Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Maximal Flow Network Modelling of Production Bottleneck Problems

Author(s): M. D. Troutt, G. P. White and S. K. Tadisina


Source: The Journal of the Operational Research Society , Feb., 2001, Vol. 52, No. 2
(Feb., 2001), pp. 182-187
Published by: Palgrave Macmillan Journals on behalf of the Operational Research
Society

Stable URL: https://www.jstor.org/stable/254146

REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/254146?seq=1&cid=pdf-
reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Operational Research Society and Palgrave Macmillan Journals are collaborating with JSTOR to
digitize, preserve and extend access to The Journal of the Operational Research Society

This content downloaded from


182.255.0.242 on Mon, 04 Oct 2021 14:28:16 UTC
All use subject to https://about.jstor.org/terms
Journal of the Operational Research Society (2001) 52, 182-187 _(t2001 Operational Research Society Ltd. All rights reserved. 0160-5682/01 $15.00

www.palgrave.com/jors

Maximal flow network modelling of production


bottleneck problems
MD Trouttl* GP White2 and SK Tadisina2
1Kent State University OH, USA; and 2Southem Illinois University, IL, USA
This paper considers the analysis of process networks with bottlenecks and shows how they may be regarded as simple
multi-source maximal flow linear programming problems. We surveyed over 30 Production/Operations Management and
management science/OR textbooks, finding that only iterative trial-and-error procedures are now being suggested for this
kind of analysis. The maximal flow network approach is easier for complex problems and also allows several advantages
not available in the trial-and-error approaches. This paper also discusses the use of a simple linear programming
sensitivity result called radial change. The modelling approach suggested here can provide new ideas for improving
system capacity following the application of Theory of Constraints techniques.

Keywords: theory of constraints; capacity decisions; linear programming; sensitivity analysis; radial change

Introduction becomes preferred as problems become larger or more


complex.
With the development of the Theory of Constraints"' and
Although our emphasis in this paper is on analytical tools
the increased interest in Business Process Reengineering,12
related to bottleneck analysis it is essential to relate this
bottleneck analysis has become important. Today, many
work to the Theory of Constraints.
P/OM textbooks provide some discussion of process
networks with bottlenecks.13-16 However, when we
surveyed more than 30 P/OM and MS/OR textbooks, we
The Theory of Constraints
found those that discussed the topic suggested only some
form of iterative trial-and-error procedure to identify the The Theory of Constraints (TOC) is a system improvement
bottleneck and resultant system capacity. Of course, bottle- philosophy developed by Goldratt and others. 1-11 TOC has
necks are often obvious in real operating systems. Thus produced a number of principles and methods for improv-
analysis may not be needed for their actual identification. ing the flow of constrained systems. Before any modelling
However, analysis becomes useful in analyzing what-if technique is applied, effort should be expended to make
scenarios for capacity and configuration changes. those improvements available from TOC considerations.
Although none of the textbooks we surveyed mentioned For example, one would not generally wish to follow a
it, such bottleneck problems can be modelled as maximal capacity increase recommendation of a model for a bottle-
flow network linear programming problems. To date, neck process in a system for which TOC results lead to
however, that approach has apparently been confined to equivalent or better performance. Unless advantage is taken
only a few advanced papers.'7 However, it offers some of the improvements available from TOC, a model may be
useful advantages and flexibilities. From our textbook solving the wrong problem.'8
survey it appears that the relevance of such models for TOC applications are numerous and are considered in
general production bottleneck problems has not been more detail in the references cited above. Here we mention
adequately stressed. We conjecture that it has not been only a few. In reality, what we think is a bottleneck often is
widely discussed in this context due to (i) the overly simple not. For example, a process may have sufficient capacity for
examples usually considered, and (ii) difficulties with post- average demand, but be unable to handle peak load demand.
optimality analysis. For simple examples, such 'back-of- That process is not a bottleneck, but is a resource that does
the-envelope' techniques are no doubt easier and faster than not have enough 'protective capacity'. There may be other
a modelling approach. However, the network model ways to solve the problem than by treating it as a bottle-
neck.4 Policies and procedures often create situations in
which the bottleneck is not fully exploited. Before examin-
*Correspondence: MD Troutt, College of Business Admini,stration,
ing ways to increase capacity of the bottleneck it is impor-
Management and Information Systems. Graduate School of Management,
Kent State University, P0 Box 5190, Kent, OH 44242-0001, USA. tant to examine these policies and procedures to see whether
E-mail: mtroutt@bsa3.kent.edu they should be changed to fully exploit the bottleneck. For

This content downloaded from


182.255.0.242 on Mon, 04 Oct 2021 14:28:16 UTC
All use subject to https://about.jstor.org/terms
MD Troutt et al-Maximal flow network modelling 183

example, transferring work to the bottleneck in smaller lot 4200 t.p.d. If processes 3 and 4 were operated at full
sizes could improve the flow to the bottleneck, thus insuring capacity they would input 4500 t.p.d. to process 5, there-
that the bottleneck does not lose time waiting for work. Also fore exceeding the capacity of process 5. Hence process 5 is
changing set-up procedures could reduce lost time due to a bottleneck. Processes 3 and 4 must therefore be operated
set-ups at the bottleneck. System flow can often be only at the levels of 2800 t.p.d. and 1400 t.p.d., respec-
increased by subordinating other processes to the bottle- tively. Finally, taking account of the mixing ratios at
neck. This may mean changing the scheduling procedures to processes 1 and 2 shows that they must be operated at
ensure that the bottleneck is never idle.6 the levels of 2100 t.p.d. and 700 t.p.d., respectively, in
order to input 2800 t.p.d. to process 3.
Linear programming and TOC To contrast this approach with that of a network flow
model, define xi as output level in t.p.d. of process i. Since
Connections between Linear Programming (LP) and TOC
the flow of the entire network system is the same as that of
have been previously discussed.9'11 Luebbe and Finch9
process 7, it follows immediately that the objective function
compare the steps of LP to those of TOC. Mabin and
is: maximise X7.
Gibson1" stress how LP and TOC were used together
One set of constraints arises due to the process capac-
effectively in a system with a product mix decision. Here
ities. These are
we consider very general networks of related processes.
Our suggested models are expected to provide additional xl < 4000 (iron ore processing)
flow improvement possibilities for such systems when they x2 < 1000 (coke ovens)
are thought to be essentially optimised with respect to TOC X3 < 3000 (blast furnace)
considerations. In the next section we introduce these X4 < 1500 (scrap handling)
models using a simple textbook example. X5 < 4200 (basic oxygen furnace)
x6 < 6000 (continuous casting)
X7 < 5000 (finishing mill).
A simple network example
The next set of constraints are material balance or
Consider the steel mill16 network flow problem shown conservation
in of flow constraints. These insure that total
Figure 1, which we assume has been optimised with respect
inputs and outputs are the same at each process. A modi-
to TOC considerations. Each numbered box represents an fication of these constraints of reflect inefficiencies or yield
individual process. Process capacities are indicated beneath
losses is considered below. These constraints are
each box in units of tons per day (t.p.d.). Arrows indicate
the directions of material flows. When a process has two or X1 + X2 = X3 (blast furnace)
more inputs the numbers on the arrows indicate required X3 + X4 = X5 (basic oxygen furnace)
mixing proportions. For example, in Figure 1 three units of
X5= X6 (continuous casting)
processed iron ore must be mixed with exactly one unit of
the coke oven output. X6= X7 (finishing mill).
For comparison purposes, we solve for the system The final set of constraints are those which account for
maximal output using a backward pass trial-and-error the necessary mixing ratios. For example, the output level
approach. (A forward pass approach is presented else-
of process 1 must be three times that of process 2. These
where. 16) Working from the end at process 7 we see constraints
that are
neither process 6 nor 7 constraints the capacity of process 5
as a starting trial maximal flow for the system; so that, xl = 3x2 (blast furnace)
following process 4, the maximal attainable flow will be X3 = 2X4 (basic oxygen furnace).

4000 tpd 3/ 3000 tpd E / 4200 tpd 6000 tpd 50 p

2 - Coke 4 - Scrap
ovens handling

1000 tpd 1500 tpd

Figure 1 Steel mill network example.

This content downloaded from


182.255.0.242 on Mon, 04 Oct 2021 14:28:16 UTC
All use subject to https://about.jstor.org/terms
184 Journal of the Operational Research Society Vol. 52, NQ. 2

Addition of non-negativity constraints completes the model.


This model was solved using STORM Personal Version
3.0.19 The optimal objective function value is 4200, which is
the earlier identified system maximal flow. Furthermore, the
constraint associated with the basic oxygen furnace capacity
is the only constraint with zero slack, indicating that
process 5 is the only bottleneck. Sensitivity analysis, used
carefully, provides additional information not easily
obtained via the customary trial-and-error approach. For
this example, the upper range for the basic oxygen furnace
capacity constraint's right-hand side (RHS) is 4500, indicat-
ing that its capacity can be increased by 300 before another
process becomes a bottleneck. Which process? That ques-
tion may be answered by parametric analysis of the RHS
side of the basic oxygen furnace constraint. Such analysis
shows that, when the RHS reaches a value of 4500, the slack Figure 2 More than two inputs.
variables associated with the constraints on capacity for the
blast furnace and scrap handling both leave the basis, Figure 3 depicts a system in which the output of process 3
indicating no slack for those processes. Thus, when the feeds both processes 5 and 6. In these cases outputs are
basic oxygen furnace capacity is increased to 4500, scrap divided and routed so that such networks possess transpor-
handling and the blast furnace also become bottlenecks. tation problem features. Such cases are also easily handled
within the present framework by creation of appropriate
new decision variables. A complete example is discussed in
Multiple inputs and relaxed mixing ratios the next section.

Figure 2 shows a network section in which three processes


feed output to another. Such constraints obviously make
trial-and-error approaches considerably more tedious, butA comprehensive example
present little difficulty for linear programming models. For This section presents the model and its solution for an
example, let xi be defined as before. Figure 2 indicates thatexample based on Figure 3 which includes all varia-
xi must be in the same proportion to x2 as 7 is to 2. tions discussed above. In Figure 3 input capacities are
Similarly, x2 must be to X3 as 2 is to 5. Hence the two given under each process box, along with yield losses as
required constraints are
percentages in parentheses. Flexible mixing ratios are
2x1 -7x2 = 0 and 5x2 - 2X3 = 0. awkward to depict in the figure and are therefore set forth
below following variable definitions.
In many processes there exists a tolerance range for Thus, let xl be the input to process i and x? be the
mixing ratios so that it is not necessary to achieve specified corresponding output. Note that the output from process 3 is
mixing ratios to a precise value. Figure 2 requires that shared between processes 5 and 6. Let xo be that portion
x2/x3 = 0.4. Suppose instead that output levels x2 and X3 used as input for process 5 and xo be that portion used as
are acceptable so long as 0.39 < x2/x3 < 0.42. Then input for process 6. To complete the problem data we
constraint 5x2 - 2X3 0 would be replaced by the two assume the following mixing ratios and ranges:
constraints:
For input to process 5: xO/X2 - 0.3, 0.05 <s x?lxo
x2-0.39x3 0 and x2 -0.42x3-0.
s 0.07
For input to process 7: x?/x = 1.5
For input to process 6: 0.07 < x06/x4 < 0.085.
Accommodating yield loss and output branching
The complete model is therefore
Processes are almost always less than 100% efficient and
the shortfall is called yield loss. Yield loss rates can be
max X0 subject to:
modelled by first interpreting process capacity as input
capacity and then defining input and output variables, xi and
x?, respectively, for each process. For example, if process i
Input xs < 2600 xl < 1400 X3 < 1200
is 98% efficient, then the constraint x? 0. 98x - 0 needs
to be included, along with the constraints which relate x ? 1000
outputs of some processes of inputs of others. Capacities: 4l ?7500 x6?2000 xl?1l0000

This content downloaded from


182.255.0.242 on Mon, 04 Oct 2021 14:28:16 UTC
All use subject to https://about.jstor.org/terms
MD Troutt et al-Maximal flow network modelling 185

Process 1

2600 tpd (2%)

1 400 tpd (3%) / 7500 tpd (4%) / 1 OOQ0 tpd (2%)

Process 3Prcs6

1200 tpd (2%) / 2000 tpd (5%)

Process 4

1000 tpd (3%)

Figure 3 Comprehensive example.

Yield 098X1
leX1O ? 0.97x
. 08I X2- 4
2 098x4
X X 3-
Table I Comprehensive bottleneck example optimal solution
4o
X40.97x4
-. ,4
detailed report
Losses: 4=0.96x4 xb-0.95x4 4=0.98x
Reduced
Number Variable Value Cost cost Status
Material xi= x? + +5
I xi 79.4370 0.0000 0.0000 Basic
Balance: 4x6=45 36 + 4 x = X0 + X60
2 x 267.5198 0.0000 0.0000 Basic
3 x3 1200.0000 0.0000 0.0000 Basic
Branched outputs: X3 -=5 xo+X 4 x 940.8216 0.0000 0.0000 Basic
5 xi 1449.4610 0.0000 0.0000 Basic
Mixing x - 0.3X2 = 0 X( - 0.085X4? < 06 xi 976.4788 0.0000 0.0000 Basic
Ratios: xo-0.07x50 < 4 O Xo -0.07x? > 0
7 xi 2319.1370 0.0000 0.0000 Basic
x? - 0.05x3O5 > 0 x? - 1.54 = 0. 8 x? 77.8483 0.0000 0.0000 Basic

This model was solved subject to non-negativity 9 X2 259.4942 0.0000 0.0000 Basic
constraints using STORM Personal Version 3.019 with 10 the x3Q 1176.0000 0.0000 0.0000 Basic
solution shown in Table 1. The reader is urged to first try xo 912.5970 0.0000 0.0000 Basic
11
13 x? 1391.4820 0.0000 0.0000 Basic
solving this example by an iterative trial-and-error approach
in order to appreciate the advantages of the LP model. 12 xQ 927.6549 0.0000 0.0000 Basic
The optimal objective function value was 2272.7540. To 14 xQ 2272.7540 1.0000 0.0000 Basic
identify bottleneck processes we need to see which of the 15 x?o 1112.1180 0.0000 0.0000 Basic
input capacity constraints are binding. It can be noted that 16 Xo 63.8818 0.0000 0.0000 Basic
only constraint 3 has zero slack. Therefore, process 3 with Slack variables
an input capacity of 1200 t.p.d. is the bottleneck. 17 CONSTR 1 2520.5630 0.0000 0.0000 Basic
18 CONSTR2 1132.4800 0.0000 0.0000 Basic
19 CONSTR 3 0.0000 0.0000 -1.8940 Lower bound
Problems with sensitivity analysis and radial changes 20 CONSTR 4 59.1784 0.0000 0.0000 Basic
21 CONSTR 5 6050.5390 0.0000 0.0000 Basic
One would hope that standard sensitivity analysis might be
22 CONSTR 6 1023.5210 0.0000 0.0000 Basic
applied to answer certain other system flow improvement 23 CONSTR 7 7680.8630 0.0000 0.0000 Basic
questions. However, it is well known that shadow prices 36 CONSTR 20 0.0000 0.0000 -6.4256 Lower bound
37 CONSTR 21 22.2424 0.0000 0.0000 Basic
and ranges are problematical for models with degenerate
38 CONSTR 22 13.6890 0.0000 0.0000 Basic
constraints.20,2 For instance, in the example above, an

This content downloaded from


182.255.0.242 on Mon, 04 Oct 2021 14:28:16 UTC
All use subject to https://about.jstor.org/terms
186 Journal of the Operational Research Society Vol. 52, No. 2

increase of the 4200 t.p.d. to 4500 t.p.d. of the basic Table 2 First changed capacity solution. Constraint 5 has new
RHS of 4500
oxygen furnace (maximal allowable increase) results in a
situation of three bottlenecks as discussed further below. Constraint Shadow RHS Allowable Allowable
Ordinary sensitivity analysis following that solution does number Slack price valtie minimum maximum
not suggest how the capacities of the new multiple bottle-
1 1750.00 0.00 4000.00 2250.00 oo
necks should be increased simultaneously. These limita-
2 250.00 0.00 1000.00 750.00 oo
tions on sensitivity analysis clearly do not reduce the 3 0.00 1.50 3000.00 0.00 3000.00
advantages described above for the network LP approach 4 0.00 0.00 1500.00 1500.00 oo
over the iterative trial-and-error methods. However, it is 5 0.00 0.00 4500.00 4500.00 oo
useful to develop a new sensitivity analysis tool. 6 1500.00 0.00 6000.00 4500.00 oo
7 500.00 0.00 5000.00 4500.00 oo
Consider the general LP model in primal canonical form,
max c'x st. Ax < b, x > 0. Suppose x* is any solution. Now,
consider the effect of doubling the RHS vector to 2b. We call basic oxygen furnace capacity constraint RHS is 4500,
this a radial change of the LP model. It is easy to see that 2x*indicating that its capacity can be increased by 300. This
will be a solution to the radially changed model. Thus the model was solved again after increasing the 4200 capacity
system output could be doubled by simply doubling the to 4500. Salient details of the solution are shown in Table 2.
capacities of all the processes in the network. More generally, Those constraints with zero right-hand side values clearly
if the radial change is to ob for o > 0 then it may be checked do not change with the radial change operation and have
that oxx* is a solution of the radially changed model and that been omitted in Table 2.
the shadow prices are the same as for the original model. Now consider the question of how to increase system
Thus radial changes are a very simple kind of sensitivity flow even further. We see first that, with slacks of zero,
result for simultaneous changes in the RHS vector. processes 3, 4 and 5 have all become bottlenecks. Degen-
The significance of radial change is as follows. Standard eracy, specifically dual degeneracy, is indicated by the zero
sensitivity analysis may show that one or more process shadow prices for the primal binding constraints 4 and 5. In
capacities are at their upper range of feasibility limits and such cases the usual interpretations of shadow prices and
cannot be increased further. But that conclusion is only in ranges may not be accurate. To see this, note that the only
respect to the corresponding LP model. In reality, all such positive shadow price is associated with constraint
bottleneck processes could be increased proportionately, at number 3. However, from its range of feasibility we see
least in principle, by managerial action. that it is at its maximal allowed value. Also the zero
shadow prices for constraints 4 and 5 suggest that system
flow cannot be increased by increasing their capacities.
Illustration Thus, the usual sensitivity analysis erroneously suggests
that the system flow cannot be increased further.
Radial change considerations can aid in postoptimality However, the current solution can be radially changed.
analysis. An example is now shown for the steel mill For example, suppose we increase the current solution so
problem. Let us first number the constraints as follows:
that the 3000 capacity for the blast furnace, process 3,
Constraint no becomes 3333.33 and the optimal system flow becomes
5000.00. This would require a management decision to
1 (iron ore processing) xl < 4000 change the capacities of processes 3, 4 and 5 by replace-
2 (coke ovens) x2 < 1000
ment or other capacity increase options. Table 3 shows the
3 (blast furnace) X3 < 3000
revised figures; it is a radial change of Table 2 in that all
4 (scrap handling) X4 < 1500
5 (basic oxygen furnace) x5 4200 values, except shadow prices, which remain constant under
6 (continuous casting) x6 ? 6000 radial change, have been increased by the factor of 50001
7 (finishing mill) X7 5000
8 (blast furnace) x + x2 -X3 = 0
9 (basic oxygen furnace) X3 + x4-x5 = 0 Table 3 A radially changed solution
10 (continuous casting) x5-x6 = 0
11 (finishing mill) X6-X7 = 0 Constraint Shadow RHS Allowable Allowable
12 (blast furnace) x- 3x2 = 0 number Slack price value minimum maximum
13 (basic oxygen furnace) X3- 2x4 = 0
1 1944.44 0.00 4444.44 2500.00 cc
2 277.78 0.00 1111.11 833.33 oo
Recall that the optimal objective function value is 3 4200,
0.00 1.50 3333.33 0.00 3333.33
which is the maximal flow. Furthermore, the constraint 4 0.00 0.00 1666.67 1666.67 cc
associated with the basic oxygen furnace capacity is the 5 0.00 0.00 5000.00 5000.00 cc
only constraint with zero slack, indicating that process 5 is 6 1666.67 0.00 6666.67 5000.00 cA
7 555.56 0.00 5555.56 5000.00 cc
the bottleneck. For this example, the upper range for the

This content downloaded from


182.255.0.242 on Mon, 04 Oct 2021 14:28:16 UTC
All use subject to https://about.jstor.org/terms
MD Troutt et al-Maximal flow network modelling 187

4500. It should be noted that ranges of feasibility are also 2 Dettmer HW (1997). Goldratt's Theory of Constraints: A
Systems Approach to Continuous Improvement. ASQC Quality
multiplied by this same factor.
Press: Milwaukee, WI, USA.
Radial change will also initially call for increasing the
3 Goldratt EM (1990). The Theory of Constraints. North River
capacities of the slack processes. However, this can be Press: Croton-on-Hudson, NY, USA.
easily corrected. In fact, for this example we see that the 4 Goldratt EM (1990). The Haystack Syndrome: Sifting Informa-
radially increased capacities of the slack processes can be tion out of the Data Ocean. North River Press Publishing
Corporation: Great Barrington, ME, USA.
returned (lowered) to their original Table 2 values without
5 Goldratt EM (1990). What is this Thing Called the Theory of
reducing the increased flow. For example, consider
Constraints and How is it Implemented? North River Press:
process 2 with Table 3 capacity of 1111.11, which was Croton-on-Hudson, NY, USA.
increased from 1000 in Table 2. From the allowable 6 Goldratt EM and Cox J (1992). The Goal: A Process of Ongoing
minimum of 833.33 in Table 3 we see that this capacity Improvement. 2nd rev. edn. North River Press Publishing
Corporation: Croton-on-Hudson, NY, USA.
can be reduced to its original Table 2 value of 1000 and, in
7 Goldratt EM and Fox RE (1986). The Race. North River Press:
fact, could be decreased even further. Similarly, RHS Croton-on-Hudson, NY, USA.
values of constraints 1, 6, and 7 may all be unilaterally 8 Kendall GI (1998). Securing the Future: Strategies for Expo-
returned to their original Table 2 values. In addition, use of nential Growth Using the Theory of Constraints. St. Lucie
the 100% Rule22 shows that the following simultaneous Press/APICS Series on Constraints Management: Boca
Raton, FL, USA.
changes can also be done: 1 and 2, 1 and 6, and 2 and 6. It
9 Luebbe R and Finch B (1992). Theory of constraints and linear
is easy to see that all of these changes can be made
programming: a comparison. Int JProd Res 30: 1471-1478.
simultaneously from Figure 1. 10 Mabin VJ and Balderstone SJ (2000). The World of the Theory
It is worth noting that certain capacity change questions of Constraints: A Review of the International Literature.
can more easily be directly modelled by LP. For example, St. Lucie Press/APICS Series on Constraints Management:
Boca Raton, FL, USA.
suppose we wish to consider what process capacity changes
11 Mabin VJ and Gibson J (1998). Synergies from spreadsheet LP
are needed to increase the system flow by a certain fixed
used with the theory of constraints: a case study. J Opl Res Soc
amount. Let the input capacity of process i be increased by 49: 918-927.
amount yi, 0 < Ya < Mi, at cost cyi, say, where ci is the cost 12 Hammer M and Champy J (1993). Reengineering the Corpora-
per unit of added capacity at that process. Then the LP tion: A Manifesto for Business Revolution. Harper Business:
New York, USA.
model for minimising the cost of modifications to achieve
13 Evans JR (1993). Applied Production and Operations Manage-
the given new flow level can easily be specified along the
ment. 4th edn. West Publishing Company: Minneapolis/
present lines. St. Paul, MN, USA.
14 Meredith JR (1992). The Management of Operations: A
Conceptual Emphasis. 4th edn. John Wiley & Sons: New
Conclusion York, USA.
15 Schmenner RW (1990). Production/Operations Management:
Simple linear programming models are useful for analyzing
Concepts and Situations. 4th edn. Macmillan: New York, USA.
bottleneck situations in production network flow systems. 16 Vonderembse MA and White GP (1996). Operations Manage-
This approach may not have been pursued so far for two ment: Con Methods and Strategies. 3rd edn. West Publishing
reasons. First, textbook-size problems can be solved by Company: Minneapolis/St. Paul, MN, USA.
17 Sahinidis NV and Grossman IE (1992). Reformulation of the
simple trial-and-error techniques. Also such models are
multiperiod MILP model for capacity expansion of chemical
problematical for postoptimality analysis due to the
processes. Oper Res 40: S 127-S 144.
presence of degenerate constraints with multiple bottle- 18 Zeleny M (1981). On the squandering of resources and profits
necks. This paper shows how linear programming network via linear programming. Interfaces 11(5): 101-117.
models are able to handle problems with realistic complex- 19 Emmons H, Flowers AD, Khot CM and Mathur K (1992).
ity. Also the use of radial or proportional change of right- STORM Personal Version 3.0. Prentice Hall: Englewood Cliffs,
NJ, USA.
hand side vectors is introduced as a simple sensitivity
20 Murty KG (1963). Linear Programming. John Wiley & Sons:
analysis tool for this class of models. It provides a way New York, USA.
of handling sensitivity analysis for a degenerate solution, 21 Rubin DS and Wagner HM (1990). Shadow Prices: Tips and
for which the usual approach is not effective. These results Traps for Managers and Instructors. Interfaces 20(4): 150-157.
provide some additional analytical tools for use with the 22 Bradley SP, Hax AC, and Magnanti TL (1977). Applied Math-
ematical Programming, Addison-Wesley Publishing Company:
Theory of Constraints methods.
Reading, MA, USA.

References

1 Cox J and Spencer MS (1998). The Constraints Management


Received April 1999;
Handbook. St. Lucie Press/APICS Series on Constraints
Management: Boca Raton, FL, USA. accepted September 2000 after two revisions

This content downloaded from


182.255.0.242 on Mon, 04 Oct 2021 14:28:16 UTC
All use subject to https://about.jstor.org/terms

You might also like