DE LANGE, T. and VENTER, P.E.

The use of simulated Tromp partition curves in developing the flowsheet

of plant extensions at Grootegeluk Coal Mine. APCOM 87. Proceedings of the Twentieth International Symposium
on the Application of Computers and Mathematics in the Mineral Industries. Volume 2: Metallurgy. Johannesburg,
SAIMM, 1987. pp. 295 - 311.

The Use of Simulation Tromp Partition Curves in

Developing the Flowsheet of Plant Extensions at
Grootegeluk Coal Mine
T. DE LANGE and P.E. VENTER
Grootegeluk Coal Mine, [scor Ltd, South Africa

Extensions to the beneficiation plant at the Grootegeluk Coal Mine are cur-
rently being planned in order to provide power station coal for Escom's Matim-
ba Power Station, presently under construction. A simulation model was
developed on the Olivetti M-24 microcomputer to facilitate the development
of the envisaged flowsheet. The computer program simulates the heavy-medium
separation and screening unit operations. The model is based on the construc-
tion of the Tromp partition curve described by an arctangent function, allowing
ideal and non-ideal separations.
The paper discusses the determination of the parameters required by the
arctangent curve from standard process parameters such as Epm, Wolf cut-
point, Tromp cutpoint and screening efficiency. The accuracy and shortcom-
ings of the model are discussed, while an overview of the application of the
model to evaluate borehole and bulk sample analyses is also given. It is con-
cluded that the model is an invaluable aid to general flowsheet development.

Introduction
The Grootegeluk Coal Mine, situated in the model involved, the accuracy of the model
Waterberg coal-field 20 km west of Ellis- for both heavy-medium separation circuits
ras, is Iscor's major source of coking and screening operations, and illustrates
coal. The beneficiation plant treats some the application of the model during the
3 000 t/h of raw coal from the Upper and flowsheet development phase.
Middle Ecca, yielding approximately 12%
coking coal and 24% of a middlings frac- System demands
tion. Borehole and bulk sampling campaigns were
Extensive plant expansion became neces- launched as the first steps in the
sary when the contract to supply 12 x 106 development of the envisaged flowsheet.
tons per annum of middlings coal to The analyses specified on these samples
Escom's Matimba Power Station, currently included, among others, fractional ash,
under construction, was awarded to Iscor. heat value ~ sink/float analyses.
A computer model was developed as part of Conventionally, these sets of data
the flowsheet design to facilitate would have been manipulated by hand to
detailed evaluation of borehole and bulk yield fractional densimetric curves and
sampling data. Mayer curves. Ideal cutpoints would then
This paper describes the simulation be utilized to calculate graphically the

THE USE OF SIMULATED TROMP PARTITION CURVES 295

 expected yields at specified heat values compute full partition curves for either a
or ash values. density or a screening operation. The
However, this conventional method had computation of separation product charac-
the following disadvantages: teristics could then be facilitated.
a) The method was tedious and time To evaluate borehole data (in addition
consuming; to the above) this model had to be able to
b) the data could not be analyzed into a) compute blended washability data for
satisfactory depth, due to time limi- different boreholes; and
tations; and b) compute expected yields at specified
c) only ideal separations could be ash, heat value or density cutpoints.
induced, while the extrapolation to By using this model it is clear that the
practical expected organic and supplied data from borehole and bulk
screening efficiencies proved to be samples could be manipulated extensively,
complicated and time consuming. maximising the usage of given data, thus
The need therefore existed to utilize a permitting the development of a more accu-
computer simulation model in order to rate and practically orientated plant
eliminate the disadvantages of the hand flowsheet than would have been possible
method. The model had to be able to otherwise.
a) accept the raw data in the format
available, calculate the information Overview of existing models
required to draw washability curves The specific requirements as discussed
and present results in a summarized above, together with the given time con-
format.; straints, virtually excluded the use of
b) simulate the heavy-medium and size an existing simulator package. Al though
separation unit processes by calcula- Iscor Wd.S negotiating the purchase of
ting partition factors, using a his- l'DDSIM from Prof. R.P. King< 1) of the
tory of standard process parameters University of the Witwatersrand, the
accumulated on the Grootegeluk coking package was not available at that stage,
coal beneficiation plant; as it was still being prepared for distri-
c) allow a choice of Epm values, organic bution.
efficiencies and other parameters in The use of other simulation packages was
order to simulate, non-ideally, ,the excluded owing to
specific heavy-media separation pro- a) inability to cater for the format of
cess required at each point in the the raw data;
flowline. Simulation of the non-ideal b) equipment information required by
screening process had to be facili- such packages, which was unavailable
tated by the choice of screening at that stage of the design; and
efficiencies; c) the time constraints involved, i.e.
d) compute the total sink/float and purchasing and commissioning.
screen analyses of both products of The decision was therefore taken to use
each separation step, in order to an available APPLE II + microcomputer and
permit a step-by-step development of develop a model in-house. The APPLE was
a logical flowline. later replaced by an Olivetti M-24.
The model therefore had to be able to

296 METALLURGY: SIMULATION

 Modelling The performance evaluation program fits
Heavy-medium separation an arctangent curve to the observed
Iscor Grootegeluk had been using a coal partition factors according to the
washer performance program for weekly following equation (4,5):

metallurgical audits for some time alrea-

dy. This program is similar to the one t = lOO(PI Arctan(P2(d P3») [1 ]
described by Wizzard(2). An extensive PI - P4
database, including Tromp curve cutpoints,
where
probable errors and Wolf curve cutpoints
was therefore available for incorporation
t = fitted Tramp partition factor (ie.
into the model.
the probability that a coal particle
Performance curve fitting of a given density, d, will report to
In order to determine the metallurgical the overflow ) ( %
performance of a coal washer, samples of d = density ( g/cm3 )

the feed, product and tailings are taken PI - P4 = parameters describing the shape
and applied to a sink-float analysis. of the specific separation curve.
Discrete Tromp partition factors are then The general shape of this curve
calculated in the conventional way< 3) • (referred to as the fitted curve) is given
(This is referred to as the observed par- in Figures 1 and 4.
tition factors). The use 01 "he:: c;.';:".' Ji ,:~ent function dif-

100-r--------------------------------------------------------~

Hx; - Shift for asymmetry

~----- Ideal Epm - no asymmetry

1------ ~-------

WWg
o
1,450 1,470 1,490 1,510
DENSITY (g/ cm3)

FIGURE 1. Plot of the overflow Tramp distribution factor against density showing the development of
the HMS simulation model
Points A = (d",,,)
B = (d so ,5o) Tramp cutpoint
C = (d'5,25)
I = (d;. t;) Arctan in flection point
W True mass based Wolf cutpoint
Wg = Two-dimensional Wolf cutpoint based on graphic integration

THE USE OF SIMULATED TROMP PARTITION CURVES 297

 fers from the method followed by Wizzavd, d = density (g/cm3)
who made use of a Weibull distribution. a = overall efficiency parameter
The arc tangent approach is preferred, b = parameter describing sharpness of
since it culminates in a two- dimensional separation
search, instead of four, thus reducing the c = horizontal shift of the arctan in-
number of iterations required from +/- flection point Under symmetrical
1000 to +/- 120, whilst still obtaining conditions, parameter e would be
correlation coefficients of 0,995 and equal to 50, in which case c would be
higher. equal to the Tromp cutpoint)
The objective of the simulation model e = parameter describing the degree of
was simple: to reverse the process of asymmetry of the partition curve
performance evaluation. This meant that
starting with calculated results avail- Four points, distributed evenly on the
able, one had to work backwards until the partition curve, are required to solve the
product and tailings streams have been four parameters a, b, c and e. Finding the
found. root of the Arctan equation, Le. the
This was approached as follows: inflection point, would solve parameter c
a) Determine which information is immediately, thus making ita logical
required to solve the four parameters choice to find. The points (d2s ,25) and
PI to P4 ; (d7 5 , 75) are distributed evenly enough
b) Solve the parameters and obtain a around the inflection point in order to
simulated Tromp partition curve; consider finding them as well. This leaves
c) Transform the continuous partition only one point not yet defined.
curve into discrete intervals, by Ideally, one would endeavour to use the
means of integration; Tromp cutpoint in order to find the in-
d) apply the discrete partition factors flection point and the Ecart Probable
to the feed stream washability data; Moyen (Epn) as a degree of sharpness to
el calculate the simulated product and solve d2s and d 75 , since these values have
t~ilings streams; found widespread application in the coal
f) compare the simulated streams to processing industry.
those observed and refine the model;
g) replace the historical performance Tramp cutpoint
data with those desired on the new At this stage it is not possible to
plant. substitute the Tromp cutpoint for the
inflection point, because of the asymmetry
Parameter solving
involved. However, if the degree of
The influence of ~ach of the parameters Pl
asymmetry is known, it can be corrected
to P4 is better understood by rearranging
for, by using the linear relationship
Equation [1] as follows:
assumed to apply in the centre region of
the partition curve.
t =a Arctan[b (d - c)] + e [2]

where Wolf cutpoint

t = overflow Tromp distribution factor, When integrating the top error area from
as before(%) the left of the curve and at the same time

298 METALLURGY: SIMULATION

 integrating the bottom area from the tI = 50,25 - 377,2 x [3]
right, one arrives at the point of equal
where
error area intersection. Under synmetrical
condi tions , this intersection will coin- tI =Abscissa of inflection point
cide with the Tramp cutpoint at t = 50. x = Tramp cutpoint - Wolf cutpoint
However, the larger the extent of asynme-
try, the larger will be the difference The relationship is considered signifi-
between these two cutpoints. Since inte- cant with a coefficient of correlation of
gration from two sides is nothing else 0,91 as exemplified in Figure 2. Further-
than a two-dimensional version of the Wolf more, if no asynmetry is present, one
cutpoint calculation, it followed that the would expect that with x =0, tI would be
the difference between the Tramp and Wolf equal to 50. The constant in Equation [3]
cutpoints was a key in the search for the is acceptably close to this theoretical
inflection point. Therefore, the relation- value, at 50,25.
ship between the difference in the Tromp
and Wolf cutpoints and the abscissa of Sharpness of separation
the inflection point had to be esta- In order to obtain the ordinat,e of the
blished. inflection point, parameter e, it is ne-
This relationship was determined by cessary to find the horizontal difference
means of linear regression on 25 observa- between this point and the Tramp cutpoint,
tions as defined as follows:

54

....
o
ti
~
§ 46
'';:::
;::l
:9
b<JJ
;a 42
0-
B
o....
E-
::: 38
o
.:;::
1i:l
:>
o 34
+

30;---~--~--~--~--~--~--~--~--~--~--~--~--~~
-0,002 0,002 0,006 0,01 0,014 0,018 0,022 0,026
Tramp cutpoint - Wolf cutpoint (g/cml)

o Included + Excluded

FIGURE 2. Plot of the difference between Tramp and Wolf cutpoints against the distribution factors of
the Arctan inflection point.
Liner regression: t; = 50,25 - 377,2 x (correlation coefficient = 0,91)

THE USE OF SIMULATED TROMP PARTITION CURVES 299

 XI = dI - dso [4] metrically around an imaginary axis lying
in an opposite direction from ds 0 than the
This can be done by determining the inflection point, dI, but with equal dis-
slope in the linear section of the parti- tance, XI. Therefore, d 7 s and dz 5 may be
tion curve and applying it to the vertical found by compensating for this shift in
difference, which is already known (tr - asymmetry:
50). Assuming that the linear relationship
holds from dz 5 to d75, the slope, m, is d7 5 = ds 0 - XI - Epm [10]
given as:
and
m = (75 - 25) I (d75 - dzs) [5]

dzs = dso - XI + Epll [11]

For coal processing, m is always negative.
Applying the definition of the Ecart Solving arctan parameters
Probable Moyen to coal beneficiation, we The parameters a., b, c and e may now be
find: solved by assigning the following initial
values:
Epm = (d.z 5 - d? 5) I 2 (positive) [6 ]
a = 100 Ire [12]
and subst,i tution thereof into Equation c = dI [13]
[5], we find m in terms of Epm: e = tI [14]

ID = -25 I hpn [7] leaving b to be solved by substitution in

[2] with (d,t) = (d7S. 75).
Solving for asymmetry
The horizontal difference between the Refinement of parameters
Tromp cutpoint and the inflection point is 'fhe above parameters must be refined,
therefore: owing to the approximation IJlade in
Equation [3] and the assumption IJlade in
5Ql [8] Equation [12]. Furthermore, an additional
m point on the Tromp curve is required since
the points dz s, ds 0, dr and d 7 5 lie in a
The inflection point, dr, may now be narrow band on the curve. Since Equation
found in terms of the Tramp cutpoint, by [12] is based on an assumption, the
rearranging [4]: additional point required must lie closely
to the end points of the distribution
dI=dso+XI [9] curve in order to incorporate the effect
of overall efficiency. The ~ximum

Solving d 75 and d25 observed distribution factor was chosen

The Ecart Probable is generally not dis- here, e.g. (1,24 ; 99), referred to as dM
tributed symmetrically around dso. Our and tH.
research has shown that the Epm tends to An i terati ve procedure is then followed
shift according to the extent of asymmetry whereby the four calculated data pairs are
present and that it is distributed sym- substituted into Equation [2], according

300 METALLURGY: SIMULATION

to following four equations: T =a (d-c) Arctan[b(d-c)] + e(d-c)

b = Tan[(t7s-e)/a] [15] a(ln[1+b2 (d-c)2) [19]

d 7s -c 2b

[16) which results into

a = (17] TJ = T(dJ-l }-T(dJ ) [20]

Arctan[b (dr.!: - c)] dJ - dJ-I

and where TJ is the discrete partition factor

for the Jth interval.
c = dso - Tan[(tso - el/a] [18]
b Size separation
Apling ( 1985) describes a method t.o mea-
applied in the order as shown. sure the performance of screens, ( 6 ) which
It was found that the set of parameters is essentially the same as that followed
converge within 5 iterations in those in the coal washer performance program.
cases where the originally observed parti- In this method the natural logaritmn of
tion factors were fitted adequately by the screen aperture is plotted on the ordinate
arctan curve. However , divergence was axis instead of the density. (See Figures
found in the cases where the observed 3 and 5 for the general form of the
partition curve had exhibited a low effi- distribution curve.)
ciency tail in the lower density region. Following the exemplary work of Apling
This tail can often be ascribed to analy- it was decided to simulate non-ideal
tical errors made in the laboratory. The screening operations too by means of the
situation was rectified by limiting the Tramp partition curve and to follow a
number of iterations to 2, thus achieving similar route to the one described above
a meta-stable convergence. in generating the partition curve. A
problem at hand was that very little plant:
Discrete partition factors history was available, since Apling's
The arctan curve describes a continuous method of computing the screening
partition factor. In practice, discrete performance curve had not been used at.
densi ty intervals are used to obtain the Grootegeluk at that stage. (It has since
partition factors. The partition factor been implemented.) The only process
thus calculated is associated wi th the parameters available were therefore (i)
midpoint of the density interval. It is undersize screen efficiency and (ii) the
therefore necessary to integrate the simu- nominal cutpoint.
lated Tromp curve across the density
interval in order to obtain the simulated Model assumptions
discrete partition factor for that parti- As less information was available than
cular interval. required to describe the partition curve
If T represents the integral of the accurately, the following assumptions had
arctan curve, then to be made (see Figure 3):

THE USE OF SIMULATED TROMP PARTITION CURVES 301

 100 .-------------------------------------------------------~_,

'§
~ 80
J-.

.2u
oj
'-
c
.9
::l 60
oD
"5en
:.a
0-
E 40
0
J-.
f-<
~
0
c A
J-.
<l)
;0-
0 20

o ~----~----~--~----_r~--~----~--~~--~--~_r~~
2,0 2,2 2,4 2,6 2,8 3,0
fn (Screen size) (mm)

FIGURE 3. Plot of the overflow Tromp distribution factor against the natural logarithm of screen size
showing the development of the screening simulation model.
Points A = (d,5.2')
B = (dsll"o) Tramp cutpoint (15,3 mm)
C = (d",,,)
= (d" t) Arctan inflection point (18,5 mm)

(1) The nominal cut size of the screen is to vary between 75 and 92% instead of the
~luivalent to the size where material 90% assumed initially.
has a 90% probability of reporting to
the overflow, i. e. (]g 0 • Parameter solving
( 2) The vertical range of the undersize Based on assumption (2) parameter a is
distribution is between 0 and 90% • defined as
(3 ) The inflection point of the Tramp
curve coincides with (]go. a = 180 /7r [21]
(4) The oversize distribution ranges
between 90 and 100% , and according to assumptions (1) and (3)
(5 ) The lower size limit, dL, is taken as parameter c is defined as
the smallest size fraction divided by
c = In (~o) [22]
2. (Zero cannot be used since the
logarithm of 0 is not defined.) and
Based on the li ttle historical data
e = 90 [23]
available all these assumptions are
largely valid. As further work was done This leaves only b yet to be determined.
it was found that the ordinate of the
inflection point does not always coincide Tramp undersize efficiency
with the nominal size but differs within a By defining f as
ratio between 1,0 and 1,25. Futhermore,
the abscissa of the inflection point tend (24)

302 METALLURGY: SIMULATION

 and substituting into [19], the error area interval halving routine was used until
of true tmdersize separation, Eu, may be EffuT EffUR is less than a preselected
calculated as follows: tolerance.

Eu = af Arctan(-bf) + ef [25] Discrete partition factors

The discrete partition factors may once
+ a In( 1 + 1>2 f2)
again be calculated as discussed previous-
2b ly. Once obtained, these are applied to
Furthermore, by defining the total the size intervals of the feed stream in
undersize area as the rectangular block order to generate the overflow (product)
100f, the Tramp undersize efficiency, and underflow (tailings) streams.

EffuT, may be defined as

Model accuracy
EffuT = 100f - Eu [26] Heavy-medium separation model
f
A typical simulated partition curve lS

By entering a required undersize shown in Figure 4, with the observed

efficiency EffuR, b can be solved by partition factors and the fitted curve
iteration until EffuR = EffuT. Since b can calculated by the performance evaluation
vary between 0,1 and 2000 a geometric program. It can be seen that both the

100lF======e=~~==~::::;:::::~==~----------------------1

Simulated

O;---~--~-.---r--'---~-'~-.---r--,---,--,~-,~~--4
1,2 1,24 1,28 1,32 1,36 1,4 1,44 1,48
l
Density (gl cm )

o Observed

FIGURE 4. Plot of overflow Tramp distribution factor against density showing the accuracy of the HMS
model
Observed Determined from sampling.
Fitted Curve generated by the performance evaluation program.
Simulated Curve generated by the HMS simulation model.

THE USE OF SIMULATED TROMP PARTITION CURVES 303

 simulated and fitted curve deviate from accurate fonns of interpolation than the
the observed points only in the high linear fonn that was used. Recent investi-
density region, by more or less equal gations at Grootegeluk showed that a
amounts. logarithmic interpolation is by far more
The accuracy of the heavy-med.ium accurate, while interpolation by using
separation model is further exemplified in Lagrange polynomials yields unpredictable
Figures 6 to 8, representing the errors results.
between simulated and actual values at
various confidence levels, the latter Size separation model
being detennined by the perfonnance Acceptable accuracies were obtained· only
evaluation program. These results, and in those cases where it MiS known from
others, are further summarized in Table experience that the assumptions made in
1. the screen model were valid. Further work
There is no doubt that other simulators showed that in other cases acceptable
might achieve better accuracies, but these accuracies could be obtained when the
were considered accurate enough for the nominal cutpoint was replaced with the
particular application, with the advantage cutpoint of inflection, and the true
that only process paramaters are utilized. parti tion factor of inflection was used
The accuracy in predicting the Wolf instead of 90%. In other words, (dI ,tI )
cutpoint may be enhanced by compensating had to be known. This is of course a
for the difference between the graphical limiting factor since usage of the co-
cutpoint and mass based cutpoint. This ordinates of the inflection point is a
value is more or less fixed for the novel concept introduced in this paper.
application at Grootegeluk at 0,003 g/cm3. More research will therefore be necessary
(See Figure 1 - points W and Wg). to establish the relationship between more
The prediction of organic efficiency may widely used parameters, such as the
also be erilianced by utilizing more nominal size, Epm values etc. and those
TABLE 1. Accuracy of HMS model mentioned above.
Furthermore, there is no reason not to
follow the same route as had been used in
Parameter Absolute errors the HMS model, apart from a lack of a
Confidence level* 50% 80% 95% database where the relevant parameters
,.
have been established. This had only ~

Concentrate ash, % 0,24 0,48 1,00 become available after the bnplementation
Misplaced material,% 0,6 1,0 2,0 of the perfonnance evaluation program for
Wolf cutpoint, g/cm3 0,002 0,004 0,006 screening operations.
Epm (Ecart) 0,001 0,002 0,004 A typical plot of the fitted and simu-
Organic efficiency,% 1,6 3,0 9,0 lated partition curves are shown in Figure
Clean coal yield, % 0,5 0,9 1,4 5, whilst Table 2 summarizes the relative
HMS errors (root mean squared) of some
To read as follows: in 50% of the typical parameters. Relative errors are
c~es the simulated concentrate ash given since a logarithmic transformation
differed less than 0,24% absolute was used to normalize the size ranges.
from the actual.

304 METALLURGY: SIMULATION

 100 ,,----------------------------------------------------~~

o ~~~~~~TrMT~~~TM~rnTrMTrn~rnTM~~~~TM~rnTM~~
1,0 1,5 2,2 3,3 5,0 7,4 11,0 16,4
Screen size (mm) (logarithmic scale)

o Observed

FIGURE 5. Plot of overflow Tromp distribution factor against screen size showing the accuracy of the
screening model
Observed Determined from sampling.
Fitted Curve generated by the performance evaluation program.
Simulated Curve generated by the screening simulation model.

TABLE 2. Accuracy of Screening Model Upper Ecca contains bright and dull coal,
suitable for the production of coking
Parameter HMS errors coal. For mining operations, 4 benches
have been developed, with bench 1 as over-
Tramp cutpoint, mm 2,1% burden. The transition zone, bench 5,
Wolf cutpoint, mm 3,5% exhibits too high a phosphorus content to
Epm (Ecart) 11,9% be rendered suitable for the production of
Tramp U/f efficiency, % 1,5% Abs coking coal. The Middle Ecca contains no
Olf yield, % 0,8% Abs bright coal and is only suitable for the
Misplaced material, % 1,3% Abs prodL'Ction of power station coal. This
section is divided into mining benches 6
----~----~----~-------------

to 14, of which bench 14 will not be mined

Application of the model as the overlying sandstone layer, bench
Geology and mining 13, is too thick.
The Waterberg coal field can be divided Ten boreholes, spaced over the planned
into the Upper and Middle Ecca series, mining operations for the next 40 years,
while the Lower Ecca is not developed. The were drilled and analysed, in order to
stratigraphic series consists of 11 coal determine the quality and expected yield
zones wi th interbedded shale layers with of the raw coal to be treated in the
each zone subdivided into samples. The envisaged beneficiation plant.

THE USE OF SIMULA;rED TROMP PARTITION CURVES 305

 Borehole evaluation peaks from the upper benches must occur,
The borehole cores were crushed to -25 DID during ini tial production stages, if the
and the -0,5 DID fraction removed. pi t is to be developed properly. This
Evaluation of the borehole analyses was implied that certain plant modules had to
done by using the model to reconstitute be designed for dual purposes in order to
the various zones from the sample analy- eliminate unusable excess capacity once
tical data, and subsequently the benches. these production peaks had been passed.
A modified version of the computer model
Bulk sample evaluation
was then used to calculate, for each
Bulk samples were collected bench by bench
bench, the
and crushed by the primary Bradford brea-
(a) in-situ characteristics;
kers in the existing plant. Analytical
(b) mass yields at a density of 2,0
results were fed into the computer and
g/crn3 , ideally and non-ideally sepa-
extensive simulations performed at the
rated;
following cutpoints:
(c) yields at other densities ranging
(a) primary screens: 35, 25 and 15 IIIID,
between 1,8 and 2,0 ;
at 88% U/f efficiency;
(d) yields at such density where a
(b) degradation screens: 30, 20 and 10
product with a 20 MJjkg heat value is
IIIID, at 92% U/f efficiency;
obtained; and
(c) feed preparation screens: 5, 3 and
(e) sensitivity analysis on the
1 DID, at 75% U/f efficiency;
yields at other heat values (ranging
(d) static bath HMS: 1,7; 1,8 1,9 and
from 12 to 22 MJjkg).
2,0 g/cm3 at 0,025 Epm and 90%
The -0, 5 DID fractions were assumed to be
beneficiated by spirals to yield a product organic efficiency;

of 20 MJ /kg and a discard of 4 MJ /kg; the (e) cyclone HMS: 1,7; 1,8; 1,9 and 2,0
results were incorporated into the above g/cm3 at 0,017 Epm and 90% organic

calculations. efficiency.

From the in-situ characteristics it was From the above simulations the optimized

possible to determine which benches could cutpoints were chosen and correlated with
be mined without any beneficiation apart the minimum and maximum expected yields
from size reduction. From the sensitivity obtained from the borehole evaluations.
and constant heat value analyses ( items d Thus the average, minimum and maximum mass
and e above) it was determined which flowrates could be established for each
benches could be beneficiated together and stream, the flowline designed in detail
which had to be beneficiated in separate and the reticulation balance completed.
plant modules.
Simulation results
Production constraints Figure 9 shows a typical simulation study
The constraints introduced by the simul ta- performed on bulk sample results from
neous adherence to product quality control bench 2, which contains 13% coking coal in
and pit development were also elicited, by situ. It shows that with pulp densities
planning production, blending options and of 1,8 g/crn 3 in the cylcone plant and 1,9
available clean coal stocks for the next g/cm 3 in the static bath plant, (the
40 years. This showed that production maximum densities achievable with con-

306 METALLURGY: SIMULATION

 32.0 32.0

----
~ 20
;>,
u
t::
<l)
;:l
0'
<l)
<.'::
<l)

.!::
~
V 10
c:.:

-0,25 0,00 0,25 0,50 0;75 1,00 1,25 1,50 1,75

Absolute error (0/0 Ash)

~ Root mean square = 0,59%

FIGURE 6. Histogram showing the relative frequency distribution of absolute error in concentrate ash
values - HMS simulation model

50,-----------------------------------------------------,

-0,005 -0,004 -0,003 -0,002 -0,001 0,0 0,001

Absolute error

(222J Root mean square = 0,0018

FIGURE 7. Histogram showing the frequency distribution of absolute error in Epm values - HMS simula-
tion model

THE USE OF SIMULATED TROMP PARTITION CURVES 307

 40.-------------------------------------------------,
36.0

30
~

-§
~

>.
u
c<1)
;:l
er
....
<1)
20
'-
<1)

.~
~
vp;::

10

-15,0 -12,0 -10,0 -7,5 -5,0 -2,5 0,0 2,5 5,0

Absolute error (070 organic efficiency)

~ Root mean square = 4,5070

FIGURE 8. Histogram showing the frequency distribution of absolute error in organic efficiency - HMS
simulation model

ventional equipnent) , the total expected be done by hand. The developnent of the
yield was 44,6%, with a product ash con- simulation model and the programming
tent of 26,1% and a heat value of 23,5 thereof took approximately 2 weeks.
MJ/kg. Since the contract specification is
35% ash and a heat value of 20 MJ/kg, it Conclusions
was clear that either higher operating (a) The simulation model described here
pulp densities were required, or that a proved invaluable for general flow-
low ash coking coal stream had to be bled sheet developnent as far as the eva-
off in order not t,o discard valuable coal. luation of borehole and bulk sample
Figure 10 shows that with the same feed analytical results were concerned.
stream, a density of 2,28 g/cm3 was re-
quired in the cyclone plant and 2,05 g/cm3 (b) The availability of such a model not
in the static bath plant, in or-deI' to only enabled the design engineers to
obtain the desired product quali ty. This meet their deadlines, but allowed the
would have resulted into a total yield of opportunity to exploit the available
55,8%, which was 11% more than in the pre- data into considerably more detail
vious situation. In this case the ash than would otherwise have been
content would have been very acceptable at possible, thus arriving at a final
84%, compared to 80% previously. flowsheet that should be much less
Over 180 separations were performed in prone to the development of bottle-
the course of 6 days, thus producing 145 necks and other design errors.
different flowsheets. This would have
taken approximately 2 months if it had to (c) The simulation accuracy achieved by

308 METALLURGY: SIMULATION

>-3
:r:tr1
c
Cl)
tr1 FIGURE 9. ROM FINAL
o"!j -150 mm PRODUCT
100,0
Cl)
Simulation study of blast 2/161 at conventional pulp densities I 56,1 44,6% Yield
12,8 43,9 26,1% Ash
~ 68,1 23,5 MJ/kg
:; 8,5

tritl +35
1'&
>-3 Cut 35mm
:;0
o LEGEND Eff 9 %
~ L--
"1:J %ROM 47,4
ASH% 49,2 -35 46,5
~
:;0 15,3
CV MJiI<g 16,2
>-3
......
>-3 +1 iill" Tromp 1,90
23 , Epm 0,012
z 7' 7 I Cut 1mm Org Eft 95%
(j
c Eff 90%
:;0
-1
-<
tr1
Cl)

Tromp 1,80 r- F 1,90

Epm 0,018 8,7
Org Eft 95% 31,9 13,1
32,4
22,8
20,2
I
I I 21,2
30,8
83,7
o 21,6

S1,80 25,6
2,9

Tailings
r--- 24,6
55,4 Product
76,2
80,4 5,7 44,6
4,2
26,1
S1,90 23,5

w
~

.1J~
w
.....
o

ROM FINAL
FIGURE 10. PRODUCT
-150 mm
Simulation study of blast 21161 at optimum product qualities 100,0
56,1 55,8% Yield
12,8 43,9 34,4% Ash
68,1 20,5 MJ/kg
8,5

+35
1"7
Cut 35mm
LEGEND Eft 9 %
L-
%ROM 47,4
ASH% 49,2 - 35 46,5
15,3
CV MJ/kg 16,2
+1 Tromp 2,05
\: 7\: 7 Epm 0,015
lUfr
Cut lmm Org Eft 95%
Eft 90%
-1

Tramp 2,28 ~ F2,05

Epm 0,020 8,7
Org Eff 95% 31,9 14,4
35,1
32,7 '
34,8
I I 21,2
29,5
84,7
o 20,5

S2,28 20,3
~ 2,6
tI:I
~ Tailings
r< .....- 14,7
44,2 Product
G 81,3
:;0 83,6 4,0 55,8
Cl 3,1 34,4
~ 20,5
[/J.
S2,05
......
~
c
~...,
o
z
T~
 the model was more than acceptable and GOTl'FRIED, B. S. The Department of
for the application in which it was Energy's coal washer performance
used. Further refinement of the model computer program. FToc. of 1st
to achieve higher accuracies and Conference on Use of Computers in the
adapt ion for other applications is Coal Industry, AIME, August 1983, ch
left to the research organizations, 24, pp. 215 - 221.
as computer modelling is not con-
sidered as the primary task of a 3. 'I'Ra1P, K.F. New Methods of Computing
production engineer based on a mine. the Washability of Coal. GlUckauf,
Vol 37, Feb 1937. pp 125-131,151-156.
Acknowledgements
4. ERASMUS, T.C. The fitting of a smooth
The authors would like to thank the man- curve to the experimentally deter-
agement of Iscor Ltd for permission to mined coordindates of a Tromp curve.
publish this paper and the personnel from Fuel Research Inst. of S.A. Report
the Ore dressing section, Research and No. 4, 1973.
Development for their valuable assistance.
5. Erasmus, T .C. Predicting the
References performance of a coal washer with the
aid of a mathematical model. Fuel
1. KING, R. P. and W<X>LLACXYIT, L. Compu- Research Inst. of S.A. Report No. 8,
ter-aided-engineering in minerals 1975.
processing. GEE course, Uni v. of the
Witwatersrand, 1986. 6. APLING, A.C. Measuring screen
performances. Mine & Quarry, April
2. WIZZARD, J.T. , KILI11EYER , R.P. 1985. pp 31 - 25.

THE USE OF SIMULATED TROMP PARTITION CURVES 311