Metallurgical Balances

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 26

Chapter 39

J. M. Richardson

Supervisor , Wineer ing System Professor,


Fluor mgineers, Inc. Mineral Process Ehgineer ing
Northern California Division University of British Co-ia
Redwod City, California Vancower, British Columbia

process. A s the design and installa-


tion of concentrating and dewatering
Metallurgical balances are circuits does not usually require heat
fundamental t o the successful design balance calculations, this chapter w i l l
and operation of concentration and be limited t o discussion of material
dewatering circuits. With caquter- balances.
ized balance techniques gaining
praninence in recent years, it is lb "balance" a system inplies that
essential that engineers understand the the system is a t steady-state. A t
basic principles that underlie the new steady-state, the mss i n p ~ trate
methods, a s well as the varicus equals the mass cutput rate, whether it
a~plicationsfor design of new plants is the entire process that is being
and analysis of existing operations. considered, or individual unit
This chapter discusses tw inportant aperations. For a system a t
types of metallurgical balance: steady-state, no m t e r i a l accurmlates
internally; each u n i t operation is
(1) steady-tate design assumed t o be functioning w i t h an
(2) operating data adjustment unchanging vollw of material already
present.
Application of carputer ized steady-
state design balancing is demonstrated A metallurgical balance for a typi-
through the exanple of a typical caFper cal concentrator may apply t o any or
concentrator f lowsheet. -rating data a l l of the follawing measurable items:
adjustment t o achieve a consistent
balance is demonstrated for a typical - total material (solids plus fluids)
concentrator aperation. - individual m t e r i a l phases
(solids, fluids)
- individual, "assayable" conponents
(cogper, molybdenum, etc .
Definition of Metallurgical Balance
- individual size fractions

The term "metallurgical balance" is Categories and Applications of


ccrmPnly applied to mean the overall Metallurgical Balances
accounting of material and energy
flawing into and out of a metallurgical With the advent of carputerized
CONCENTRATIONAND DEWATERING CIRCUITS

balance calculations, two distinct cates which system appear to be


categories of metallurgical balance process- or plant-specific and which
have arisen: are general-plrpose. Some of the work
referenced includes discussion of
- process design operating data adjustment as related to
- operating data adjustmnt developnent of process-specif ic models .
Each category contributes signifi- DaQ Miustment Data adjustment
cantly, but in a different manner, to procedures for metallurgical balances
the engineering goal of optimization of involve the non-model-based statistical
process plant design and operation. adjustment of inconplete or inaccurate
This section takes a closer look at operating plant data to obtain a
both categories and identifies specific consisitent material balance. Data
applications found in each. adjustmnt is one step involved in the
developnent of models (Mular and Herbst
Process Desian In this balance I197811 and should not be confused with
category, engineers endeavor to sinulation programs enployed for pro-
accurately calculate material balances cess design. me latter may utilize
for individual unit operations, cir- adjusted data in a suitable rmnner, so
cuits within a plant, or an entire that data reconciliation programs and
process flowsheet. Typically, the process design programs are closely
calculations are based on the use of connected.
predictive mathematical models, usually
of individual unit operations. Such The need for data adjustment pack-
models normally function to transform ages a w r s to be unique to the
specified or k n m inplt stream data minerals industry, where the physical
into outplt stream values. In and chemical nature of the materials
practice, the use of such models is treated and the processes used have
further dependent on parameters ihich corrbined to make accurate measurement
help define operating conditions and of process conditions extremely
performce of the particular unit difficult, if not inpossible. The
operation. Model sophistication varies inherent inability to measure or infer
widely: flow of material results in either
under-defined (inconplete) balances or
- highly theoretical, mechanistic overdefined (inaccurate) balances
around individual equipnent and even
models which attenpt to explain the
unit operation from first principles entire plants. By providing a
reasonable estimate of what the actual,
- semi-enpirical models having consistent balance around a unit
operation, circuit, or plant should be,
grounding in both theory and
operating or experimental data data adjustment program can be used
in the following applications:
- fully-enpirical models based on - metallurgical accounting
engineering assunptions and opera-
ting or experimental data - process instrument / control system
calibration
Steady-state design balance results - data verification for enpirical
are used for the following applications: model developnent
- equipnent sizing
- evaluation of alternative processes Table I1 gives a very brief listing
- modification to existing plants of sane of the most recent and m r e
frequently referenced efforts in the
Table I represents a -ling of the area of steady-state data adjustment.
recent literature on conplterized
steady-state design balance systems for
minerals industry processes. The table,
which is bv no means conplete, indi-
METALLURGICAL BALANCES

A m E I G CF DESIGN BAGANCE SYETENS

DEVFXOPIEX; UTILITY
QBatwam
general-plrpose,
minerals industry
ASPEN MIT

ASPEN-PLUS general-plrpose,
nulti-industry
Fluor mgineers general-purpose,
minerals industry
U. of Idaho general-plrpose,
nulti-industry
(Gotffried, et. al.) U. of Pittsburgh/ coal-preparation
USEP1/US#)E plants only
r'WmAL*-II SAmWv. of -tern general-plrpose ,
Chtar io minerals industry
Mineral Processing Jblius Kruttschnitt general-plrpose ,
Sirmlator Mineral Research Centre minerals industry
U. of Witwatersrand general-plrpose,
minerals ind.~stry
Pimock, Allen c general-plrpose ,
Aolt lkgineers minerals industry
(Rular, et. al.) U. of British plant-specif i c
Colmbi&d Brenda Mines
u. of Utah general-plrpose,
minerals industry
Lava1 University general-plrpose,
minerals industry
U. of California, general-plrpose,
Berkeley minerals industry
SWC CAEMET general-pllrpcee,
minerals industry
CONCENTRATION AND DEWATERING CIRCUITS

TABLE 11
A SAMPJ,IK OF RECENT DATA ADJUSJMWl SYSl'FNS

SYSTEM
0
Lava1 University
U. of Utah

U. of Minnesota

U. of Minnesota

Michigan Tech. University


(Mular, e t . a l . ) U. of British Colurtbia/
Brenda Mines
U. of Arizona

U s e of Complters for Metallurgical EARCY PROCESS DESIGN SYSTPIS: De-


Balances sign balance programs have been carmDn
i n t h e Chemical Process Industry ( B I )
&j& m i l e the need for since the mid-1960's. Program
calculating metallurgical balances has designed specifically for minerals
always been apparent, several factors industry flowsheets did not begin
have ccmbined historically t o retard appearing u n t i l the l a t e 1970's
t h e developnent and widespread use of (Gottfried and Abara t19781, Bartlett
conplterized system for automating t19791, Gupta and Messa t19801,
calculations. A key factor was t h e Richardson, et. a l . t1980, 19811,
lack of econanic incentive t o develop
s@isticated techniples for process
.
Izurieta and Edwards l19801) Most such
system were written for use on large
cptimization. The availability of high "mainf ramen c-ters. Recently,
grade ores t h a t were easy and h m v e r , new system a r e appearing more
inexpensive t o process limited the need frequently (King t19851, Laguitton
t o understand and model the processes r19851, W e e l19851 , Richardsan and
used. Conplter equipnent was expensive Sredanovic t19851 and most a r e de-
t o own and operate. Personnel with signed t o run on t h e present generation
adequate carputer training were scarce. of popllar "micro" carplters.
A noticeable change occurred in the DWA AIUuSMEW PRXWW: m i l e de-
1960's and 1970's. Ore grades had been sign balances have been calculated
steadily declining bile processes were since long before the advent of canplt-
becoming ever more carplex. Universi- ers, the c a x e p t of data adjustment for
t i e s began t o broaden t h e exposure of operating flowsheets did not really
engineering students t o conputer progress u n t i l c c q u t e r s were available
science. In t h e l a t e 197OWs,t h e per- t o permit t h e lengthy, canplex statis-
m i t cost of conputer hardware plunged t i c a l calculations involved. Reviews
dramatically, starting a rapid proli- of t h e subject, detailing early and
feration of small, inexpensive, but recent efforts, a r e provided by Mular
b u t powerful conplters. (1979) and by Reid, et a l . (1982).
major change.
DESIGN BRLRNCES
While mthermtical techniques abound
% i s section explains the basic for solving almast any given s e t of
principles of design balance programs. s h l t a n e o u s equations, a satisfactory,
easily programable method for auto-
Basic Principles of Conplterized Design mating the specification and set-up of
Balances equations has yet t o emerge from the
literature. The fact that process flow-
A design balance is the starting sheets can vary widely has nearly
point from which logical selection of eliminated the use of sinultaneous
alternatives can be made. In some equation-solving in general-plrpose
cases, marginal f lawsheet ar rangernents design balance program.
can be eliminated from consideration
by carparing balance results for such Modular Method. Most of
item a s circulating loads, quality of today's design balance program use
product, waste-product build-up, etc. some variation of t h i s technique, which
me design balance is also the basis grew from the traditional m u a l "trial-
for further calculations such a s e q u i p and-errorn method. Sam key concepts:
ment sizing and determination of
reagent and power consumption. This PFWESS PICrORlAL ANI B W F m
leads t o develapnent of capital and A pictorial diagram is
D1PIC;RAMS:
operating cost estimates which are then essential t o understanding the inter-
used t o further narrow the field of relationship of material flows and unit
viable process alternatives. operat ions. A typical capper/mlykde
num concentrator is shown a s Figure 1.
IterativePsrJroach~Simultaneous The unit operations shown in t h i s f l o w
There are two basic sheet mst be modelled (described
techniques which have been applied t o mathematically) in order t o solve the
the task: the iterative (trial-and- balance. The mathematical models are
error) approach, and the method of implemented a s program subroutines
simltaneous equations. m i l e the called modules. These modules usually
l a t t e r has not disa~pearedaltogether replace the unit operations cne for
from balance system, developers aFpear one, but there may also be reason for
t o favor the use of iterative tech- having several unit operations rep-
niques over s h l t a n e o u s equations. resented by one module, or vice versa.
%is chapter focuses on the iterative A drawing which shows the modules
approach; but f i r s t , a brief review of chosen is defined here a s a "block flow
why sinultaneous equations a r e rejected: diagram". Figure 2 is the block flow
diagram representing the flowsheet of
With sinultaneous equations, the Figure 1. Each module and stream is
engineer mst s e t up and solve a system labelled with a unique n&r for easy
of equations that exactly describes the reference and so that no stream or
exchange of m t e r i a l in a f lowsheet. module nunber occurs twice.
Conplex flowsheets are thus attacked in
their entirety, rather than in syste- UNIT OPERATION MC6XlLES: Before
mat ic, manageable chunks. Either continuing with the exarrple flowsheet,
the user or the program designer rmst it is useful t o f i r s t look a t a typical
ensure that a l l possible equations are module in claser detail. The module
specified. This makes it difficult t o chosen is the flotation module ( W T ) .
design a general-purpose program which
can accept the rmlti-equation defini- . .
DescrlPtlon: The F W module
tion of many different flowsheet perform a balance about a flotation
arrangements. Dedicated programs c e l l , a bank of cells, or an overall
written t o balance a specific flow- circuit such a s rougher, scavenger,
s h e e t a r e easier t o achieve, but rmst cleaner, etc. The module accepts a
be rewritten for each new flowsheet or single inplt stream and calculates two
CONCENTRATION AND DEWATERING CIRCUITS

n
STOCK PILE

APRONFEEOER

BELT COIJVEIOR

4 7
WATER

111. CLEARER

I
WATER
REGRINDMILL

FIGURE 1. Copper / Molybdenum Concentrator Pictorial Diagram


METALLURGICAL BALANCES

MIXER

MIXER

- -

FIGURE 2. Copper / Molybdenum Concentrator Block Flow Diagram


614 CONCENTRATION AND DEWATERING CIRCUITS

outplt streams (concentrate and XsG = solid conpnent mass


tailings). concentration, stream n,
ith cofiponent
The overall solids flowrate in the
concentrate is based on the concentrate Wf, = total fluid mass flaw, stream n
grade and recwery of a designated
"primryn solid ccargonent. Up to two Parameter Definitions: The nuher of
additional "secondary " ccnponents may equipnent parameters required for the
be designated in the concentrate. The operation of EUXT will vary depending
"seccndaryn conponent recoveries do not on the n&r of solid capnents
affect the overall solids balance but defined as "recoverable".
do determine the balance for each such
conponent. The overall fluid flowrate Percent solids in the concentrate
in the concentrate is based on a user- stream:
specified concentrate percent solids.
All tailings stream solid flowrates are
calculated by difference. Any solid
canpcnents not designated for recovery,
and any size fractions present, are
assumed to be split in the same ratio
as total solids. Grade of p r h r y ccipmnt in
concentrate:
Diaaram: A flaw diagram for
the FLQRT module is s h m in Figure 3.

Recwery of primary conponent in


concentrate:

Recovery of secondary cmponent in


cancentrate:

~Calculations:
Concentrate Solids Tailings Solids
FIGURE 3. F m T Mcdule Flow Diagram ma= x BL m3=W8,-
p3
The variables show with each stream
are defined as:
m, = total solids flwr stream n
Cshi = solid conponent mass flow,
stream n, ith conponent
A1 BALANCES 615

a single recycle group my contain a l l


Concentrate Fluids Tailings Fluids the modules of a f lawsheet. f i e
exanple of Figure 2 has three recycle
groups: group 1, which contains the
semi-autogenous grinding circuit;
group 2, which contains the ball m i l l
circuit; and group 4, which is mde up
of the regrind circuit, cleaner
lumHEEz TOPOIM;'Y: If material flotation, and scavenger flotation.
streams do not recycle or loop back in
a f lorJsheet, mas balance calculations
are single. Unfortunately, moat
processes contain recycle loops that
make mass balance calculations dif-
f icult and t Wmwning. me sequen-
t i a l modular method sirrplifies calcu-
lations by subdividing the flawsheet
into groups of unit operation modules
.
that can be solved indivudally Groups
are categorized according to certain
characteristics, and each is solved by
a calculation method amropriate for
that group. The type and nmber of
module groups occurring in a f lawsheet
is also referred t o as the flowsheet
topology. The two categories of module
groups are k n m as 'serial" and
"recycle".
SerialnngRSYGkGroMs: A
serial group i s the largest group of
modules that can be calculated in
sequence, working from the beginning of
the group. All input stream flawrates
to each module are known before the FIGURE 4. Recycle Group
output flaws are calculated. Serial
groups can be solved in a single pass, ?he calculation of
without repetition of the calculations. serial groups i s straightforward.
In Figure 2, there are only two serial Hawever, t o understand the calculation
groups in the exanple f lowsheet. me of recycle groups, it is f i r s t
such group is the rougher flotation necessary t o understand the concept of
(module 110) and the Launder water "tearing." lb illustrate this concept
addition mixer (module 111). 'Ihe other refer t o one of the recycle groups of
serial group contains the second the exanple flowsheet. The singlest
cleaner launder water addition mixer case i s group 1, the semi-autogencus
(module 119) and the final tailings grinding circuit, which is ahown
sunp (module 122). isolated in Figure 4. The definition
of a recycle group is satisfied; that
A recycle group i s the m l l e s t is, there is no module for which a l l
group of a l e s that cannot be solved input stream are k m before
in a single calculation pess. I n a calculating. A l q i c a l calculation
recycle group, there is no apparent path through the circuit is not readily
starting point for orderly calculation, apparent. Tb resalve this, one or mre
since no module exists for which a l l streams in the circuit met be 'torn,"
inplt stream are known prior t o any i.e., arbitrarily and artificially
calculations. A recycle group mat
contain a t least two modules; however,
divided into two new streams as shown equal before the circuit is "converged"
in Figure 5. After tearing, stream 7 (balanced). If the circuit has not
becomes two streams: stream 7 and converged, a new guess w i l l be
stream 801. Stream 7 is the oversize required, and the calulation sequence
from the screen and stream 801 is an nust be repeated. The entire procedure
i n w t t o the feed conveyor. w i l l repeat until convergence is
achieved. The principles of tearing and
ordering as shown in t h i s exanple are
the same for any recycle group, and the
objective is always the same, t o
provide a straightforward, serial path
through the group, with a l l module
inputs known or assumed known when each
module is calculated.

FI(;URE 5. "Tearing" a Recycle Group

Orderina=m-Seauence:
The objective of tearing is t o a l l w a
recycle group t o be re-ordered in a
sequential manner. This is demn-
strated for the exanple circuit in
Figure 6. By assuming that the flows
in stream 801 are known, there is n w a FI- 6. "Serial" Structure After
serial calculation path through the Tearing
group. Using an assumed (guessed)
value for stream 801, the conveyor Conversence Modules:. Design
(MIXER) module can calculate f l w balance program use convergence
values for stream 2. The water addition modules t o help solve recycle groups.
(MIXER) module subiequently r d v e s mis type of module conpares torn
known-valued input stream 2 and 3, and stream guessed and calculated values
can in turn calculate stream 4. T h i s and generates new guesses (if the guess
allows the m i l l (C(mIN) module t o next is not equal t o the calculated value).
calculate stream 5 values, and the Tb demonstrate, Figure 7 shows the
screen ~~) module can, in its exanple circuit reassenbled with a new
turn, calculate stream 6 and 7. module, the convergence module, insert-
However, unless the i n i t i a l guess for ed be- stream 7 (the calculated
stream 801 was accurate, stream 7 w i l l stream) and stream 801 (the guessed
not be equivalent t o strean 801. Since stream). Since the cmvergence module
these two stream are really the same is like any unit aperation module,
stream, they nust be equal or nearly given an input stream, it w i l l calcu-
METALLURGICAL BALANCES

late an outplt stream by projecting new where x is the guessed value and f(x)
values for the outplt stream based on is the calculated value. Wen f (x) = x
the recent history of calculated for any particular stream variable, the
values. circuit is converged for that variable.
The Wegstein method requires suc-
cessive substitution t o give it the
f i r s t two points from which to nrake a
projection. After that it can generate
i t s awn points and w i l l solve in many
fewer iterations than successive
substitution.
For the exanple circuit, the
calculational order begins with the
convergence module and ends w i t h the
screen. Any initial f lawrate values,
including zero, for stream 801 are
sufficient t o start the iterative
calculation process. The order is
repeated until the convergence module
"signalsn the program that recycle group
1 has converged. Then the next group in
the flowsheetcan be calculated. If the
recycle group is corrplex and contains
mre than one torn stream, and therefore
mre than one converqence module, the
calculation sequence-will be repeated
until a l l converqence modules "signal"
FIGURE 7. Recycle Group with Converge that convergence-has been achiwd.
Module
The a m t of error remaining a t
The numerical method used by many convergence is knawn as convergence
design balance program t o canverge tolerance, defined as the absolute value
recycle loops is actually a conbination of the relative ratio of guessed and
of two techniques known respectively as calculated values:
succesive substitution and the
unbounded Wegstein convergence accel-
eration method. In successive substi- convergence tolerance =
tution, the calculated value of the
torn stream obtained from the previous
iteration becomes the guessed value
which starts the current iteration.
This method usually works but often
requires many iterations t o reach a For each stream variable being
solution. The kgstein method solves balanced, the convergence module
considerably faster. It uses a secant carpltes the current value of this
projection from the previous two expression and conpares it against a
calculated values t o determine what the value supplied initially by the user.
current guessed value should be. The If it is less than or equal t o the
objective of the projection is t o solve user-sumlied nunber, convergence is
for the following : achiwed for that variable. All stream
variables rmst be converged before the
module w i l l stop projecting new
guesses.
618 CONCENTRATION AND DEWATERING ClRCLllTS
. .
--TowsraPhv: convergence until every variable is
Some balance program require extensive converged. The nm&r of individual
m u a l network analysis of a flowsheet conponents from each phase that can be
before the actual balance can be per- balanced depends on h i c h program is
formed. This means that the engineer being used, as do the maxirmm nuher of
m s t f i r s t organize the modules into unit operations and streams allowed for
serial and recycle groups and determine each flowsheet. Concentrations are
*ich stream in each recycle group dependent values calculated from
should be torn. He then has t o place individual conponent mass flows and
convergence modules into the flowsheet total phase flow rates on a stream by
and determine a calculational order. stream basis.
With certain balance programs, these
t a s k s are performed automatically by The item balanced in the
the program, with the user only cgper/n-olybdenum flowsheet example
required t o specify overall flawsheet include:
topography (interconnection between
streams and modules). Special program - t o t a l solids
network analysis subroutines perform - t o t a l fluids
grouping, tearing, insertion of - solid-phase c o m r
convergence modules, and calculational - solid-phase molybdenum
ordering. The algorithm used in these
types of subroutines are discussed in -
MEEXIE SPECIFICA!CIOW THE USE CF
greater detail by Coles (1982) and CDNl!ROUEEt MODULES: When a l l process
Wsterberg, e t . al. (1979). feed stream flow rates and unit
operation module parameters are known
A c m n convention is used t o in advance, a balance is easily
specify process stredmodule inter- achieved. Hawever, in m y cases
connections: For each module, a l l several of these variables are not
module input streams are given positive known i n advance. For exanple, two
nunber labels, and a l l outplt streams parameters required by the FU3.W module
are given negative nunber labels. A l l are the concentrate primary conponent
module input stream are always listed grade and the per cent recovery of the
before output streams. lb specify the primary conponent. In complex copper
topography of the entire exanple flow- circuits, m y flotation unit
sheet of Figure 2, a table, such a s the operations are enployed, but for some
data sheet shown in Figure 8, m s t be of these, such a s cleaner or scavenger
constructed. circuits, it is often difficult t o
determine beforehand what the actual
STREAM M A R48JIREMENl'S: When grade and recovery of copper might
calculating the balance , the flow actually be. Thus, the system is
variables balanced usually include underspecified. However, scavenger
t o t a l solids, t o t a l fluid, and circuits often recycle copper toward
conponents of the solid and fluid the front of a process, and it is
phases. "Conponent" means the likely that a circulating load of
individual chemical elements or copper, or the grade of a process
conpounds that make up the solid or tailing of copper, w i l l be known.
fluid phases; anything that might be Mten, an overall process recovery of
assayed for in real life. The copper w i l l be k n m . Any of these
definition can also be extended t o values can be used t o replace an
solid particle size fractions. unknown d u l e parameter or feed stream
value, and the system w i l l be
Same programs such as FLEXMF2. conpletely specified.
consider each variable as independent
with respect t o the overall flowsheet. mConceotnfConttollers: To
Therefore, convergence modules work enable users t o specify internal or
separately and independently on each product stream flowrates and concen-
variable, but do not "signal" trations, design balance programs use a
METALLURGICAL BALANCES

FIGURE 8. Data Sheet Specifying Flowsheet Layout

special-plrpose module called the There are usually two types of


"controller". Controller d u l e s can controller modules: feedforward and
be inserted into a flOW6heet to modify feedback. Each type has several
feed stream or module parameters to parameters associated with it that
m e t a constraint. In effect, a allw the user to specify how it is
controller is instructed by the user: used.
" m r ing the calculational sequence,
specified internal or product Stream Feedforward Controller Module:
values nust be made to occur." Wile This type of moclule monitors a stream
the use of controller variable ahead of the feed etream
d u l e s aFpears analogous to the variable or unit operation module pa-
utilization of process cantrollers in rameter that is to be adjusted. The
an actual operating plant, the end plrpose of the adjustment is to meet a
plrpose here is only to solve a specified stream value, or setpoint,
steady-state mass balance. Transient downstream fram the adjustment. This
conditions are not considered, even is illustrated by examining the rougher
though the algorithm enployed by launder water addition (MMER) module
controller modules cause them to behave ran the example flowsheet. The module
like real process controllers when is shown in closeup in Figure 9. A
converging towards a balance solution. water stream is being added to dilute
It is the end result achiwed by these the solids concentration in the rougher
modules that is most inportant. concentrate from 40 weight percent
solids down to 25 percent solids.
620 CONCENTRATION AND DEWATERING CIRCUITS

Usually, an engineer w i l l know these requires only one algorithm t o solve


solids concentrations rather than the any control situation, but the
exact t o n ~ g eof water t o be added in algorithm is considerably more conplex
stream 18. Therefore, a feedforward than any of the feedforward control
controller can be applied as shown i n algorithm. It is similar to the
Figure 9. In the calculational unbounded Wegstein method used by the
sequence, the controller is called convergence nodule and uses the same
before the MIXER nudule. The convergence tolerance to determine i f
controller monitors the value of the the setpoint has been achieved.
solids concentration of stream 16. Several adjustments by a feedback
Using the setpoint of 25 percent solids controller are usually necessary t o
supplied by the user, the controller meet a constraint; therefore a fe-ck
enplays a special dilution equation, or controller w i l l always be a part of a
algorithm, t o calculate the amxlnt of recycle group, even if there are no
water in stream 18 that w i l l cause the material streams recycled. The
setpoint value t o be met for stream 19. "recycle" of control information alone
Different f eedforward control w i l l cause a recycle group t o be
algorithms are required to meet other formed.
situations, such as dilution of fluid
st ream containing dissolved
conpments, proportional addition of
reagents, and so forth.

M r a mlr lPl
DILUlB WCmRlll
I I1 44 Yl :! SOLIDS I1 LI m 1 lOLIDI

FIGURE 10. m l i c a t i o n of Feedback


Controller Module
An exanple of a f-ck controller
FICXlRE 9. Fgplication of Feedforward i s the ball mill-qclone-sunp portion
Controller Module of the e x q l e flowsheet. The module
representing the cyclone is a
FeeQackControllerModules:A circulating load classifier (CLCLAS)
This module has two parameters: (1)
.
feedback controller can monitor a
stream -re in the f l m h e e t and circulating load (defined as the
adjust a module parameter or feed percentage ratio of underflow solids to
stream value in order t o cause the overf low solids) and (2) fluid split
monitored value t o equal a s e t p i n t (defined as a percentage of the total
supplied by the user. Feedback control fluid flow in the cyclone feed) which
METALLURGICAL BALANCES

reports t o the overflow stream. Using well as controller strategy chosen.


these parameters, it is likely that the Table I11 shows the c i t e r i a for the
circulating load value w i l l be known, example f lowsheet.
but that the fluid s p l i t w i l l not.
m v e r the solids concentration of the Controller s e q y , Before developing
underf low w i l l also be known. Figure 10 controller strategy for a flowsheet, it
shows how a feedback controller can be is of p r h r y irrportance t o f i r s t
applied t o resolve the situation. 'Lhe def ine the design criteria, as
controller monitors the solids discussed above. Criteria that are not
concentration of stream 15, the obtainable with defined equipnent
underflow, which has been calculated by modules are met through the use of
C X L S . 'Ihe controller then canpares controllers, as follows:
the monitored value with the
user-sugplied setpoint. If the two 1. Solids concentration i n the
values are not equivalent (within the feed t o the SAG m i l l , b a l l m i l l ,
limits of the user4efined convergence and cyclone are specified as 70,
tolerance), the controller calculates a 60 and 50 % solids, respectively.
new value for the fluid s p l i t 'Ihese particular solids
parameter. 'Ihis new parameter w i l l be concentrations can be achieved by
used on the next iteration through the adding feedforward controllers
group, resulting in a new underflow h i c h use a special algorithm t o
solids concentration. Since t h i s calculate how mch dilution water
circuit is also a conventional recycle should be added t o each feed
group, a convergence module is stream. 'Ihe controllers are
necessary. 'Ihe calculations continue indicated in the block flow
u n t i l both the controller and the diagram of Figure 2.
convergence module signal convergence
in the same iteration. 2. Operation of the screen and
cyclones i n the f lawsheet was
Performing Design Balances modelled by use of a circulating
load classifier (QICLAS) module.
'Ihe actual solution of a conplete This module has two parameters: 1)
design balance is derronstrated using the circulating load, which was
the c w r /mlybdenum f lowsheet example defined in the criteria, and 2)
of Figures 1 and 2. The solution was the weight % of feed fluids
achieved using the FLI9MET design reporting t o the overflow, which
balance program system (Richardson, et. was u n k m .
a l . 1198111.
!his second parameter is
Flowsheet Iaw>ut. 'Ihe flowsheet calculated for each cyclone
layout, as explained previously, is through the use of feetback
specified for inplt as in the data controller modules. Criteria
sheet of Figure 8. 'Ihe program then used were: 90 % solids for the
perform an organizational analysis , SAG m i l l screen oversize, 35 %
s a i v i d i n g the flowsheet into s e r i a l solids for the ball mill cyclone
and recycle groups, inserting overflow, and 60 % solids for the
convergence modules &re required, and regrind m i l l cyclone underflow.
r-rdering recycle groups t o prepare In each case a controller adjusts
for iterative calculations. the s d QI=LRS parameter,
fluid split, t o achiwe the
Desiqn Criteria. The reminder of the desired solids concentration in
information required t o initialize the the product stream.
program is obtained from the design
criteria. These are the specifications 3. Determination of launder Mter
governing the particular f lowsheet addition t o dilute flotation
under study. 'l%ey w i l l have an inpact concentrates is a calculation not
on unit operation module parameters, as performed by the flotation
CONCENTRATION AND DEWATERING CIRCUITS

W L E 111

DESIGN CRITERIA FOR C O F P ~ L Y B COKXNTFWDR


D ~

Plant feed Feed, mtph


Moisture, %
Copper grade, %
Molybdenum grade, %
sffimill Feed, % solids
Circulating load, %
Screen Oversize, % solids
Ball mill Feed, % solids
Circulating load, %
Cyclone Feed, % solids
Cwerflow, % solids
bugher flotation Copper recovery, %
Molybdenum recovery, %
Concentrate grade (ccpper), %
Concentrate, % solids at lip
Diluted concentrate, % solids
Regrind mill Circulating load, %

Regrind cyclone Underflow, % solids


1st Cleaner Copper recovery, % varied
flotation Molybdenum recovery, % varied
Concentrate grade (copper), % 15
Concentrate, % solids at lip 25
Diluted concentrate, % solids 15
2nd Cleaner Copper recovery, %
flotation Molybdenum recovery, %
Concentrate grade (copper), %
Cancentrate, % solids at lip
Diluted concentrate, 8 solids
Scavenger flotation Copper recovery, %
Molybdenum recovery, %
Concentrate grade ( c w r ), %
Concentrate, % solids at lip
Diluted concentrate, % solids
Overall plant Copper, %
recovery Holybdenum, %
(FLOAT) module. The f i r s t By adding controllers t o achieve
parameter of the FLQbT module these special criteria, there was no
defines the percent solids of the need for prior manual calculations or
concentrate a t the c e l l lip. The the design of special new equipnent
values of the f i r s t parameter and Rodules.
the concentrate percent solids
after being diluted with launder &&rxg Results. The final report
water are established by the obtained by the program is shown as
design c r i t e r i a of Table 111. Tables IV and V. These final balance
The amount of launder water results, which converged t o within a
addition required t o achieve the tolerance of 0.0001, were obtained
specified dilution can be after some 40 iterations of the most
achieved by use of feedforward corrplex recycle group.
controllers enploying the same
dilution algorithm as in the
exanple of Figure 9. OPERATINS DlUN AIUUslMENI?
4. Cverall process recovery Steady s t a t e material balances
c r i t e r i a for each cmponent calculated from operating data gathered
(cogper and mlybdenum) can be a t key sanpling points are inportant
met in two ways. Cne way is t o for many purposes including metallur-
hand calculate a l l recoveries gical accounting and rmdel building.
that mst be achieved by the Often, raw data acquisition is such
individual FUW modules t o that the system is over-defined (mea-
achieve the specified overall surement information is in excess or
recwery. lhese recoveries can there is redundancy in the data), so
then be entered as parameters of that a balance can be calculated in
the individual FIBAT modules. several different ways. In the absence
of error, such balances would agree
The second option requires a with each other. mfortunately, var-
review of a l l flowsheet modules ious sources of error in measurement
t o determine i f any of their create serious disagreement. This has
parameters w i l l affect overall led t o the developnent of s t a t i s t i c a l
conpcnent recovery. For t h i s data adjustment program which account
case, it is found that only FZ,aAT for measurement errors. Data points,
rrodules affect cmpnent adjusted i n a minimal sense, lead t o
recovery. A t t h i s point it is consistent balances. Programs usually
helpful t o determine h t minimize a suitable form of R for n
individual flotation stage data points:
recoveries may be best known,
perhaps based on testwork or
experience. For t h i s exanple, the
rougher, scavenger and second
cleaner recovery values were f e l t
t o be more reliable than the where w, is a weighting factor for M;,
estimate for f i r s t cleaner the i t h data point and 8; is the cor-
recovery of both ccmponents .
Controllers were used t o adjust
responding adjusted data point. Each Ri
is such that a l l mass balance con-
the individual recovery straints are satisfied.
parameters of the EUlW module
representing f i r s t cleaner Techniques t o minimize o m of R
flotation. In t h i s way, the can be placed in two major categories,
specified overall recwery was namely m e which involve Lagrange
achiwed. Any of the EUXT rwd- m l t i p l i e r s and derivatives and those
ules could have been selected as which involve direct search methods.
the adjusted equipnent, i f jus-
t i f ied by testwork or experience.
CONCENTRATIONAND DEWATERING CIRCUITS

TABLE IV
B&?WCE RESULTS - WTAL SOLID5 / FLUIDS

PROBLEM: CASE 21 BENEFlCIlTION CIRCUIT-COPPER MIE (ORIND-FLOTATION) DATE: 9130185

TOTlL STREAM INFORMlTIOH

STREAM STREW DESCRIPTION SMlD FLOW FLUID FLOW SOLIDS COlCENTRlTION


LABEL TIH TIH YT. Z SOLIDS
ttttttttttttttttt1ttttttttttttttttttttttttttttttttttttt+t+ttt~+t+ttt*tttttttttttttttttttttttttttttttt~tttttt*tt
1 New F e ~ d 500.00 20.83 96.00
7 Screen O'Size 75.00 8.33 90.00
2 Hew Feed L Screen O'Sife Undiluted 575.00 29.16 95.17
3 Process Water .OO 217.30 .OO
4 Total Sag H i l l Feed 575.00 246.40 70.00
5 Sag H i l l DischarqelScreen Feed 575.00 246.40 70.00
6 Screen U'Size 500.00 238.10 67.74
15 Cyclone U'FIor 1498.00 1070.00 58.33
8 Screen U'Sife ICyclone U'FIor 1997.00 1308.00 60.43
9 Process Water .OO 23.94 .OO
10 B a l l H i l l Feed 1997.00 1332.00 60.00
11 B a l l H i l l Discharge 1997.00 1332.00 60.00
12 Process Water .00 665.80 .OO
13 Cyclone Feed 1997.00 1997.00 50.00
14 Cyclone O'Flor 499.30 927.30 35.00
I6 Rwgher Concentrate 64-29 96.43 40.00
17 Rougher T a i l 435.00 830.90 34.37
19 Rougher Concentrate Diluted 64.29 192.90 25.00
23 Regrind H i l l Discharge 73.53 49.02 60.00
36 Scavenger Concentrate Diluted 9.26 52-46 15.00
20 Regrind Cyclone Feed 147.10 294.30 33.32
21 Regrind Cyclone O'Flor 73,53 245.30 23.06
22 Regrind Cyclone U'Flow 73.53 49.02 60.00
30 2nd Cleaner T a i l 14.49 127.50 10.21
24 Total 1st Cleaner Feed 88.02 372.80 19.10
25 1st Cleaner Concentrate 31.50 94.50 25.00
26 1st Cleaner T a i l 56.52 278.30 16.88
27 Launder Water -00 84.00 .OO
20 2nd Cleaner Feed 31.50 178.50 15.00
29 2nd Cleaner Concentrate 17.01 51.03 25.00
31 Launder Water ,00 45.36 -00
32 Final Concentrate 17.01 96.39 15.00
33 Scavenger Concentrate 9,26 27.77 25.00
34 Scavenger Tail 47.26 250.50 15.87
35 Launder Y a t r ,00 24.69 .OO
37 Final Tails 482.30 1081.00 30.84
18 Launder Water .OO 96.43 .OO
METALLURGICAL BALANCES

TABLE V
BALANCE RESUETS -

PROBLERI CASE 21 BENEFlClIITIDll CIRCUIT-COPPER ORE (GRIND-FLOTIITION) DATE: 9130105

SOLID COHPOWENT FLOHRRTES CONCElTRRTlOlS

STREM STREIIH DESCRIPTION COPPER MLY COPPER lMLY


LLBEL TIH TIH 2 2
ftttfttfttttttftfft+t#fttttt+tftfttftftfttffftftfff+tftftttt+tttttttttftftftfffffftffffffHIfffffftff+ffftftfffftff
I New Feed 5.00 -25 1.00 -05
7 Screen O'Size .75 -04 1.00 -05
2 N e Feed k Screen O'Size Undiluted 5.75 .29 1.00 .05
3 Proce~sHater -00 .00 .OO -00
4 Total Sag H i l l Feed 5.75 .29 1,OO -05
5 Sap H i l l DischarpelSEreen Feed 5.75 .29 1.00 .05
6 Screen U'Size 5.00 .25 1.00 -05
15 Cyclone U'Flor 15.00 .75 1.00 ,05
0 Screen U'Size k Cyclone U'Flw 20.00 1.00 1.00 .05
9 Process Hater .00 -00 $00 .OO
10 Ball H i l l Feed 20.00 1.00 1,OO .05
11 Ball H i l l Discharge 20.00 1.00 1.00 .05
12 Pracns Hater .00 .OO .OO .OO
13 Cyclone Feed 20.00 1.00 1.00 .05
I4 Cyclone O'Flow 5.00 .25 1,OO -05
16 Rougher Concentrate 4.50 -20 7,OO .31
17 Rougher Tail .50 05 , I1 -01
19 Rougher Concentrate Diluted 4.50 .20 7.00 -31
23 Rrgrind H i l l Discharge 4.07 .27 6,62 .37
36 Scavengtr Conctntrate D i l u t t d .37 .07 4.00 .76
20 Regrind Cyclone Feed 7,74 ,54 6.62 .37
21 Regrind Cyclone O'Flolr 4,87 -27 6.62 .37
22 Regrind Cyclone U'Flon 4.07 .27 6.62 .37
30 2nd Cleaner Tail .47 .04 3.26 29
24 Total 1st Cleaner F t t d 5.34 .31 6-07 -36
25 1st Cleantr Concentrate 4.72 .21 15.00 .67
26 1st Cltantr Tail .62 .I0 1.09 .I0
27 L d e r Hater -00 .OO .OO -00
28 h d Cltaner F t d 4.72 .21 15.00 .67
29 2nd Cleaner Concentrate 4.25 .17 25.00 1.00
31 Launder Yater .OO .00 .OO .00
32 Final Concentrate 4.25 -17 25.00 1.00
33 Scavenger Concrntrate ,37 a 07 4.00 .76
34 Scavenger T a i l .25 03 .52 .06
35 Laundrr Hater -00 .00 -00 -00
37 Final Tails
10 Laundw Yater
,75
.OO
.00
.DO
.I5
.OO .
-02
00
Over-Defined Data Sets

Consider a rougher flotation circuit


a t steady state. From a mass balance,
F = C + T, where F, C and T are mass
flow rates of feed, concentrate and
tailing respectively. If a l l three
flaws are measured in some way, then i n
general F w i l l not equal C+ T. nis
situation arises because of errors in
each measured data point and the prob-
lem is noticed because the data s e t is
over-defined. Only two of the terms are
necessary - the third is calculated
from the other two. It is tenpting t o
drop, say, F and calculate it (e.g.,
= C + T, where means calculated)
A

m v e r , this is arbitrary - why not


.
drop C or T? In a l l fairness, despite
the fact that one of the measured data
p i n t s is redundant for balancing
Furposes, a l l three should be permitted
t o have some influence on the "true"
balance. In fact, data adjustment pro- FIGURE 11. Sinple Flotation Circuit
cedures "thrive" on r-dancy, and
accuracy w i l l increase as the nlmber of A standard procedure has been de-
redundant data points increases. vised (Mular, e t . al., t19761) t o ad-
j u s t raw data by means of search
Program Not Involving Derivatives methods. It involves (a) choosing
synbols for a l l flaws and measurements
To minimize R for a s e t of raw data, in the circuit, (b) tabulating measured
methods which involve neither deriva- data points, (c) writing a l l mass
tives nor Lagrange rmltipliers have balance relationships t o be satisfied,
been fornulated. Black (1968) enployed (dl choosing search variables, (el
the Hooke-Jeeves direct search routine showing how t o calculate measured data
t o adjust measured flows of solids and points from search variables, (f
water in a bank of rougher flotation writing an objective function t o mini-
cells. The objective function mini- mize, (g) picking a search method
mized was R with w inversely propor- and (h) writing a canplter program.
tional t o the square of the correspond-
ing i t h data point. A unique feature Qloose Synbols selected t o
was the use of a direct search. Using represent mass flows and assays are
different search techniques, Mular shown in Figure 11 and Table VI. Mte
(1976,1979,1983) and m i t e (1977) have that the tailings flow rate was not
adjusted metallurgical data with suc- measured.
cess. To illustrate camon features
of such methods. consider the sinple Tabulate Table VI shows the raw
flotation circuit shown in Figure 11. data a s measured from sanples/sensors
A bulk concentrate of Cu-Mo is t o be a t suitable locations, the tailing
separated where Mo minerals w i l l float. stream mass flow rate being excluded.
Cu minerals constitute a tailing. Note that synbols w i t h a bar above them
Elements t o be assayed in each stream mean raw data popints (i.e., measured
are a,Mo, Fe and C. Mass flow rates data p i n t s ) .
of each stream, except the tailing, are
t o be measured.
METALLURGICAL BALANCES

TABLE VI
RAW FLUl'ATION CIRCUIT DATA

3 = 30.00WJFH of dry feed


l!7 = 1.12 WJFH of dry Conl
gs = 1.51 WJFH of dry Con2
C3 = 0.79 MTPH of dry Con3
T = - of dry Tail

Balances. Balances that m s t be There are 20 degrees of freedom, so


satisfied a t steady s t a t e involve t o t a l there are 20 search variables which may
mass flows and element mass flows as be chosen arbitrarily. Choose F, CI, C,
follows: , C,. cfi (i = 1 t o 4 and j = 1 t o 3)
and t i = 1 t o 4).
CalculatemEEQmsearchVariables.
Each data point shown i n Table VI rnay
be calculated from some suitable c m
bination of the search variables,
although not a l l w i l l be required in a
particular conbinat ion. Bcpressions
for calculation of data points mst be
written in preparation for the objec-
Equation (2) above introduces 5 new t i v e function. Letting the synt>ol A

variables; equation (3) involves 4 mean a data point calculated from


equations each with 5 new variables. search variables, then:
Hence there are 25 variables and 5
equations - 20 degrees of freedom.
This means that 20 variables mst be
kncrwn t o determine the balance.

Choose Search Variables. Search


variables are the mininuin nlmber of
variables that are necessary t o A
determine the abwe balances. For C" = CiJ
example there are 5 new variables "(j = 1,2,3 and i = 1,2,3,4)
introduced in equation (2) and one
-
equation. Hence 5 1 = 4 variables
are necessary t o define the balance,
since the 5th can be found from the
other 4.
Note that each "hatted" term w i l l
appear i n the objective function and
CONCENTRATIONAND DEWATERINC CIRCUITS

that each depends only upon some with n t l s e t s of n search variables.


conbination of search variables. The sinplex mves on the R surface by
reflection into an adjacent sinplex and
Construct Cbiective Function. Equation is forced t o adapt itself t o the local
(1)may be written as follows for the landscape, elongating dawn inclined
adjustment problem being considered: planes, changing direction on encoun-
tering a valley a t an angle, and con-
tracting in the neighborhood of a mini-
m R. 'Ihe search is rapid a t the
s t a r t , but slows considerably in the
vicinity of R.
The m i t e search involves three
routines, namely, a step search,
repetitive quadratic f i t t i n g , and a
where the weighting factor, wt for the search along a vector defined by two
i t h data point has been set t o one. vertices most likely oriented in the
OFten, w i is estimated as the inverse direction of m i n i m R. The step
of the error variance associated with search and repetitive quadratic f i t t i n g
the i t h data point. For t h i s exanple, are patterned after Coggin (Kuester and
error variances (24 of them, one for Mize 119731) and enployed for each
each variable measured) were not search variable t o find two vertices,
estimated. Vl and V2. Then a vector V is f i t t e d
t o Vl and V2 by:
In same situations, w ; has been s e t
t o the inverse of the measured data
point squared. In t h i s cbse, the frac-
tional dwiatiom squared are t o be
minimized.
where V2 is a vector of the current
Search A search method 5 e e t " search variables, and V1 is a
is a ma-tical procedure which finds vector of the prwious "best" search
that m e conbinatiar of the search variables. The variable a is treated
variables which causes R t o be a mini- l i k e a search variable in the Coggin
mnn. M m the unique set is known, then routine but is constrained between
a l l data points (hatted terms) calcu- limits that make elements of V greater
lated from it would constitute the than or equal t o zero. A new best
adjusted data. vertex, V, Occurs at OI = an;,,
routine has been described in detail
.
The
h e way t o find the s e t , would be t o ( m i t e 119771 and Mular, et al.
guess a t their values (20 of them) and
calculate R. Anather guess muld be
[I98311 .
nede and R again calculated. If the For less than about 30 search
prwious R is larger, then the second variables the sinplex search is not
gusss is best. Clearly, with 20 search t b w x n ~ ~ u m i n g .Above 30, the m i t e
variables t o the set, it w i l l be many search is preferred because it is
years before an answer is found1 faster by a factor of 10. For the
Direct search methods are the prefer red exanple herein, the sirrglex search
route t o find the minimmi R. These routine was selected.
have been reviewed (Kcmlik and m r n e
[I9741 ), and a t least tvo, namly, the W Proarrrm. The Sinplex
Sirrplex search @War and farsen 119821) search routine may be programed f ram
and the m i t e march ( m i t e [I9771) the original paper by Nelder and Mead
have been used for data adjustment in (19651. Alternatively, most conplting
recent years. centers w i l l have the program available
as a general minimization routine. 'Ib
The sirrplex direct search begins set up the starting sinplex, s t a r t
values for search variables a r e esti- a s follows:
mated from raw data. Starting s t e p
sizes shwld be selected t o make t h e 1. Adjust tonnages such t h a t a
sinplex a s "regularn a s possible. mininum adjustrent occurs using
element balances a s constraints.
A program was written in BASIC t o If one tonnage is grossly i n
adjust the raw data shown in Table VI. error, it may be assigned a zero
EQsults fromthe adjustment a r e tab- weight (k = 0).
ulated in Table VII, where it is a p r -
ent that adjustments tend t o be m i n h l . 2. Close the tonnage balance using
'Ihe R value a t the m i n i m was 2.04. the i n i t i a l l y adjusted tonnages
M t e t h a t the largest discrepancy is with the tonnage balance a s the
with the mass flow r a t e of Con3 constraint.
(approximately a 30 % difference) . 3. Close each element balance with
Programs Involving Lagrange Ftultipliers its respective element balance
a s the constraint.
Nielsen and Diaz (1970) described a
method t o adjust data for metallurgical For the f i r s t step, element assays
balances. Equation (1) ffom above was a r e assumed t o be correct and then
employed with w; = ;;k / S; (k is a gross tonnages a r e adjusted t o best f i t
weighting factor with value of zero or element material balances with assays
one and ST is an estimate of the error held constant. 'Ihe objective function
variance associated with the i t h data t o minimize is:
point). Lagrange multipliers were
introduced to locate the extremes of a
differentiable objective function with
the nlmber of multipliers equal t o the
nunber of constraint equations (element
balances). 'Ihe procedure was adopted

TABLE VII
WARISON BETWEEN MEASURED AN) A I U U m QATA

Feed
Con1
Con2
Con3
Tai 1

Feed call Con2 Con3 Tail


Meas. 30.00 1.12 1.51 0.79 -
Adjust. 30.00 1.21 1.83 1.03 25.93
630 CONCENTRATION AND DEWATERING CIRCUITS

where h; is a rmltiplier for the i t h which included error terms were


balance, XjI is the assay of element i
in stream j, S i is the variance as-
linearized in approximate fashion. An
approximate solution was obtained and
ciated with a measured gross tonnage Wi inproved upon in an iterative fashion
for the jth stream, Wj is a f i r s t t o whatever accuracy was desired. The
estimate of gross tonnage for the jth program was generalized t o permit
stream, and rj is f 1 (-1 if the stream balancing of any circuit.
is an o u t p t , +1i f the stream is an
.
inplt) Partial derivatives of 2, with 'Ihe lagrange rmltiplier method has
.
.
respect t o hi (i = 1,2,. ,n) and W e ( j
= 1,2,. ,b) are determined and se6 t o
been used t o minimize R in a different
way by Smith and Ichyen (1973). They
zero. The resulting normal equations eliminated by substitution the hi in
are solved s h l t a n ~ u s l yt o obtain the n o m l equations by solving them
.
i n i t i a l estimates, W j for assay adjustments, O X j i . If an
inihial estimate of an assay is denoted
In the second step, the tonnage by X j i then a better estimate w i l l be
balance is closed by minimizing: Xji = 4+ . q i Yji was tha replaced
by XJiand the Rosenbrock direct search
method was enplayed t o find a s e t of W j
values that minimized 4;. A new s e t of
AXJ\values were then determined and the
procedure repeated t o suitable comer-
where ho is a rmltiplier and W j w i l l gence. A disadvantage is that measured
be the final adjusted tonnage in the tonnage values are not permissible data
jth stream. points.
The third step is analogous t o step Hockings and Callen (1977 enployed
two, except that the objective function the Lagrange method for a general cir-
for the i t h element balance is: cuit using an incidence m t r i x nota-
tion, where process units were called
nodes and streams were referred t o as
branches. 'Iheir objective function was
written as:

where Xji is the measured assay of


element i in stream j, X,; w i l l be the
corresposlding final adjusted value and
$1 is the corresponding error variance.
Xjiis determined for each balance (i =
.
1,2,. ,n) by minimizing the correct 4; . where fju are weighting factors, Fij are
mveday (1971) extended the above +1 or -1 from the incidence matrix, and
method t o incorporate measured size the greek syrrbols are the rmltipliers.
analyses along with a m p and mass The Wj values in the second and fourth
flows. In addition, a technique t o terms of the equation are treated sepa-
estimate standard errors for f lowrates rately as a s t a r t (similar t o Nielson
was introduced. and Diaz a t t h i s point). If any of the
W j are fixed, such as the feed rate, a
Because each measured tonnage and subset of those that are independent is
assay is subject t o error, the element determined. Start values for the sub-
balances are essentially non-linear
Weigel (1972) ) utilized a technique
. s e t a r e selected and the dependent WJ
are calculated t o give a conplete t r i a l
such that R could be minimized wing set. The t r i a l set is tenporarily
Lagrange mltipliers. Cross-product regarded as constant. Next, the equa-
t e r n in material balance equations tion is differentiated w i t h respect t o
:A1BALANCES

each Xj* and each of the nultipliersk,,


Each equation is set equal t o zero.
The s e t of equations are solved s h l -
tanecusly and t h i s results in an ad-
justed s e t of X ~ l u e s . These are now Bartlett, J. T., 1979, "Metallurgical
the best estimates. The independent Wj Process Simulator", Paper presented t o
then beccane search variables for the 18th Annual Conference of Metallurgists,
Pawell search routine. Each objective CIM, Sucbury.
function evaluation called for by the
search uses estimates of X j u found from lack, K., 1968, "Search Methods in
the nonnal equations, until is mini- @tirnization", Procesees: 'Iheir
mized t o a suitable value. The problem ODtimizatlan.dControl,
of missing analyses and size reduction Mlar , A., and Bull, W., ed., Queens
equipnent was recognized. Univ., Kingston, Chap. 13.
The above method, although general- Coles, D. R., 1982, "Metallurgical
ized for any circuit, is really similar Flowsheet Solution by a New Sequential
t o that praposed by 9nith and Ichiyen, Modular -roachn, Proceedinss.
except that measured mass flows m y be m,SME-Am, pp. 63-77.
introduced a s constants and the Fuwell
search replaces the Rosenbrock search. FLB(MET User's GulQ, 1983, Fluor
Haever, i f mass flows are measured mgineers, Inc., Redwood City,
variables (i.e., such as by combined California.
density/flow meter sensors), the errors
associated with the measurements are Gottfried, B.S., and Abara, J., 1978,
not taken into account. " m t e r Simulation of Caal Preparation
Plantsw, Carrxlters a d Chemical
Since 1978, data adjustment programs -, MI. 2, no. 213, m.
involving Lagrange multipliers have 99-107.
became extremely general. A user
enters h i s circuit and raw data in Oupta, V. S. , and Messa, C: J., 1980,
standard format and runs the program. "ISMPX Process Evaluation Executive
Because minly analytical solutions are (-1 : A General Purpose Sirmlator for
enployed by the programs, carrplting Canplting Steady-State Material and
time is reduced substantially. Thus mergy Balances of Hetallurgical Process
program such as BILMAT and MATBAL Flmheets", Preprint 80-136, Annual
(Laguitton l19851) written for the SPOC AIME Meeting, Kas Vegas.
project are accanpanied by colrplete
user-oriented documentation. Statis- Hockings, W. A,, and Callen, R. W.,
tical routines have been incorporated 1977, "Con-pter Program for Calculating
to assess raw data reliability. mss Flow Balances of Continuous Process
Streamw, Paper 77-8-372, 9uIE Fall
Meeting, St. Louis.
Izurieta, H., and Edwards, L. L., 1980,
The authors wish t o thank the management "The General mergy and Materials
of Fluor Engineers for permission t o use Balance Conplter System (GEMS) Applied
certain m t e r i a l Aich was excerpted t o Hydrometallurgical Processesw,
from the FrS(MET User's Q,I& (Fluor JourMl pf &&&, MI. 32, no. 12,
Engineers [I98311. D e c e b r , pp. 23-31.
King, R. P., 1985, "MBSIw: A Powerful
Modular Ore Dressing Plant Sirmlatorw,
Paper presented t o 24th Annual
Conference of Metallurgists, CIM,
Vancouver .
CONCENTRATIONAND DEWATERING CIRCUITS

Rawalik, J., and Osborne, M.R., 1974, Minimization", m,vo1.7,


l!whxkfnr-m p. 308.
-, Elsevier, New York.
Nielson, W. H., and Diaz, L., 1970,
Wester, J. L., and Mize, J. H., 1973, "Techniques for Lncreasing Accuracy in
Techniauesa-, M i l l Performance Calculations", Paper
McGraw-Hill, New York, p. 275. 70-AR-333, SME Fall Meeting, St. Louis.

bguitton, D., 1985, SPOC: A Reid, et a l . , 1982, "A Survey of


conplter-Based Methodology for Ore and Material Balance Canplter Packages i n
Coal Process Evaluation and the Mineral Processing Industry",
@timizationn, Paper presented t o 24th - P llth
Annual Conference of Metallurgists, CMI
Vancouver.
York . New

Richardson, J. M., et al., 1980,


Loveday, B. K., 1973, "Mass-Balance -
"FLWIWr A Flexible Coqmtation System
Smoothing of Flotation Plant Data", for Steady-State Analysis of
Internal Report, Nat. Inst. for Met., Metallurgical Process Flowsheets" , Paper
Johannesburg, S. A. presented t o APlE Arizona Conference
Annual Meeting, mcson.
McKee, D, J., 1985, "General Purpose
Sirmlator of Mineral Processing Richardson, J. M., Coles, D. R., and
Operations", Paper presented t o 24th M i t e , J. W., 1981, "Fluor Mining and
AMual Conference of Metallurgists, CIM, Metals Introduces FLEOQT", awineerincl
Vancouver. &-Journal, vol. 182, no. 10,
October, pp. 88-97.
Mular, A. L., et a l . , 1976, "Mass
Balance of a Grinding Circuit", C U Richardson, J. M., and Sredanovic,
&U&i& vol. 69, no. 776, D s e s r , M. S., 1985, "A New Generation Process
~ p 124-129.
. Simulator for the Minerals Industry",
Bm!+Technolowinmineral
Mular, A. L. , and Herbst, J. A., 1978, I- Spisakr J - I &.I S ~ E - ~ I
"Digital Simulation: An Aid for Mineral Littleton, Chap. 14.
Processing Plant Design", Mineral
Mular, A. L.r Smith, H., and Ichiyen, N.r 1973,
and Bha~pl,R. B., ed., SMl+AIME, "Conplter Adjustment of Metallurgical
Littleton, Chap. 14. Balances", aBulletin, Sept .
Mular, A. L, 1979, "Data Adjustment Ti-, R., et al., 1978, "Mass Balances
Procedures for Mass Balances", Cgmats i n M i l l Metallurgical CQerationsn, 10th
!!&$K&fpE~Uin*ELinerals Annual Meeting of Canadian Mineral
-, A. Weiss,ed., SMD-AIME, Processors, CANEC, Ottawa.
Littleton.
Westerberg, A. W., e t al., 1979, Bxxs%
Mular, A. L., and h r s e n , C. R., 1982, ,- Canbridge Univ. Press, New
"Acquisition of Steady-State Grinding York.
Circuit Data for Model ~ u i l d i n g " , SgM:
w,CAEFIET, Ottawa, Chap. 2-2. Wiegel, R. L., 1972, "Advances i n
Mineral Processing Material Balancesn,
Mular, A. L, e t al., 1983, "Ihe Quart&, vol. 11, no. 2,
s t a t i s t i c a l Adjustment of Potash p. 413
Flotation mta for Mass Balancesn,
Potash B ,W e r c h e r , R., ed., M i t e , J. W., et al., 1977, "A Useful
Pergamon Press, Toronto. Technique for Metallurgical Mass
Balances - m l i c a t i o n s in Grinding",
Nelder, J. A., and Mead, R., 1965, "A D L L Mineral Processirm, vol. 4,
Sinplex Method for Function Elsevier, p. 39.

You might also like