INFLUENCE OF CHEMISTRY AND MICROSTRUCTURE
ON PORTLAND CEMENT CLINKER GRINDABILITY
C.E. BUCHANAN JR., ROAN INDUSTRIES INC.
H.J. BAYLES, COPLAY CEMENT COMPANY
ABSTRACT
A simple test is presented which can detect differences in clinker grindability. The results
obtained correlate fairly well with clinker chemistry and with the clinker structure as determined by the scanning electron
microscope.
INTRODUCTION
A simple test was needed to evaluate various materials for efficiency as grinding aids.
From this evolved the current test procedure which can be used to predict the relative grindability of Portland cement clinkers.
PREVIOUS TESTS
In all probability soon after man broke his first rock, he began wondering how much energy
was required to split it, and why some rocks split easier than others. It was not until 1867 however that Rittinger proposed
that the new surface produced was directly proportional to the work input, and thus began the first quantification of these
relationships. In 1885, Kick stated that the reduction in volume of the particles was directly proportional to the work input.
In 1951, the Hardgrove-Machine method, developed by the Babcock & Wilcox Company, was accepted by ASTM as a standard for
coal grinding. (1) The Hardgrove-Machine is now used for a variety of materials, with an index of from 26 to 50 being reported
for 23 Portland cement clinkers with an average of 37.8. Also in 1951 Mr. Fred Bond of the AllisChalmers Co. proposed his
third theory of communition which states that "The work input is proportional to the new crack tip length produced in particle
breakage and equals the work represented by the product minus that represented by the feed." (2) From this theory, Mr. Bond
developed his work index (Wi), which is used extensively today. Eight Portland cement clinkers showed a low of 10.8 a high
of 18, with an average of 13.2.
DEVELOPMENT OF PROCEDURE
Both of these tests are good and have been used extensively to determine clinker grindability
and in turn to size finish mills when commercial grindability was not available. However, they are relatively expensive, require
a considerable equipment investment, need a large sample, and are time consuming. Needing a quick method of evaluating a series
of materials to determine their relative worth as a grinding aid, I decided to investigate the Bleuler mill. This mill uses
a small sample size, produces Portland cement fineness in a short period of time, and is available to most Portland cement
laboratories.
The first attempt was to use a fixed grinding time. It soon became apparent that a precise
grinding time could not be achieved and that only one fineness level was reported. It was then decided to use four different
times, each of approximately the same duration, but with the actual time measured accurately. The first times selected were
1,2,3 and 4 minutes. After evaluating the data, it was apparent that the 1 and 4 minute time were higher than anticipated.
An examination of the procedure showed that the 3 and 4 minute grinds were sticking to the mill, and for clean out, sand with
a good amount of grinding aid was being used. Apparently there was enough residual grinding aid in the container to influence
the result of the following grind. The procedure was then changed to 15, 30, 60 and 120 seconds. The mill is cleaned after
120 seconds but with sand only.
Consequently, the procedure used is to take 47 grams of clinker which has been passed through
a jaw crusher, add 3 grams of gypsum, grind for the prescribed time, measure the time accurately, and determine Blaine surface
area. A typical work sheet is shown in Table 1. A simple program has been written for a Hewlett-Packard Model 86 computer
using VisiCalc in which the grinding time in seconds, and Blaine surface area are entered for the four grinds. The natural
logarithm of the grinding time is calculated, and linear regression computed. Blaines are then calculated at exact grinding
time of 15, 30, 60 and 120 seconds. An evaluation is made principally on the basis of the correlation coefficient being above
.985 of whether additional grinds are needed or not.
TEST RESULTS
Thirty three clinkers have been tested in this program. The grindability data is shown in
Table 2 with the calculated Blaine results at 15, 30, 60 and 120 seconds, the slope of the regression curve (A1), the standard
error of estimate (SY.X), the correlation coefficient (r), and the coefficient squared (R), being
given. It can be noted that a very good relationship was obtained between the natural log of grinding time and Blaine, with
an average SY.X of 80 and a maximum of 154 Blaine points being obtained. The data for sample A190 is shown on figure 1, along
with the regression lines for the easiest, hardest and average grinding clinkers.
The chemical analyses are shown in Table 3 with their potential composition being given
in Table 4. These show a wide variation in chemical composition as well as being manufactured in wet kilns, grate and four
stage preheaters, and flash calciners. The compound composition was calculated using the equations found in ASTM C-114, the
lime saturation factor (LBS..) from the formula CaO-(1.65 x A12O3 + 0.35 x Fe2O3 + 0.7 x SO3) x 100 all divided by 2.8 x SiO2.
(3) The percent liquid was calculated using 2.95 x A12O3 + 2.2 x Fe2O3 + MgO -- alkalies. The silica ratio (S.M.) was simply,
the silica divided by the sum of the A12O3 and Fe2O and the (A.M.) was the alumina divided by the iron.
Simple correlation is shown on the 32 clinkers in Table 5. It is interesting to note that
when grinding time is considered, A12O3, Fe2O3, CaO, Loss, Free CaO, and C3S all had increasingly positive coefficients with
increased grinding times while SiO2, MgO, SO3, K2O, and the S.M. all had increasingly negative ones.
Multiple regression was then run on the data, using up to 22 independent variables. It soon
became apparent that sample A254did not fall on the regression curves. Consequently it was dropper from the regression and
32 observations used. This can be justified because A254 was clinker dust from a clinker cooler dust collector rather than
straight clinker. The 120 second grinding time gave the best model fit and it was used to draw all of the conclusions concerning
hardness of grind. A summary of the stepwise multiple regression is shown in Table 7. In addition the 15 second grind summary
is shown in Table 6, and the slope, (Al), is shown in Table 8.
In this particular program, a F to add and a F to delete is given and the program continues
until no more independent variables can be added or deleted.
If you observe Table 7, you can note that the S.M. correlated best, with the Loss, Fe2O3,
SO3, K2O, L.S.F., CaO, Liquid, A.M. and MgO following. The last step then eliminated the Fe2O3 from the model. The best model
however probably would use just the S.M., Loss, and the Fe2O3.
Consequently, conclusions can be drawn that there is a definite relationship between grindability
and chemistry, that it improves as the Blaine surface area increases, and that Type I fineness approximately 65% of the variation
can be explained by the chemical analyses.
S.E.M. RESULTS
In order to see if the clinker structure and physical appearance played a part in grindability,
five clinkers were selected for examination by the Scanning Electron Microscope (S.E.M.) based on the grindability data determined
above. The two samples with hardest grindability and the two easiest grinding samples were selected as well as a fifth sample
located between the hardest and the softest grinding ones. Specimens were studied at 20x, 200x, 300x, and 1300x.
Figures 2a, through 2e illustrate the 20x series arranged from easiest to hardest grinding.
We can observe more porosity and a more prominent structure in the harder grindability compared to the easier one.
Figures 3a through 3e at 200x illustrates more dramatic differences with the harder grindability
exhibiting a well formed prominent crystal structure compared to the easier one which shows less apparent and poorer defined
crystals.
Figures 4a through 4e at 300x illustrates the same relationship of crystallinity as does
the 200x.
Figures 5a through 5e at 1300x gives the closest look at the relationship of crystal structure
and grindability for these samples.
CONCLUSIONS
In conclusion, a simple grinding test is presented that can correlate clinker hardness to
crystal structure and appearance as seen in the S.E.M. and chemistry as shown by multiple regression analysis.
REFERENCES
(1) Pit & Quarry Handbook, Seventy Third Edition, 1980-1981, page B-86
(2) Don Olson, Grinding Mill Theory, Portland Cement Association Mill Grinding Short Course,
Paper No. 2, 1977
(3) A.K. Chatterjee, Cement Raw Materials and Raw Mixes, Pit & Quarry, September 1979,
page 103
FIGURE 1, SAMPLE A190 COMPARED TO OTHER CLINKERS

TABLE 1-WORK SHEET FOR REGRESSION CALCULATIONS