caracteizacion de la porocidad en el coque por analisis de imagen

7/30/2019 caracteizacion de la porocidad en el coque por analisis de imagen

1/7

CHARACTERIZATION OF POROSITY IN COKES BY IMAGE ANALYSIS

Stein Rrvik1, Harald A. ye2, Morten Srlie3

1 SINTEF Applied Chemistry, Inorganic Chemistry, Trondheim N-7465 Norway2

Norwegian University of Science and Technology, Department of Chemistry, Trondheim N-7491 Norway3 Elkem ASA Research, Kristiansand N-4675 Norway

ABSTRACT

A fully automatic method for image analysis of porosity of cokes

has been developed. The method outputs a continuous pore size

distribution from 1 m to 10 mm, and will therefore cover a larger

range than mercury porosimetry. The method measures only pores

inside the coke grains; voids between coke grains in the sample

are ignored. A selection of calcined commercial cokes in different

fraction sizes has been analysed. There are considerable

differences in the pore size distributions of the different cokes.

INTRODUCTION

A fully automatic method for image analysis of porosity in carbon

materials has been developed within the frame of the Expomat /

Prosmat research program during the last years. The method is

based on computerised image analysis and optical microscopy,

and is capable of analysing large sample areas (several cm2). It

provides a logarithmic size distribution of pores in the range of 1

m to 10 mm pore radius. In addition to this size distribution, a

relative measure of pore surface area and pore connectivity is

given.

The main aspects of this image analysis method have been

published earlier [1], as a method to analyse porosity in anodes.

This paper describes how the previous method has been modified

to be suitable for analysis of coke grains.The main problem to resolve with respect to porosity analysis of

cokes is how to separate pores inside coke grains from the voids

between grains. Vibrated bulk density (VBD) of a narrow coke

fraction is a common measure of macroporosity in cokes. VBD

includes all voids between grains, and the result will then be

dependent on grain size and shape. Mercury porosimetry ignores

voids between grains (provided that the coke particle size is not

too small) at the same time as it fills large pores inside the coke

grains. Hence, both these methods have disadvantages. This paper

shows how image analysis can be used to ignore voids between

grains and include all pores inside grains. Comparison with VBD

and mercury porosity will be given. The analysed cokes presented

in this paper are petroleum cokes used for anodes in the

aluminium industry, made by six different producers.

METHOD DESCRIPTION

A brief summary of the previously published method [1] follows

here:

1. Coke grains are sieved to different fraction sizes and

impregnated with a fluorescent epoxy1 under vacuum. The

sample surfaces are ground and polished after curing of the

epoxy. Pores filled with fluorescent epoxy light up brightly if

viewed with ultraviolet light in a microscope. This reduces

the error of creating false pores when cuttingand polishing

the samples.

2. The samples are examined using a standard inverted reflected

light metallurgical microscope (Leica MeF3A), equipped

with a motorised XY- stage and focus controller. The stage

movement and focus is controlled directly by the computer

image analysis software. Digital images are acquired using

an electronic 3-chip CCD2 video camera (Sony DCX 930P)

and a frame-grabber card.

1 The epoxy is two-component and consists o f Bisphenol-A-Diglycidyl-Ether and Tri-Ethylene-Tetramin with Sodium-Fluorescein added as fluorescent dye. Product trade

names are Epofix and Epodye; both made by Struers, Denmark.

2 CCD is an abbreviation for Charge Coupled Device. These cameras use a chip withan array of sensors that accumulates an electrical charge proportional to the amount of

light exposed onto them.
http://quit/http://../NO_INDEX/AUTHORS.PDFhttp://../NO_INDEX/CONTENTS.PDFhttp://../MAINMENU.PDF


2/7

3. A grid of adjacent images, sufficiently large to cover most of

the sample is acquired automatically by the computer and

stored on disk. The images are then analysed in batch. When

the adjacent frames are analysed, the pores entirely inside each

frame are measured while the pores cut by the image edges are

not. These pores are saved and measured after four frames have

been analysed and merged. This process continues recursively,

allowing arbitrarily large pores to be measured. There is no

upper limit to the measured pore size. The advantage of thismethod is obvious: Mercury porosimetry does not give reliable

values above 40-50 m, while the pore size that can be

analysed by the image analysis method is only limited by the

movement of the stage, which is 50 mm. The disadvantage

with the image analysis method is that it does not include

microporosity below 1 m pore radius.

4. The resulting data is merged using Microsoft Excel macros

and presented using templates (Figure 1). The image analysis

outputs the porosity values as thesum of the areas pores with a

specified inscribed radius cover, as a percentage of total

analysed sample area. The sum of all porosity values is thus

equivalent to the total porosity. The overview image in Figure

1 shows pores greater than about 50 m radius and gives useful

visual information such as the apparent homogeneity of thecoke grains.

5. The image analysis procedure is fully automated, it only

requires an operator to place the sample on the microscope and

start the procedure using the desired parameters. With

sufficient storage space, images of a series of cokes can be

acquired during daytime and be analysed in batch by the

computer during the night. At medium magnification (80x)

16x24 = 384 frames are required to cover a 30 mm sample. It

takes about half an hour to acquire these frames and 2 hours to

analyse them. See Figure 2 for an overview of how the sample

is covered by the adjacent frames.

The computer software used was the general image analysis

Macintosh application NIH image, developed by Wayne Rasbandat the National Institute of Health in the US. This software is

available in the public domain, and can be downloaded freely from

ftp://rsbweb.nih.gov/pub/nih-image/3. The source code has been

customised by adding support for the microscope hardware and

some extra image analysis procedures. The NIH image software has

a Pascal-like macro language, which was used to control the

analysis. A proper macro programming language is essential for this

kind of work.

3Alternate sources for NIH image are from Library 9 of the MacApp forum

on CompuServe, and on floppy disk from NTIS, 5285 Port Royal

Rd.,Springfield, VA 22161, part number PB93-504868.

Porosit (smoothed) [%]

1 0.0000

1.26 0.0000

1.58 0.0000

2 0.0000

2.51 0.0637

3.16 0.1275

3.98 0.0637

5.01 0.0764

6.31 0.1826

7.94 0.2121

10 0.2687

12.59 0.3484

15.85 0.416119.95 0.5255

25.12 0.6604

31.62 0.8309

39.81 0.9369

50.12 0.9147

63.1 0.9182

79.43 0.9522

100 0.9512

125.89 0.9672

158.49 1.1726

199.53 1.5764

251.19 1.7463

316.23 1.5828

398.11 1.5538

501.19 1.4824

630.96 1.2417

794.33 0.9543

1000 0.3751

1258.93 0.0000

1584.89 0.0000

1995.26 0.0000

2511.89 0.0000

3162.28 0.0000

3981.07 0.0000

5011.87 0.00006309.57 0.0000

7943.28 0.0000

Sum 21.1016

SF pores 0.07639

SF carbon 0.02336

Porosity 21.10

Connectivity 0.10

SampleName 17.7.1

Figure 1: Standard diagram for pore size analysis of a single

sample, created by a Microsoft Excel template.

Figure 2: Schematic view of the order the images are acquired at

80x magnification, with the sizes involved for a 30 mm sample.

Sample - Coke

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

10

Pore Radius [m]

Porosity[%]

Porosity [%]

Porosity (sm) [%]


3/7

PORES VS. VOIDS BETWEEN GRAINS

As explained in the introduction, the main problem with measuring

porosity in cokes is to distinguish between pores inside grains and

voids between grains. Figure 3 shows an example image with coke

grains (from the 1.0-2.0 mm fraction) embedded in epoxy. The

carbon in the coke appears grey in this image, while the epoxy with

the fluorescent dye is black. The grains were embedded in epoxyunder vacuum. The epoxy wets the carbon well, so the epoxy fills

the pores almost completely. The fraction of closed pores is less

than a percent and these pores are usually smaller than 5 m. Closed

pores will be ignored by the image analysis.

Most image analysis programs have a function to discard pores

touching the edge of an image. Figure 4 shows the effect of this

function. Pores connected to the image edge (the surrounding epoxy

will always cross the image edge) are here rendered in grey, while

unconnected pores are black. The large pore inside the grain on the

right is connected to the surrounding epoxy through some cracks.

This is very typical. Most cracks in the coke grains are connected to

the outside. It is obvious that regular image analysis of such images

will give a too low porosity. The larger the pores, the lower the total

porosity will be.

A common way to resolve the problem with connected pores is by a

technique called watershed segmentation [3]. This technique is

based on an erosion of features until one point is left, and then a

conditional iterative growth that avoids joining features. The effect

of this segmentation is that white lines will be drawn across all parts

of features that are narrower than the neighbourhood. Figure 5

shows the result of applying watershed segmentation to Figure 4.

The black areas show the features that will be disconnected from the

edge and analysed. In this case, far too many pores will be analysed.

All areas between the coke grains will be measured as pores.

As Figure 5 shows, regular watershed segmentation will not give

the desired result. The implementation was therefore changed with a

condition that does not create separation lines longer than aspecified width. The value of this variable was called

MaxDisconnect (MD). This is in some applications called limited

watershed erosion. Figure 6 shows the result of the modified

watershed segmentation, using a value of 200 m in MD. Some of

the areas between grains will now not be included, but there is still

too much intergrain porosity measured.

Figure 7 shows the result of the modified watershed segmentation

with a value of MD decreased to 50 m. Most of the intergrain

porosity is now ignored, except the area in the middle of the image.

The grains are in this case closer than 50 m. Decreasing the value

of MD further will cause the cracks inside the coke to be ignored as

well, which is not desired. A different approach to the problem is

therefore needed.

Some condition must be added that will separate the areas based on

maximum size, in addition to the maximum separation width.

Simply setting a limit to the maximum size of the pores that will be

measured has been shown not to work very well. Especially with

sponge-like coke grains, the size of the pores inside the grains are

sometimes as large or larger than the size of the voids between the

grains. Because of this, the implementation was changed to includea condition based on relative size rather than absolute size.

Separation lines will only be drawn between features where the

smallest feature has a maximum inscribed circle radius smaller than

a specified percent below the largest features radius. This

percentage was called RelativeSizeDifference (RSD). A RSD value

of 0% means that the condition has no effect; all features will be

disconnected as usual. 100% means that no features will be

disconnected. 33% means that features will be separated if the

smaller feature has a size less than 2/3 of the larger feature. Tests

have shown that values between 10% and 40% work well. Figure 8

shows the result of the modified watershed segmentation with a MD

value of 50 m and a RSD value of 20%. Now the large area

between the coke grains is excluded from the measurement, because

its maximum radius is about the same as the maximum radius of the

other areas between grains. The crack inside the grain on the left is

measured, because its maximum radius is much smaller than the

areas between the grains. Some small short segments between flat

edges of adjacent grains will still be included, but that is difficult to

avoid without setting the limits too tight.

The value of the MaxDisconnect variable must be increased at

increasing grain size. Larger grains have larger pores and cracks

along edges that should be included in the measurement, but also a

larger distance between the particles. The larger distance between

particles appears because the surface examined is a plane cutting the

coke grains at random sections. For spherical grains, the intersected

area will be a sinus function of the maximum diameter. The average

distance between grains in an intersection will therefore be roughly

proportional to the grain size.

In this work, a MD value of 25 m was used for measurements of

the 0.5-1.0 mm fraction; 50 m was used for the 1.0-2.0 mm

fraction and 75 m was used for the 2.0-4.0 mm fraction. These

values gave a satisfactory separation between pores inside grains

and voids between grains.


4/7

Figure 3: Coke grains embedded in epoxy

Figure 4: Pores connected to surrounding epoxy (grey)

Figure 5: Pores split by regular watershed segmentation (black)

Figure 6: Modified watershed segmentation, MD=200

Figure 7: Modified watershed segmentation, MD=50

Figure 8: Modified watershed segm., MD=50, RSD=20%


5/7

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

1 10 100 1000

Radius [m]

Porosity[%]

Coke A, 2.0-4.0 mm

Coke A, 1.0-2.0 mm

Coke A, 0.5-1.0 mm

Figure 9: Pore size distributions for fractions of coke A

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

1 10 100 1000

Radius [m]

Porosity[%]

Coke B, 2.0-4.0 mm

Coke B, 1.0-2.0 mm

Coke B, 0.5-1.0 mm

Figure 10: Pore size distributions for fractions of coke B

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

1 10 100 1000

Radius [m]

Porosity[%]

Coke C, 2.0-4.0 mm

Coke C, 1.0-2.0 mm

Coke C, 0.5-1.0 mm

Figure 11: Pore size distributions for fractions of coke C

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

1 10 100 1000

Radius [m]

Porosity[%]

Coke D, 2.0-4.0 mm

Coke D, 1.0-2.0 mm

Coke D, 0.5-1.0 mm

Figure 12: Pore size distributions for fractions of coke D

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

1 10 100 1000

Radius [m]

Porosity[%]

Coke E, 2.0-4.0 mm

Coke E, 1.0-2.0 mm

Coke E, 0.5-1.0 mm

Figure 13: Pore size distributions for fractions of coke E

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

1 10 100 1000

Radius [m]

Porosity[%]

Coke F, 2.0-4.0 mm

Coke F, 1.0-2.0 mm

Coke F, 0.5-1.0 mm

Figure 14: Pore size distributions for fractions of coke F


6/7

RESULTS AND DISCUSSION

Figure 9 to Figure 14 show the pore size distributions for 6

different cokes. Three different fraction sizes were analysed: 0.5 to

1.0 mm, 1.0 to 2.0 mm and 2.0 to 4.0 mm. As expected, the porosity

decreases with decreasing fraction size. The difference is larger

between the 0.5-1.0 and the 1.0-2.0 fraction than the 1.0-2.0 and

2.0-4.0 mm fraction. Most cokes have a similar size distribution up

to 10 m, where the smallest fraction starts to decrease in porosity.

The two larger fractions decrease in porosity from 20 to 50 m

radius, and then increase again up to 100 m. This two peak

distribution is typical for calcined cokes. The largest diameter peak

is due to large, round gas entrapment pores that are present in the

green coke. The smallest diameter peak is due to the slit-like pores

and cracks that evolve during the calcination process.

One of the cokes, B, has a much different distribution compared to

the other cokes (Figure 10). The 0.5-1.0 mm fraction is similar to

the other cokes, but the two coarser fractions have a much lower

porosity and only one peak, around 30 m. The 2.0-4.0 mm fraction

has a lower porosity than the 1.0-2.0 mm fraction up to 50 m.

Coke B is known to have a higher content of shot coke than the

other cokes. The shot grains, which have a lower porosity than the

other grains, partially survive the crushing and get concentrated in

the coarser fractions.

Figure 15 compares mercury and image analysis porosity for the

middle fraction. It is seen here that coke B does not have lower

porosity than the other cokes as measured with mercury porosity,

but the difference in macro-porosity is easily seen with image

analysis. Coke B also has the largest difference between porosity as

measured by mercury porosimetry and by image analysis.

coke 1.0-2.0 mm

0

2

4

6

8

10

12

1416

18

20

Coke

A

Coke

B

Coke

C

Coke

D

Coke

E

Coke

F

Porosity[%]

Total Porosity, Hg [%]

Total Porosity, IA [%]

Figure 15: Comparison of total porosity measured by image

analysis (IA) and mercury porosity (Hg)

Figure 16 and Figure 17 shows a 12.5 x 12.5 mm overview area of

the 1.0-2.0 mm fraction of coke A and B, respectively. The black

areas are pores that have been measured, while the grey areas are

ignored. It is evident from these images that the technique described

above is quite successful in separating pores inside coke grains and

voids between grains.

Figure 16: Overview of 1.5 cm2 area of coke A

Figure 17: Overview of 1.5 cm2 area of coke B

The shot coke grains of coke B are easily seen in Figure 17. These

grains do not have the thin calcination pores that can be seen in

coke A. Coke C through F are visually similar to coke A.


7/7

The only coke other than coke B that shows a lower porosity below

30 m for the largest fraction is coke E (Figure 13). This coke is a

mixed source coke. It probably behaves similarly to coke B: Some

grains are less porous than the other grains, so that these stronger

grains get concentrated in the largest fractions.

Another coke that deviates from the other cokes in size distribution

is coke F (Figure 14). This coke is known to be more brittle and to

have a more fibrous structure than the other cokes. (Coke F is

calcined in a rotary hearth calciner, while the rest are calcined inrotary kilns.) This coke has a larger difference in porosity between

the two largest fractions above 100 m. This indicates that the

porous grains do not survive the crushing as well as in the other

cokes.

The pores below 50 m seem to be most important for strength.

Figure 18 compares coke porosity with Hardgrove Grindability

Index (HGI), which can be considered an inverse measure of coke

grain strength. HGI values were not available for coke A to F, so

Figure 18 shows data for a different set of cokes, analysed at a

different laboratory. These cokes are also commercial petroleum

cokes used in the aluminium industry. There is a trend that the HGI

increases with increasing porosity in the 5-50 m interval, while the

other porosity intervals do not have any significant effect.

coke 1.0-2.0 mm

0.0

2.0

4.0

6.0

8.0

10.0

12.0

20 25 30 35 40

HGI (0.60-1.18mm) [%]

Porosity[%]

Porosity [%] 0-5 m

Porosity [%] 5-50 m

Porosity [%] 50-10000 m

Figure 18: Comparison of coke porosity and HGI

Figure 19 shows a comparison of coke porosity and mill time. Mill

time is the time a given fraction of coke (1.0-2.0 mm) requires in a

ball mill to get 70% of the mass below 200 mesh. It is therefore a

measure of grain strength. The trend is that a decreasing amount of

small pores (0-25 m radius) will give a longer time in the mill. The

larger pores (above 25 m radius) do not seem to correlate with mill

time. Both Figure 18 and Figure 19

indicates that the smallest pores are more important for coke

strength than the larger pores.

coke 1.0-2.0 mm

0

2

4

6

8

10

12

4 24 44 64 84

Mill Time 70 % -200# [min]

Porosity[%]

Porosity [%] 0-25 m

Porosity [%] 25-250 m

Porosity [%] 250-10000 m

Figure 19: Comparison of coke porosity and mill time

CONCLUSION

Image analysis is useful for analysing macroporosity in coke grains.

Image analysis can be used for measuring pores up to several mm

size, and is able to separate between pores inside coke grains and

voids between grains. Image analysis has shown to be a good

alternative to mercury porosimetry analysis of anode cokes.

ACKNOWLEDGEMENTS

Financial support from The Research Council of Norway and the

Norwegian aluminium industry (via the EXPOMAT and

PROSMAT research programs) is gratefully acknowledged. Thanks

are also due to Elkem Aluminium ANS and Hydro Aluminium for

providing coke samples and analysis data.

REFERENCES

[1] Stein Rrvik, Harald A. ye:A Method for Characterizationof Anode Pore Structure by Image Analysis. The Minerals,Metals and Materials Society (TMS); Light MetalsProceedings 1996, p. 561-568.

[2] Kjell Kvam, Harald Schreiner, Stein Rrvik, M. Srlie, H.A. ye: Porosity Development in Sderberg Anodes

Laboratory Simulation. Extended Abstract 24th

BiennialConf. on Carbon, Charleston (South Carolina, USA)American Carbon Society 1999.

[3] J.C. Russ: The Image Processing Handbook; CRC Press,USA (1995) ISBN 0-8493-2516-1, page 476-477.

caracteizacion de la porocidad en el coque por analisis de imagen

Documents