Network Planning and Analysis

Introduction
Network analysis is a family of related techniques developed to aid management in the
planning, co-ordination and controlling of large complex projects using limited resources like
personnel, material, money, time etc. in order to achieve some objective.
Definition of key terms
Project: It is a human undertaking which consists of a series of interrelated tasks geared toward
a definite objective. It involves a considerable amount of time, human and financial resources for
completion.
Task/activity: This is an individual identifiable activity in a project which has a definite
beginning and end. A task requires time and other resources for its completion. An activity or
task is considered non divisible. An activity is represented in a network by an arrow.
Event: This denotes the beginning or the ending of an activity. It is just a moment in time hence
does not require any resources or time. An event is represented in a network by a circle or node.
Industry context
Business modelling has traditionally been done by operations research (OR) and management
science professionals. It involves the use of quantitative and computer methods for planning the
efficient allocation of resources in business, industry and the agencies of government. It overlaps
economics as a discipline but it primarily focuses on the internal operations of organizations
emphasizes planning rather than market mechanisms for allocating resources makes more
extensive use of quantitative methods.
Operations management often presents complex problems that can be modelled by linear
functions. The mathematical technique of linear programming is instrumental in solving a wide
range of operations management problems.
Linear programming is used to solve problems in many aspects of business administration
including:
•• product mix planning
•• distribution networks
•• truck routing
•• staff scheduling
•• financial portfolios
•• corporate restructuring
The most recognizable feature of PERT is the "PERT Networks", a chart of interconnecting
timelines. PERT is intended for very large-scale, one-time, complex, non-routine projects.
This is a system of interrelationships between jobs and tasks for planning and control of
resources of a project by identifying critical path of the project.
In Network analysis attention should be directed to some activities. These include:
Dummy activity
This is used in a network for
i. improving clarity of the network
ii. To facilitate a logical flow of activities in the network.
A dummy activity consumes no time or resources. Its duration is zero. A dummy activity is
represented by a dotted arrow/line.
275
STUD Y TEX T
Network Planning and Analysis
Activity floats
Floats give a measure of the flexibility which exists for either individual activities or a path of
activities in a network in relation to whether we can extend project completion time as measured
by the critical path. These are the total, free and independent floats.
a) Total float (TFij)
This indicates the amount of time by which a non-critical activity can be delayed without
affecting the project duration dates. It is equal to the difference between the total time allowed on
a performance of an activity and the actual time required for its performance. For critical
activities, the total float is zero. For non-critical activities, the total float can be computed using
the following formula:
TFij = LCj – Esi – Dij
Where
TFij = total float
LCj= latest completion time at the head of the activity
ESi = earliest start time at the tail of the activity
Dij = means the duration of the activity.
b) Free float (FFij)
This indicates how far a non-critical activity can be delayed beyond its earliest start time without
affecting the earliest start time of the activities immediately following it. Critical activities have
zero free float. For non-critical activities, free float can be determined as follows:
FFij = ESj – Esi – Dij
Where
FFij = free float
ES j = earliest start time at the head of the activity.
ES i = earliest start time at the tail of the activity.
Dij = duration of the activity.
c) Independent float (IF)
This is the amount of time an activity can be delayed without delaying the project completion
time if all preceding activities are completed as early as possible and subsequent activities are
started as soon as possible.
It is given as follows:
IFij = ESj –LCi – Dij
Where
IFij = independent float
ES j = earliest start time at the head of the activity.
LCi = latest completion time at the tail of the activity
Dij = duration of the activity
2 7 6 quantitative techniques
Event slack
An event slack is the maximum time an event can be delayed without delaying the overall project
completion time. For an event i ,it given as LCi – Esi
Note:
Total float ≥ free float ≥ independent float
When asked to calculate float or slack without specifying, take that to mean the total float.
Role of network analysis
Network analysis in project management will help:
Define the job to be done, breaking it into individual activities;
Integrate them in a logical time sequence;
Afford a system of dynamic control over the progress of the plan

How would you use network analysis to solve a management problem?

Phase I: Formulation
Identify the project
Determine and list all the activities in the project.
Estimate the duration and cost of all activities listed above listing all resources required.
Establish a logical flow of all the activities – first, second, and so on.
Phase II: Scheduling
Draw the network diagram of the project indicating various events and their duration.
Determine the earliest and the latest occurrence of each event.
Calculate activity floats.
Determine the critical path for network – the critical path is defined as an unbroken continuity
of activities sequence from start to finish which is the longest path (and) which represents the
minimum time the project can be completed. A delay of an activity on the critical path inevitably
delays the whole project.
Note: There can be more than one critical path in one network.
Phase III: Implementation and control
i) Analyze the effects of delays in the project time, reduction of costs (activity crashing).
ii) Check off the progress of implementation with the plans.
iii) Reassign or reschedule the resources that are used by the project as appropriate.
iv) Revise the network and set a new schedule where necessary.
There are two approaches to pictorially represent a network:
Activity on arrow approach
Activity on node approach
Rules to construct network diagrams
1. As a rule, no two activities can begin and end at the same event node as they would not be
uniquely identified. (Use a dummy activity to ensure uniqueness).

A complete network can have only one beginning event node and one ending event node.
Diagram 8.2
2. E very activity must have one preceding or “tail” event and one succeeding or “head” event.
Note that many activities may use the same tail event and many may use the head event.
Diagram 8.3
3. “Loops” are a series of activities which lead back to the same event and are NOT allowed
because a network is a progression of activities always moving onwards in time.
A
B
Dummy
Start Stop
Tail
Event
Haed
Event
2 7 8 quantitative techniques
STUD Y TEX T
Diagram 8.4
4. A ll activities must be tied into the network i.e. they must contribute to the progression or
be discarded as irrelevant. Activities which do not link into the overall project are termed
“danglers”. Danglers should not be used.
Diagram 8.5
5. N etworks proceed from left to right.
6. N etworks are not drawn to scale
7. U se straight lines, not bent or curved ones.
8. T he arrows should not cross each other unless it is completely necessary.
9. T he length of the arrow is not proportional to its duration.
10. When the drawing is complete, use code number to number the events. The code number at
the beginning of an activity must be smaller than the code number at the end of an activity.
These code numbers ensure that each activity can be uniquely identified in the diagram
Two techniques were developed almost simultaneously in the period 1956-1958 by different
firms for different reasons and independently:
(a) T he critical path method (CPM)
(b) T he programme evaluation and review technique (PERT )
The main difference is that CPM is a certainty (deterministic) model while PERT is a stochastic
(uncertainty) model with respect to the project completion time.
Other initial differences:
Dangler
279
STUD Y TEX T
Network Planning and Analysis
i) Mechanic of drawing network
PERT used AOA approach where AOA = Activity-on-Arrow approach.
Diagram 8.6
CPM used AON approach where AON = Activity-on–Node approach.
Diagram 8.7
ii) H ow to estimate activity time
CPM used a single time estimate therefore it was deterministic in the estimate.
PERT on the other hand used multiple time estimates allowing for any doubt to be included in
the
estimates therefore probabilistic.
These estimates were reduced to three for examination purposes.
a) Optimistic time estimate = a
This is the shortest time an activity can take to be complete. It represents uder real estimate such
that the probability is small that the activity can be completed in less time.
b) Pessimistic time estimate = b
This is the longest time an activity can take to be completed. It is the worst time estimate
representing bad luck, such that the probability is small that the activity will take longer.
Most likely time estimate = m
This refers to the time that would be expected if you work under normal conditions. Used to give
the activity expected (mean) time using the following formula:
Expected activity time, te =
6
a + 4m + b
Critical activity
1. I t must be critical for the completion of the project.
2. I t must be started and completed as originally scheduled otherwise the project will be late.
3. It must have the first priority in the use of resources.
4. R equires keen management attention and control.
ABC
ABC
2 8 0 quantitative techniques
STUD Y TEX T
Three conditions to be satisfied simultaneously
1. ES i = LCi
2. ES j = LCj
3. ES j – Esi = Dij or LCj – Lci = Dij
Critical path calculations
Fast Forward: Today, it is commonly used with all forms of projects, including construction,
software development, research projects, product development, engineering, and plant
maintenance, among others.
Two phases are computed:
a) Phase I is the Forward Pass computations. It is used to determine the ESi for each
activity.
b) Phase II is the Backward pass computations. It used to determine the LCj for each
activity.
Phase I – Forward Pass = ESi
Computation begins from the initial event. Move forward until you reach the terminal event.
The initial event time is zero which is base time.
Earliest start time of the activity beginning from the immediate next event which is base time
plus
the duration of the activity.
If there is more than one activity going into an event, for the ESi from that event, take the
maximum
allowable time using the following formual:-
ES j = max ESi + Dij for all (ij) acitivities.
Phase II – Backward pass = LCj
Start from the terminal event and move backwards to reach the initial event.
Terminal event LCij= ESi
For the next event it is the difference between the maximum allowable time and the duration of
the activity.
Take minimum allowable time, LCi = min LCj – Dij for (ij) all activities
281
STUD Y TEX T
Network Planning and Analysis
Crashing the project
Crashing means trying to perform an activity in a time shorter than the activity required i.e.
putting in extra effort to deliberately reduce the activity time. Consequently, the duration of some
activities can be reduced if some additional resources are employed but by introducing such
additional resources the cost of performing such activities increases and the normal time by
which an activity is required to be reduced the greater the amount of resources required to be
employed.
Normal duration (Dn)
This is the time required to perform an activity under normal circumstances and with minimum
direct costs.
Crash duration (Dc)
This is the time taken to perform an activity if the duration is reduced or shortened.
Normal costs (Cn)
This is the absolute minimum direct costs required to perform an activity within the normal
duration.
Crash costs (Cc)
This is the cost incurred to achieve the reduced performance time. It is generally more than the
normal costs because of introduction of additional resources.
Incremental costs (Ic)
This is the increase in cost incurred per unit of time that is used. It is determined by the following
formula:
Incremental costs (Ic) = crashcost − normalcost = Δ cost
Normalduration − Crashduration Δ duration
Guideline for crashing
Identify the critical activities and the critical path(s)
Crash the durations of the critical activities only because they are the ones that determine the
project duration.
Determine the incremental cost of all the critical activities.
If there are several critical activities to be crashed select the critical activity with the least
incremental cost to be crashed first.
For each crashing step a unit of time is the maximum amount that can be used because there is
a possibility of new critical activities and critical path emerging.
Check to be sure that the critical path being crashed is still critical. This is because a reduction
in activity time along the critical path may cause non-critical activities and paths to become
critical.
If a project has more than one critical path then all critical paths must be reduced simultaneously
by an equal amount.
2 8 2 quantitative techniques
STUD Y TEX T
When to stop crashing
For technical reasons the duration of an activity cannot be reduced indefinitely. If crash time
represents the minimum duration that is allowable consequently stop crashing activity time or
duration when:
The management wish has been achieved.
When all crashable activities have reached their climax crash limits.
When it is no longer economical to continue crashing.
Resource scheduling and profiling
So far we have assumed that an activity can begin as soon as all the activities which must
precede it have been completed. This assumes that there are sufficient resources to perform all
work defined by the activity even if concurrently. Such an assumption may not always be true in
practice. Further, resource utilisation needs to be planned so that it is as steady (smooth/even)
as possible.
Planning resource requirements is done through the use of a time (Gantt) chart levelling of
resources usage which will lead to the construction of a resource profile. A resource profile is
used as a calendar of time schedule by the personnel who will implement the project.
We perform resource scheduling and profiling because:
We accept that concurrent activities may not be possible due to resource constraints.
For other resources such as manpower, the application should be planned for in advance to
minimize losses due to shortages at certain time or idleness at other times.
Smoothing a profile
This is the process of attempting to reduce the peaks and troughs in the resource allocation
so that we have a more even usage of personnel. Smoothing requires that we make use of
activity floats since we generally cannot extend completion time. We cannot reschedule critical
activities. We can normally reschedule only non-critical activities because they have floats. Such
rescheduling should be for as little as possible since it is prudent to delay an activity for as short
a time as possible. Smoothing is conducted on the basis of trial and error because there is no
analytical method for it.
283
STUD Y TEX T
Network Planning and Analysis
Illustration
Consider the following project network and activity times (in weeks).
Activity Activity Times
(weeks)
A
B
C
D
E
F
G
H
5
3
7
6
7
3
10
8
Required:
(i) How long will it take to finish this project?
(ii) C an activity D be delayed without delaying the entire project? If so, by how many
weeks?
(iii) What is the schedule for activity E?
Solution
i) We can determine the project duration by drawing a schedule to ascertain the critical
activities.
C
C
D
BEG
F Finish
Start
H
2 8 4 quantitative techniques
STUD Y TEX T
Activity Duration ES EF LS LF Slack Critical
Activity
A
B
C
D
E
F
G
H
5
3
7
6
7
3
10
8
0
0
5
5
3
11
11
14
5
3
12
11
10
14
21
22
0
1
7
5
4
11
12
14
5
4
14
11
11
14
22
22
0
3
7
0
7
0
11
0
Yes
No
No
Yes
Yes
Yes
No
Yes
The critical path as identified using the float method (see table) is ADFH.
It thus takes 22 days to complete the project.
ii) A ctivity D can’t be delayed without delaying the entire project because it is a critical
activity.
iii) T he schedule for activity E is as below:
EST 3rd week
LST 4th week
EFT 10th week
LFT 11th week.
N ote:
T he PERT diagram, can also be drawn as below:
5 14
3 14
11 F
11
11
10 10 21
21
10
F
G
D
5A
0
0
3
37
4E
B
14
H
12 8
22
22
22
7
C
5
285
STUD Y TEX T
Network Planning and Analysis
T he critical path is A – G – H – J.
Clearly, the project cannot be completed within the remaining time of 1 week. There is need to
increase the duration of critical path by ½ week.
The assumption of PERT as used in network analysis is that the activity time duration can be
best
described by a beta probability distribution.
Illustration
Lilian Wambugu is the project manager of Jokete Construction Company. The company is
bidding
on a contract to install telephone lines in a small town. It has identified the following activities
along with their predecessor restrictions, expected times and worker requirements.
Activity Predecessors Duration Weeks Crew Size Workers
A-44
B-72
CA32
DA34
EB26
FB23
G D, E 2 3
H F, G 3 4
Lilian Wambugu has agreed with the client that the project should be completed in the
shortest duration.
Required:
(i) Draw a network for the project.
(ii) Determine the critical path and the shortest project duration.
(iii) Lilian Wambugu will assign a fixed number of workers to the project for its entire duration
and so she would like to ensure that the minimum number of workers is assigned and
that the project will be completed in 14 weeks.
Draw a schedule showing how the project will be completed in 14 weeks.
4
8
8B
A
00
3
9 9 12 12
G3
3
ED
6
8 25
5J
17 17 25
43
13
2 8 6 quantitative techniques
STUD Y TEX T
(iv) C omment on the schedule you have drawn in (iii) above.
Solution
(ii) C ritical path is B – E – G - H
S hortest project duration = 14 weeks.
(iii) Time float : TF = LFT – EST – D
TF (A) = 6 – 0 – 4 = 2
TF (B) = 7 – 0 – 7 = 0
TF (C) = 14 – 4 – 3 = 7
TF (D) = 9 – 4 – 3 = 2
TF (E) = 0
TF (F) = 11 – 7 – 2 = 2
TF (G) = 0
TF (H) = 0
H4
G3
F33
E6
D4
C22
B2
A4
Time (wk) 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Workers 6 6 6 6 8 8 8 9 9 3 3 4 4 4
Reschedule
F by 2 wks
66668886666444
Reschedule
C by 7 wks
66666666666666
(iv) C omment
Minimum number of workers needed throughout the duration is 6 workers.
C and F are pushed to the end of their slack times to ensure that there is a smooth engagement
of the resources.
Illustration
James Mutiso is a computer engineer in an information technology firm. The firm has decided to
install a new computer system to be used by the firm’s help desk. James Mutiso has identified
the activities required to complete the installation.
The table below provides a summary of the activities’ durations and the required number of
technicians:
287
STUD Y TEX T
Network Planning and Analysis
Activity Duration (weeks) Required number of technicians
1–232
1–314
2–434
2–522
3–424
3–644
4–522
5–622
6–722
Required:
(i) Draw a gantt chart for the project.
(ii) Mr. Mutiso would like to reschedule the activities so that not more than 6 technicians
are required each week.
Determine if this is possible and how it can be achieved by rescheduling the activities.
Solution
12
7 12
I
442
3
B
0
03
3
3
CG
H
2
2
2
3
8
8
5
D
2
A
1
12F
1
E
10
10
6
2 8 8 quantitative techniques
STUD Y TEX T
Critical path = A – C – G – H - I
To have at most 6 technicians Mr. Mutiso should reschedule activity F to start on the 6th week
i.e. delay it by 5 weeks.
8.2 Transportation and Assignment Models
The linear programming methods discussed in Chapter Seven are suitable for a range of
allocation
problems. The work involved in solving the model can be drastically reduced by use of a
computer
package. This leaves the decision maker free to concentrate on the interpretation and evaluation
of final solution. However, the package still requires the formulation of the linear programming
model. This can be a major task for large problems. The variables must be identified and the
constraints formulated.
For certain types of allocation problems, the use of specially designed algorithm simplifies the
building of the initial model. In this chapter, we will look at two related examples of this type of
algorithm which are suitable for solving transportation and assignment problems.
In both cases the allocation concerns items which are transferred from a number of origins
to a set of destinations according to a particular objective. The objective is often one which
minimises the total cost of the transfer. For example, a company has three factories and five
6
6
1
6
2
6
4
4
2
2
F
E
D
D
Personnel 2
Activities
Critical activities ACGHI
Now critical activities BDEF
4
CGHI
222
T/F
D
E
F
G
H
I
T/F
3
3
5
0
0
0
3
6
4
10 10 10 10
6
2
56
4
2
7
4
6
2
8
6
2
9
6
2
10
6
2 Duration (wks)
rescheduled
Unbalanced resource profile
11 12
289
STUD Y TEX T
Network Planning and Analysis
regional distribution centres. The management requires the transfer of the finished goods from
the factories to the distribution centres to be achieved at minimum cost. This is a situation in
which the transportation method would be appropriate.
Assignment is a particular case of the transportation problem. For each combination of origin
and destination, the transfer involves one item only. For example, a machine shop has six lathes
of varying ages and design. On a given morning, the machine shop manager has six jobs to
allocate. The jobs will take different lengths of time on different lathes. The manager wishes to
allocate one job to each lathe to minimise the total working time. The assignment algorithm may
be used to solve problems such as this.
In this chapter we will describe the application of these two algorithms using small problems. It
should be borne in mind that, in practice, the problems will he much larger and are solved using
computer packages. In addition, transportation models often involve several stages, for example,
factory-to depot-to-retail outlet. In these cases the basic algorithm has to be modified and more
sophisticated methods used.
Transportation Problem and Algorithm
This problem is concerned with the allocation of items between suppliers (called origins) and
consumers (called destinations) so that the total cost of the allocation is minimised. The problem
can he solved using either linear programming methods or the special transportation algorithm.
The linear programming method is illustrated in Example 1.
The transportation problem
Example 1: To Illustrate a basic transportation problem
Ace Foods Ltd manufacture soft drinks at two plants, A and B. Bottles for the two plants are
supplied by two firms, P and Q. For the month of November, plant A requires 5,000 bottles and
plant B requires 3,500 bottles. Firm P is able to supply a maximum of 7,500 bottles and firm Q
is able to supply a maximum of 4,000 bottles. The cost per bottle of transport between each
supplier and each plant is shown in the
Table 1: costs, requirements and availabilities of bottle supply
T ransport cost, pence/bottle to plant available bottles
A B
From supplier P 4 4 7500
Q 3 2 4000
Bottles required 5000 3500
How should the bottles be supplied to the plants to minimise the total transport cost?
2 9 0 quantitative techniques
STUD Y TEX T
Solution
It is always useful with transportation to see if there is an obvious solution. Ideally, we would
like to use only the cheapest routes. Supplier Q will be preferred by both plants, since it is
cheaper
than P. Unfortunately Q has only 4,000 bottles, compared with the total requirement of 8,500.
The cheapest solution will probably be to use the 2p per bottle route from Q to plant B, to supply
all the requirement at B (3,500). The balance from Q (500) should be sent to A at 3p per bottle.
The rest of the demand at plant A will come from P at a cost of 4p per bottle. The total cost of
this
allocation is:
0.02 x 3500 + 0.03 x 500 + 0.04 x 4500 = £265/month
We have no proof that this is the most economic allocation. One of the important aspects of
the model which we are about to consider, is that it provides a solution, demonstrates that it is
the optimum solution and shows the effect on the solution of any changes which arise in the
problem.
We will now solve the above problem using a conventional linear programming model with a
graphical solution.
Suppose firm P supplies x bottles to plant A and y bottles to plant B. Firm Q must supply the
remaining (5000 — x) bottles to A and the remaining (3500 — y) bottles to B. The objective is
to
minimise the total transport cost, C, pence, where:
C = 4x + 4y + 3(5000 — x) + 2(3500 — y)
Therefore:
C = x + 2y + 22,000
and:
Z = C - 22,000 = x + 2y
Z will take its minimum value, when C takes its minimum value. The values of x and y which
minimise Z = x + 2y, will also minimise C. The objective function is minimised subject to:
Requirement at A: x <5000 bottles
Requirement at B: y <3500 bottles
Supply from P: x + y <7500 bottles
Supply from Q: (5000 — x) + (3500y) <4000 bottles
i.e.:
x + y >4500 bottles
x, y> 0
291
STUD Y TEX T
Network Planning and Analysis
These constraints are illustrated in Figure 1.
The point x = 4000, y = 2000 is a feasible solution. At this point:
Z = 4000 + 2 x 2000 = 8000 p
The trial objective function is 8000=x+2y. This is shown in figure.1. Moving in the direction of
decreasing values of Z, corner A is the optimum. At this corner, x = 4500 and y = 0. Therefore
the
optimum solution is for P to supply 4500 bottles to A and none to B, while 0 supplies 500 bottles
to A and 3500 bottles B. At this solution, the minimum cost is:
Cmin = 4500 + 2 x 0 + 22,000 26,500 p = £265
The only spare capacity is at firm P which retains 3,000 bottles. This is the solution that we
thought would be the minimum. We have now shown that this is the case.
The transportation algorithm
The problem in section1 may be solved using the transportation a1gorithm. To use this
algorithm,
a number of conditions must be satisfied:
1. The cost per item for each combination of origin and destination must be specified.
2. T he supply of items at each origin must be known.
3. T he requirement of items at each destination must be known.
4. T he total supply must equal the total demand.
Example.1 satisfies the first three conditions but not the last one. However, a dummy plant can
be included for which the requirement is the difference between the total available and the total
required. In Example 1 the dummy plant would have a requirement for (11,500 - 8500) =3,000
bottles.
Any items allocated to a dummy destination represent items which do not leave the supplier. In
a similar way, if the total supply is less than the total demand, a dummy supplier is included to
supply the shortfall. Any items allocated from this dummy represent items not supplied.
x=5000
x + y=7500
000
No. of
bottles
supplied
from P
to B
y
8
6
4
2
0
012468
Typical objective function
x+2y=8000
y=3500
A No. of bottles supplied from P to A
x + y=4500
2 9 2 quantitative techniques
STUD Y TEX T
The transportation algorithm has four stages:
Stage 1: Arrange the data in tableau format and find any feasible allocation. A feasible allocation
is
one in which all demand at the destinations is satisfied and all supply at the origins is allocated.
Stage 2: Test the allocation to see if it is the optimal solution.
Stage 3: If the first allocation is not optimal, re-allocate in order to move to a better, lower cost,
solution.
Stage 4: Test again for optimality.
Repeat this iterative process until the optimum allocation is found.
Finding an initial allocation
The initial allocation can be made using any method which will produce a feasible solution.
However, a systematic approach tends to produce more useful solutions. We will look at two
methods of finding an initial allocation, the minimum cost method and Vogel’s method. The
procedure is explained in Example 2.
Example 2: To Illustrate methods of finding an Initial allocation
Three warehouses P, Q and R can supply 9, 4 and 8 items respectively. Three stores at A, B and
C require 3, 5 and 6 items respectively. What is the minimum cost of allocating the items from
the
warehouses to the stores if the unit transportation costs are as shown in table below?
Table.2 Costs, requirements and availabilities for Example 2
Transport costs £/item, to
stores
A BC
Total available
From warehouses P
Q
R
10 20 5 9
4
8
2 10 8
1 20 7
Total required 3 5 6
Solution
The information on costs, availability and requirements are given but total supply is bigger than
demand. The warehouses have 21 items available but the stores require only 14, a dummy store
is needed to absorb the 7 items which are surplus to requirements. These 7 items will never
actually leave the warehouse, and therefore their transportation costs are assumed to be zero.
The first transportation tableau is given below:
293
STUD Y TEX T
Network Planning and Analysis
Table.3 Balanced transportation tableau
Transport costs £/item, to
stores
A B C dummy
Total available
From warehouses P
Q
R
10 20 5 0 9
4
8
2 10 8 0
1 20 7 0
Total required 3 5 6 21
To find the first feasible allocation we will use the minimum cost method and Vogel’s method in
turn. However, it should be remembered that only one method is actually required.
Method 1: Minimum Cost Method
1. We allocate as much as possible to the cell with the minimum unit cost.
2. We adjust the remaining availabilities and requirements.
3. We choose the next smallest cost and allocate as much as possible to this cell, and so
on, until supply and demand are all zero.
4. I f more than one cell has the smallest value of unit cost, then we choose at random.
In table.4, the transport costs are placed in the separate boxes in the upper right of each cell. The
subscripts indicate the order in which the allocations are made and should help you to follow the
explanation. The dash in a cell indicates that this cell is no longer available.
Table 4 First allocation using minimum cost method
To store
A B C Dummy
Total available
P
From
warehouse Q
R
10 20 5 0
920
- - 23 71
2 10 8 0
40
- 45 - -
1 20 7 0
8510
32 16 44 -
Total required
3
0
5
1
0
6
4
0
7
0
21
2 9 4 quantitative techniques
STUD Y TEX T
Key:
The smallest cost is zero. We choose any one of cells (P, Dummy), (0, Dummy) or (R, Dummy).
Cell (P, Dummy) is chosen and we allocate the maximum amount, 7 units, to the cell. We reduce
by 7 the amount available at P and the amount required by the dummy. Cross out the cells which
now cannot be used, that is, (0, Dummy) and (R, Dummy).
Neither of the other zero costs are available, so the next smallest cost is 1 in cell (RA). We
allocate as many units as possible, 3, to this cell. Adjust the row and column totals and cross out
cells which are no longer available, that is (P, A) and (Q, A).
The smallest cost, still available, is 5 in cell (P, C). The remaining 2 items at P are allocated to
this
cell. The row and column totals are adjusted and we cross out the remaining cell in row P.
The final allocations are, in order, (R, C), (Q, B) and (RB).
If the allocation is feasible, the total available at each warehouse and the total required at each
store should now be zero. The above allocation is feasible.
Cost £ ((3 x 1) + (4 x 10) + (1 x 20) + (2 x 5) + (4 x 7) + (7 x 0) = £101
We do not know yet whether this allocation is the cheapest but it should give a reasonable cost.
Method 2: Vogel’s Method
This method uses penalty costs. For each row and column the penalty cost is the difference
between
the cheapest available route and the next cheapest. We try to minimise these penalties.
To calculate the penalty cost for each row and column, we look at the least cost cell and next
smallest cost cell. For each row and column subtract the smallest cost from the next smallest.
This gives the penalty cost of not allocating into the cell with the cheapest cost.
We choose the row or column with the largest penalty cost and allocate as much as possible
to the cell with the smallest cost in that row or column. In this way, the high penalty costs are
avoided as far as possible.
As with the previous method, we adjust the row and column totals.
We cross out the remaining cells in any row or column for which the supply or demand is now
zero, since these cells are no longer available.
Return to 1 and re-calculate the penalty costs, ignoring the cells which have been used or crossed
out.
We repeat these steps until all demand is satisfied. The subscripts in table.5 show the order of
choosing the penalty costs and making the allocations.
Unit transport cost
Units
allocated
295
STUD Y TEX T
Network Planning and Analysis
Table 5 First allocation using Vogel’s method
To store
A B C Dummy
Total available
Penalty
123
P
From
warehouse
Q
R
10 20 5 0
9,8,2,0 5 5 5
-162
2 10 8 0
4,0 2 - -
- 41 - -
1 20 7 0
8,5,0 1 1 73
32 - - 53
Total required
3
0
5
1,0
6
0
7
2,0
21
1st penalty
2nd penalty
3rd penalty
1
92
-
101
0
0
2
2
2
0
0
0
After the third allocation, there is only one way of completing the solution. The remaining items
are allocated as follows—(P, B), (P, C) and (P, Dummy).
Cost = (1 x20+6x5+2x0+4x 10+3x 1+5x0)’93
Again, we do not yet know whether this is optimal but we do know that it is a cheaper allocation
than the £101 for the minimum cost method.
Testing for optimality
To test for optimality, we must first determine whether the initial allocation is basic, that is,
whether
it is a solution at a corner of the feasible region. The tableau shown in table 4 gives a feasible
solution, that is, a solution inside or on the edge of the feasible region. If the allocation is basic,
there should be one basic variable for every constraint. In a problem with m warehouses and n
stores (including the dummy), there are (m + n - 1) independent constraints. A basic solution will
therefore have (m + n - 1) allocated cells. These (m + n - 1) variables must be in independent
positions. It is not necessary to worry about independence at this stage since any problems will
emerge during the test for optimality.
If the allocation has (m + n - I) independent variables, the methods for testing for optimality may
be applied directly. If there are fewer variables, the tests have to be modified, as will be
illustrated
in table 6. However, if there are more than (m + n - 1) variables, then the allocation procedure
has
been used incorrectly. It should be possible to modify the allocation to give a cheaper cost with
the correct number of variables.
2 9 6 quantitative techniques
STUD Y TEX T
Refer to Example.2. We will test each of the allocations for basicness. The tableau has 3 rows
and 4 columns, therefore a basic solution will have (3 + 4 -1) = 6 allocated cells. We can see that
this is the case for both allocation methods. The two methods have given solutions at different
corners of the feasible region. The testing procedures may be used without modification.
The initial allocation is tested to determine whether it is the cheapest solution and, if it is not, it
should be changed. We will illustrate two methods of testing for optimality.
In the stepping stone method, the costs of using the unallocated cells - the shadow costs - are
calculated. The procedure is long and rather clumsy but the physical meaning is clear.
The MODI (modified distribution) method is a mathematical procedure which gives the same
shadow costs much more quickly, although the physical meaning is not so obvious.
In both methods, if the allocation is not optimal, a stepping stone procedure is used to move to
the next basic allocation. Once a basic solution has been found, the algorithm enables us to
move from corner to corner of the feasible region, until the optimum solution is found.
Example.3: To illustrate the test for optimality using the stepping stone method
We will use the allocation produced by the minimum cost method to illustrate the procedure. The
allocation is repeated in table 6 below
Table 6 First allocation using minimum cost method
To store
A B C Dummy
Total available
P
From warehouse
Q
R
10 20 5 0 9 - - 2 7
2 10 8 0 4 - 4 - -
1 20 7 0
314-8
Total required 3 5 6 7 21
Key:
The stepping stones are the cells which have allocations in them - (P, C), (P, dummy), (0 B), (A,
A), (R, B) and (A, C). We take one of the empty cells and pretend that we move one item into
it. This move upsets the totals for the row or column in which the empty cell lies. The amounts
in some of the allocated cells are then adjusted to restore the balance. We use these allocated
cells, the stepping stones, to calculate the cost of the transfer of this one item into the empty cell.
If the cost is positive, using the empty cell will increase total costs and we do not want to do this.
If the cost is negative, using the cell will reduce costs. This means that the present allocation is
not optimal and we can find a better solution using this cell.
P
R
P
R
Unit transport cost
Units
allocated
297
STUD Y TEX T
Network Planning and Analysis
It does not matter which empty cell is chosen as the start. We will choose (P, A). We add 1 item
to (P, A). The allocation is no longer correct. Store A is receiving 4 items, when it wants only 3.
Warehouse P is supplying 10 when it has only 9 items. We must adjust the A column and the
P row. To balance the A column, we must deduct 1 item from the stepping stone (R, A). This
corrects column A, but unbalances row R, reducing its supply from 8 to 7.
We can re-balance row P by subtracting 1 item from either (P, C) or (P, Dummy). If we choose
(P,
Dummy), there is no other allocated cell in the dummy column which could be used to re-adjust
the dummy column, therefore we do not make this choice. Adjustments can be made using only
those cells which have allocations already. We must use (P, C). We deduct 1 item from (P, C).
This corrects row P, but unbalances column C. We now have problems with row A and column
C. Both can be adjusted simultaneously by adding 1 item into (R, C). The physical effect of
using
the empty cell (P, A) and returning to a balanced allocation is shown in table 7 below.
The net cost effect of moving 1 item into (P, A) is:
+1 x (P, A) cost—1 x (R, A) cost +1 x (R, C) cost—i x (P, C) cost
= + (1 x 10) – (1 x 1) + (1 x 7) – (1 x 5) = +£11/item
Table 7 Testing empty cell (P, A)
Physical change- items
Using (P, A) would cost an extra £11 for each item sent from P to A. The shadow price is
positive
therefore we do not choose to use this empty cell.
We return to the original allocation and repeat the procedure for the other empty cells in turn.
Look next at the cell (R, Dummy) and use the stepping stones (P, Dummy), (P. C), and (A, C) to
show the physical and cost changes of moving 1 unit into (P, Dummy):
Table 8 Testing empty cell (R, Dummy) T able9 testing empty cell (R,dummy)
Physical change—Items
cost change, £
P
R
Test cell
+1
Allocated cell
-1
Allocated
cell
-1
Allocated
cell
+1
A C
Test cell
+1
Allocated cell
-1
Test cell
+1
Allocated cell
-1
Allocated
cell
-1
Allocated
cell
+1
Allocated
cell
+1
Allocated
cell
+1
P
R
2 9 8 quantitative techniques
STUD Y TEX T
The net cost change of adding 1 item to (R, Dummy) is:
+0—0 + 5—7 = —2 per item
By allocating into cell (R, Dummy), it is possible to reduce the costs, therefore, the present
allocation is not optimal. We can find a cheaper allocation, saving £2 per item, by using (R,
Dummy) and this stepping stone route. We must, however, complete the testing of all the empty
cells, since there may be a cell which gives an even better saving.
Let us next construct the stepping stone path for the empty cell (0, Dummy). We must remember
that to re-balance rows and columns we can step on allocated cells only. A four step circuit is not
possible this time. We must look for a more complex route. We allocate 1 item to cell (0,
Dummy).
There is only 1 allocated cell in row Q and only 1 allocated cell in the Dummy column. Suppose
we choose to move from (Q, Dummy) to (Q, B). We deduct 1 item from this cell which balances
row Q. Column B can be balanced by (R, B) only, therefore we add 1 item to this cell. We can
balance row R via (R, A) or (R, C) but (R, A) is the only allocation in column A, so we do not
use
this cell. If we did, we would not be able to balance column A. We deduct 1 item from (R, C).
The
route home should now be clear. We balance column C by adding 1 item to (P, C) and balance
row P by deducting one item from (P, Dummy). This last move also balances the Dummy
column
and the circuit is complete. We should remember that as long as the initial allocation is basic, it
is
possible to find a suitable route round the tableau which starts and ends with the chosen empty
cell. The physical effects and cost changes are summarised in Tables.10 and 11.
Table 10 testing empty cell (Q, dummy)
Physical change - items
BC Dummy
Empty
Allocated
+1
allocated
-1
Allocated
-1
empty
test
+1
Allocated
+1 Allocated -1 Empty
Table 11 testing empty cell (Q, dummy)
Cost change £
B C Dummy
Empty Allocated +5 allocated
-0
Allocated
-10
empty
Test +0
Allocated
+20
Allocated -7 empty
299
STUD Y TEX T
Network Planning and Analysis
The net cost affect of adding 1 item to the empty cell (Q, Dummy) is:
+0—0+5--7+20—10= +8/item
The total cost of allocation will increase by £8 per item if we allocate into this empty cell. We do
not choose to make this change. The shadow costs for the remaining empty cells are calculated
in a similar way to that described above. The full set of values is shown in table 12:
Table 12
To store Total
available
P
From warehouse
Q
R
10 20 5 0
9
+11 +2 2 7
2 10 8 0
4
+11 4 +11 +8
1 20 7 0
8
3 1 4 -2
Total required 3 5 6 7 21
Key:
This is not the optimal allocation because cell (R, Dummy) has a negative shadow cost of -.E2.
The present cost of the allocation, £101, can be reduced by using this cell and the stepping stone
circuit which gave the net saving of £2 per item.
We will continue with this example, to find the optimum allocation, in section.5, but first we will
discuss the MOD1 method of calculating the shadow costs. The stepping stone procedure is
a clumsy method and it is easy to make errors. It is more sensible to use the mathematical
approach of the MODI method to test for optimality. The procedure does not give an insight into
the physical problem but it does produce the same shadow prices with much less effort.
Shadow cost
of empty cost
Unit transport
cost
Units
allocated
m
3 0 0 quantitative techniques
STUD Y TEX T
To begin with, consider only the allocated cells. Each unit cost, Cjj, for these cells is split into
two components, ui for the row and vj for the column. For example, cell (R,B), which is in row 3
and column 2, has a unit cost c32 = £20. This is split into the row component u3 and the column
component v2, i.e.
c32 = 20 = U3 + v2
For each empty (non-basic) cell, we calculate the shadow cost from:
S ij = cij – (ui + vj)
This shadow cost is the extra cost of transporting one item by the route i to j. If all of the shadow
costs are positive or zero, that is, Sij > 0, then the solution is optimal. In this case, if an allocation
is moved into an empty cell for which the shadow cost is positive, the total costs will increase, if
the shadow cost is zero, the total costs remains unchanged.
Example.4: To test a basic allocation (or optimality using the MODI method
Refer again to the initial allocation obtained using the minimum cost method. We will test this
allocation for optimality using the MODI method. The initial allocation is repeated below:
Table 13 First allocation using minimum cost method
P
From warehouse
Q
R
To store
A B C Dummy
Total
available
10 20 5 0
9
--27
2 10 8 0
4
-4--
1 20 7 0
8
314-
Total required 3 5 6 7 21
The row components, u and the column components, v,are calculated using the allocated cells.
The allocated cells are (P, C), (P, Dummy), (0, B), (R, A), (A, B), and (A, C), which give the
following six simultaneous equations. These six equations contain seven variables, hence there
is no unique solution. The actual values given to the components are not important as long as
the set of values is consistent,
c13 = 5 = U1 + v3 for allocated cell P. C)
c14 = 0 = U1 + v4 for allocated cell (P, Dummy)
c33 = 7 = u3 + v3 for allocated cell (A, C)
c31 = 1 = u3 + v1 for allocated cell (A, A )
c32 = 20 = u3 = v2 for allocated cell (R, B)
c22 = 10= u2 + v2 for allocated cell (0, B)
301
STUD Y TEX T
Network Planning and Analysis
A value is assigned to any one of the components, then the values of the other components are
found from the equations. We will choose to set u1 = 0. It follows that v3 = 5, v4 = 0, U3 2, v1 =
-1, v2 18, and u2 = -8. We can now calculate the shadow costs for the unallocated cells, from
the equation:
s c,1 — (u, + v)
Substituting gives the following shadow costs:
s11 = 10 – (0 – ( -1)) = + 11 for empty cell (PA)
s12 = 20 – (0 + 18) = +2 for empty cell (P,B)
s21 = 2 – (-8 – 1) = + 11 for empty cell (0, A)
s23 = 8 – (-8 + 5) = + 11 for empty cell (Q, C)
s24 = 0 – (-8 + 0) = + 8 for empty cell(Q, Dummy)
s34 = 0 - (2 + 0) = - 2 for empty cell (R, Dummy)
These values are entered on the tableau in table 14:
Table 14 Testing the Initial allocation - MODI method
P
From
warehouse
Q
R
To store
A B C Dummy
Total available
10 20 5 0
9 u1 = 0
+11 +2 2 7
2 10 8 0
4 u2 = -8
+11 4 +11 +8
1 20 7 0
8 u3 = 2
3 1 4 -2
Total required 3 5 6 7 21
V1 = -1 V2 = 18 V3 = 5 V4 = 0
The shadow costs are the same as those found using the stepping stone method in table 12.
Route (R, Dummy) has a negative shadow cost of - £2 per item; therefore the solution is not
optimal. Items must be re-allocated using this cell and the associated stepping stone circuit to
reduce the costs.
Finding the optimum solution
The iterative procedure for finding the optimum allocation is as follows:
If there is more than one empty cell with a negative shadow cost, choose the cell with the largest
negative value.
Find the stepping stone circuit for this empty cell, as described above.
3 0 2 quantitative techniques
STUD Y TEX T
Identify those cells from which items are to be deducted and determine the amount which could
be deducted from each cell, without any of the allocations becoming negative. The minimum
value of these figures gives the maximum amount that can be allocated to the chosen cell.
Reallocate around the circuit.
There is no guarantee that further improvements cannot be made. The new allocation must
be checked for optimality using the MODI method. The minimum cost is found when all of the
shadow costs are positive or zero.
We will continue with Example.4.
Example.4 continued: To test a basic allocation for optimality using the MODI method
Cell (R, Dummy) is the only one with a negative shadow cost, -—2. We wish to allocate as much
as possible into this cell.
The stepping stone circuit for (R, Dummy), which gives the —2, is shown below with the
existing
allocations and Unit costs.
Table 15 Stepping stone circuit for (R, Dummy)
C Dummy
P
R
+ denotes items are to be added to this cell. – denotes that items are to be deducted from this
cell
The - cells are (P, Dummy) and (R, C), which contain allocations of 7 and 4 items. The minimum
value in the — cells is 4, which means that 4 items can be moved round the circuit, into the +
cells and out of the — cells. The total saving in cost is (2 x 4) = £8. The revised tableau is given
in table 16.
Table 16 Revised allocations
P
From warehouse
Q
R
To store
A B C Dummy
Total available
10 20 5 0
9
- - 2+4 7-4
2 10 8 0
4
-4--
1 20 7 0
8
3 1 4-4 0+4
Total required 3 5 6 7 21
+5-0
27
-7+0
4 -2
303
STUD Y TEX T
Network Planning and Analysis
The solution is still basic since there are 6 allocations. We will re-check for optimality, using
the MODI method. Using the allocated cells, (P, C), (P, Dummy), (Q, B), (R, A), (R, B) and (R,
Dummy):
c13 = 5 = u1 + v3 Choose u1 = 01 then v3 = 5
c14 = 0 = u1 + v4 v4 = 0
c34 = 0 = u3 + v4 u3 = 0
c31 = 1= u3 + v1 v1 = 1
C32 = 20 = u3 + v2 v2 = 20
c22 = 10 = u2 + v2 u2 = -10
The shadow costs for the empty cells are:
Sij = cij – (ui + vj)
S11 = 10 – (0 + 1) = + 9
S12 = 20 – (0 + 20) = 0
S21 = 2 – (-10 + 1) = +11
S23 = 8 – (- 10 + 5) = + 13
S24 = 0 – (- 10 + 0) = + 10
S33 = 7 – (0 + 5) = + 2
None of the shadow costs is negative, therefore the allocation is optimal.
Table 17 Testing the optimum allocation using the MODI method
P
From warehouse
Q
R
To store
A B C Dummy
Total available
10 20 5 0
9 u1 = 0
+9 0 6 3
2 10 8 0
4 u2 = -10
+11 4 +13 +10
1 20 7 0
8 u3 = 0
3 1 +2 4
Total required 3 5 6 7 21
v1 = 1 v2 = 20 v3 = 5 v4 = 0
The minimum cost is:
£101 + (4 x (— 2)) = £93
Solution:
• Warehouse P sends 6 items to store C and retains 3 items.
• Warehouse Q sends 4 items to store B.
• Warehouse R sends 3 items to store A, 1 to store B and retains 4 items.
3 0 4 quantitative techniques
STUD Y TEX T
If this second allocation is not optimal, the re-allocation procedure is repeated as many times as
is necessary.
It should be noted that the minimum cost was achieved on the first allocation using Vogel’s
method. This will frequently happen for small scale problems. Vogel’s method tends to produce
a better first tableau but there is no guarantee that it will give the optimum immediately. It should
also be noted that the allocation produced by Vogel’s method is different to the one above (see
Example.2). There is an alternative optimum solution:
• Warehouse P sends 1 item to Store B, 6 items to Store C and retains 2 items.
• Warehouse Q sends 4 items to Store B.
• Warehouse R sends 3 items to Store A and retains 5 items.
The existence of an alternative optimum solution was signaled to us by the zero shadow cost for
cell (P, B). These zero shadow costs are associated with alternative allocations which generate
the same total cost.
Sensitivity analysis
The final allocation, together with the shadow costs of the empty cells, can be used for sensitivity
analysis. The shadow cost tells us by how much the total cost will increase, if we are forced to
allocate one item to that empty cell. If we are forced to send one item from warehouse Q to store
C, the extra cost will be £13, much more than the unit cost of £18 for the (Q C) route itself. The
extra cost arises because we have to re-balance the allocation using the following stepping stone
circuit.
Table 18 Stepping stone circuits for (Q, C)
Physical change - items
BC Dummy
Table 19 Stepping stone circuit for (Q, C)
BC Dummy
empty Allocated
-5
Allocated +0
Allocated
-10
Test +8 Empty
Allocated
+20
empty Allocated
-0
P
Q
R
empty Allocated
-1
Allocated
+1
Allocated
-1
Test +1 empty
Allocated
+1
empty Allocated
-1
P
Q
R
305
STUD Y TEX T
Network Planning and Analysis
The net change in cost is:
+8 - 5+0 + 0 + 20 -10= +£l3/item
The maximum number of items which would be moved round this circuit is the minimum
quantity
in the - cell that is:
(P, C) = 6, (R, Dummy) = 4 or (Q, B) = 4
Four items is the maximum number which could be moved.
The zero shadow cost in cell (P, B) was mentioned in the previous section. The stepping stone
circuit for this empty cell is:
Table.20 stepping stone Circuit for (P, B) T able.21 stepping stone circuit for (P, B)
Items can be allocated to cell (P, B) and the net effect on the cost is zero. This means that there is
another allocation which will give the same minimum cost of £93. The maximum number of
items
which can be added to (P, B) is the minimum quantity in the — cells, (R, B) = 1 and (P, Dummy)
= 3. Only one item can, therefore, be moved around the circuit into (P, B).
The shadow costs may also be used to indicate how the cost for an empty cell must change
before the optimum allocation is affected; the shadow cost for the empty cell (R, C) is + £2, and
the actual cost of transfer is £7 per item. The actual cost would have to reduce to at most (7 - 2)
= £5 per item, before we would use this cell to reduce the overall costs.
It is more difficult to determine the effect of cost changes in the allocated cells. If the costs
reduce,
we are encouraged to put more items into that cell. If the costs of an allocated cell increase, at
some point we will wish to stop using that cell and transfer to another.
If we look at the allocated cell, (P, C), it has an actual cost of £5 per item. If this cost is reduced,
it will not affect the physical allocation since this cell already has the full requirement for Store
C
allocated to it.
If the cell cost is increased from £5, we must look at the stepping stone circuits which use (P,
C). These are the circuits which give the shadow costs of £3 for (Q, C) and £2 for (R, C). In
both of these circuits (P, C) is a — cell and any increase in the £5 cost will reduce the shadow
cost of these empty cells. The physical allocation will change when the unit transfer cost of (P,
C) increases by more than £2, from £5 to more than £7. The shadow cost of (R, C) will then
become negative. At this point, it will be advantageous to use this empty cell, changing the (P,
C) allocation.
For the current optimum allocation, the cost of (P, C) has an upper limit of £7 and a lower limit
of
£0. Between these limits the physical allocation is the same but the total costs will change.
test
+20
allocated
-0
allocated
-20
allocated
+0
test
+1
allocated
-1
allocated
—1
allocated
+1
3 0 6 quantitative techniques
STUD Y TEX T
Variations in the transportation problem
Forbidden Allocations
If an allocation from a particular origin to a particular destination is impossible for some reason,
the algorithm can be forced to avoid this allocation by assigning a large cost to the cell. The
exact
value is unimportant but it must be much larger than the other costs in the tableau. The algorithm
will then automatically avoid this cell.
Example 5: Forbidden allocations
This example also illustrates how the transportation algorithm can be used to solve problems
which are not the straight forward transfer of goods from origins to destinations. In this example,
we will deal with the transfer of goods through time. We have a four-month production schedule
to meet. The demand and production capacities are given below.
Table 22 Demand and production capacity
Month Production capacity Demand items
1 300 300
2 350 275
3 325 400
4 375 300
There is an initial stock of 50 items held at the beginning of month 1. Items can be made to meet
immediate demand or for stock to meet future demand. If orders are not met during the required
month, the sales are lost. The variable costs are £100 per item. The stockholding costs are £2
per item per month. What is the optimum production schedule?
Solution
The situation can be modelled using a transportation tableau with the rows representing the
initial stock and monthly production, and the columns representing monthly requirements. The
forbidden cells are those which involve meeting past orders from future production. These cells
are given an infinite cost in table.23.
Table 23 Data for production schedule for months 1 - 4
Cost £ per item months
M1 M2 M3 M4
Total Available
stock M1 2 4 6 8 50
M1
Production
100 102 104 106 300
M2 ∞ 100 102 104 350
M3 ∞ ∞ 100 102 325
M4 ∞ ∞ ∞ 100 375
Total demand 300 275 400 300
The normal procedure is now followed to solve this transportation problem to minimise the cost
of meeting the schedule (See Example 8).
307
STUD Y TEX T
Network Planning and Analysis
Degeneracy
A solution is degenerate when there are fewer than (m + n - 1) allocations in the tableau. This
problem can be overcome by allocating a very small amount, essentially zero, to an independent
cell. The number of allocations is then increased to (m + n — 1). The procedure in the MODI
test
for optimality will show us which empty cells to use.
Example 6: Degenerate solutions
Three warehouses (X, Y and Z) can supply 6, 3 and 4 items to 3 shops (L, M and N), which
require 4, 5 and 1 items respectively. The unit costs of transport are given in the tableau.
Table 24 Data for example.6
To shop (£/item)
LMN
Total Available
X
From Warehouse Y
Z
6496
5323
2364
Total Required 4 5 1
How should the items be allocated in order to minimise the total cost of transport?
Solution
There is a total of 13 items available which is more than the total requirement for 10 items,
therefore, we include a dummy shop which absorbs the surplus from the warehouses. We will
use Vogel’s method to find an initial allocation:
Table.25 initial allocation for example 8 using Vogel’s method
The cost of the allocation is:
4 x 3+0 x 3+3 x 2+2 x l +2 x4=28
To shop
L M N Dummy Total available
Penalty cost
123
6 4 9 0 41 2 2
X - 3 - 31 6 3 0
From warehouse 5 3 2 0 2 1 2
Y - 2 12 - 3 2 0
2360211
Z 43 - - - 4 0
Total required 4
0
5
30
1
0
3
0 13
1st penalty
2nd penalty
3rd penalty
3
3
33
0
0
0
4
42
-
0
-
-
3 0 8 quantitative techniques
STUD Y TEX T
For a basic solution, there should be (3 + 4 — 1) = 6 allocations, but here there are only 5.
The allocation is degenerate. As the MODI method proceeds, we will have to make one zero
allocation, to convert an empty cell into a pretend allocated cell. This gives the required total of 6
allocated cells. It will then be possible to calculate all of the u and v components and hence the
shadow costs.
We begin the MODI procedure using the initial 5 allocated cells. The extra zero allocation is
made when we can go no further. See table.26.
The allocated cells are used to find the row and column components from cij = ui + vj with = 0.
We can calculate v2, v4, u2 and v3 without any problem but we cannot find u3 or v1. We need
an
additional allocated cell. We can put the zero allocation into any empty cell in the v1 column or
the u3 row. It does not matter which of these cells is chosen. Cell (Z, N) is used. The procedure
can be completed and the shadow costs found for the empty cells from sij cij - (ui + vi). The
figures are given in the table:
Table 26 Testing a degenerate allocation—MODI method
Two of the shadow costs are negative. The allocation is not optimal. It is necessary to re-allocate
into cell (Z, M) or (Z, Dummy). We will start with (Z, M), since this has the larger negative
shadow
cost. The stepping stone circuit for (Z, M), showing the items allocated to each cell is:
Table 27 Stepping stone circuit for (Z, M)
MN
Allocated -2 Allocated +1
Test + Zero allocation 0
To shop
L M N Dummy Total available
6 4 9 0 u1=0
X +7 3 +6 3 6
From warehouse 5 3 2 0 u2=-1
Y +7 2 1 +1 3
2 3 6 0 u3=3
Z 4 -4 0 -3 4
Total required 4 5 1 3 13
v1=-1 v2=4 v3=3 v4=0
309
STUD Y TEX T
Network Planning and Analysis
To find the number of items to move round the circuit, we look at the — cells, (Y, M) and (Z,
N),
which contain 2 and zero units. This means we must move the zero allocation round the circuit
so that cell (Z, N) becomes empty again, cell (Z, M) takes the zero allocation and becomes the
pretend allocated cell. The other allocations remain unchanged. When we test for optimality this
time, we find that all of the shadow costs are positive. The allocation is optimal. This means that
the initial solution with the 5 allocations is actually the optimal solution. See the tableau in table
28:
Table 28 T esting the optimal allocation – MODI method
To shop
L M N Dummy Total available
6 4 9 0 u1=0
X +3 3 +6 3 6
From warehouse 5 3 2 0 u2=-1
Y +3 2 1 +1 3
2 3 6 0 u3=-1
Z 4 0 +4 +1 4
Total required 4 5 1 3 13
v1=3 v2=4 v3=3 v4=0
This type of result does occur in some problems in which then an ordinary allocation is
degenerate.
There will be some problems in which the re-allocation process adds items to the zero allocation
cells and it becomes an ordinary allocated cell. The degeneracy of the solution then disappears.
The usual procedures will then lead to a better allocation.
Maximisation
The transportation algorithm assumes that the objective function is to be minimised. However, if
a suitable problem requires the objective function to be maximised, the algorithm can be
modified
slightly to deal with this. For example, we may wish to transfer the items in Example 5 in such
a way that the total contribution is maximised. In such a case the data required are the unit
contributions between each origin and destination. The procedure is to multiply all of the unit
contributions by (-1_ and then to proceed in the usual way.
The Assignment Problem
The assignment problem is a special case of the transportation problem, in which the number
of origins must equal the number of destinations, that is, the tableau is square. Also, at each
destination, the ‘demand’ = 1 and at each origin the ‘supply’ = 1. Any assignment problem may
be
solved using either linear programming or the transportation algorithm. However, the particular
structure of this problem has resulted in the development of a specially designed solution
procedure called the Hungarian algorithm
3 1 0 quantitative techniques
STUD Y TEX T
The assignment algorithm
The algorithm has three stages to it.
Stage 1:
1. S et out the problem in tableau format, as in the transportation algorithm.
2. For each row in the tableau, find the smallest row element and subtract it from every
element in the row
3. R epeat for the columns.
There is now at least one zero in every row and every column. The assignment problem
represented by this ‘reduced’ tableau is equivalent to the original problem and the optimum
allocation will be the same for both, the objective of the Hungarian algorithm is to continue to
reduce the matrix until all of the items to be assigned, can be allocated to a cell with a zero value.
This means that the total value of the reduced objective function will be zero. Since negative
values are not allowed, an objective function values of zero is the optimum.
Stage 2: For a feasible solution, there must be exactly one assignment in every row and every
column.
1. I f the assignments are made only to cells with zero values, this will give us the minimum
value of the objective function.
2. F ind a row with only one zero in it, and make an assignment to this zero. If no such row
exists, begin with any zero.
3. C ross out all other zeros in the same column.
R epeat 1 and 2 until no further progress can be made.
I f, at this stage, there are still zeros which are not either assigned or deleted, then:
4. F ind a column with only one zero and assign to it.
5. C ross out all other zeros in the same row.
6. R epeat 4 and 5 until no further movement is possible.
If all zeros are still not accounted for, repeat I to 6. If the solution is feasible, that is, all of the
allocations have been made to zeros, and then the solution must also be optimal. If the solution
is not feasible, go on to Stage 3.
Stage 3:
1. Draw the minimum number of straight lines through the rows and columns (not diagonals)
that all zeros in the tableau are covered.
2. F ind the smallest element without a line through it.
3. S ubtract this number from every element without a line through it.
4. A dd the chosen number to every element with two lines through it.
5. Leave alone all elements with one line through them.
This procedure has now created at least one new zero. Return to Stage 2 and repeat the procedure
until the optimum solution is reached.
Example 7: To illustrate the application of the assignment algorithm
A company has 4 distribution depots and 4 orders to be delivered to separate customers. Each
depot has one lorry available which is large enough to carry one of these orders. The distances
between each depot and each customer are given in table.29:
311
STUD Y TEX T
Network Planning and Analysis
Table 29 Distance from depots to customers
Distance in miles
Customers
I II III IV
A 68 72 75 83
Depot B 56 60 58 63
C 38 40 35 45
D 47 42 40 45
How should the orders be assigned to the depots in order to minimise the total distance
travelled?
Solution
It will help you to understand the problem if you try to find a solution to it using a familiar
technique,
before applying the mechanics of the Hungarian algorithm. Try using Vogel’s penalty cost
method.
See how close you are to the optimum given at the end of the section. The availabilities and
requirements are one for each row and column.
Stage 1 of the Hungarian algorithm: Find the smallest row elements.
Table 30 To find the smallest row elements
Customers S mallest row element
I II III IV
A 68 72 75 83 68
Depot B 56 60 58 63 56
C 38 40 35 45 35
D 47 42 40 45 40
Subtract the smallest element from each element in its row
Table 31
0 4 7 15
0427
3 5 0 10
7205
0 2 0 5 Smallest column
element
3 1 2 quantitative techniques
STUD Y TEX T
Subtract the smallest column element from each element in its column
Table 32
0 2 7 10
0222
3305
7000
Make assignments as described in Stage 2 above; (0) denotes an assignment
Table 33
(0) 2 7 10
0222
3 3 (0) 5
7 (0) 0 0
We can make only three zero assignments and we require four. This is not a feasible solution. Go
on to Stage 3. We draw the least number of lines to cover all of the zeros.
Table 34
(0) 2 7 10
0222
3 3 (0) 5
7 (0) 0 0
The smallest number without a line through it is 2. Adjust the tableau as described in Stage 3
above, that is, deduct 2 from any number without a line through it, add 2 to any number at the
intersection of two lines and leave all the remaining numbers, which are cut by one line. We then
re-allocate depots to customers.
Table 35 Adjusted tableaus with assignments to zeros
I II III IV
A (0) 0 7 8
B 0 (0) 2 0
C 3 1 (0) 3
D 9 0 2 (0)
313
STUD Y TEX T
Network Planning and Analysis
We have now made the required four allocations to zeros and therefore the solution is optimal.
We allocate depot A to customer I, depot B to customer II, depot C to customer Ill and depot
D to customer IV. The solution, though optimal, is not unique. (C, Ill) must always be allocated
because it is the only zero in the C row. There are two other optimum allocations.
Table 36 First alternative optimum allocation
I II III IV
A (0) 0 7 8
B 0 0 2 (0)
C 3 1 (0) 3
D 9 (0) 2 0
Table 37 Second alternative optimum allocation
I II III IV
A 0 (0) 7 8
B (0) 0 2 0
C 3 1 (0) 3
D 9 0 2 (0)
The minimum mileage for each of these three allocations can be calculated from the original
tableau:
Allocation 1 = 68 + 60 + 35 + 45 = 208 miles
Allocation 2 = 68 + 63 + 35 + 42 = 208 miles
Allocation 3 = 72 + 56 + 35 + 45 = 208 miles
All three solutions give the same total mileage.
Note: For larger problems than the one in Example.7, it may be more difficult to be sure that,
at Stage 3, step 1, the minimum number of lines has been drawn to cover all of the zeros. The
following rule of thumb’ may be helpful:
i. C hoose any row or column which has a single zero in it.
ii. I f a row has been chosen, draw a line through the column in which the zero lies.
iii. I f a column is chosen, draw a line through the row in which the zero lies.
iv. R epeat steps 1 to 3 until all of the zeros have been covered.
Special cases of the assignment problem
Maximise the objective function
The assignment algorithm is designed to minimise the objective function. If we have an
assignment
problem, but we wish to maximise the objective function, we deal with this as we would for the
transportation algorithm. We set up the first tableau and multiply all the values in the cells by -1.
3 1 4 quantitative techniques
STUD Y TEX T
Example 8: To use the assignment algorithm to maximise an objective
A company has 6 sales areas and 6 salesmen. From past experience it is known that the salesmen
perform differently in the different areas. The company’s sales director has estimated sales for
each person in each area. These are given in table 38:
Table 38 Sales by area and salesmen
Sales, £ 000
Area
I II III IV V VI
A 68 72 75 83 75 69
B 56 60 58 63 61 59
salesman C 35 38 40 45 25 27
D 40 42 47 45 53 36
E 62 70 68 67 69 70
F 65 63 69 70 72 68
How should the sales director assign the salesmen to the areas to maximise total sales?
Solution
We multiply all of the values in the tableau by (-1):
Table 39 Modify data and find smallest row element.
Area
Smallest element
I II III IV V VI
A -68 -72 -75 -83 -75 -69 -83
B -56 -60 -58 -63 -61 -59 -63
C -35 -38 -40 -45 -25 -27 -45
D -40 -42 -47 -45 -53 -36 -53
E -62 -70 -68 -67 -69 -70 -70
F -65 -63 -69 -70 -72 -68 -72
We subtract the smallest (most negative) element from each element in the row.
315
STUD Y TEX T
Network Planning and Analysis
Table 40 Subtract row elements and find smallest column elements
Area
I II III IV V VI
A 15 11 8 0 8 14
B735024
C 10 7 5 0 20 18
D 13 11 6 8 0 17
E802310
F793204
702000
Smallest element
We subtract the smallest column element from each element in the column.
Table 41 Subtract smallest column element
Area
I II III IV V VI
A 8 11 6 0 8 14
B033024
C 3 7 3 0 20 18
D 6 11 4 8 0 17
E100310
F091204
Forbidden Allocations
Again this problem is solved in the same way as it was in the transportation algorithm. If a
particular assignment is impossible for some reason, we insert a value into the relevant cell
which
is much bigger than any other value. The algorithm will then automatically avoid this allocation.
Unequal numbers of origins and destinations
If the tableau is not square, then additional dummy rows or columns must be included to make
it square. The values assigned to these dummy cells will usually be zero. Destinations which
receive allocations from dummy rows (origins) are the ones which, in practice, will not receive
an allocation. Allocations which are made to dummy columns represent items which are not
allocated.
3 1 6 quantitative techniques
STUD Y TEX T
Exercise 1
In the Kingdom of the Republic of Idion there are five coal mines which have the following
outputs
and production costs:
Mine output (tonnes/day production cost (£/tonne)
1 120 25
2 150 29
3 80 34
4 160 26
5 140 28
Before the coal can be sold, it must be ‘cleaned’ and graded at one of three coal preparation
plants. The capacities and operating costs of these three plants are as follows:
Plant capacity (tones/day) operating cost (£/tonne)
A 300 2
B 200 3
C 200 3
All coal is transported by rail at cost of £0.5 per tonne kilometer, and the distances (in
kilometers)
from each mine to the three preparation plants are:
Preparation
plant Mines (Distance)
12345
A 22 44 26 52 24
B 18 16 24 42 48
C 44 32 16 16 22
Required
a) U sing a transportation model, determine how the output of each mine should be
allocated to the three preparation plants.
b) F ollowing the installation of new equipment at coal mine number 3, the production cost
is expected to fall to £30 per tonne. What effect, if any, will this have on the allocation
of coal to the preparation plants?
c) I t is planned to increase the output of coal mine number 5 to 180 tonnes per day which
can be achieved without any increase in production cost per tonne. How will this affect
the allocation of coal to the preparation plants?
(ACCA, June 1986)
317
STUD Y TEX T
Network Planning and Analysis
Answer
a) F ive mines supply three preparation plants
The total mines output per day = 650 tonnes/day
The total preparation plant capacity = 700 tonnes/day
T herefore introduce a dummy mine to indicate which plant will not be fully used.
The unit costs for each combination of mine and preparation plant comprise:
Unit variable production cost at the time
+ unit operating cost at the plant
+ unit transport cost
The values shown in the following tableau. Vogel’s penalty cost method is used to find the first
allocation and the MODI method to test for optimality.
There must be (m – n - 1) = 8 entries for a basic solution. There are 8 entries. All the shadow
costs are positives, therefore, this is the optimum allocation.
Mine 1 supplies 70 tonnes per day to A and 50 to B; mine 2 supplies 150 tonnes per day to B;
mine 3 supplies 40 tonnes per day to A and 40 to C; mine 4 supplies 160 tonnes per day to C;
and mine 5 supplies 140 tonnes per day to A.
To prep plant Tons/day
available
Penalty
costs
MODI
ABC
1 38 37 50 1 125 u1=0
705 504
+16- 120,70,0
2 53 40 48 83 u2=3
+12- 1503
+11- 150 0
3 49 49 45 4 u3=11
From mine 40 +1- 40 80 0
54 50 37 132 u4=3
4 +13- +10- 1602 160 0
42 55 42 0 u5=4
5 140 +14- +4- 140 0
0 0 0 0 u6=-38
Dummy 501
+1- +4- 50 0
Tones/day
required
300,250,
180,40,
0
200
50
0
200
40
0
Penalty costs 381
47
37
3 124
37
53
V1=38 V2=37 V3=34
3 1 8 quantitative techniques
STUD Y TEX T
Preparation plant A has 50 tonnes per day spare capacity even though it has the cheapest
operating costs. The total costs of the above allocation are:
70 x 38 + 50 x 37 + 150 x 40 + 40 x 49 + 40 x 45 + 160 x 37 + 140 x 42 = $26,070/day
a) Production costs at Mine 3 fall from $34 to $30 per tonne. All mine output is already
taken by the plants and production costs are like a fixed cost and do not affect the
allocation, therefore total cost will be reduced by 80 x 4 = $320 per day
b) Mine 5 plans to increase output by 40 tonnes per day from 140 to 180. All of Mine 5’s
output is allocated to Plant A which has 50 tonnes per day spare capacity. The extra
40 tonnes per day output will go from Mine 5 TO Plant A, increasing costs by 40 x 42 =
$1,680 per day
Exercise 2
a) Briefly describe and contrast two methods of finding an initial feasible solution to a
transportation problem
b) T he Braintree Electronics Company produces video cassette tapes for purchase by the
general public. The demand (in hundreds) and production capacities (in hundreds) for
the three months in the forth quarter of the year are shown below
Month Normal time Overtime Demand capacity
October 300 400 150
November 450 400 150
December 800 400 150
Note that capacity exists to produce the video cassette tapes by working normal and overtime
hours, where capacity remains constant over time but demand increases for the Christmas sales.
The company does not have any initial inventory and does not wish to have any inventory on
hand after December.
The costs for production of the video cassette tapes are £150 (per hundred) if produced during
normal working hours and £180 (per hundred) if produced during overtime. It has been
determined
that inventory costs are £20 (per hundred) per month. You should assume that all orders are
satisfied on time and that all demands and outputs occur at the midpoint of each month.
Required
i) Formulate this production scheduling situation as a transportation problem with six
‘sources’ and three ‘destinations’, showing the unit cost associated with each source
destination combination
ii) Use the transportation algorithm to find the optimum production schedule over this
period. State the total production cost of your solution
(ACCA, June 1988)
a) Describe two methods of finding an initial feasible solution. We describe two methods
these are the minimum cost and Vogel’s penalty cost methods
b) Braintree Electronics – video tapes. There are six sources of production: October
normal (400) and December overtime (150). Total capacity is 1650 (’00 tapes) and total
demand is 1550 (’00 tapes). Therefore, we need a dummy demand column. The costs
in the tableau are production and inventory costs.
319
STUD Y TEX T
Network Planning and Analysis
(£/`00 video tapes0
To demand
Total
capacity
Oct Nov Dec Dummy MODI
Oct: normal 150 170 190 0 400 100 u1=0
300 50 50 +30_ 50 0
Overtime 180 200 220 0 150 100 u2=30
0_ 0_ 50 100 0
Nov: normal ∞ 150 170 0 400 0 u3=-20
_ 400 0_ +50_
Overtime ∞ 180 200 0 150 0 u4=10
_ _ 150 +20_
∞ ∞ 150 0 400 0 u5=-40
Dec: normal _ _ 400 +70_
∞ ∞ 180 0 150 0 u6=-10
Overtime _ _ 150 +40_
Total demand
300
0
450
50 0
800 400
200 50
0
100
0
1650
MODI v1=150 v2=170 v3=190 v4=-30
The initial allocation was done using the minimum cost method. There are 9 entries in the table
giving a basic solution. The MODI method is used to test for optimality.
All the shadow costs are positive or zero, therefore, this allocation will give the minimum cost.
The total cost is:
300 x 150 + 50 x 170 etc = £251,000
The zero shadow costs mean that there are other allocations (found using the stepping stone
method) that will give this same cost.
Exercise 3
a) E xplain the terms
i) Degeneracy
ii) I nequality of supply and demand
N on-unique solution,
iv) in transportation problems, explain how the transportation algorithm is adapted to
overcome these difficulties
b) T he total Wedgetown Pottery Company has orders to be completed next week for three
of its products – mugs, cups, bowls – as given in the table below
3 2 0 quantitative techniques
STUD Y TEX T
Product Order units
Mugs 4000
Cups 2400
Bowls 1000
There are three machines available for the manufacturing operations, and all three can produce
each of the products at the same production rate. However, the unit costs of these products
vary depending upon the machine used. The unit costs (in £) of each machine are given in the
following table
Mugs cups bowls
A 1.20 1.30 1.10
Machine B 1.40 1.30 1.50
C 1.10 1.00 1.30
Furthermore, it is known that capacity for next week for machines B and C is 3000 units and for
machine A is 2000 units
Required
i) Use the transportation model to find the minimum cost production schedule for the
products and machines. Determine this minimum cost
ii) I f this optimal solution is not unique, describe all other production schedule with the
minimum cost. If the production manager would like the minimum cost schedule to
have the smallest number of changeovers of production on machines, recommend the
optimal solution
Answers
(a) (i) Degeneracy:
I n the transportation algorithm, an allocation is degenerate if the number of routes used
is less than (number of rows + number of columns - l). The difficulty may be overcome
by the use of dummy routes. For example, if the actual number of routes used is one
too few, then an empty cell is selected and treated as if it were an allocated cell, if we
wish, we can allocate a very small amount to this cell. The amount allocated should be
so small that the associated transport costs can be ignored. If the transport allocation
is degenerate by two routes then we use two empty cells in this way, and so on.
(ii) I nequality of supply and demand:
T he transportation algorithm requires the total amount demanded at all of the destinations
to be equal to the total amount available at all of the origins. If this is not the case, then
the algorithm cannot be used until a dummy origin or destination has been added to
satisfy this condition.
I f the total supply exceeds the total demand, then a dummy destination is added which
has a demand equal to the excess supply. If the total supply is less than the total
demand, then a dummy origin added which has a supply equal to the shortfall. The
transportation costs for all routes involving the dummy are zero.
(iii) N on-unique optimal solution:
T he optimal solution is non-unique if there is more than one allocation of routes which
produce the minimum cost. The existence of other optimal allocations can be identified
by examining the shadow cost. If any of the shadow costs are zero, then alternative
optima exist.
321
STUD Y TEX T
Network Planning and Analysis
A zero shadow cost means that items may be moved into that cell without increasing
the total cost of the allocation. Non-unique optimal solutions are not a difficulty in the
transportation problem, but it does mean that the decision maker must use some
criterion, other than cost to choose the allocation of routes to be used.
(b) (i) F ind the minimum cost production schedule using the transportation method:
T otal capacity of the machines is 8,000 units.
T otal requirement is 7,400 units.
T herefore (here is excess capacity of 600 units. A dummy product must be included in
the transportation tableau, for which the demand is 600 units next week.
S et up a transportation tableau and use Vogel’s method to make the initial allocation.
Product capacity penalty
mugs cups bowls dummy
A 1.20 1.30 1.10 0.00 2000 1.10,0.10
10004 _ 10003 _ 1000 0.10
0
B 1.40 1.30 1.50 0.00 3000
2400
0
1.30,0.10
Machine 24004 _ _ 6001 0.10
C 1.10 1.00 1.30 0.00 3000
600
0
1.00,0.10
6004 24002 _ _ 0.20
Demand 4000
0
2400
0
1000
0
600
0
Penalty 0.10
0.10
0.10
0.30
0.302
0.20
0.20
0.203
0.00
For a basic allocation we require:
No of allocated cells = no of rows + no of columns -1
In this case, no of rows + no of columns 3 + 4-1= 6, which is the same as the number of allocated
cells. The allocation is basic.
Test for optimality
For each allocated cell, we split the unit cost, cij, into a row component, ui, and a column
component, vj
For each empty cell, we calculate the shadow cost, sij, where suj = cij – (ui + vj)
3 2 2 quantitative techniques
STUD Y TEX T
Allocated cells:
C11 = 1.20 = u1 + v1 let u1 = 0.00, then v3 = 1.20
C13 = 1.10 = u1 + v3 v3 = 1.10
C21 = 1.40 = u2 + v1 u2 = -0.20
C24 = 0.00 = u2 + v4 v4 = -0.20
C31 = 1.10 = u3 + v1 u3 = -0.10
C31 = 1.10 = u3 + v1 u3 = -0.10
C32 = 0 = u3 + v2 v2 = -1.10
For the empty cells:
S12 = 1.30 – (0.00 + 1.10) = 0.20
S14 = 0.00 – (0.00 + 0.20) = 0.20
S22 = 1.30 – (0.30 + 1.00) = 0.00
S23 = 1.50 – (0.20 + 1.10) = 0.20
S33 = 1.30 – (-0.10 + 1.10) = 0.30
S34 = 0.00 – (-0.10 – 0.20) = 0.30
All of the shadow costs are positive or zero, therefore the allocation is optimal. Since there is
a zero shadow cost, we know that it is a non-unique optimum. The minimum cost production
schedule is.
Machine A makes 1,000 mugs and 1,000 bowls and is fully used;
Machine B makes 2,400 mug and is under-used by 600 units;
Machine C makes 600 mop and 2,400 cups and is fully used.
The minimum cost is:
1.20 x 1000 + 1.10 x 1000 + 1.40 x 2400 + 1.10 x 600 + 1.00 x 2400 = £8720
(ii) S elect the optimum allocation which has the least number of machine changeovers:
The current optimum schedule requires 2 machine changeovers (1 on each of machines A and
C)
Using the stepping-stone method we can allocate N items to cell (B cups) and
Adjust the row and column allocations. We find that N 2400 and that the solution has become
degenerate. However, the solution is still optimum. The total cost is £8720. The alternative
optimum is:
323
STUD Y TEX T
Network Planning and Analysis
Product
mugs cups bowls dummy
A 1.20 1.30 1.10 0.00
1000 _ 1000 _
B 1.40 1.30 1.50 0.00
Machine 2400 _ 600
C 1.10 1.00 1.30 0.00
3000 _ _ _
This solution requires only one changeover (machine A) and therefore is the preferred choice.
The preferred minimum cost production schedule is:
Machine A makes 1,000 mugs and l000 bowls and is filly used;
Machine B makes 2,400 cups and is under-used by 600 units;
Machine C makes 3,000 mugs and is fully used.
Exercise 4
a) Explain briefly how the transportation algorithm can be modified for profit maximisation
rather than the minimisation of costs.
b) T he Orange Computer Company manufactures one product, a dot-matrix printer, which
is currently in short supply. Four of Orange’s main outlets, large specialty computer
shops at Abbotstown, Beswich, Carlic and Denstone, already have requirements which
in total exceed the combined capacity of its three production plants at Rexford, Eadon
and Tristron. The company needs to know how to allocate its production capacity to
maximize profits
Distribution costs (£) per unit from each production plant to each speciality shop are given in the
following table.
From
To
Abbotstown To Beswich To Carlic To Denstone
££££
Rexford 22 24 22 30
Seadon 24 20 18 28
Tristron 26 20 26 24
Since the four specialty shops are in different parts of the country, and as there are differing
transportation costs between the production plants and the specialty shops, along with slightly
three of its different production costs at each of the production plants, there is a pricing structure
that enables different prices to be charged at the four shops. Currently the price per unit charged
is £230 at Abbotstown, £235 at Beswich, £225 at Carlic, and £240 at Denstone. The variable unit
production costs arc £150 at plants Rexford and Tristron, and £155 at plant Seadon.
3 2 4 quantitative techniques
STUD Y TEX T
Required:
i) Set up a matrix showing the unit contribution to profit associated with each production
plant/speciality shop allocation.
ii) T he demands at Abbotstown, Beswich, Carlic, and Denstone are 850, 640, 380, and
230 to produce respectively. The plant capacity at Rexford is 625, at Seadon is 825,
and at Tristron is 450. Denstone vary. Use the transportation algorithm to determine the
optimal allocation.
iii) Determine the contribution to profit for the optimal allocation.
(ACCA, June 1990)
Answers
(a) How can the transportation algorithm be modified to maximise rather than minimise?
Instead of minimising the positive unit costs of all of the cells, calculate the unit profits,
make them negative and put these in each cell. Use the transportation algorithm as
usual to minimise these negative profits.
Alternatively, load the cells with the largest profits (instead of smallest costs) to give an
initial allocation. Test the empty cells as usual, but use any cell which has a price. If all
the shadow prices are negative or zero, then allocation give the maximum profit.
(b) Orange produces printers in three factories and sells them to four main outlets. How
many of the available printers should be supplied from each factory to each outlet in
order to maximise profit?
Factories R, S and T supply shops A, B, C and D.
The contribution per printer = selling price at the shop for a particular cell
- variable cost at the factory
- Factory to shop transport cost
For example the contribution per printer supplied A = 230 – 50 = £180 per printer.
The matrix of contributions is given below:
Contribution £ per printer
To Shop
From
Factory
ABCD
R 58 61 53 60
S 51 60 52 57
T 54 65 49 66
The total demand from the four shops is:
850 - 640+380+ 230 = 2100 printers
The total supply t the three Factories is:
625+825+450=1900 printers
325
STUD Y TEX T
Network Planning and Analysis
There is a shortfall of 200 printers and a dummy factory is needed in the transportation algorithm
to take up the 200 shortfall. In the transportation algorithm, printers supplied by the dummy
factory represent printers which are not supplied to that particular shop.
To shop T o t a l
Available penalty
ABCD
R
58 61 53 60 1 3
6253 _ _ _ 625 0
S
From
factory
51 60 52 57 825 405 3 8
256 4204 3805 _ 25 0 8 1
T
54 65 49 66 450 220
1
1
12
_ 2202 _ 2301 0
Dummy
000000
2007 _ _ _ 200 0 0 0
Demand
850
200
225
0
640
0
420 380 0 230 0 2100
Penalty
4
73
516
4
51
44
1 604
11
1 52
525
61
The penalty cost method will be used to give a first allocation, except that the difference between
the largest profit and the next largest is taken to give the penalty instead of the smallest cost - the
next smallest cost.
This m allocation is b where there are (r + c-1 =) seven allocated cells. There are seven entries
therefore the MODI method may be used to check for optimality without any problems. We use
each allocated cell to calculate each row component, v, and each column component, v. For each
allocated cell:
cij = ui + vj
We choose to sit u = 0 and the other values follow. For each empty cell we calculate
the shadow profit using:
sij = cij – (ui + vj)
3 2 6 quantitative techniques
STUD Y TEX T
This gives the following table:
To shop U
ABCD
R 58 61 53 60 u1=0
625 -6 -6 -8
S
From
factory
51 60 52 57 u2=-7
25 420 380 -4
T 54 65 49 66 u3=-2
-2 220 -8 230
Dummy 0 0 0 0 u4=-58
200 -9 -1 -10
V v1=58 v2=67 v3=59 v4=68
All of the shadow prices are negative, therefore any change would reduce the contribution,
consequently this is the optimum allocation.
The Rexford factory supplies all of its 625 printers to Abbotstown.
Seadon supplies 25 to Abbotstown, 420 to Beswich and 380 to Garlic.
Triston supplies 220 to Beswich and 230 to Ocustonc.
Abbotstown does not receive 200 oldie printers it asked for.
(iii) The contribution to profit of this allocation is:
58 x 625 +51 x 25 + 60 x 420 + others
Exercise 5
(a) T he assignment problem can be regarded as a special case of a transportation
problem.
Describe these special features of the assignment problem and explain why the
transportation algorithm tends not to be used to solve such problems.
(b) The Midland Research Association has recently been notified that it has received
government grant for the research grants to undertake four major projects. The
managing director has to assign a research officer to each of these projects. Currently
there are five research officers - Adams, Brown, Carr, Day, Evans who are available to
carry out these duties. The amount of time required to complete each of the research
projects is dependent on the experience and ability of the research optimal officer who
is assigned to the project. The managing director has been provided with an estimate
of the project completion time (in days) for each officer and each project.
327
STUD Y TEX T
Network Planning and Analysis
Research
officer
Project
1234
Adams 80 120 60 104
Brown 72 144 48 110
Carr 96 148 72 120
Day 60 108 52 92
Evans 64 140 60 96
As the four projects have equal priority, the managing director would like to assign research
officers in a way that would minimise the total time (in days) necessary to complete all four
shops
at projects.
Required:
(i) Determine an optimal assignment of research officers to projects, and hence determine
the minimum total number of days allocated to these four projects.
(ii) S tate any further allocations that would result in the same total number of days. If
research officers Brown, Carr and Day express a preference for projects 2 or 3, whilst
officers Adams and Evans express their preference for projects 1 or 4, which of the
optimal allocations seems to be the most sensible for the managing director to make?
iii) What feature of this particular project duration matrix could be exploited to simplify the
problem?
(ACCA, June 1989)
Answers
(a) T he special features of the assignment problem:
T he assignment problem must have an equal number of origins and destinations. The
supply is one unit at each origin and the demand is one unit at each destination
T he transportation algorithm can not to be used to solve transportation problems under
the above conditions because of difficulties due to degenerate allocations. There
will usually be several degenerate stages which increases the number of iterations
required.
(b) (i) Determine the optimal allocation of people to projects and the minimum total time
required for the four projects
Since there are 5 research officers and 4 projects, we must include a dummy project
which takes zero time no matter which officer is assigned to it. However, if we
look at the data in more detail, we can see that Carr is estimated to require the longest
time to complete each of the projects. It follows, therefore, that the algorithm will allocate
the dummy project to Carr. Knowing this we can reduce the problem to the allocation of
4 projects to 4 people.
3 2 8 quantitative techniques
STUD Y TEX T
Research
Officer 1 2 3 4
Smallest value in the row
Adams 80 120 60 104 60
Brown 72 144 48 110 48
Day 60 108 52 92 52
Evans 64 140 60 96 60
Step 1: For each row, deduct the smallest value in the row from each value in the row.
Research
Officer 1 2 3 4
Adams 20 60 0 44
Brown 24 96 0 62
Day 8 56 0 40
Evans 4 80 0 36
Smallest value in column 4 56 0 36
Step 2. For each column, deduct the smallest value in the column fro each value in the column.
Research
Officer
Project time, days
1234
Adams 16 4 0 8
Brown 20 40 0 26
Day 4 0 0 4
Evans 0 24 0 0
Step 3: Beginning at row 1, if there is only one available zero in a row, make an allocation to it
and cross out all other zeros in the same column. When no further progress can be made with
the rows, repeat the procedure for the columns
Research
Officer
Project time, days
1234
Adams 16 4 0 8
Brown 20 40 0 26
Day 4 0 0 4
Evans 0 24 0 0
329
STUD Y TEX T
Network Planning and Analysis
There are three allocations only, therefore we know we have not reached the optimum solution.
Step 4: Draw the minimum number of lines through the cells with zeros. Find the smallest
number
in a cell without a line through it. Subtract this number from the values in the unlined’ cells and
add it to the values in the cell, with two lines through them.
Research
Officer
Project time, days
1234
Adams 16 4 0 8
Brown 20 40 0 26
Day 4 0 0 4
Evans 0 24 0 0
Research
Officer
Project time, days
1234
Adams 12 0 0 4
Brown 16 36 0 22
Day 4 0 4 4
Evans 0 24 4 0
Step 5: Repeat the allocation procedure with the new tableau
Research
Officer
Project time, days
1234
Adams 12 0 0 4
Brown 16 36 0 22
Day 4 0 4 4
Evans 0 24 4 0
The allocation is still not optimal. Repeat steps 4 and 5 above
Research
officer
Project time, days
1234
Adams 12 0 0 4
Brown 16 36 0 22
Day 4 0 4 4
Evans 0 24 4 0
The smallest
Number in an
Uncovered cell
Is 4.
The smallest number in a cell,
not covered by a line is 4.
3 3 0 quantitative techniques
STUD Y TEX T
Research
Officer
Project time, days
1234
Adams 8 0 0 0
Brown 12 36 0 18
Day 0 0 4 0
Evans 0 28 8 0
Repeat the allocation procedure
Research
Officer
Project time, days
1234
Adams 8 0 0 0
Brown 12 36 0 18
Day 0 0 4 0
Evans 0 28 8 0
We now find that the solution is optimal but not unique. Brown must be allocated to project 3,
but
Adams, Day and Evans can be allocated to more than one of the other projects. The total time
used will not be affected.
We will take the option in which Adams does project 2, Day does project I and Evans does
project
4. The total time used is 120+48+60-,-96=324 days.
(ii) State the alternative optimum and select the most sensible one:
The alternative optimum allocations are:
Officer Project Project Project
Adams 2 2 4
Brown 3 3 3
Day 1 4 2
Evans 4 1 1
Officer Brown must have his preference for project 3 which means that Officer can have his
preference for project 2. We must, therefore, choose alternaiivenumba3, which gives project 4 to
Adams and project I to Evans.
(iii) How can the problem be simplified?
The elimination of Officer Carr from the project has already been exploited in
part (b)(i).
331
STUD Y TEX T
Network Planning and Analysis
Chapter Summary
Two techniques were developed almost simultaneously in the period 1956-1958 by different
firms for different reasons and independently:
(a) T he critical path method (CPM)
(b) T he programme evaluation and review technique (PERT )
The main difference is that CPM is a certainty (deterministic) model while PERT is a stochastic
(uncertainty) model with respect to the project completion time.
Time Estimates
(a) O ptimistic time estimate = a
This is the shortest time an activity can take to be complete. It represents real estimate such that
the probability is small that the activity can be completed in less time.
(b) Pessimistic time estimate = b
This is the longest time an activity can take to be completed. It is the worst time estimate
representing bad luck. Such that the probability is small that the activity will take longer.
(c) Most likely time estimate = m
This refers to the time that would be expected if you work under normal conditions. Used to give
the activity expected (mean) time using the following formula:
Expected activity time, te =
6
a + 4m + b
Chapter quiz
1. What denotes the beginning or the ending of an activity?
2. T hese desriptions relate to a certain activity. Which activity is it?
(a) used to improve clarity of the network
b) Used to facilitate a logical flow of activities in the network.
3. T his is the longest time an activity can take to be completed
4. T his is the process of attempting to reduce the peaks and troughs in the resource
allocation so that we have a more even usage of personnel.
5. A solution is ……….. when there are fewer than (m + n - 1) allocations in the tableau.
3 3 2 quantitative techniques
STUD Y TEX T
Answers to chapter quiz
1. E vent
2. Dummy activity
3. Pessimistic time estimate
4. Smoothing a profile
5. Degenerate
questions from previous exams
June 2000 Question 8
a) E xplain the following terms as used in network analysis:
(i) Backward pass (2 marks)
(ii) C rashing (2 marks)
(iii) S tock (2 marks)
(iv) E arliest start times (2 marks)
(v) C ritical-path activities (2 marks)
b) XYZ Construction Company is building a 250-unit apartment complex in Embakasi,
Kenya. The project consists of hundreds of activities involving excavating, framing,
wiring, plastering, painting, landscaping and more. Some of the activities must be done
sequentially and others can be done simultaneously. Also, some of the activities can be
completed faster than normal by acquiring additional resources.
Required:
(i) H ow would Quantitative Techniques be used to solve this problem? (2 marks)
(ii) What would be the uncontrollable inputs? (2 marks)
(iii) What would be the decision variables of the model? The objective function? The
constraints? (3 marks)
(iv) I s the model deterministic or stochastic? (1 mark)
v) S uggest assumptions that could be made to simplify the model. (2 marks)
(Total: 20 marks)
333
STUD Y TEX T
Network Planning and Analysis
December 2000 Question 8
a) A small construction project involves the following activities:
Normal Crash
Activity Time
(days) Cost (Sh.) Time (days) Cost (Sh.)
1 – 2 Clear ground 6 60,000 5 70,000
1 – 3 Lay foundation 5 30,000 3 50,000
2 – 4 Build walls 3 10,000 2 15,000
3 – 4 Roofing &
piping 7 40,000 4 55,000
3 – 5 Painting 4 20,000 3 30,000
4 – 5 Landscaping 2 10,000 1 17,500
Required
(i) Determine the shortest time and the associated cost to finish this project (8 marks)
(ii) I f a penalty of Sh4,500 must be charged for every day beyond 12 days, what is the most
economical time for competing the project? (4 marks)
b) Explain the four different methods or approaches for organising and displaying project
information. (8 marks)
(Total: 20 marks)
December 2001 Question 8
Consider a project which has been modelled as follows:
Activity I m m e d i a t e
Predecessor(s)
Completion Time (hours)
A-7
B - 10
CA4
D A 30
EA7
F B,C 12
G B,C 15
H E,F 11
I E,F 25
J E,F 6
K D,H 21
L G,J 25
3 3 4 quantitative techniques
STUD Y TEX T
Required:
a) Determine the project’s expected completion time and its critical path. (12 marks)
b) C an activities E and G be performed at the same time without delaying the completion
of the project? (2 marks)
c) C an one person perform A, G and I without delaying project? (2 marks)
d) By how much time can activities G and L be delayed without delaying the entire
project? (2 marks)
e) By how much time would the project be delayed if activity G were delayed by 3 hours
and activity L by 4 hours? (2 marks)
(Total: 20 marks)
June 2004 Question 8
Mrs. Mwangi wants to open a cafeteria in Kisumu. A small business enterprise adviser whom
she
approached, listed for her six major activities to be carried out. The table below gives a summary
of the normal time estimates of each activity, crash time and the cost reduction per day.
Activity Predecessor Normal time
(weeks)
Crash time
(weeks)
A Procurement of materials - 3 3
B Plumbing A 6 4
C Masonry - 5 3
D Electrical works C 8 7
E Carpentry C 6 4
F Finishing B,D,E 4 2
A ctivity C ost Slope (Sh.)
A -
B 45,000
C 30,000
D 60,000
E 22,500
F 75,000
Required
(a) T he normal completion time of the project and the critical activities (6 marks)
(b) (i) T he shortest time the project can be completed. (8 marks)
(ii) T he additional cost to be incurred if the project is crashed. (2 marks)
(c) E xplain the meaning of the cost slope and how it is computed. (2 marks)
(d) A ssumption made when crashing (2 marks)
335
STUD Y TEX T
Network Planning and Analysis
December 2004 Question 8
The following data relate to activities in the development of a two-stage audit software by
Mwangi
and Onyango associates for Mwitu Ltd.
Activity Description Immediate
predecessor
Expected time
(days)
A Evaluate the current system - 7
B Adjust the current system A 3
C Develop a new system - 6
D Dry-run the new system C 3
E Test the system B,D 2
The recommended crashed activity times and total costs are shown below:
Activity Expected time (days) Cost (Sh.‘000’)
A 6 600
B 3 200
C 6 500
D 3 200
E 1 550
At the start of day 8, the following activity status report is received:
Activity Actual cost (AC, sh. ‘000) Percent completion
A 800 100
B 100 67
C 450 100
D 250 50
E00
Required:
a) (i) I n terms of cost, is the project on schedule? Make a recommendation on the action
to be taken. (10 marks)
(ii) C learly explain the assumptions that are made in calculating the ‘crash’ costs for the
activities. (2 marks)
(b) E xplain how the crash cost per unit is computed. Why is it important to compute
them? (3 marks)
(c) E xplain the purpose of a cost status report. (5 marks)
(Total: 20 marks)