Quantitative analysis of microstructure of carbon materials by HRTEM

Available online at www.sciencedirect.comsCIENCEs doInEoT.Transactions of.Nonferrous MetalsSociety of ChinaScienceTrans. Nonferrous Met. Soc. China 16(2006)s796- s803Presswww.csu.edu.cn/ysxb/Quantitative analysis of microstructure of carbon materials by HRTEMYANG Jun-he', CHENG Shu-hui?, WANG Xia', ZHANG Zhuo', LIU Xiao-rong', TANG Guo-hua'1.Department of Materials Science and Engineering, Shanghai Institute of Technology, Shanghai 200235, China;2.School of Chemical Engineering, East China University of Science and Technology, Shanghai 200000, ChinaReceived 10 April 2006; accepted 25 April 2006Abstract: The main object of the present research is to make a quantitative evaluation on the microstructure of carbon materials interms of microcrystal. The digitized images acquired from finely pulverized carbon materials under HRTEM at a high magnificationwere processed by the image processing software so as to extract the fringes of (002) lattice of graphite crystal from thebackground image, and an FFT-IFFT filtering operation was performed followed by processes as binarization for the image andskeletonization for the fringes. A set of geometrical parameters including position, length and orientation was set up for every latticefringe by calculating the binarized image. Then, the above obtained fringe parameters were put into an algorithm, which wasespecially developed for such fringe images so as to find fringes that could be regarded as those belonged to one single graphitemicrocrystal. The fringe was subjected sequentially to comparing procedures with every other fringe on aspects as parallelism,relative position and spacing, and the above comparisons were repeated till the last fringe. Eventually, the microcrystal size, itsstacking number, and the distribution of the microcrystal in the whole sample, as well as other related structure information of suchmicrocrystal in carbon materials were statistically calculated. Such microstructure information at nanometer level may contributegreatly to the interpretation of the properties of carbon materials and a better correlation with the same macrostructure.Key words: carbon materials; microcrystal; digitized images; quantitative analysis; HRTEMquantitativeness. OSHIDA et al discussed the influence1 Introductionof conditions of HRTEM to the image processing andanalysis with respect to carbon materials as fluorinatedCarbon materials have been confirmed to havegraphite fiber[7]， activatedcarbon fiber[8] andturbostratic structure constituted by many graphiteamorphous carbon film[9]. HUANG Z H et al[10]microcrystals. XRD technique is conventionally used todisplayed the nanopore structure of activated carboninvestigate the crystastructurematerialsfiber under HRTEM and obtained its fractal featureproperties[1- -3]. Some crystal parameters such as La, L。based on fractal theory. Other researchers[11- 13] triedand d-spacing can be calculated based on the results ofon the quantization of the crystal parameters of carbonXRD. Recently， with the occurrence of HRTEMmaterials, where some structural parameters were also(high-resolution transmission electron microscopy)obtained. Additionally, based on the results from imagetechnique, a visual view of such structure can beanalysis, SHIM et al[14] focused on the orientationpresented. SHAMA et al[4] reported layeredbehavior of the graphite layer in carbon materials.graphite-like microcrystal of bituminous coal underA main object of the present study is to develop aHRTEM. RUSSELL et al[5] observed the cokingnew statistical algorithm to determine the microcrystaltransformation of Pittsburgh coal using HRTEM. Inunits from HRTEM images of carbon materials. Basedaddition, quantitative image analysis has become one ofon the output of the algorithm, quantitative structuralthe most successful laboratory techniques in materialsinformation, especially the distribution of microcrystalsscience[6]. Based on stereology theory, data from ain carbon materials, is obtained. Another object of thistwo-dimensional image can be converted into reliableresearch is to establish a new crystallinity index toand accurate structural information of three-dimension,characterize中国煤化工ls in carbonwhich is a big stride from qualitativeness tomaterials so aYHCNMHGetweentheCorresponding author: YANG Jun-he; E mail: jhyang@shsit.edu.cn.YANG Jun-he, et al/Trans. Nonferrous Met. Soc. China 16(2006)s797microstructure and the properties of carbon materials.procedure that can separate fringes with excellentreproducibility is developed. Thereby, the difference2 Experimentalbetween fringes can be revealed, and identification onthese fringes can be carried out effectively andCarbon materials samples, which were dispersed inobjectively. Fig.2 shows the specific processing stepsethanol, were obtained from Liulin coal (a bituminousespecially designed for fringes image of the presentcoal) by coking process. JEM-2010(HITACHI, Japan)research, wherein step of filtration, step of conversionwas work at an accelerating voltage of 200 kV. Carbonand step of skeletonization are described in detail asmaterials powders that were suspended in ethanol werefollows. The processing software used here is ImageJchosen as carbon materials samples and were carried byv1.30 from National Institutes of Health, USA.a copper grid and transferred into the observationchamber of JEM-2010 microscope. After the system was3.1 FFT Filtrationevacuated to a vacuity of 1.33X10 4- -1.33X10-2 Pa, theOne of the key means for extraction of fringes usedsamples were first examined at a moderate magnificationhere is FFT filtration, which is capable of removing thesuch as 100 KX or 200 KX to find wedge- shapednoise of no interest without losing the information offragments that had edges thin enough (tens offringes. Fig.2 gives the flow diagram of imagenanometers typically) for the electron to pass through so processing on lattice image of carbon materials. Fig.2(b)as to meet the imaging condition. Under the highgives a frequency domain showing a frequencymagnification of 500 KX, 10 or more such edge regionsdistribution of the original image (Fig.2(a)) after beingfor one sample were then selected and photographedsubjected to a Fast Fourier Transform (FFT), whereinwhen a clear fringe image was imaged.farther it is from the center, higher the frequency is. IrImages from the microscope were stored directlyFig.2(b), there are two concentric arcs with a relativelyinto a PC as a digital format (RAW) with 1 024X1 024higher brightness, which are actually two segments ofpixels and a 8-bit grayscale. The size of the digital imagene similiar circular ring. It is a typical diffractionpattern of the multicrystal, which is the result of themeans that one pixel on the image represents a real sizediffactions superposition of countless micro singleof about 1/26 nm. Fig.l shows some example imagescrystal grains. In a case where the arrangement of thesethus-obtained. Due to the limitation of the current samplesingle crystals is totally random, there will be an intactpreparation technique, however, the superpositions of ring shape for thediffraction pattern. In the present case,fringes are sometimes unavoidable, as shown in Fig.1(c)however, only two symmetrical segments of the ring aretypically.presented. That is to say, the arrangement of singlecrystals exhibits an orientation preference, which is3 Image processinginversely proved by the original crystal latticeimage(Fig.2 ().The first challenging task for the quantitativeBased on the above facts, by using the reversibilityanalysis of TEM fringe images is the conversion of theof Fourier Transform, a filtering process can be easilycomplex original image into a set of distinct, identifiableoperated. To achieve such object, with respect to thefringes that can be analyzed by object-oriented imagefrequency domain, the portion (frequency) other than theanalysisalgorithms. Here, based on the prior imageregion where the circular ring represented the orderedprocessing methodology, an improved processingstructure to be located was deleted, i.e. covered by purea)中国煤化工CnmMYHCNMHG一二Fig.1 Some typical fringe images of carbon materials by HRTEM.s798YANG Jun-he, et al/Trans. Nonferrous Met. Soc. China 16(2006)(a6)FFTFilteringIFFTTranfromBinarizationFig.2 Flow diagram of image processing on lttie image of carbon materialsdarkness shown in Fig.2(c). Then, an inverse fast Fourierrepresented by binary digit system, which conforms toTransform (IFFT) was performed to rebuild the crystalthe operation of computer.lattice image as shown in Fig.2(d), where the backgroundTo perform binarization, a threshold value betweenwas largely suppressed and incidentally, although the0 and 255 needs to be set firstly. In a lattice fringe image,inclination angle of these two symmetrical arcs withthe grayscale value for fringes is not at the same level asrespect to the image indicates the orientation preferencefor the background. A threshold is the optimum value forof lattice fringes, there is no comparability of one withdistinguishing fringes from the background. Pixels .others since the acquirement of lattice images is nothaving a grayscale value more than the threshold arecontinuous.assigned a maximum value l, i.e. white, while thosehaving a grayscale value no more than the threshold are3.2 Binarizationassigned a minimum value 0, i.e. black. Thus, a fringeThe imageimage in which fringes are black and the background issubjected to a binarization operation so as to convert thewhite is obtained. The threshold can be easily set by useoriginal gray image to a binary image， i.e.;of histogram of the image that shows the distribution ofblack-and-white image. In a 256-grayscale image, everythe grayscale of pixels. An image after binarization is .pixel has a grayscale corresponded to a value between 0shown in Fig.2(e), from which it can be seen that theand 255. In a binary image, however, there are only twofringes are totall中国煤化工xkground.gray scales that are black and white. The reason for this"YHCNM HGoperation lies not only in the increase of the image3.3 Skeletonizaunt anu pust-pruessiigcontrast, but also in that a binary image can be directlyBefore the skeletonization can be performed, an.YANG Jum-he, et al/Trans. Nonferrous Met. Soc. China 16(2006)s799open operation that comprises eroding step followed bythe fted ellipse is assigned as the (x, y) coordinates anddilating step is performed to the above- obtained image.its inclination angle to the X-axis is assigned as theEroding can, on one hand, remove the noise signals of noinclination angle of the fringe. Additionally, the totalmore than four pixels, which are maintained during thenumber of fringes N in one image can also be easilybinarization process, and on the other hand, separatedobtained. All the above-mentioned parameters will befringes that joint together but substantially belong touseful for analyzing and reproducing the structure ofdifferent fringes effectively. The following dilatingmicrocrystal of carbon materialss mentioned after.process is to revert the shape of eroded fringes back. In aword, an open operation can separate adjacent fringes to4.2 Statistical algorithman appropriate extent without changing their shape.The above feature parameters obtained for theSkeletonization, which is also referred as medialaxisfringe are just some preliminary parameters fortransform, is the last step for image processing. Inindividual fringe only and do not show relationshipskeletonization，eroding operation is continuouslybetween fringes, thus cannot reflect the structuralperformed until the skeleton of the fringe having a widthinformation of microcrystal from an integral point.of only one pixel is obtained. Such fringes are easier toHowever, based on these parameters, it is possible tocalculate their feature parameters. A skeletonized latticeachieve such objective through a certain algorithm. Thefringe image is shown in Fig.2(f).first thing that needs to be made clear is to determinewhat kind of fringes (layers) is to be assigned to one4 Quantitative analysis of structurestack (one microcrystal). Only when such rule is set, mayit be possible to find these fringes and thus calculate theConventionally quantitativeinformation ofparameters concermning the stack.microcrystal structure of carbon materials obtainedReferring to the structure of a perfect graphitemainly includes average interlayer spacing (d-spacing)crystal structure, such rule for determining fringe stackand average size (La, L) of lattice fringes. In the presentcab be roughly figured out, i.e. microcrystal from theresearch, however, parameters (d-spacing, La and Le) offringes image. In a perfect graphite crystal, aromaticindividual fringe is firstly attermpted to obtain, whichlayers are paralleled with each other at a constantmeans the distribution information of the microcrystalinterlayer spacing of 0.335 nm. The HRTEM latticeconcerning the d-spacing and size besides the averagedimage of carbon materials, however, shows that mostones can also be obtained. Further, based on thefringes are non-linear and stacked at an uncertaindistribution information, a new parameter crystallinityinterlayer spacing. Therefore, it is necessary to widenthat can reflect the microcystal structure informationsuch conditions appropriately for microcrystals far frommore objectively and more of actual utilization isperfect here represented. Three conditions wereattempted to establish.concluded to determine whether two randomly selectedfringes belong to one similiar stack preliminarily, which4.1 Feature parameters of lattice fringeare:ImageJ is provided with a particle analysis function,1) The difference of inclination angle of them doeswhich searches for the pixel having a grayscale of zero not exceed a certain range, that is, they seem to be(black) and attributes it and its adjacent black pixels toparallel.one set. Regarding pixels in one set as an integrated2) The midpoint of each fringe does not go beyondobject, i.e. one fringe, ImageJ can be used to calculate the region demarcated by two perpendicular lines passingand/or obtain some parameters directly concerning theeach end of the other fringe, that is, they seem to befringes such as the area A, the true length L, the positionstacked.in the coordinate (x, y), Peret diameter (the maximum3) The interlayer spacing therebetween is within atangential diameter), and inclination angle, etc.specific range, that is, they have bonding force with eachThe area A of a fringe is obtained by counting theother.number of pixels. The true length L of the fringe isWhen all the three conditions are satisfied, theseconverted from the area A by the following equationtwo fringes are assigned to one stack. Next, the specificsince the fringe has a width of only one pixel:determining method will be described in detail withL=26A.reference to a HRTEM image of carbon materialswhere number 26 is the scaling ratio for real size of thepowder from Liulin coal shown in Fig.3.image. Since most fringes are tortuous to varied extend,As seen in中国煤化Ilage is namedthe above coordinates (x, y) and inclination angle of theby a serial numEd fringe i, its .fringe are obtained by fiting an ellipse to every fringe, in coordinates (x;,YHCNMH G。r axis ra),which the midpoint coordinates of the major axis (ra) ofinclination angle 0， and the length L are known.s800YANG Jun-he, et al/Trans. Nonferrous Met. Soc. China 16(2006)beforehand. When compared the next fringe j with fringesatisfy the constitution of graphite crystal or the abovei, the following three determining conditions can be set:conditions into one stack (unit), calculate the size, height,1) Parallelism determination. It is considered to beof every unit and further to obtain theparallelism when the absolute difference between the distribution of units in an image.inclination angles of fringe i and fringe j is no more thanFirst, a random fringe is selected from the lttice15°, shown in the following equation: |0 -θ|≤π/18.image as the reference fringe. A second fringe is selectedfor being compared with the reference fringe inaccordance with the above three conditions sequentially.If all the three are satisfied, both of the two fringes aremerged into one group and together serve as referencefringes of this group. Then, such comparison of everyother fringe that has not been compared is made one byone with respect to every fringe that has been merged to中the present group (unit). This fringe group is defined asne graphite microcrystal unit. During the abovecomparison, if a fringe that fails to meet any one of theconditions, a new comparison circulation is performedtaking this fringe as a starting point. When thecirculation is finished, another fringe group/microcrystalunit is obtained. When all fringes within an image aremerged into independent fringe group, such fringegroups are classified into single-fringe group, 2-fringeFig.3 HRTEM image of carbon materials powder from Liulingroup and 3-fringe group according to the number ofcoalfringe in it.The above algorithm conception can be shown by a2) Stacked position determination. If the centerof flow diagram shown in Fig.4. Since thereis nofringe j falls within the region demarcated by tw(ready-made for this algorithm, a merging program withperpendicular lines that pass through the ends of fringe i,C++ is written. As shown in Fig.4, an array variable I isthe condition is satisfied. This condition is converted to,set up at the beginning of the program for depositingin the practical operation, that an intercept bj of thefringes that meet the determining conditions. Theperpendicular line passing through the center of fringe jprogram comprises a triple circulation represented by①，on the Y-axis is between intercepts b1 and b2 of two②and③, respectively. In the inner circulation ①,perpendicular line passing through two ends of fringe i.fringe j is compared with every fringe that has beenIt is noted that the center and ends of the fringe aremerged into I respectively. Noted that once fringe japproximated by the midpoint and the ends of major axissucceeds with one fringe of I, the current circulation isr]x of the fitted ellipse. More specifically, an inclinationceased and another fringe j=j+1 is selected for a newnner circulation. In mid circulation ②, fringe j otherangle for both of the fringe i and j needs to be newlydefined since these two fringes may not be parallelthan fringes from I steps goes sequentially intoabsolutely θ=(0;+0)/2. On the Y-axis intercepts that linescirculation①for comparison until all fringe j iare perpendicular to the fringes and pass through the twocompared. The current fringe group I is thus obtainedends of fringe i and center of fringe j respectively areand the number of the fringes within I is counted.Meanwhile, fringes in the current group I are marked ascalculated by the following equation:compared fringes. In outer circulation ③, fringe I isb1.2./)1.2+x1.2Cot0selected as a reference fringe. When there is no leftwherein, (V1,2, x1,2) is the coordinates of the two ends.fringe for selection, it indicates that all finges have beenThus the determining condition is as follows:merged. Classification is operated in accordance with theb,∈[b),b2] (brSelecing fringe jj+1Marking fringes in/ as comparedCounting number of fringes in ISelecting fringe网i+1Merging, calculating and outputingEndFig.4 Flow diagram of the algorithm for fringe merge4.3 Establishment of crystallinity index of carbonx=Zx;m;(1)materials crystallizationA new index for indicating the degree ofwhere X is crystallinity index, i is stacking number ofcrystallization in carbon materials is defined in twothe unit, n is maximum stacking number of the unit, m;considerations, the dimensional size of microcrystal unitis fraction of i-fringe unit accounting the whole sample.and the ordered degree. The ordered degree aThe crystallinity fraction index of i-fringe unit is definedmicrocrystal unit is mainly indicated by the difference中国煤化工between the d-spacing of microcrystal of carbonCNMHGmaterials and that of graphite. The crystallinity index canYHbe expressed by the following equation:X;=()]台2 (0.34(2).s802YANG Jun-he, et al/Trans. Nonferrous Met. Soc. China 16(2006)Table 1 Statistical results of Liulin carbon materials by the algorithm program and its crystallinity XStackingNumber ofFraction ofAverage diameter ofAverage spacingXnumber ilayerslayers m/%layers l/nmd/nm111 97846.589 721.014 7260.777 97217.836 801.072 680 .0.383 7241.019 5312 58012.221 791.235 8280.375 5561.382 67541 5567.598 1811.273 9230.372 2181.482 4081 0905.786 9941.385 0630.370 2131.761 8393.586 9711.417 8390.368 6291.854 1444412.430 0991.437 5680.368 4121.907 2252001.202 8161.568 9610.368 2652.272 7031530.868 5101.480 903 .0.367 8382.027 1011(1000.527 5321.376 2340.364 0621.768 837_SUM23 32200X=186.156 5where l, l and d; are average length (diameter) of needs to be big enough. However, it is not easy to geti-layer unit, average length of 2-layer unit and averageimages that are proper for image process, which alsospacing of i-layer unit, respectively. d-spacing betweenbecomes a challenge for the modern technology. As forgraphite layer is 0.34 nm. (/l2)2:i/2 means thethe image processing, some of the parameters used forcontribution of i-fringe unit on volume, (d/0.34)厂meansdetermination are set artificially. Different parameterthe contribution of i-fringe unit on degree of microcrystalvalues may bring different results. Since there is noorder.standard concerned, these results have no absolutemeaning and are comparable only within the same5 Results and discussionoperation system.In order to ensure the results to be representative,6 Conclusionstotal 14 images were acquired from powder sample ofLiulin carbon materials by HRTEM. For every image,Lattice fringe image of carbon materials byimage processing was performed to obtain a set ofHRTEM technique at a high magnification directlyparameters for fringes within. Such data were input intoshows the structure of the graphite microcrystal inthe merging program. All output data of these 14 images carbon materials. By use of image process andare summed up and calculated by using the equations foranalysis, together with the statistical algorithm withcrystallinity X. The results are shown in Table 1. It isrespect to the HRTEM image of carbon materials,shown that there is no d-spacing for single-fringe group.quantitative microstructure information including notHere, the crystallinity is an index of no unit. The largeronly the size and d-spacing but also the distribution ofthe crystallinity, the higher the degree of crystallizationmicrocrystals in carbon materials is obtained. Further,of carbon materials.statistical results were used to establish a newVhen combining HRTEM technology with imagecrystallinity index, which has a great potential fcprocessing technology, quantitative informationcorrelating the microstructure with the knownconcerning the microcrystal of carbon materialsproperties of carbon materials， thus may haveobtained. Compared with the; corresponding informationmeaningful application in the carbon materialsby conventional XRD technology, the distribution of theindustry.graphite microcrystal in carbon materials is obtainedbesides the averaged size and d-spacing. Based on theReferencesdistribution data, a crytalinity index is set up, whichmakes it possible to correlate the microstructure with the[1] ANDREW K. K, DENNIS C. N. Microstructural evolution duringcharcoal carbonization by X-ray diffraction analysis []. Carbon,macro properties of the carbon materials.2001 41.15- 77Due to the inherent feature of HRTEM technology,ZHENG XJ, YANG z y, ZHOU Y C. X-ray ifrction measurementthe sample to be observed must be very small, whichof residual stress in PZT thin films prepared by pulsed laserchallenges the current sample preparation techniques. Ondeposition []. Acta Mater, 2004, 52(11): 3313- -3322.ZHENG )the other hand, since the actual size of a HRTEM image中国煤化idul stes iemetal organicis within the nanometer scale, the results from such adecompositionMHCNMHG_rvation of layeredsmall viewing field may be doubted. In order to make the [4] SHAMAA, Kstructure of coals by a transmission electron microscope [0]. Energyquantitative information be typical, the sampling number.YANG Jun-he, et al/Trans. Nonferrous Met. Soc. China 16(2006)s803& Fuels, 200, 14:515- 516.cUI M L, CHENG z Y. Pore structure and fractal charateristics of[5] RUSSELL N v, GIBBINS J R, WILLAMSON J. Structural orderingactivated carbon fibers characterized by using HRTEM [] J Colloidin high temperature carbon materialss and the effect on reactivity [J].1] YOSHIZAWA N, YAMADA y, SHIRAISHI M. TEM lttice[6] LIFSHIN E. Characterization of materials, Part II [M]. Beijing:images and their evaluation by image analysis for activated carbonsScience Press, 1998.with disordered microtexture []. J Mater Sci, 1998, 33: 199- 206.[7] OSHIDA K，ENDO M, NAKAJIMA T, s. L. di Vittorio,[12] DOBB M G GUO H, JOHNSON D J Image analysis of ltticeDRESSEIHAUS M S, DRESSEIHAUS G Image analysis of TEMimperfections in carbon fibers [J]. Carbon, 1995, 33: 115-1120.pictures of furine-intercalated graphite fibers [0]. J Mater Res, 1993，[13] ARPAD B P, LENORE C R, ADEL F S. Effect of oxidation on the8(3): 512-522.microstructure of carbon blacks 小Energy & Fuels, 1996， 10:[8] OSHIDA K, ENDO M, DRESSEIHAUS M s, DRESSEIHAUS G254- 259.Analysis of pore structure of activated carbon fibers using high14] SHIM H s, ROBERTtH. H, NANCY Y. C. YANG A methodologyresolution transmission electron microscopy and image processingfor analysis of 002 lttice fringe images and its application to[]. J Mater Res, 1995, 10(10): 2507- -2517.combustion-derived carbons []. Carbon, 2000, 38: 29 -45.[9] OSHIDA K, NAKAZAWA T, MIYAZAKI T, ENDO M. Application15] KENNETH R C. Digital Image Processing [M]. Beiing: Electricalof image processing techniques for analysis of nano- andIndustrial Press, 2002.micro-spaces in carbon materials[]. Synthetic Metals, 2002, 125:(Edited by CHEN Can-hua)223- -230.[10] HUANG zZ H, KANG F. Y, HUANG W L, YANGJ B, LIANG K M,中国煤化工MYHCNMH G.