Circulating load estimation in closed circuits

Andr Carlos Silva* Elenice Maria Schons Silva**

Ricardo Antonio de Rezende***

* Gois Federal University, Catalo, GO, 75704-020

Brazil (Tel: 55(64)3441-5327; e-mail: [email protected]).
** Gois Federal University, Catalo, GO, 75704-020
Brazil (e-mail:[email protected])
*** Gois Federal University, Catalo, GO, 75704-020
Brazil (e-mail: [email protected])
Abstract: A problem for solving mass balances in mineral processing plants is the calculation of
circulating load in closed circuits. A family of possible methods to the resolution of this calculation is the
iterative methods, consisting of a finite loop where each iteration the initial solution is refined in order to
move closer to the exact solution. The present work shows a low-complexity iterative algorithm for
circulating load calculation in mineral processing closed circuits, thus enabling the construction of mass,
metallurgist and water reliable balances. The proposed equations on the algorithm were obtained through
the analysis of many industrial systems, taking into account the process operational parameters. A
validation was performed with real industrial data, in order to ensure a greater reliability of the obtained
results. Four different types of closed circuits are presented, each one with different levels of complexity,
to clarify the proposed algorithm. With the obtained results it is possible to affirm that the proposed
iterative algorithm can be successfully applied to any kind of closed circuit in mineral processing. The
results obtained were satisfactory with respect to processing speed, convergence of the solution and the
number of iterations required for the circulating load calculation.
Keywords: Process modeling, control and optimization; Advanced process control; Mineral processing.

1. INTRODUCTION
The mass balances in mineral processing circuits are based on
the equation of mass conservation, given by:
(1)

Where F is the flow (mass or volume) entering the system; C

is the flow (mass or volume) of concentrate and T is the flow
(mass or volume) both coming out of the system.
The equation 1 assumes theoretical conditions streaming
operation, whose fundamental assumption is that the mass that
enters the system is equal to the mass that comes out of the
system (Lavoisiers law of mass conservation). Data obtained
in technological characterization tests or through sampling
campaigns in mineral processing plants must be consistent
and reliable, being common the reconciliation of experimental
data using simulation software. To this end, the mass balance
of the circuit is used to confirm the quality of the data and the
use of computational simulation aims to extend the knowledge
of the industrial processes function and as a tool for studies of
process optimization.
The circulating load can be defined as a process flow (mass or
volumetric) of a given material that returns to a unit operation
after failing in some selection criteria. This disapproval may
be due to an imperfection in the selection equipment or due
the material being out of the specification accepted by any
subsequent step in the process. So, while the material does not

suit the specifications of subsequent steps it will be forwarded

to this unit operation. Taking the most common circulating
load as example, a closed-circuit grinding (see figure 1), after
going through the mill the pulp feeds a hydrocyclone for a
particle size classification. The hydrocyclone overflow heads
for the subsequent step of the process and the underflow,
which size is above the acceptable to follow the process,
return to the mill to be comminuted again. In mineral
processing plants is common practice operating mills with
circulating load above 400%, that is, the material flow
returning to the mill is about four times the flow of new
material fed into the mill.

Fig. 1. Grinding circuit closed by a hydrocyclone.

The circulating load correct calculation in a mass balance is
essential to the understanding of the studied circuit, to build
reliable simulations of the circuit and for the operational
control. Chen et alli (2008) working with predictive control

models applied to ball mills stated that the stable control of

grinding process has great importance to achieve
improvements in equipment operation efficiency, for the
valuable minerals recovery and significant costs reductions in
production in concentration. Lestage et alli (2002) showed a
supervisory system for real-time milling circuits optimization,
where the circulating load was one of the most important
configuration parameters of the system and its value was
dependent of operational parameters such as: feed rate, pulp
density and particle size distribution of the feed and product.
White et alli (1977) studied algorithms used for data
reconciliation from field measurements with the mass balance
results. According to the authors a simple algorithm based in
the least squares method could be applied to the minimize
errors and should be enough to adjust the real to the
theoretical data. Grinding circuit data had to be reconciled due
to inability to calculate the circulating load of these circuits
with good accuracy.
Complex circuit analysis with data redundancy in mineral
processing historically requires a wide sampling campaign
and the use of matrix calculation for each mineral component
present in each considered flow. Wills (1986) demonstrated
how complex circuits can be solved by transforming the
productive flowchart in a graph, where each node represents a
unit operation.
The present work shows a low complexity algorithm with
high convergence speed for circulating load calculation in
mineral processing closed circuits. The obtained results
indicate that the proposed algorithm can be satisfactorily used
for circulating load calculation in any closed circuit,
regardless the complexity of the circuit, with low computing
demand and high convergence speed of the result.

2. METHODOLOGY
2.1. General formulation of the algorithm
Analyses of many existing mineral processing circuits were
carried out, observing the behavior of minerals flows through
the variations of operating parameters, such as grades,
partitions and metallurgical recovery. From these
observations, a correlation between all kinds of circuits was
established, regardless the complexity level of the circuit.
Yingling (1990) used a Markov chain to model the mineral
flow in mining operations. From this work, the conception of
an iterative method for the calculation of the circulating load
was thought. At each iteration the circulating load (CL) is
calculated using the following equation:

The circulating load calculation shown in equation 2 differs

from the one proposed by Tsakalakis (2000) by combining the
unit operations efficiencies in a single parameter (pi). The
exact result obtained by the author with the use of
monographic can also be obtained by using the equation 2,
eliminating the need of this chart.
The linearity hypothesis proposed by Meloy (1983) was
considered to developed the iterative algorithm proposed,
which assumes that in a separation process there is not
particle-particle type interactions that may affect the
probability of a given particle be selected for an output stream
of the process. In other words, this hypothesis indicates if the
feed flow of a given unit operation is doubled or tripled the
fraction of particles which possess a given selection feature
will continue the same in each unit operation output stream.
However, this hypothesis is not real, since an increase in the
feed flow will affect the behavior of the operation itself,
whether it be separation or classification. The adoption of the
linearity hypothesis proposed by Meloy (1983) simplifies the
proposed algorithm, since the unit operations partition varies
linearly with the operation feed. However, any other model
that allows the calculation of the unit operations feed partition
can be adopted without any change in the proposed algorithm.
Jankovic and Valery (2013) correlated the grinding and
classification efficiency with the circulating load in closed
circuit grinding operations. The same authors also showed
that the hydrocyclones classification efficiency does not vary
linearly with the circulating load in this circuit type.
The error calculation is given by the difference between the
circulating load calculated in k-th iteration and the circulating
load calculated in iteration (k-1)-th. At the first iteration the
error will be given only by the circulating load calculated on
this iteration. At the beginning of each iteration the feed of
each unit operation must be calculated, since they depend on
the circulating load and the flow partition.
Figure 2 presents the proposed iterative algorithm for the
circulating load calculation in closed circuits. It can be noted
that the iterations are repeated while a preset tolerance limit is
not reached. This is because of the solution obtained by this
method is not exact, but an approximation of the real solution.
The convergence of the method is provided by calculating the
error, which can be both positive and negative. The Meloy
(1983) linearity hypothesis applies here, since at each iteration
the feed circuit varies, keeping constant the partition of each
unit operations involved.
It stands out that the water flows in a given circuit can be
calculated using the proposed algorithm without any
additional modification. In this way, the feed flow (fi) to be
used in the algorithm can be a solid, water or pulp flow.

(2)
2.2 Application of proposed method for grinding circuits
Where fi is the flow (volumetric or mass) which feeds the i-th
unit operation that contributes directly to the circulating load
and pi is the flow partition in the same unit operation,
calculated according to the unit operation and its operational
parameters. The product pi.fi is nothing more than the output
stream of the i-th unit operation that contributes directly to the
circulating load.

For analysis and validation of the proposed iterative algorithm

four circuits with different levels of complexity were tested.
All analyzed circuits were built based on actual industrial
processes present in mining companies installed in
Catalo/GO/Brazil in order to be possible the comparison of
results obtained by the proposed algorithm with the data
provided by the companies. The first considered circuit this is

a simple closed-circuit grinding, as shown in figure 3, which

is used as a secondary ball mill by VALE Fertilizers
Company, Catalo/GO, Brazil. According to Furuya et alli
(1971) closed-circuit comminution systems typically involve
mills, classifiers and material transport equipment in various
combinations, being the combination presented in figure 1 the
most common. The coarse particles are separated by a
classifier (in this case a hydrocyclone) and feed back into the
mill.

circulating load calculation, because another output flow and

two other partitions were generated. The circulating load
calculation in this case will take into account the
hydrocyclone partition, the pipeline partition and the magnetic
separator partition. This circuit is used industrially for
separation of magnetite present in the phosphate ore.

Fig. 4. Closed circuit grinding with a low-field magnetic

separation.

2.3 Application of proposed method to flotation circuits

In a froth flotation circuit the flow partition (p) can be
estimated by the metallurgical recovery and by the minerals
grade present in flows. So, for this type of circuit the
circulating load can be calculated by the j-th mineral species
by:
(3)
Fig. 2. Iterative algorithm for circulating load calculation in
closed circuits.

(4)

Where fij is the flow (volumetric or mass) of the j-th mineral

specie in the i-th unit operation that directly contributes to the
circulating load; rij is the metallurgical recovery the j-th
mineral specie in the i-th unit operation; tinputij is the feed
grade of the j-th mineral specie in the i-th unit operation and
toutputij is the output grade in the considered flow of the j-th
mineral specie in the i-th unit operation.

Fig. 3. A simple grinding closed circuit.

The circuit shown in figure 4 is used as secondary ball mill in
AngloAmerican Phosphate, Catalan/GO, Brazil and it is a
grinding circuit where a low-field magnetic separator receives
part of the hydrocyclone underflow. The flow partition is
physically made by valves installed in the pipeline. The
addition of a magnetic separator in the circuit changes the

The circuit shown in figure 5 describes the apatite froth

flotation used in AngloAmerican Phosphate, Catalo/GO,
Brazil, and consists of four froth flotation steps: rougher,
cleaner, scavenger and recleaner. Notice that in this circuit the
circulating load is influenced by four metallurgical recoveries
and five grades for each mineral specie, since the output grade
of an unit operation is the subsequent operation feed (for
example toutput1j = timput4j and toutput2j = timput3j).

Circulating load
Overflow do hidrociclone
Iterations
Time spent
Error

Fig. 5. Circuit with four froth flotation cells with the variables
required for the circulation load calculation for the j-th
mineral specie.

2,482.33 t/h
515%
482 t/h
177
3.01 s
0,00E+00

To the grinding circuit with low-field magnetic separator

described in figure 4 the hydrocyclone partition parameter (p1)
adopted was equal to 76.29% and the low-field magnetic
separator partition parameter (p2) was equal to 15%.
Considering the new feed equals to 303.74 t/h the calculated
circulating load was 947.14 t/h, requiring 118 iterations (2.72
s). The results can be seen in the table 2.
Table 2. Results of the iterative algorithm applied to the to
the figure 4 milling circuit.

The last proposed circuit is used by VALE Fertilizers

Company, Catalo/GO, Brazil, in the apatite froth floatation
(figure 6) and it is a battery consisting of two rougher cells,
two scavenger cells and a froth flotation column operating as
cleaner. For the calculation of the circulating load in this
circuit it is necessary to know seven metallurgical recoveries
and six grades.

New feed
Hydrocyclone partition (p1)
Low-field magnetic separator partition (p2)
Circulating load
Hydrocyclone overflow
Low-field magnetic separator concentrate
Iterations
Time spent
Error

303.74 t/h
76.29 %
15.00 %
947.14 t/h
311.82%
296.58 t/h
7.16 t/h
118
2.72 s
0.00E+00

Table 3 presents data (recovery and grade) required for the

calculation of the circulating load of the froth flotation battery
shown in figure 5. The results are shown in table 4. For a new
feed equals to 267.02 t/h the calculated circulating load was
138.87 t/h, needed 33 iterations (2.17 s) for the exact
calculation of the circulating load.
Table 3. Input data for circulating load calculation of the
forth flotation circuit (figure 5).
Fig. 6. Circuit with four froth flotation cells and a forth
flotation column.

3. RESULTS AND DISCUSSION

Table 1 summarizes the results of the proposed iterative
method applied to a simple grinding (figure 3). The
hydrocyclone partition parameter adopted was 83.74%,
indicating that 83.74% of the hydrocyclone feed was
forwarded for the mill. Considering the new feed equals to
482 t/h the calculated circulating load was equal to 2,482.33
t/h, being necessary 177 iterations for the calculation. The
time spent in the calculation was 3.01 seconds in a Samsung
notebook RV411 Intel i3 processor 2.53 GHz, 3.0 GB RAM
and Windows 7.
Table 1. Results of the iterative algorithm applied to the
figure 3 milling circuit.
New feed
Hydrocyclone partition (p1)

482 t/h
83.74 %

Metallurgical recoveries
66.41%
Rougher
65.41%
Scavenger
91.42%
Cleaner
90.31%
Recleaner

Grades
Rougher feed
Scavenger feed
Scavenger concentrate
Recleaner feed
Recleaner concentrate

24.15%
15.30%
22.70%
35.45%
36.64%

Table 4. Results of the iterative algorithm applied to the

froth flotation circuit (figure 5).
New feed
Circulating load
Rougher feed
Scavenger feed
Cleaner feed
Recleaner feed
Iterations
Time spent
Error

267.02 t/h
138.87 t/h
52%
405.89 t/h
215.21 t/h
190.68 t/h
167.88 t/h
33
2.17 s
0.00E+00

Table 5 presents the data (recovery and grade) required for the
circulating load calculation of the froth flotation battery
shown in figure 6. The results are shown in table 6, as well as
a comparison with the results for the mass balance calculated
using the Caspeo BILCOTM software, version 3.0, which
uses the Lagrange methodology to calculate mass balances
and it was admitted an error of 5% for the mass balance
closure. For a new feed equals to 287.17 t/h the calculated
circulating load was equal to 97.838 t/h by the iterative
method and 97.833 t/h by BILCO.
Table 5. Input data for circulating load calculation of the
forth flotation circuit (figure 6).
Metallurgical
recoveries
Rougher 1 75.79%
Rougher 2 72.05%
Scavenger 1 27.91%
Scavenger 2 29.77%
77.07%
Cleaner

Grades
Roughers feed
Rougher 1 concentrate
Rougher 2 concentrate
Scavenger 1 feed
Scavenger 2 feed

8.257%
18.000%
18.260%
24.213%
27.946%

Table 6. Results of the iterative algorithm applied to the

froth flotation circuit (figure 6).
Iterative
algorithm
New feed

within a preset tolerance limit (here adopted as 5%). So, if the

solution found is in this tolerance limit, this solution is
adopted as a solution of the Lagrange method which is
basically an error minimization method, similar to the method
of the least squares.

Note that the number of iterations needed to calculate the

circulating load changed in each of the tested circuits. This is
due to the fact that the method iterates the feed value while
the error converges to zero. Therefore, the smaller the
circulating load, closer to the solution the adopted initially
value will be and less iteration should be made. In a numerical
sequence of this type, the first approximation of the solution is
called seed. The proposed method is analogous to the bisector
method for algebraic equations root determination. In this
method the number of iterations depends on the amplitude
between the two extreme points of the considered range. In
the case of the proposed algorithm the number of iterations
depends on the amplitude between the seed and the solution.

4. CONCLUSION

BILCO Difference

The proposed iterative algorithm was valid and acceptable in

relation to the iteration number and convergence speed to
different circuit types and complexities. It is noticed that the
iterative method converges faster to the exact solution in froth
flotation circuits than in other circuits. This fact can be proven
by examining the iteration number needed spent on each
circuit, being the cause of such behavior the fact that the
initial solutions (the algorithm seed) in froth flotation circuits
were closer to the exact solutions. For all circuits studied the
results were compared with the results obtained by Caspeo
BILCOTM software, version 3.0, and similar results between
the software and the proposed algorithm were found.

287.175720 t/h

287.175723

-3.00E-06

97.837758 t/h

97.833147

4.611E-03

34.068%

34.06%

0.008

24

0.45 s

Rougher 1 feed
Rougher 1
concentrate
Rougher 1 tailings

192.506739 t/h

192.504435 t/h

2.304E-03

62.722467 t/h

62.719372 t/h

3.095E-03

129.792810 t/h

129.785063 t/h

7.747E-03

Rougher 2 feed
Rougher 2
concentrate
Rougher 2 tailings
Scavenger 1
concentrate
Scavenger 1
tailings
Scavenger 2
concentrate
Scavenger 2
tailings
Cleaner feed
Cleaner
concentrate
Cleaner tailings

192.506739 t/h

192.504435 t/h

2.304E-03

66.925493 t/h

66.921965 t/h

3.528E-03

125.569093 t/h

125.582470 t/h

-1.33E-02

14.679437 t/h

14.679263 t/h

1.74E-04

5. ACKNOWLEDGEMENTS

115.112238 t/h

115.105800 t/h

6.438E-03

The authors thank financial support from the Brazilian

agencies CNPq, CAPES, FAPEG and FUNAPE.

13.262072 t/h

13.260988 t/h

1.084E-03

112.326512 t/h

112.321483 t/h

5.029E-03

129.647861 t/h

129.641337

6.524E-03

59.751602 t/h

59.748441

3.161E-03

69.896259 t/h

69.892896

3.363E-03

Circulating load
Iterations
Time spent

The present work shows that the calculation of the circulating

load can be understood as a mathematical function where it is
wanted to minimize the error and, therefore, the any errors
minimization techniques or mathematical optimization can be
successfully used to calculate the circulating load.

6. NOMENCLATURE
C is the flow (mass or volume) of concentrate;

The difference between the circulating load calculation by the

proposed iterative algorithm and the Lagrangean method used
by BILCO is due to the fact that BILCO searches a solution

F is the flow (mass or volume) entering the system;

fi is the flow (volumetric or mass) which feeds the i-th unit
operation unit that contributes directly to the circulating load;
fij is the flow (volumetric or mass) of the j-th mineral specie in
the i-th unit operation that directly contributes to the
circulating load;
pi is the flow partition in the same unit operation, calculated
according to the unit operation and its operational parameters.

rij is the metallurgical recovery the j-th mineral specie in the ith unit operation;
T is the flow (mass or volume) both coming out of the system;
tinputij is the feed grade of the j-th mineral specie in the i-th unit
operation;
toutputij is the output grade in the considered flow of the j-th
mineral specie in the i-th unit operation.

REFERENCES
CHEN, X., LI, Q., FEI, S. Constrained model predictive
control in ball mill grinding process. Powder Technology,
v. 186, n. 1, p. 31-39, 2008.
FURUYA, M., NAKAJIMA, Y., TANAKA, T. Theoretical
Analysis of Closed-Circuit Grinding System Based on
Comminution Kinetics. Industrial & Engineering
Chemistry Process Design and Development, v. 10, n. 4,
p. 449-456, 1971.
JANKOVIC, A., VALERY, W. Closed circuit ball mill
Basics revisited. Minerals Engineering, v. 43-44, pa. 148153, 2013.
LESTAGE, R., POMERLEAU, A., HODOUIN, D.
Constrained real-time optimization of a grinding circuit
using steady-state linear programming supervisory
control. Powder Technology, v. 124, n. 3, p. 254-263,
2002.
MELOY, T. P. (1983) Analysis and optimization of mineral
processing and coal-cleaning circuits circuit analysis.
International Journal of Mineral Processing, v. 10, n. 1,
pp. 61-80.
TSAKALAKIS, K. (2000) Use of a simplified method to
calculate closed crushing circuits. Minerals Engineering,
v. 13, n. 12, pp. 1289-1299.
WHITE, J. W., WINSLOW, R. L. ROSSITER, G. J. (1977) A
useful technique for metallurgical mass balances
applications in grinding. International Journal of Mineral
Processing, v. 4, n. 1, pp. 39-49.
WILLS, B. A. (1986) Complex circuit mass balancing A
simple,
practical,
sensitivity
analysis
method.
International Journal of Mineral Processing, v. 16, n. 3-4,
pp. 245-262.
YINGLING, J. C. (1990) Circuit analysis: optimizing mineral
processing flowsheet layouts and steady state control
specifications. International Journal of Mineral
Processing, v. 29, n. 3-4, pp. 149-174.