Grain Size Measurements Using Circular or Rectangular Test Grids

George F. Vander Voort

Consultant – Struers Inc.
Wadsworth, Illinois USA

Abstract

Experiments were conducted to determine the influence of the number of grains counted using
the Jeffries planimetric procedure of ASTM E 112 with a single test circle compared to the
number counted using rectangles as suggested by Saltykov. Both planimetric methods produced
similar estimates of the ASTM grain size number when the counts per field placement were high.
The Saltykov rectangle method is a viable test method and produced good data down to low
count numbers. Bias was not observed at low counts, only data scatter. The intercept count
method, using only a single test circle per field, yielded slightly higher estimates of the ASTM
grain size number, but the difference was relatively insignificant.

Introduction

Measurement of the grain size of metals and alloys [1-5] is the most important quantitative
measurement performed due to the influence of grain size upon properties and service
performance. Grain shapes must be space filling and the grain surfaces must meet certain laws
regarding minimum surface area and surface tension. Simple shapes, such as spheres and cubes
may meet one of these rules but not all rules. Theoretical studies of space filling shapes with
minimum area and surface tension began with the work of Lord Kelvin in 1887. He postulated
that the ideal shape meeting these rules is a polyhedron known as a tetrakaidecahedron which has
14 faces, 24 corners and 36 edges. The pentagonal dodecahedron, which is not a space-filling
shape, correlates well with actual grain measurements. Several researchers have made actual
measurements of three-dimensional metallic grains. Desch, for example, used liquid-metal
embrittlement of coarse beta brass to study grain shapes. He found that the number of faces per
grain varied from 11 to 20, with 14.5 as the average. Larger grains were more complex than
smaller grains, not surprisingly. The number of sides per face varied from three to eight and five-
sided (pentagonal) grains were most common. The average number of edges per face was 5.14.

But, metallographers are working with a two-dimensional plane-of-polish through the three-
dimensional structure. Furthermore, they may be observing several types of grain structures:
non-twinned “ferrite” grains from BCC metals and FCC aluminum; FCC grain structures
containing twin boundaries; and, prior-austenite grain boundaries in heat treated steels. To also
complicate the situation, the grains may be equiaxed in shape, for example, in recrystallize
annealed or solution annealed specimens with a single uni-modal grain size distribution, or a
duplex, bi-modal grain structure condition may be present where there are distinct (or
overlapping) populations of smaller and larger sized grains. Figure 1 shows examples of uni-
modal and bi-modal grain structures. If the metal has been cold worked, the grain structure will
be elongated in the deformation direction, Figure 2, and the grain size and shape will vary the
three principle planes; longitudinal, planar and transverse, thus adding to the difficulty of
measuring grain “size.” So far, we have been discussing single phase microstructures, but there
are many classic examples of duplex, or two-phase microstructures, Figure 3. If those were the
only problems to deal with, the metallographer’s life would be easier. However, revealing the
grain structure fully for all of these cases is far from simple.

Measuring Grain Size

ASTM Committee E-4 on Metallography was founded in 1916. Their first standard, ASTM E 2
was proposed in 1917 [6], it included a detailed description of how to measure grain size using
the planimetric method written by Zay Jeffries [7,8], based upon earlier work done by his
graduate school advisor, Albert Sauveur [9], plus a brief account of the intercept method
developed by Emil Heyn [10]. At that time, grain size was defined in terms of the number of
grains per mm2 at 1X by the planimetric method and by the average intercept length by the
intercept method. The ASTM equation relating the number of grains per in2 at 100X and the
ASTM grain size number was introduced in 1951 by an unknown Timken member of E-4 when
E 91-51T was introduced [11]. The equation is:

N = 2G-1 (1)

where n is the number of grains per in2 at 100X and G is the ASTM grain size number.

The Jeffries planimetric grain size method utilizes a test circle with a diameter of 79.8 mm which
is superimposed over the microstructure. The magnification is chosen to give at least 50 grains to
be counted. This wording could be improved. The rater must count all the grains that are
completely inside the test circle, ninside, and all the grains that are intercepted by the circle,
nintercepted. It is assumed that, on average, half of the intercepted grains are inside the test circle
and half are outside. To get an accurate count, the operator must mark off the grains as they are
counted using a felt tip pen, etc. This, however, makes the method slow and less popular. The
calculation is:

NA = f(ninside + 0.5nintercepted) (2)

where NA is the number of grains per mm2 at 1X and f is the Jeffries multiplier:

f = M2/5000 (3)
and M is the magnification. If the test area is different than 5000 mm2 (from a circle 79.8 mm
diameter), then the alternate area used is divided into the magnification squared.

The average grain area, A, is the reciprocal of NA. The ASTM grain size number is calculated
by:

G = 3.321928 LogNA - 2.954 (4)

G is rounded off to the nearest tenth value. In practice, more than one field must be evaluated to
obtain a good estimate of G.

The intercept method suggested by Heyn is considerably faster to perform manually which has
made it popular, despite the fact that there is no direct mathematical connection between the
mean lineal intercept length and G. Both straight lines and circles have been used as templates,
plus other shapes. For many years, E 112 stated that in case of disputes, the planimetric method
is the referee method. But, in the development of the Precision and Bias section for E 112 [12], it
was observed that the intercept method, because it is faster, developed the required degree of
statistical precision (relative accuracy  10%) in less time, the intercept method was declared to
be the referee method. Subsequent experiments, however, have shown that the planimetric
method yields a better estimate of the actual grain size for a given test area, as all of the grains
within the 5000 mm2 area are being sampled. The three-circle intercept method samples more of
the grains in an area than using a single circle, but does not sample all of the grains within the
periphery of the outer test circle. Hence, in a dispute, the planimetric method is preferred despite
the fact that more work is required.

In the 1976 revision of E 112 by Halle Abrams, he introduced the three-concentric circle test grid
and a more formal methodology for performing intercept grain size measurements. The idea was
that the total circumference of the three circles was 500 mm. He suggested adjusting the
magnification so that, on average, about 100 grain boundary intersections, P, or grain
interceptions, N, would be obtained. Then, five random applications of the three-circle grid
would yield ~500 N or P hits which would give ~10% relative accuracy. This strategy was
verified in the round-robin [12]. However, counting errors seemed to be greater when each
placement yielded ~100 counts compared to a magnification that yielded ~50 counts. So, the
standard was modified to recommend using a magnification that yielded ~50 counts per
placement. The goal was still to get ~500 total counts. Ten randomly chosen fields would give a
more representative estimate of the grain size than five fields.

To do the intercept method, one counts either grain boundary intersections, P, or grains
intercepted, N, by the circles. For a single phase structure, it is easier to do P counts. For a two-
phase structure, one must do N counts. For a single phase grain structure P = N and either count
can be made. The P or N count is divided by the true line length, LT, which is the line length
divided by the magnification, L/M. This yields PL or NL, the number of intersections per unit
length or the number of interceptions per unit length. The reciprocal of PL or NL is the mean
lineal intercept length, L3, which may be designated as l. The mean lineal intercept is related to G
by the following empirical equation:

G = (-6.6457 log L3) – 3.298 (5)

Saltykov [13] suggested using a rectangular test grid, rather than a circle, to minimize bias in
grain counts as the number of grains inside the test figure, ninside, decreases. The grains that
intersect the four corners are ignored in the count of nintersected. As there are four corners, and each
grain intersecting the corners is considered to be 1/4th inside the rectangle, Saltykov adds one to
the quantity:

NA = f[ninside + 0.5nintersected +1] (6)

The purpose of the experimental work reported here is to determine how the number of counts
inside the test figure, whether it is a circle or a rectangle, affects the calculated grain size, G. At
the same time, the circles used for the E 112 planimetric measurements will be used for intercept
counts. These are single circles, however, not three concentric circles as recommended in E 112.

Experimental Program

Two images of a ferritic stainless steel, etched electrolytically with aqueous 60% nitric acid,
were used for the experiments. The magnifications were 100 and 400X and they were located at
the same area. These images were printed on paper with different sizes and circles or rectangles
of various size were superimposed over the images to generate a wide variation in counts. The
images are shown in Figure 1 (left) and in Figure 4.

The 100 and 400X images were printed at a variety of sizes and test circles of varying diameter
were printed on the grain structures to facilitate counting. Forty measurements were made using
the planimetric method of E 112, Figure 5, where the (ninside + 0.5nintercepted) varied from 5 to 444.
The linear trend line shows a slight upward slope as this value decreases. At high counts, the
mean grain size is ~6.7 (6.695  0.059 for the 9 highest count values). For the same 40 test
circles, the intercept method of E 112 (using one test circle rather than three concentric circles)
gave very similar results, Figure 6. Again, the linear trend line shows a slight upward slope with
decreasing Pintersections. For the same nine test circles that yielded the highest counts, the mean
grain size is ~6.75 (6.759  0.108 for the nine highest count values). The agreement between the
two methods is very good.

Figure 7 shows the effect of eliminating (ninside + 0.5nintercepted) values below 30. Note that the
linear trend line is now flat across this range of counts. The mean grain size for the
measurements >30 is ~6.72 (6.718  0.043 for the 21 test counts). Figure 8 shows the effect of
eliminating Pinside values below 20. The mean grain size for measurements >20 is ~6.76 (6.756 
0.10 for the 23 values >20). Note that the trend line slows a slight downward trend as the count
decreases, the reverse of Figure 6 for all data. Figure 9 shows the planimetric data plotted against
the intercept data for the same test circles. The data falls around the 1 to 1 correspondence line.
Such data will always show some scatter because the intercept technique is only sampling the
grains that intersect the circumference of the circle while the planimetric samples those, plus all
of the grains within the test circle. Overall, the data are in good agreement between the two
methods.

Figure 10 shows the planimetric grain size measurements using rectangular test grids and
Saltykov’s counting approach (equation 6). This is based upon 45 measurements using rectangles
of varying size where [ninside + 0.5nintersected +1] varied from 2.5 to 790.5. Note that the linear trend
line is nearly flat over the full range. The mean grain size for the eight highest measurements
(>100 counts) was 6.7 (6.7  0.06). For counts >50, the mean grain size, G was ~6.72 (6.718 
0.052) for 17 measurements. For counts 29, the mean grain size was also ~6.72 (6.722 
0.049) for 20 measurements. Figure 11 shows a plot of the planimetric grain size using
rectangles and Saltykov’s procedure when [ninside + 0.5nintersected +1] is >10 (23 measurements).
Note that the linear trend line is flat across the graph. The mean grain size for counts >10 is
~6.74 (6.737  0.052). Figure 12 shows a comparison of the planimetric method according to
Jeffries using test circles and by Saltykov using rectangles. Over the data range, the agreement is
excellent.

Table 1 summarizes the test results using the planimetric method of Jeffries, as defined in E 112,
the Saltykov planimetric method using rectangles and the modified counting method, and the E
112 intercept method of Heyn using the Jeffries circle, that is, a single test circle, rather than the
Abrams three concentric circle method (which would yield higher counts per field). This shows
that the Saltykov method yields data equivalent to the Jeffries method, with lower 95%
confidence values and better G estimates at lower counts. The Heyn intercept method yielded
slightly higher estimates of G and higher 95% confidence limits, although this is affected by the
lower count rates due to the use of only one test circle.

Conclusions

The experiment showed that the best planimetric grain size results were obtained with a
relatively high count of the number of grains that are completely inside and intersecting the
circle or rectangle. For the classic E 112 Jeffries planimetric method, where the number of grains
inside and number intersecting the circle was 120, the grain size was 6.7. If that number is
lowered to 30, the grain size estimate increased slightly to 6.72. The intercept method, using
only the single test circle yielded a slightly higher estimate of the grain size, 6.76 for both the 9
highest, 42 intersected grains, or for the 23 highest counts, 20 intersected grains.
The planimetric grain size estimates using rectangles yielded the same grain size, G = 6.7, for the
8 highest counts (>100 inside and intersecting). When that number was lower to 50, the
estimate of G was 6.72 and when lowered to  10, G was 6.73 – very consistent results. The use
of rectangles produced the best estimate of the grain size at low counts. The difference in the
grain size estimate using low counts with either a circle or rectangles is quite small. Intercept
counts, although none exceeded 89 intersected grains per placement, yielded slightly higher
estimates of the mean grain size. However, the single test circle does not sample as many grains
using the intercept method as when using the planimetric method. This shows the need to use
several concentric test circles per field, as with the Abrams three-circle intercept method. Bias
was not detected at low counts of the number of grains inside the test circle plus half of the
number of grains intersecting the test circle. Instead, only data scatter was detected.

References

1. G.F. Vander Voort, “Grain Size Measurement,” Practical Applications of Quantitative

Metallography, ASTM STP 839, ASTM, Philadelphia, 1884, pp. 85-131.
2. G.F. Vander Voort, Metallography: Principles and Practice, McGraw-Hill Book Co., NY,
1984; ASM International, Materials Park, OH, 1999, pp. 219-223 and 435-472.
3. G.F. Vander Voort, Committee E-4 and Grain Size Measurements: 75 Years of Progress,”
Standardization News, Vol. 19, May 1991, pp. 42-47.
4. G.F. Vander Voort, “Examination of Some Grain Size Measurement Problems,
Metallography: Past, Present and Future, ASTM STP 1165, ASTM, Philadelphia, 1993, pp.
266-294.
5. G.F. Vander Voort and J.J. Friel, “Image Analysis Measurements of Duplex Grain Structures,”
Materials Characterization, Vol. 29, No. 3, October 1992, pp. 293-312.
6. ASTM E 2-62 (reapproved 1974), “Standard Methods of Preparation of Metals and Alloys.”
7. Z. Jeffries, A.H. Kline and E.B. Zimmer, Trans. AIME, Vol. 54, 1916, pp. 594-607.
8. Z. Jeffries, Trans. of the Faraday Society, Vol. 12, 1916, pp. 40-56.
9. A. Sauveur, Trans. AIME, Vol. 22, 1894, pp. 546-557.
10. E. Heyn, The Metallographist, Vol. 5, 1903, pp. 39-64.
11. ASTM E 91-51T, “Method for Estimating the Average Grain Size of Non-Ferrous Metals
Other than Copper and Their Alloys.”
12. G.F. Vander Voort, “Precision and Reproducibility of Quantitative Measurements,”
Quantitative Microscopy and Image Analysis, ASM International, Materials Park, OH, 1994,
pp. 21-34.
13. S.A. Saltykov, Steremetricheskaya Metallograpfiya (Stereometric Metallography), 2nd
revised and supplemented edition, Metallurgizdat, Moscow, 1958, 444 pgs.
Figure 1: Examples of an equiaxed, uni-modal grain size distribution in a ferritic stainless steel
(left, magnification bar is 25 μm long) and a bi-modal grain size distribution in a ferritic stainless
steel (right, magnification bar is 100 μm long). Both specimens etched with aqueous 60% nitric
acid, ~1 V dc for 120 s (left) and for 20 s (right).

Figure 2: Examples of fully annealed, equiaxed cartridge brass (left) compared to the same
material reduced 60% in thickness (right). Both etched with equal parts of ammonium hydroxide
and hydrogen peroxide (3% conc.); longitudinal planes shown with magnification bars at 100
μm.
Figure 3: Examples of duplex (two-phase) microstructures in (left) a high-carbon version of 430
stainless steel (each phase has a different grain size). The coarse white grains (some are
elongated) are ferrite while the small grains are martensite. Specimen etched with aqueous 60%
nitric acid at 1 V dc for 40 s (magnification bar is 25 μm). The specimen on the right is Cu –
40% Zn (“Muntz” metal) that was cold worked and annealed at 843 C. It was etched with
Klemm’s I reagent and viewed with Nomarski DIC (magnification bar is 100 μm).

Figure 4: Low magnification view of the specimen shown in Figure 1 (left), etched for 2 minutes
at 1 V dc using aqueous 60% nitric acid. These two images were used for the study.
 Planimetric Grain Size, E 112
9
ASTM Grain Size, G

4
0 100 200 300 400 500
(ninside + 0.5nintercepted)

Figure 5: Results for 40 measurements using the E 112 planimetric method with (ninside +
0.5nintercepted) varying from 5 to 444. The linear trend line is almost flat.

Intercept Grain Size, E 112

9
ASTM Grain Size, G

4
0 10 20 30 40 50 60 70 80 90 100

Pintersections
Figure 6: Results for 40 measurements using the E 112 intercept method (but with a single test
circle) with Pintersections varying from 6 to 89. The linear trend line is almost flat.
 Planimetric Grain Size, E 112
9
ASTM Grain Size, G

4
0 50 100 150 200 250 300 350 400 450 500

(ninside + 0.5nintercepted) > 30

Figure 7: Plot of (ninside + 0.5nintercepted) for values greater than 30. Note that the linear trend line is
virtually flat over this range.

Intercept Grain Size, E 112

9
ASTM Grain Size, G

4
10 20 30 40 50 60 70 80 90 100

Pintersections > 20

Figure 8: Plot of Pintersections for values above 20. Note that the linear trend line has a slight
downward slope now (compare with Figure 6 for all data).
 E 112: Planimetric G vs Intercept G
9

8
Intercept G

4
4 5 6 7 8 9

Planimetric G

Figure 9: Correspondence between the grain sizes measured by the E 112 planimetric and
intercept methods. The planimetric method is sampling all of the grains within the test circle plus
those intersecting the circle while the intercept method samples only the grains that intersect the
circle. The line is a 1:1 correspondence line.
 Saltykov Rectangles:
(ninside + 0.5nintersected) +1
9
ASTM Grain Size, G

4
0 100 200 300 400 500 600 700 800

(ninside + 0.5nintersected) + 1

Figure 10: Planimetric grain size measurements using rectangles and Saltykov’s counting
method for 43 rectangles with (ninside + 0.5nintersected) + 1 values ranging from 2.5 to 790.5. Note
that the linear trend line is nearly flat.

Saltykov Rectangles:
(ninside+0.5nintersected +1)10
9
ASTM Grain Size, G

4
0 100 200 300 400 500 600 700 800

(ninside + 0.5 nintersected) + 1

Figure 11: Planimetric grain size measurements using rectangles and Saltykov’s counting
method for 23 rectangles with (ninside + nintersected) + 1 values above 10. Note that the linear trend
line is flat over this range.
 Comparison of Planimetric Data
9
ASTM Grain Size, G

4
0 100 200 300 400 500 600 700 800

(ninside +0.5nintersected +1) or (ninside + 0.5

nintersected)

Saltykov Rectangles Linear (Saltykov Rectangles)

Jeffries Circles Linear (Jeffries Circles)
Figure 12: Although the linear trend line for the Jeffries planimetric method shows a slight
downward trend with increasing counts, over the mutual measurement range for the two
methods, the data agreement is excellent.

Table 1: Summary of the Test Results

E112 Planimetric Circle (n1 + 0.5n2) No. Avg. G  95% CL Std. Dev.
>120 9 6.695  0.059 0.07689
>30 23 6.72  0.0378 0.08747
<30 17 6.964  0.204 0.39729
<10 11 6.886  0.313 0.46615
Saltykov Planimetric Rectangle (n1 + 0.5n2 + 1)
>100 8 6.703  0.063 0.07555
>50 17 6.719  0.0519 0.10091
29 20 6.722  0.049 0.10382
>10 24 6.732  0.047 0.11129
<10 19 6.817  0.199 0.41267
E112 Intercept Circle Pi
42 9 6.76  0.108 0.14264
>20 23 6.76  0.101 0.23256
<20 17 7.016  0.211 0.41083
<10 8 6.851  0.437 0.52314