Professional Documents
Culture Documents
Coal Preparation
Publication details, including instructions for
authors and subscription information:
http://www.tandfonline.com/loi/gcop19
This article may be used for research, teaching, and private study purposes.
Any substantial or systematic reproduction, redistribution, reselling, loan,
sub-licensing, systematic supply, or distribution in any form to anyone is
expressly forbidden. Terms & Conditions of access and use can be found at
http://www.tandfonline.com/page/terms-and-conditions
INTRODUCTION
Test data collected from operating coal preparation plant circuits are
often untrustworthy. There are several possible reasons for this problem
including improper sampling procedures, unreliable laboratory analyses,
and unstable equipment operation. Questionable data may result in
Received 8 March 2004; accepted 23 March 2004.
Address correspondence to Gerald H. Luttrell, Department of Mining and Minerals
Engineering, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061,
USA. E-mail: luttrell@vt.edu
35
36
G. H. Luttrell
misleading conclusions that can result in lost productivity and unnecessary expenditures of capital and operating funds. The traditional
approach of assessing the reliability of experimental data is to repeat
the test program or analyses and to compare the reproducibility.
Unfortunately, this approach is generally impractical due to the high
costs of plant sampling programs and the mathematical complexities
associated with interpreting statistical information.
A simpler and less costly approach of assessing the reliability of circuit
and plant test data is to assess and adjust the data using mass balances.
Mass balances can be readily used to analyze the reliability of the test
data and to make statistically sound estimates of the true performance
indicators [19]. Unfortunately, the mathematics associated with mass
balances can be cumbersome when conflicting data sets and=or complex
circuits are evaluated. To overcome this problem, a spreadsheet-based
mass balance program has been developed using optimization tools
embedded in modern spreadsheet programs. To utilize this program,
users simply enter a description of the processing circuit using a simple
connection matrix followed by the input of the assay values (i.e., ash,
sulfur, heat content, etc.) for each stream. The built-in minimization
routines then generate the best estimates of the assay values and flag
values that may be unreliable. This approach requires no formal
knowledge of advanced mathematics or statistical analyses. In addition,
the mathematical routines incorporated within the spreadsheet-based
platform can be readily modified to handle specific problems without the
need for any formal programming experience.
37
38
G. H. Luttrell
TABLE 1 Flowsheet Connection Matrix for the Two-stage
WOC Circuit
Node ( j)
Stream (i)
1
2
3
4
5
6
7
8
1
1
0
0
0
1
0
0
0
1
1
1
0
0
0
0
0
0
1
0
1
1
0
0
0
0
0
1
0
0
1
1
Node 1, Feed Sump; Node 2, Primary WOC; Node 3, Secondary WOC; Node 4, Clean Coal Cyclones.
Stream 1, Circuit Feed; Stream 2, Primary WOC Feed; Stream
3, Primary WOC U=F (or Secondary WOC Feed); Stream 4, Primary WOC O=F (or Clean Coal Cyclone Feed); Stream 5, Secondary WOC U=F (or Refuse); Stream 6, Secondary WOC
O=F; Stream 7, Clean Coal Cyclone U=F (or Clean); Stream 8,
Clean Coal Cyclone O=F (or Flotation Feed).
Cij Mi 0;
Cij Aki Mi 0;
i1
m
X
i1
where Mi is the mass flow rate of each stream i and Aik is the assay for
each component k in stream i. According to Smith and Frew [11], the
minimum number of streams (N) which must be sampled and assayed to
obtain an unambiguous mass balance can be calculated from
39
N 2F S 1;
where F is the number of feed streams entering the circuit and S is the
number of simple separators (nodes having one stream in and two
streams out). For the case shown in Figure 1, there are three simple
separators (S 3) and one feed stream (F 1). Therefore, in addition to
the feed mass flow rate, samples from 7 different streams must be
collected and assayed for a single component in order to obtain a mass
balance that is exactly defined for this circuit.
! 0;
WSSQ
si
ski
k1 i1
i1
where sik and si are standard deviations of the measured assay values
and measured flow rates, respectively [1]. The superscript * is used to
distinguish estimated values from experimental values. In most cases, sik
40
G. H. Luttrell
s i ei Mi ;
where eki and ei are the respective relative errors in the measurements of
the kth component assays and mass rates for stream i.
Several mathematical methods can be used to perform the mass
balance minimization. The earliest approaches included the use of
Lagrangian multipliers [2] and direct search techniques [3]. Historically
these approaches have been limited to simple systems (often only one
node) and have typically required the construction of computer programs
to perform the minimization. More recently, complex algorithms have
been developed that can be used to simultaneously compute the best
estimates of mass flow rates and component assays for relatively
complicated circuits. Some of the most noteworthy contributions made to
this scientific area were made by Hodouin and associates [1214].
Unfortunately, many of the advanced mass balance algorithms are
difficult to apply for arbitrary circuit configurations and require computational capabilities that often exceed those of personal computers [10].
More importantly, most practicing process engineers simply lack the
advanced mathematical skills required to solve the constrained minimization problem defined by Equation (4). As a result, commercial
software packages have been specifically developed for this purpose.
Popular software packages include Bilmat (Algosys, Canada), JKMBal
(JKTech, Australia), and Bilco (BRGM, France). All of these packages
are capable of handling relatively complex mass balance problems
commonly encountered in the coal industry. However, the commercial
programs are costly and cannot be distributed among other users without
the purchase of additional licenses. These programs can also be incompatible with other software packages, require considerable user training, and
cannot be modified by the user. More importantly, the commercial packages
have capabilities that tend to exceed the general needs of personnel responsible for the routine evaluation of common coal plant circuits. As a result, the
vast majority of coal processing engineers simply avoid conducting mass
balances when evaluating the performance of coal processing circuits.
41
process engineers do not realize that spreadsheet developers have incorporated powerful optimization routines into their software. These built-in
routines can be used to quickly solve constrained minimization problems
such as those encountered in mass balances, quickly. This integrated
capability makes it possible for nearly anyone to construct mass balances
without the need for extensive programming skills.
Many of the leading producers of spreadsheet software make use of
global optimization routines developed during the past decade by
Frontline Systems, Inc. These routines include SOLVER for Microsoft
Excel, new SOLVER for Lotus 1-2-3, and OPTIMIZER for Corel
Quattro Pro. These routines, which are typically provided as an add-in
to the basic spreadsheet program, use the GRG2 (Generalized Reduced
Gradient) nonlinear optimization code developed by Leon Lasdon
(University of Texas at Austin) and Allan Waren (Cleveland State
University). These routines make it possible to quickly identify an optimal value for a formula in a target cell that is related, either directly or
indirectly, to other cells in the spreadsheet. These cells can be selected and
adjusted by SOLVER (or OPTIMIZER) to produce the result specified
in the target cell subject to user-defined constraints. For example, consider the generalized mass balance problem discussed previously.
SOLVER can be used to minimize a target cell calculated using Equation
(4) by adjusting mass flow (Mi) and component assay (Aik) values subject
to constraints calculated from Equations (1) and (2). SOLVER allows the
user to have direct control over many of the criteria used in the adjustment of the data (e.g., degree of precision, error tolerance, degree of
convergence, iteration method, etc.). The ability to perform the high-level
minimization without extensive programming makes this approach very
powerful. More importantly, the optimization routine is extremely robust
and capable of dealing with many of the pitfalls commonly encountered
in other mathematical approaches. Because of these advantages,
spreadsheets now offer an attractive alternative to commercial software
packages for solving many industrial mass balance problems.
In order to demonstrate the advantages of spreadsheet-based mass
balances, a generalized program has been developed in the present work
using the Microsoft Excel for Windows. The present form of the
generalized mass balance spreadsheet has been configured to handle plant
circuits with up to 20 streams and 10 nodes. The spreadsheet can be easily
expanded by the user to handle a greater number of nodes and streams.
Custom spreadsheets which balance size-by-size washability (float-sink)
data can also be constructed for relatively complicated mass balancing
problems using the same basic approach. Some of the balances
constructed to date include more than 500 constraints and variables, yet
they can be handled with relative ease by the SOLVER routine. In each
case, the user simply enters the experimental mass rates and component
assays for each size and gravity class. From these input values, all data
42
G. H. Luttrell
are converted to mass flow rates (i.e., ton per hour of total dry solids, ton
per hour of ash, ton per hour of sulfur, ton per hour of slurry, etc.). The
mass balance is performed by using SOLVER to adjust all input values
(mass flows and assays) by the smallest amount possible such that the
sum of all component mass flows entering a node minus those leaving the
node are equal to zero. These constraints are relatively easy to configure
in a spreadsheet format once the assay values have been converted to
mass flow rates.
In order to demonstrate the utility of mass balance spreadsheets, it is
best to consider a relatively simple example. Consider the flowsheet
43
Stream
Circuit feed
Cyclone oversize
Cyclone undersize
Spiral concentrate
Spiral reject
Screen oversize
Screen undersize
Flotation feed
Flotation concentrate
Flotation reject
Rate (tph)*
Ash (%)
Sulfur (%)
67.89
55.52
22.25
41.44
20.03
33.32
8.69
20.76
14.23
10.92
27.90
26.10
33.50
8.60
76.20
7.81
21.40
32.20
8.10
71.80
0.89
0.71
1.33
0.56
1.21
5.50
0.71
1.25
0.86
1.71
Step
Step
Step
Step
Step
44
G. H. Luttrell
45
FIGURE 5. Completed DATA page showing measured mass and assay values
(after cut-and-paste of measured values into estimated values).
46
G. H. Luttrell
provided on the right-hand side of this page. Large changes suggest that
the measured values are not reliable or indicate severe problems with the
internal consistency of the overall data set. Prior to making adjustments,
the percentage change for each value is zero.
Minimization Routine
The total WSSQ must be minimized to obtain the best statistical estimates of the circuit mass flow rates and assays. The value of WSSQ is
calculated internally and displayed at the bottom of the DATA page. In
the Microsoft Excel platform, the WSSQ minimization is conducted using
a built-in tool known as SOLVER. This powerful add-in is provided as a
standard feature of Microsoft Excel, although it may need to be activated
during installation in order to be functional (see the Tools menu).
Figure 7 shows the dialog box that appears when the SOLVER add-in
is called from the Excel toolbar. When invoked, this integrated routine
allows the user to specify a target cell to be minimized. In the present
example, the cell containing the value of the total WSSQ at the bottom of
the DATA page is selected. The Min button is then toggled within the
dialog box to indicate that a minimization problem is to be undertaken.
Finally, the range of cells to be adjusted by the minimization routine is
specified within the By Changing Cells box. The mouse pointer can be
used to select these cells for adjustment. The selected values normally
include all the numerical values to be estimated in the right-hand side of
the DATA page. In most cases, the experimentally measured values serve
as the best approximations of the true values. Therefore, good initial
estimates can be obtained by simply copying the measured values from
47
the left side of the DATA page and pasting them onto the right side of the
same page (see Figure 5). The limitations imposed by Equations (1) and
(2) must also be entered as constraints within the SOLVER dialog box.
For convenience, these constraints have been entered already when the
spreadsheet was constructed and need not be changed. However, these
constraints can be readily manipulated by the user to customize the
spreadsheet to solve very specific types of mass balance problems. Special
constraints, such as upper or lower limits on dependent values, can be
readily entered by the user as new mathematical constraints within
SOLVER. These special constraints are typically difficult to handle using
commercial mass balance software. If necessary, the Options button
can be selected prior to minimization. These include adjustable parameters for numerical precision and convergence limits. Solutions can also
be restricted to include only nonnegative values (which is typically
appropriate for mass balances). After entering all of the requested input
values, the minimization routine is initiated by choosing the SOLVE
button.
48
G. H. Luttrell
FIGURE 8. DATA and ERROR pages after solving for estimated values.
the right side of the DATA page. To assist in the comparison of the
measured and adjusted values, the percentage change is shown on the right
side of the ERROR page. This figure shows that the spreadsheet made
significant changes (> 25%) to several of the mass flow rates (i.e., spiral
refuse, sieve undersize, and flotation refuse). These large changes are
expected since the relative errors are set very high for these data entries
(i.e., 10%). However, the data also show that a self-consistent data set
could only be obtained by making a significant change to the sulfur assay
for the sieve oversize stream (i.e., from 5.50% to 0.59%). This indicates
that the sulfur assay for this particular stream may be unreliable. In fact, a
follow-up discussion with the analytical laboratory indicated that this
particular sulfur assay had been entered incorrectly by the laboratory staff
as 5.5% instead of the experimentally measured value of 0.55%.
Figure 9 shows the results of a second mass balance conducted after
properly entering the correct numerical value for the sulfur content of the
49
FIGURE 9. DATA and ERROR pages after changing the data entry for an
incorrectly entered value of sieve oversize sulfur content.
sieve oversize stream. This change dropped the total WSSQ down to 111,
indicating a much better data fit. More importantly, the relative changes
to the ash and sulfur assays are all well below 3% for this particular
solution. Therefore, these adjusted assay values for ash and sulfur are
believed to be good estimates of the actual circuit performance. On the
other hand, significant changes are still required in order to obtain
reasonable balances for the mass flow rates. As shown in Figure 10, most
of the measured mass flow rates are typically higher than the adjusted
values. These larger changes are expected since the relative standard
deviations for the mass flow rates were set at 10% compared to 1% for
the ash and sulfur assays. This difference was deemed appropriate given
the experimental means and conditions under which the mass flow rates
were determined. The adjusted mass flow rates are believed to be good
estimates of the actual circuit performance since the mass balance
50
G. H. Luttrell
procedure uses the entire data set (and not just isolated values) to
establish these values.
SUMMARY
The establishment of a consistent mass balance is a key step in the
analysis of data obtained from coal processing circuits. Unfortunately,
this task can be difficult when redundant or conflicting data are obtained
experimentally. To help resolve this problem, a spreadsheet-based
approach for reconciling mass balance data has been presented. This
technique makes use of minimization tools that are now available in most
modern spreadsheet programs. These tools, which are typically incorporated as add-ins called SOLVER or OPTIMIZER, make it possible for
plant engineers to perform relatively complicated mass balance equations
that were once considered to be possible only through the use of commercial software packages specifically designed for this purpose. The
51
REFERENCES
[1] R. L. Wiegel, Advances in Mineral Processing Material Balances, Canadian Metallurgical Quarterly, Vol. 11, pp. 413419 (1972).
[2] G. W. Cutting, Estimation of Interlocking Mass Balances on Complex Mineral
Beneciation Plants, International Journal of Mineral Processing, Vol. 3, pp. 207218
(1976).
[3] A. L. Mular, Data Adjustment Procedures for Mass Balances, Computer Methods for
the 80s in the Mineral Processing Industry (A. Weiss, ed.), SME=AIME, New York,
pp. 843849, 1979.
[4] R. L. Wiegel, The Practical Benets of Improved Metallurgical Balance Techniques,
AIME Annual Meeting, New Orleans, LA, Preprint 7992, 1979.
[5] D. Hodouin and M. D. Everell, A Hierarchical Procedure for Adjustment and
Material Balancing of Mineral Processing Data, International Journal of Mineral
Processing, Vol. 7, pp. 91116 (1980).
[6] K. J. Reid, K. A. Smith, V. R. Voller, and M. Cross, A Survey of Material Balance
Computer Packages in the Mineral Industry, Proceedings of the 17th APCOM
Symposium, New York, pp. 4162, 1982.
[7] B. A. Wills and R. J. Manser, Reconciliation of Simple Node Excess Data, Trans.
Institute of Mining and Metallurgy, Section C, Vol. 94, C209C214 (1985).
[8] B. A. Wills, Complex Circuit Mass BalancingA Simple, Practical, Sensitivity
Analysis Method, International Journal of Mineral Processing, Vol. 16, pp. 245262
(1986).
[9] D. Hodouin, T. Kasongo, E. Kouame, and M. D. Everell, Bilmat: An Algorithm for
Material Balancing Mineral Processing CircuitsApplications to Comminution,
Desliming and Flotation Circuits, CIM Bulletin, Vol. 74, pp. 123131 (1981).
[10] B. A. Wills, Metallurgical Accounting, Control and Simulation, Mineral Processing
Technology, 6th ed., Butterworth-Heinemann, Boston, Chapter 3, pp. 4291, 1997.
[11] H. W. Smith and J. A. Frew, Design and Analysis of Sampling ExperimentsA
Sensitivity Approach, International Journal of Mineral Processing, Vol. 11, 267284
(1983).
52
G. H. Luttrell