingmethod is suitable for image segmentation because differentregions tend to have similar 119869 values A set of initial seedcan be set by calculation of the standard deviation of local119869 if certain numbers (usually set to about 1 of the pixelsof calculating window) of pixels have 119869 value less than theaverage and are connected They shall be set as a seed Thenthe nonseed pixel grows one by one the one with minimum

Table 1 Parameter set for different scale

Scale(pixels)

Window(pixels)

Region size(pixels)

Min seed(pixels)

0 33 times 33 256 times 256 512

1 17 times 17 128 times 128 128

2 9 times 9 64 times 64 32

3 5 times 5 32 times 32 8Min seed refers to the minimum number of seed pixels

local 119869 value is assigned to its adjacent seed until all pixels areassigned [22]

3 Results and Discussion

In this section the results of the proposed method werepresented on two experiments JSEGrsquos parameter analysis andits comparison with other methods

To JSEG algorithm the scale of the window for calcu-lating 119869 is significant Figure 6 gives some example of 119869-image computing under different scale Table 1 gives differentwindow size for computing 119869 values and region size for seeddetermination As can be seen in Figure 6 Figure 6(a) isoriginal image and the following images are their 119869-imagescalculated under scales 1 2 and 3 The smaller the scale isthe more details the 119869-image retains Bigger scale will neglectthe small details while edges become rather coarse

Since the whole seed determination and seed growingprocess are based on 119869-image the result of segmentationvaries by different scale As demonstrated in Figure 7Figure 7(a) refers to the seedsrsquo generation and growth by scale1 Figure 7(b) refers to scale 2 and Figure 7(c) refers to scale 3119869-image in scale 3 containsmore edged details yet it also leadsto oversegmentation The scratches on the particlersquos surfacewere not taken as one unified region That is because when


(a) (b) (c) (d)

Figure 6 Different 119869-image calculation under different scales

the calculationwindow is too small the 119869 value of local regionrises and falls too quickly especially when the texture is notso fine

Now We will compare the proposed method with tradi-tional thresholding and edge detection method A programwritten by interactive data language is developed to separatethe particle from the background During the process theedge of the particle was obtained using morphology dilationand erosion operation through human-computer interactionfollowing the steps below

(i) Set a gray intensity threshold to separate particle 119860from the background

(ii) Using a structure ( 0 1 01 1 10 1 0

) perform erosion to the tar-get (particle) one can get 119861 the four-neighborhoodedge is 119860 cap 119861

(iii) Select the edge by region growth

Some typical examples are shown in Figure 8 Figure 8(a)is edge acquired by thresholding Thresholding involves

separating the dark and light regions of the image andthus identifying dark objects on a light background (or viceversa) When objects are large and do not possess very muchsurface detail segmentation can be imagined as splittingthe image into a number of regions each having a highlevel of uniformity in some parameter such as brightnesscolor texture or even motion In the second example thethresholding outperformed the proposed method Howeverwhen the texture is complicated and the color intensity ofparticle and background are similar thresholding lost itsadvantage because it cannot make full use of texturersquos infor-mation Figure 8(b) is edges detected by Sobel operator Edgeprovides intrinsicallymore rigorousmeans than thresholdingfor initiating image segmentation yet the edge we get is notclosed and its oversegmentation cannot be resolved unlessusing Canny operator and yet its parameter set requireshumanrsquos experience

Generally speaking JSEG performed well on wear parti-clesrsquo imagesThe algorithmmakes use of both color informa-tion and texture information Yet it still has some limitationsone major problem is shades of wear particlesThe shapes are


(a) Initial seed area (b) Regions after initial growing (c) Segmentation result

Figure 7 An example of final segmentation results under 3 scales

usually part of the particle yet due to lack of illuminationit often causes oversegmentation However this problem canbe solved by improving hardware for example taking severalpictures under different illumination and then using them tocompose a new picture Other problem is that when particleswith similar texture are attached the algorithm will not takethem as two but one big particle This problem might besolved by combining watershed algorithm or enhancing edgeinformation

4 Conclusion

In this work a segmentation algorithm for wear particlersquosimage is presented The segmentation consists of colorquantization and spatial segmentation The advantage of this

method is that it is fully unsupervised and can be used directlyfor color images Different parameters are tested Experi-ments showed that for regular ferrograph image large scaleis more suitable small scale could lead to over segmentationCompared to early approaches of image segmentation andedge detection the proposed method holds great promisein the autoclassification system for wear particles yet thereare still some problems that need to be solved For instancethe performance is poor when facing accumulated particlesespecially when the overlapping particles share similar tex-ture property To solve this problem other algorithms such aswatershedmethod or other image properties such as intensivecolor changes should be taken into account to improve thesegmentation result The further study will focus on solvingthese problems and will be tested in real-time machinersquoslubricant system


(a) (b) (c)

Figure 8 The comparison between proposed method and traditional method

Competing Interests

The authors declare that they have no competing interests

References

[1] G W Stachowiak and P Podsiadlo ldquoTowards the developmentof an automated wear particle classification systemrdquo TribologyInternational vol 39 no 12 pp 1615ndash1623 2006

[2] J Wang and X Wang ldquoA wear particle identification methodby combining principal component analysis and grey relationalanalysisrdquoWear vol 304 no 1-2 pp 96ndash102 2013

[3] C Kowandy C Richard Y-M Chen and J-J Tessier ldquoCor-relation between the tribological behaviour and wear particle

morphology-case of grey cast iron 250 versus Graphite andPTFErdquoWear vol 262 no 7-8 pp 996ndash1006 2007

[4] S Raadnui ldquoWear particle analysismdashutilization of quantitativecomputer image analysis a reviewrdquo Tribology International vol38 no 10 pp 871ndash878 2005

[5] H Liu H Wei L Wei J Li and Z Yang ldquoAn experimenton wear particlersquos texture analysis and identification by usingdeterministic tourist walk algorithmrdquo Industrial Lubricationand Tribology vol 67 no 6 pp 582ndash593 2015

[6] J A Williams ldquoWear and wear particlesmdashsome fundamentalsrdquoTribology International vol 38 no 10 pp 863ndash870 2005

[7] R K Upadhyay ldquoMicroscopic technique to determine variouswear modes of used engine oilrdquo Journal of Microscopy andUltrastructure vol 1 no 3 pp 111ndash114 2013


[8] B J Roylance ldquoFerrographymdashthen and nowrdquoTribology Interna-tional vol 38 no 10 pp 857ndash862 2005

[9] J Wang L Zhang F Lu and X Wang ldquoThe segmentation ofwear particles in ferrograph images based on an improved antcolony algorithmrdquoWear vol 311 no 1-2 pp 123ndash129 2014

[10] Z Yu O C Au R Zou W Yu and J Tian ldquoAn adaptiveunsupervised approach toward pixel clustering and color imagesegmentationrdquoPattern Recognition vol 43 no 5 pp 1889ndash19062010

[11] G P Stachowiak G W Stachowiak and P Podsiadlo ldquoAuto-mated classification of wear particles based on their surfacetexture and shape featuresrdquo Tribology International vol 41 no1 pp 34ndash43 2008

[12] G P Stachowiak P Podsiadlo and G W Stachowiak ldquoEval-uation of methods for reduction of surface texture featuresrdquoTribology Letters vol 22 no 2 pp 151ndash165 2006

[13] G P Stachowiak P Podsiadlo and G W Stachowiak ldquoAcomparison of texture feature extraction methods for machinecondition monitoring and failure analysisrdquo Tribology Lettersvol 20 no 2 pp 133ndash147 2005

[14] C Q Yuan J Li X P Yan and Z Peng ldquoThe use of thefractal description to characterize engineering surfaces andwear particlesrdquoWear vol 255 no 1ndash6 pp 315ndash326 2003

[15] T Wu Y Peng H Wu X Zhang and J Wang ldquoFull-lifedynamic identification of wear state based on on-line weardebris image featuresrdquo Mechanical Systems and Signal Process-ing vol 42 no 1-2 pp 404ndash414 2014

[16] X Liao H Xu Y Zhou et al ldquoAutomatic image segmentationusing salient key point extraction and star shape priorrdquo SignalProcessing vol 105 pp 122ndash136 2014

[17] M Jungmann H Pape P Wiszligkirchen C Clauser and TBerlage ldquoSegmentation of thin section images for grain sizeanalysis using region competition and edge-weighted regionmergingrdquo Computers amp Geosciences vol 72 pp 33ndash48 2014

[18] T Wu H Wu Y Du N Kwok and Z Peng ldquoImaged weardebris separation for on-line monitoring using gray level andintegrated morphological featuresrdquo Wear vol 316 no 1-2 pp19ndash29 2014

[19] F Li C Xu G-Q Ren and J-W Gao ldquoImage segmentationof ferrography wear particles based on mathematical morphol-ogyrdquo Journal of Nanjing University of Science and Technologyvol 29 no 1 pp 70ndash72 2005

[20] Y Deng and B S Manjunath ldquoUnsupervised segmentation ofcolor-texture regions in images and videordquo IEEE Transactionson Pattern Analysis and Machine Intelligence vol 23 no 8 pp800ndash810 2001

[21] R O Duda and P E Hart Pattern Classification and SceneAnalysis John Wiley amp Sons New York NY USA 1970

[22] C Deng C Kenney M S Moore and B S ManjunathldquoGroup filtering and perceptual color image quantizationrdquo inProceedings of the IEEE International Symposium onCircuits andSystems vol 4 pp 21ndash24 Orlando Fla USA May 1999

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



(a) Analytical ferrogram

0

10

20

30

40

50

Exit end Drain tube

Nonwettingbarrier

Wear particlesnear exit are

submicroscopic

Entry region wherefluid first touches down

on the ferrogram

Dist

ance

from

exit

end

(mill

imet

ers)

(b) Deposition pattern on a ferrogram

Figure 1 Picture of analytical ferrogram and its schematic diagram

adhesive fatigue and abrasive particles using area perimeterand elongation parameters study shows that these simpleparameters can be effective for certain types of wear particlessuch as abrasive particles and sphere like fatigue particlesThesame research group also compared the different dimensionreduction methods and texture features [12 13] In the studyby [14] fractal dimension had been used to analyze weardebris surface In the study by [15] a small online monitoringsystem for gearbox have been built and gray level andintegrated morphological features have been used for wearparticlesrsquo image segmentation

The possibility of using a pattern recognition system forwear particle analysis without the need of a human expertholds great promise in the condition monitoring industryStill there are several problems that need to be resolved Oneof the first and most important of them is autosegmentationfor wear particlesrsquo image In order to autoclassify wearparticles one must first demarcate the object appearing indigital images The accuracy of segmentation will directlyaffect the subsequent feature extraction classification andidentification of wear particles Therefore wear particlesegmentation is the vital first step for wear particlesrsquo imageanalysis However the existing segmentationmethods are notadaptable for complicated ferrograph images for two reasons

(1) Wear particles deposited on the glass substrate oftenhave complicated background Large abnormal parti-cles such as severe sliding particles or chunky fatigueparticles are main target yet they are surrounded bycountless microparticles which often share similarcolor making accurate segmentation by thresholdtechnique a hard job

(2) Since the pictures are microscopic image the edges ofthe object are frequently blurred leading to difficultyin segmentation by edge detection algorithm

As demonstrated in Figure 2(b) these particles havecomplicated texture features and their background are filled

withmicroparticles and some of the edges were blurred Evenin the area of computer vision segmentation is actually oneof the central and most difficult practical problems [16 17]Yet until recently not much work has been done In [18]gray level and integrated morphological features are used tosegment the agglomerated particles yet the background ofthe image is rather simple and clean algorithms adaptive tocomplicated background shall be tested It is reported in [9]that ant-colony algorithm can be used to complete the taskyet computing efficiency needs to be improved Li et al usedmorphological erosion and dilation operations on binaryimages the edge of a single wear particle could be detectedwith the Laplace operation [19] Yet how to get accurate binaryimage in itself is a difficult problem

In this research attempts have been made to solvethe problems mentioned above improving the result ofautomatic analysis for ferrography images We represent theimproved JSEG algorithm for wear particlesrsquo color imagesThe object of this study is to see whether this algorithmcan be used in particle image segmentation Parameters weretested according to the characteristics of wear particles Acomputer program is developed to preprocess images Theimage samples used are generated using tribology testingmachines Experiment results showed that JSEG is suitablefor wear particlersquos image segmentation

2 Experiment and Method

21 Collection of Wear Particlersquos Images Wear testing is oftencarried out to study tribological performance of certainmaterials or lubricants In this research however it is donefor collecting wear particles and building a database of theirimages

The adhesive wear testing was carried out on a pin-on-disk machine which consists of a horizontal rotating discand a deadweight-loaded pin The stationary pin is stainlesssteel (hardness 175HV) and the disk is made of cast iron


OptM 500x100120583m

(a)

OptM 500x100120583m

(b)

OptM 500x100120583m

(c)







Color class-map


J-image

Region growing







119894(119899) The


119889119894(119899) =

10038171003817100381710038171199090(119899) minus 119909

119894(119899)

1003817100381710038171003817 119894 = 0 119896

1198890(119899) le 119889


119896(119899)

(1)


119875 (119899) = 119909119894(119899) 119894 = 0 119898 (119899) minus 1 (2)


119889119898(119899)minus1

(119899) le 119879 (119899)

119889119898(119899)

(119899) gt 119879 (119899)

(3)



119867(119894) =

10038161003816100381610038161198861(119894) minus 119886

2(119894)1003816100381610038161003816

2

1199041

2

(119894) + 1199042

2

(119894)

119894 = 1 119896 (4)

where

1198861(119894) =

1

119894

119894minus1

sum

119895=0

119889119895(119899)

1198862(119894) =

1

119896 + 1 minus 119894

119896

sum

119895=119894

119889119895(119899)

1199041

2

(119894) =

119894minus1

sum

119895=0

10038161003816100381610038161003816119889119895(119899) minus 119886

1(119894)

10038161003816100381610038161003816

2

1199042

2

(119894) =

119896

sum

119895=119894

10038161003816100381610038161003816119889119895(119899) minus 119886

2(119894)

10038161003816100381610038161003816

2

(5)


119898(119899) = arg max119894

119867(119894) (6)


119894(119899)


119891119894(119899) are calculated

119891119894(119899) = 119889

119894+1(119899) minus 119889

119894(119899) (7)


119895(119899) for 119895 le 119894 or 119895 gt 119894




119909new (119899) =sum119898(119899)minus1

119894=1

119908119894119901119894(119899)

sum119898(119899)minus1

119894=1

119908119894

119901119894(119899) isin 119901 (119899) (8)



0(119899)



(a) (b) (c)





119898 =

1

119873

sum

119911isin119885

119911 (9)


119894be the


119894

119898119894=

1

119873119894

sum

119911isin119885

119911 (10)




Let

119878119879= sum

119911isin119885

119911 minus 1198982

119878119882=

119862

sum

119894=1

119878119894=

119862

sum

119894=1

sum

119911isin119885119894

1003817100381710038171003817119911 minus 119898

119894

1003817100381710038171003817

2

(11)


class Define

119869 =

119878119879minus 119878119882

119878119882

(12)




Scale(pixels)

Window(pixels)

Region size(pixels)

Min seed(pixels)

0 33 times 33 256 times 256 512

1 17 times 17 128 times 128 128

2 9 times 9 64 times 64 32








(a) (b) (c) (d)















4 Conclusion




(a) (b) (c)


Competing Interests


References



























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










OptM 500x100120583m

(a)

OptM 500x100120583m

(b)

OptM 500x100120583m

(c)







Color class-map


J-image

Region growing







119894(119899) The


119889119894(119899) =

10038171003817100381710038171199090(119899) minus 119909

119894(119899)

1003817100381710038171003817 119894 = 0 119896

1198890(119899) le 119889


119896(119899)

(1)


119875 (119899) = 119909119894(119899) 119894 = 0 119898 (119899) minus 1 (2)


119889119898(119899)minus1

(119899) le 119879 (119899)

119889119898(119899)

(119899) gt 119879 (119899)

(3)



119867(119894) =

10038161003816100381610038161198861(119894) minus 119886

2(119894)1003816100381610038161003816

2

1199041

2

(119894) + 1199042

2

(119894)

119894 = 1 119896 (4)

where

1198861(119894) =

1

119894

119894minus1

sum

119895=0

119889119895(119899)

1198862(119894) =

1

119896 + 1 minus 119894

119896

sum

119895=119894

119889119895(119899)

1199041

2

(119894) =

119894minus1

sum

119895=0

10038161003816100381610038161003816119889119895(119899) minus 119886

1(119894)

10038161003816100381610038161003816

2

1199042

2

(119894) =

119896

sum

119895=119894

10038161003816100381610038161003816119889119895(119899) minus 119886

2(119894)

10038161003816100381610038161003816

2

(5)


119898(119899) = arg max119894

119867(119894) (6)


119894(119899)


119891119894(119899) are calculated

119891119894(119899) = 119889

119894+1(119899) minus 119889

119894(119899) (7)


119895(119899) for 119895 le 119894 or 119895 gt 119894




119909new (119899) =sum119898(119899)minus1

119894=1

119908119894119901119894(119899)

sum119898(119899)minus1

119894=1

119908119894

119901119894(119899) isin 119901 (119899) (8)



0(119899)



(a) (b) (c)





119898 =

1

119873

sum

119911isin119885

119911 (9)


119894be the


119894

119898119894=

1

119873119894

sum

119911isin119885

119911 (10)




Let

119878119879= sum

119911isin119885

119911 minus 1198982

119878119882=

119862

sum

119894=1

119878119894=

119862

sum

119894=1

sum

119911isin119885119894

1003817100381710038171003817119911 minus 119898

119894

1003817100381710038171003817

2

(11)


class Define

119869 =

119878119879minus 119878119882

119878119882

(12)




Scale(pixels)

Window(pixels)

Region size(pixels)

Min seed(pixels)

0 33 times 33 256 times 256 512

1 17 times 17 128 times 128 128

2 9 times 9 64 times 64 32








(a) (b) (c) (d)















4 Conclusion




(a) (b) (c)


Competing Interests


References



























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











Color class-map


J-image

Region growing







119894(119899) The


119889119894(119899) =

10038171003817100381710038171199090(119899) minus 119909

119894(119899)

1003817100381710038171003817 119894 = 0 119896

1198890(119899) le 119889


119896(119899)

(1)


119875 (119899) = 119909119894(119899) 119894 = 0 119898 (119899) minus 1 (2)


119889119898(119899)minus1

(119899) le 119879 (119899)

119889119898(119899)

(119899) gt 119879 (119899)

(3)



119867(119894) =

10038161003816100381610038161198861(119894) minus 119886

2(119894)1003816100381610038161003816

2

1199041

2

(119894) + 1199042

2

(119894)

119894 = 1 119896 (4)

where

1198861(119894) =

1

119894

119894minus1

sum

119895=0

119889119895(119899)

1198862(119894) =

1

119896 + 1 minus 119894

119896

sum

119895=119894

119889119895(119899)

1199041

2

(119894) =

119894minus1

sum

119895=0

10038161003816100381610038161003816119889119895(119899) minus 119886

1(119894)

10038161003816100381610038161003816

2

1199042

2

(119894) =

119896

sum

119895=119894

10038161003816100381610038161003816119889119895(119899) minus 119886

2(119894)

10038161003816100381610038161003816

2

(5)


119898(119899) = arg max119894

119867(119894) (6)


119894(119899)


119891119894(119899) are calculated

119891119894(119899) = 119889

119894+1(119899) minus 119889

119894(119899) (7)


119895(119899) for 119895 le 119894 or 119895 gt 119894




119909new (119899) =sum119898(119899)minus1

119894=1

119908119894119901119894(119899)

sum119898(119899)minus1

119894=1

119908119894

119901119894(119899) isin 119901 (119899) (8)



0(119899)



(a) (b) (c)





119898 =

1

119873

sum

119911isin119885

119911 (9)


119894be the


119894

119898119894=

1

119873119894

sum

119911isin119885

119911 (10)




Let

119878119879= sum

119911isin119885

119911 minus 1198982

119878119882=

119862

sum

119894=1

119878119894=

119862

sum

119894=1

sum

119911isin119885119894

1003817100381710038171003817119911 minus 119898

119894

1003817100381710038171003817

2

(11)


class Define

119869 =

119878119879minus 119878119882

119878119882

(12)




Scale(pixels)

Window(pixels)

Region size(pixels)

Min seed(pixels)

0 33 times 33 256 times 256 512

1 17 times 17 128 times 128 128

2 9 times 9 64 times 64 32








(a) (b) (c) (d)















4 Conclusion




(a) (b) (c)


Competing Interests


References



























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










(a) (b) (c)





119898 =

1

119873

sum

119911isin119885

119911 (9)


119894be the


119894

119898119894=

1

119873119894

sum

119911isin119885

119911 (10)




Let

119878119879= sum

119911isin119885

119911 minus 1198982

119878119882=

119862

sum

119894=1

119878119894=

119862

sum

119894=1

sum

119911isin119885119894

1003817100381710038171003817119911 minus 119898

119894

1003817100381710038171003817

2

(11)


class Define

119869 =

119878119879minus 119878119882

119878119882

(12)




Scale(pixels)

Window(pixels)

Region size(pixels)

Min seed(pixels)

0 33 times 33 256 times 256 512

1 17 times 17 128 times 128 128

2 9 times 9 64 times 64 32








(a) (b) (c) (d)















4 Conclusion




(a) (b) (c)


Competing Interests


References



























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












Let

119878119879= sum

119911isin119885

119911 minus 1198982

119878119882=

119862

sum

119894=1

119878119894=

119862

sum

119894=1

sum

119911isin119885119894

1003817100381710038171003817119911 minus 119898

119894

1003817100381710038171003817

2

(11)


class Define

119869 =

119878119879minus 119878119882

119878119882

(12)




Scale(pixels)

Window(pixels)

Region size(pixels)

Min seed(pixels)

0 33 times 33 256 times 256 512

1 17 times 17 128 times 128 128

2 9 times 9 64 times 64 32








(a) (b) (c) (d)















4 Conclusion




(a) (b) (c)


Competing Interests


References



























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










(a) (b) (c) (d)















4 Conclusion




(a) (b) (c)


Competing Interests


References



























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation













4 Conclusion




(a) (b) (c)


Competing Interests


References



























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










(a) (b) (c)


Competing Interests


References



























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation



























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Documents

Research Article The Segmentation of Wear Particles …downloads.hindawi.com/journals/at/2016/4931502.pdf · The Segmentation of Wear Particles Images Using-Segmentation Algorithm