SORS 2104 OPERATION RESEARCH TECHNIQUES

COURSE OUTLINE
Chapter1 NETWORK MODELS

(a) Shortest route

(b) Minimal Spanning tree
(c) Maximal flow

Chapter 2. Project Management

(a) Critical path analysis

(b) Deterministic activity times
(c) Probabilistic activity times
(d) Gantt charts resources scheduling
(e) Cost crashing by observation
(f) Formulation of linear programming

Chapter 3 Inventory Models

(a) Deterministic and probabilistic demand models

(b) Economic order quantity
(c) Economic production lot size
(d) Economic order quantity with back orders quantity, re order point model, periodic review model,
safety stock

Chapter 4 Material requirements planning

(a) Product structure

(b) Gross requirement planning,
(c) product structure and gross requirements, net requirements
(d) Time of order

Chapter 5 Decision Analysis

(a) Decision making without probabilities

(b) Optimistic approach conservative approach
(c) Minimax regret approach
(d) Decision making with probabilities
(e) Decision trees expected monetary values, value of perfect information. Value of survey
information
(f) Bayer’s posterior probabilities. Expected utilities attitudes of risk, procedures for obtaining utility
function from a decision tree

1
Chapter 6 Queuing Theory

(a) Single channel,

(b) multi- channel,
(c) cost analysis.

Text books
1 Operations Research by Hamdy A. Taha
2 Introduction to Operations Research By C.W. Churchman, R.L Ackoff and E.L. Arnoff
3 Management Science by David R Anderson Dennis J Sweeney and Thomas A Williams
4 Any other Operations research textbook.

2
 Chapter 1 Network Models

1.1 Shortest Route Problem

The primary objective is to determine the shortest route or path between any pair of nodes in a
network.

Example

Nyamz Construction Company has several construction projects located throughout

Zimbabwe. Some construction sites are sometimes located as for as 400km from their main
offices in Bulawayo. With multiple daily trips carrying personnel, equipment and supplies to
and from the construction locations, the costs associated with transportation activities are
substantial. The network shown below describes the travel alternatives to and from six of
Nyamz’s newest construction sites.
The circles, or nodes of the network corresponds to the site locations. The roads, streets, and
highways appear as the arcs in the network.
The company would like to determine the routes or paths that will minimize the total travel
distance ( in tens of km) from the office to each site.

2 17 7

15 6 5
4 6
1 3 4
10 6
Main Office 3 5 2
4

We need to determine the shortest route from the main office (node1) to each of the other
nodes in the network. We are going to perform a labeling procedure by giving node one the
permanent label {0,S}. The first number in the label for a particular node indicates the distance
from node 1 to that node, while the second number indicates the preceding node on the route
from node1 to that node.

[14,5 ]
Preceding node on the route from node1 to this node is node 5

Distance from node 1 to this node is 14.

3
To perform the first step of the labeling procedure, we must consider every node that can be reached
directly from node 1, ie node 2 & 3.

[15;1]

2 17 7

15 6 5
4 6
1 3 4
10 6
[0;S] 3 5 2
4

[10;1]

We say node 2 & 3 are tentatively or temporally labeled.

We now consider all tentatively labeled nodes and identify the node with the smallest distance value in
its label, thus node 3 with a travel distance of 10 km.
Node 3 can be permanently labeled since there is no any other route to node 3 which has a distance less
or equal to 10.
We can shade node 3 to show that it is now permanently labeled or bold the brackets at node 3.
We now consider all nodes that are not yet permanently labeled that can be reached directly from node
3, and these are node 2 & 5.
Tentative label at node 2 is revised to [10 + 3] = [13] and that at node 5 is [10+4] = [14].

We next consider all tentatively labeled nodes in order to find the node with the smallest distance value
in its label. Node 2 which has the shortest distance out of all the temporally labeled nodes is now
permanently labeled. We now know that node 2 can be reached from node1 in the shortest possible
distance of 13 km by going through node 3. The next step or iteration begins at node 2 , the recently
permanently labeled node. Nodes that can be reached from node 2 are node 4 and7.
Node 4 become [4;3] node 7 is [30;2]. From among the tentatively labeled nodes 4,5 &7, we select the
node with the smallest distance value and declare that node permanently labeled node.
Continuing this process summarise the solution as in the following table

Table 1.1 Shortest route from node one to all the other nodes.

Node Shortest Route from Node1 Distance

2 1-3-2 13
3 1-3 10
4 1-3-5-4 18
5 1-3-5 14
6 1-3-5-6 16
7 1-3-5-6-7 22

4
Example 1.2
Find the shortest route between nodes 1 and all the other nodes in the following network.

2 2 5 3
14 9
2 6 6 8
5
1 5 3 10 6 1
4 12 5 2 0
1 13 8
9
4 7 10
11

1.2 Minimal Spanning Tree Problem

This involves using the arcs of the network to reach all nodes of the network in such a fashion that the
total length of all the arcs used is minimized.

A spanning tree for an N- node network is a set of N-1arcs that connects every node to every other node.
A minimal spanning tree is the set of arcs that does this at the shortest distance. The steps of the
algorithm of the shortest route problem is as follows:

Step 1 Arbitrarily begin at any node and connect it to the closest node in terms of the criteria being
used. (eg, time, cost or distance). The two nodes are referred to as connected nodes and the
remaining as unconnected nodes.

Step 2 Identify the unconnected node that is closest to the set of the connected nodes. Break ties
arbitrarily if two or more nodes quality as the closest node. Add this new node to the set of
connected nodes.

Repeat this step until all nodes have been connected.

EXAMPLE 1.3

N.U.S.T. I.T computer center must have special computer communications lines installed to connect
five satellite users with a new central computer. A Telephone Company will install the new
communication network. However, the installation is an expensive operation. To reduce costs, the
center’s management group wants the total length of the new communication lines to be as small as
possible. While the central computer could be connected directly to each user, it appears to be more
economical to install a direct line to some users and let other users tap into the system by linking them

5
with users already connected to the system. The network for this problem with possible connection
alternatives and distances (in km) is shown below.

2 40
20 5
50
1 40 3
30 40
Computer 40 10 30
Center 4
20

Solution
Starting at node1, the closest node is node 2 with a distance of 20.
Connect nodes 1&2 using bold lines.
In step 2 we find that the closest unconnected node to one of the connected nodes is node 4 with a
distance of 30.
Repeating the steps the minimal spanning tree solution is as below.

2 40
20 5

1
3
30
10
4
20

The minimal length of the spanning tree is given by the sum of distances on the arcs forming the
spanning tree which is 110 km in this case and this is the optimal solution.

EXAMPLE 1.4

The Bulawayo City Council has recently acquired land in the just ended land redistribution exercise for
a new council park, and park planners have identified the ideal locations for the lodges, cabins picnic
groves, boat dock, and scenic points of interest. These locations are represented by the nodes of the
network below. The arcs of the network represent possible road alternatives in the park.

6
If the council designers want to minimize the total road distance (km) that must be constructed in the
park and still permit access to all facilities (nodes), which road alternatives should be constructed.

2 6 12
1 12
5 3 8
6 6 3
8 5 8 7
11
11 4 6 15
3 7 9
4 1 2 2
2
1 4 6 8 3 5
10 14
4 4 7
9 2
3 5 3 4
3 6 13
9

1.3 MAXIMAL FLOW PROBLEM

This involves a network with one input or source node and one output or one sink node.
The problem tries to answer the maximum amount of flow (vehicles, messages, fluids etc) that can enter
and exit the network system in a given period of time.
The amount of flow in any arc is limited due to capacity restrictions on the various arcs of the network.
Flow capacities are based on the direction of the flow.

eg

1 6 0 2

This indicates that a flow of 6 in the direction of 1-2 and no flow from 2-1.
On roads this can be interpreted as a “one way” road.

Steps of the maximal flow algorithm

Step1 Find any path from the source (input) node to the sink (output) node that has flow capacities in
the direction of flow greater than zero for all arcs on the path. If no path is available, then an
optimal solution has been reached.

7
Step 2 Find the smallest arc capacity Pf on the path selected in step 1. Increase the flow through the
network by sending an amount Pf over the path selected in step1.

Step 3 For the path selected in step 1, reduce all arc flow capacities in the direction of flow by Pf
and increase all arc flow capacities in the reverse direction by Pf . Go to step1.

EXAMPLES 1.5

The West to East flow of vehicles moving in Rusape town reaches a level of 1500 vehicles per hour at
peak times. Due to some maintenance program, which calls for the temporary closing of lanes and lower
speed limits a network of alternative routes through the high way has been proposed by a transport
planning committee. The alternative routes include other highways as well as city streets.
Due to differences in speed limits and traffic patterns, flow capacities vary, depending on the particular
streets or roads used. The proposed network with arc capacities is shown below.

No of vehicles(00) passing per hour

3 0
2
0 5
2 0 8
5 1
2 0
3
1 6 0 6
3
5 7 1 0
3 0 7
0 6
0 4 5 0

Applying step 1 to step 3 the total number of vehicles that can pass through the network is 1400 per hour
using the following path.

Path Pf

1-3-6-7 6
1-2-5-7 3
1-2-3-5-7 2
1-4-6-7 1
1-4-6-5-7 1
1-4-6-3-5-7 1

Total 14

8
Consider the 3-6 arc with initial and final flow capacities shown below

Initial Capacities
7 0
3 6

Final Capacities

2 5
3 6

The arc has a flow of 7-2 = 5 in the 3-6 and the flow is summarized as follows

3 5 6

The complete network will be as below

2 3
5
5 7
3

1 6 1 6
3
5
3 7
6
4 3

9
Example 1.6

BP Oil Company ours a pipeline network that is used to transmit oil from its source to several storage
locations. A portion of the network is as follows.
Due to varying pipe sizes, the flow capacities also vary. By selectively opening and closing sections of
the pipeline network, the firm can supply any of the storage locations
(a) if the firm wants to supply storage location 7 (node 7) and fully utilize the system capacity how long
will it take to satisfy a location 7 demand of 100, 000 gallons. What is the maximal flow for this
pipeline system.
(b) If a break occurs on time 2-3 and it is closed down, what is the maximal flow for the system. How
long will take to transmit 100,000 gallows location 7.

3 3
2 5
0 2 5
2

6 2 2
1
3
6 2 2 6 5 0
2 1
2
0 1 0
7
4 2 0

Exercises

10
 Chapter 2

PROJECT SCHEDULING: PERT/CPM

2.1 Introduction

PERT - Program Evaluation and Review Technique.

CPM – Critical Path Method

Uses of Pert and CPM

1 Construction of plants, buildings and highways.

2 Research and development of new products and processes.
3 Maintenance of large and complex equipment.
4 Design and installation of new systems.

A complicating factor in carrying out scheduling of various jobs or activities using

PERT/ CPM is the interdependence of the activities ie same activities depend on the
completion of other activities before they can be started.

2.2 PERT/CPM Network

The first step in the PERT/CPM project scheduling process is the determination of the
specific activities that will make up the project.
As a simple example consider the table below which involves the building of a daga hut
by a new resettled farmer.

Activity Activity Description Immediate Predecessor

A Cutting of poles -
B Clearing of the ground -
C Digging of holes B
D Erection of poles A,C
E Roofing D

The immediate predecessors for a particular activity are the activities that when
completed, enable the start of the activity in question eg. Activity A and B can start

11
anytime since they do not depend on the completion of any prior activity, but activity D
cannot start until activity A and C have been completed.

A network that shows the activities and also shows the predecessor relationships among
the activities is shown below.

Arcs indicate project activities

11111
1 A D E
3 4 5

B
C

2 2222

Nodes correspond to the beginning and ending of activities.

The completion of all the activities that lead into the node is referred to as an event. For
example node 2 corresponds to the event that activity B has been completed and node 3
corresponds to the event that both activity A and C have been completed.

Example 2.1

Develop the network for the project having the following activities and immediate
predecessors.

Activity Immediate Predecessor

A -
B -
C B
D A,C
E C
F C
G D,E,F

12
Activity E should not start at node 3 as this will imply that A and C are the immediate
predecessors for activity E, which is not. E has only C as the immediate predecessor.
We insert a dummy activity, which is rather fictitious activity used to show proper
precedence among the activities.

A 3 D
1

4
B G 6
E
F
2 C 5

Note: Activity E and F both starts at node 5 and end at node 4. This situation causes
problem for certain computer programs that use starting and ending nodes, and they
recognize activities E and F as the same activity. Dummy activities can be added to
make sure that no two or more activities have the same starting and ending nodes.

A 3 D
1

4
B G 6
E
2 C 5

F
7

13
 2.3 PROJECT SCHEDULING WITH PERT/CPM

Example2:
The owner of Bulawayo Shopping Center is considering modernizing and expanding the
current 32 business Shopping complex. If the expansion project is undertaken, the ower
hopes to add 8 to 10 new business or tenants to the Shopping complex. The specific
activities that make up the expansion project are listed below. The objective is to try and
finish the project as quickly as possible. How long will it take to finish the project?

Activity Activity Description Immediate Predecessor Completion

Time(weeks)
A Prepare architectural  5
drawings
B Identity potential new  6
tenants
C Develop prospectus for A 4
tenants
D Select Contractor A 3
E Prepare building A 1
permits
F Obtain approval for E 4
buildings permits
G Perform construction D,F 14
H Finalize contracts with B,C 12
tenants
I Tenants move in GH 2
Total 51

Pert/CPM Network

5
2 D3

A 5 E1 4 4 F
1
C 4 G 14 7

B 6

3 H 12 6 2 I

14
 NB: Total time to complete the project is 51 weeks. However several activities can be
conducted simultaneously eg A and B and this will reduce the total completion time for the
whole project.

2.4 THE CRITICAL PATH

To determine the project completion time we must identify the critical path.
A path is a sequence of connected activities that leads from the starting node (1) to the end node.
The longest path through the network determines the total time required to complete the whole
project, and is called the critical path. The longest path activities are called critical path activities
of the project.
If managers wish to reduce the project completion time they will have to reduce the length of the
critical path by shortening the duration of the critical path activities
To get the critical path we must compute an earliest start and earliest finish time ES and EF
respectively.
ES = earliest start time for a particular activity
EF = earliest finish time for a particular activity
t = expected activity time for the activity

EF  ES  t

For activity A. ES  0 t 5 EF  0  5  5

On the network this will be written as

ES
A
1 [0;5] 2
5
EF

RULE 1

The earliest start time for an activity leaving a particular node is equal to the largest of the earliest finish
times for all the activities entering that node.
Proceeding in a forward pass through the network we can establish the ES and EF for each activity

15
 5
2 D [5;8]

A[0;5] E[5;6] 4 F [6;10]

1
C [5;9] G [10;24] 7

B [0;6]

3 H [9;12] 6 I[24;26]

The latest finish time for activity I is 26 weeks so the total time required to complete the project is 26
weeks. The critical path is obtained by making a backward pass calculation.
Starting at the completion point (node7) and using a latest finish time of 26 for activity 1 we trace back
through the network, computing a latest start and latest finish time for each activity.

LS = Latest start time for a particular activity.

LF = Latest finish time for a particular activity.

LS  LF  t

eg LS for activity I = 26 - 2 = 24

Rule 2

The latest finish time for an activity entering a particular node is equal to the smallest of the latest start
times for all activities leaving that node.
Calculating backwards the LF and LS times are as below.

16
 {7;10} 5
2 D [5;8]

A[0;5] E[5;6] 4 F [6;10]

{0;5} {5;6} {6;10]
1
C [5;9] G [10;24] 7
{8;12} {10;24}
B [0;6]
{6;12}
3 H [9;12] 6 I[24;26]
{12;24} {24;26}

Slack is defined as the length of time an activity can be delayed without affecting the total time required
to complete the whole project.

ie Slack = LS  ES  LF  EF

eg For activity C Slack = LS – ES = 8 – 5 = 3 weeks

This means activity C can be delayed up to 3 week (staring anywhere between week 5 and 8) without
changing the completion time of the whole project of 26 weeks.
The activity C is not a critical path activities.
For critical path activities slack = 0 ie Activities A, E, F, G and I form the critical path.
The critical path is A-E-F-G-I

Activity ES LS EF LF Slack LS- Critical

ES path
A 0 0 5 5 0 YES
B 0 6 6 12 6
C 5 8 9 12 3
D 5 7 8 10 2
E 5 5 6 6 0 YES
F 6 6 10 10 0 YES
G 10 10 24 24 0 YES
H 9 12 21 24 3
I 24 24 26 26 0 YES

17
2.5 PROJECT SCHEDULING WITH UNCERTIAN ACTIVITY

This involves the scheduling of a new product with the activities in this project never been previously
attempted.

Examples 2.3
The H and S Company have been manufacturing industrial vacuum cleaning system for a number of
years. A member of the company’s new- product research team submitted a report suggesting that the
company considers manufacturing a cordless vacuum cleaner. The company’s management would like
to study the feasibility of manufacturing the new product. The researcher suggests the activity list of the
new product as below.

Activity Description Immediate Predecessor

A Develop product design
B Plan market research
C Prepare manufacturing engineering A
D Build prototype model A
E Prepare marketing brochure A
F Prepare marketing brochure C
G Prepare cost estimates D
H Preliminary product testing B,E
I Complete pricing and forecast report H
J Prepare final report F.G.I.

Network diagram for the New product

2 C 5
F
A D
G J
1 E 4 7 8

B I
3 6
H

18
2.5.1 UNCERTIAN ACTIVITY TIMES

Accurate activity time estimates are important in the development of an activity schedule.
When activity times are uncertain, three time estimate (1) optimistic time, (ii) Most probable time and
(iii) pessimistic time allow the project manager to take uncertainty into consideration in determining the
critical path and the activity schedule.
Optimistic time a = the activity time if everything progressed in an ideal manner.
Most probable time m = Most probable activity time under normal conditions
Pessimistic time b = the activity time if significant delays are encountered.

The average or expected time (t) for any activity is determined by the formula

a  4m  b
t


eg for activity A the manager estimates the most probable time to be 5 weeks with a range of 4 weeks
(optimistic) to 12weeks (pessimistic) therefore;

4  4(5)  12 36
tA    6 weeks
6 6

Also with uncertain activity times the variance or variation in the activity time is given by

ba
2

 
2

 6 

The difference between the pessimistic (b) and optimistic (a) time estimates greatly affects the value of
the variance.

 12  4 
2 2
8
So  
2
A      1.78
 6  6

1
NB The variance equation is based on the notion that a standard deviation is approximately of the
6
difference between the extreme values of the distribution. Variance is the square of the standard
deviation.

19
Activity Optimistic
A 4
B 1
C 2
D 3
E 2
F 1,5
G 1,5
H 2,5
I 1,5
J 1
Most Pessimistic
Probable (m) (b)
5 12
1,5 5 2
3 4
4 11
3 4
2 2.5
3 4.5
3,5 7.5
2 2.5
2 3
Expected Variance
(t) σ2
6 1.78
0.44
3 0.11
5 1.78
3 0.11
2 0.03
3 0.25
4 0.69
2 0.03
2 0.11
32

2.5.2 CRITICAL PATH

To get the critical path we treat the expected activity times as the fixed length or duration of each
activity.
The critical path calculations will only determine the expected or average time to complete the project.

2 C[6;9] 5 {13;15}
{10;13} F[9;11]
A[0;6] D[6;11]
{0;6} {6;9} {7;13} G[11;14] J[15;17]
1 E[6;9] 4 {12;15} 7 {15;17} 8

B[0;2] I
{7;9} 3 6 {13;15}
{9;13} [13;15]
H[9;13]

20
The expected completion time for the entire project is 17 weeks and the critical A-E-H- I-J.

2.5.3 VARIABILITY IN THE PROJECT COMPLETION TIME

Variation in critical path activities can cause variation in the project completion time .If a non- critical
path activity is delayed longer enough than its slack time, then if becomes part of a new critical path and
may affect the project completion time. Let T denote the total time required to complete the project. The
expect value of T which is determined by the critical activities A- E- H- I-J is

ET   t a  t E  t H  t I  t J
 6  3  4  2  2  17

The variance of T is given by

 2   A2   E2   H2   I2   J2
 1.78  0.11  0.69  0.03  0.11
 2.72

Stdev    2  1.65
The normal distribution tends to be the better approximation of the distribution of total time for larger
projects where the critical path has many activities. Suppose that management has allocated 20 weeks
for the project. What is the probability that the 20 weeks deadline will be meet.

Using the normal distribution shown below we want P[T  20]

The Z- value for the normal distribution at T = 20 is given by

20  17
Z
1.65
 1.87

Using Z = 1.82

P[T  20]  0.4656  0.5

 0.9656

There is an excellent (96.6%) chance that the project will be completed before the 20 week deadline.

21
2.6 CONDISERING TIME-COST TRADE –OFFS

Use of more resources (such as more workers, overtime and so on) generally increase project costs, and
reduce activity times.
A manager has to make a decision that involves trading decreased activity time for increased project
costs.
In the previous example the 17- week completion time could be reduced if management were willing to
add more resources to shorten any of the critical path activities

Consider the following example

Activity Description Immediate Expected

Predecessor Time(days)
A Overhaul machine I  7
B Adjust machine 1 A 3
C Overhaul machine II  6
D Adjust machine II C 3
E Test machine II B,D 2

The network diagram and the critical path is as below;

A B
1 4 5
E

C D

The critical path is A-B-E and the project is taking 12 days.

22
2.6 CRASHING ACTIVIY TIMES

Suppose the above project was declared to last for 10 days. This can be achieved by shortening selected
activity times by adding resources such as labor or overtime and is referred to as crashing. Crashing
results in added project costs.
We have to identify activities, which cost less to crash and then crash on them to meet the desired
project completion time.

We need information like

1. Estimated activity cost under normal or expected activity time.

2. Estimated time to complete the activity under maximum crashing (ie, the shortest possible activity
time)
3. Estimated activity cost under maximum crashing.

Let  j = normal time for activity j.

 'j = time for activity j under maximum crashing
M j = maximum possible reduction in time for activity j due to maximum crashing.
C j =estimated cost for activity j under normal time
C 'j = estimated cost for activity j under maximum crashing

M j   j   'j
The per unit time basis crashing cost K j is given by

C 'j  C j
Kj 
Mj

Eg If the normal time for activity A is 7 days at CA = $500 and the time under maximum crashing is 4
days at C’A = $800, then
M A  7  4  3 days and

C A'  C A 800  500

KA    $100 per day
MA 3

1 1
So if we decide to crash activity A by 1 days the added cost is 1 ($100)  $150
2 2

23
Total cost for activity A will be $500+150 = $650

Activity Time (Days) Total Cost Maximum Crash cost

normal Crash Normal Crash (cj) reduction per day Kj
(cj) in time = Insert
(mj) formulae
A 7 4 $500 800 3 100
B 3 2 $200 350 1 150
C 6 4 $500 900 2 200
D 3 1 $200 500 2 150
E 2 1 $300 550 1 250

Which activities must be crashed and by how much.

As one crash the current critical path activities other paths may become critical as well.

2.7 A LINEAR PROGRAMMING MODEL FOR CRASHING DECISIONS

Since we have 5 nodes or events in the above example we need five decisions variables to identify the
time of occurrence for each event.
We also need 5 decision variables to represent the amount of crash time used for each of the five
activities. So let
x i = time of occurrence of event i for i  1, 2,.... 5
y j = amount of crash time used for activity j = A,B,C,D and E
Since the total cost for a normal completion time is fixed at $1 700, we can minimize total project cost
(normal cost plus crash cost) by minimizing the total crashing cost.

The objection function became

E
Minimise Z   k j y j k j crash cost coefficient.
j A

Constraints are based on;

1. The time of occurrence of event i ( x i ) must be greater than or equal to the activity completion time
for all activities leading into a node.
2. An activity start time is equal to the occurrence time of its preceding node.
3. The time to complete an activity time is equal to its normal time less the length is crashed.

24
 Chapter 3

Inventory Models

Introduction

Inventory refers to idle goods or materials that are held by an organizations for use in the
future. While inventory serves as a buffer against uncertain and fluctuating usage and
keeps a supply of items available in case the items are needed by the organization or its
customers, the expense associated with financing and maintaining inventories is so high.

Two important questions that need to be addressed in inventory management are;

(i) how much to order when the level of inventory drops,
(ii) and when to order a given item,
should be answered to effectively manage inventories.

3.2 Economic Order Quality (EOQ) model

This is the best known and must fundamental inventory model and is applied when the
demand for an item has a constant, or nearly constant rate.
Also the entire ordered quantity arrives in inventory at one point in time.

To illustrate the use of the EOQ model let us consider the following example.

Examples 3.1

National Breweries is a distributor of beer, wine and soft drink products. From a main
warehouse located in Bulawayo the company supplies nearly 1000 retail stores with its
beverage products. The beer inventory which constitutes about 40% of the company’s
total inventory, averages around 50 000 crates. With an average cost per crate of
approximately $8, the company estimates the value of its beer inventory to be $400 000.
The warehouse manager has decided to do a detailed study of the inventory costs
associated with castle beer the number one selling beer for the company. The purpose of
the study is to establish the how much to- order and the when to order decisions. The
manager has obtained the following historical demand data for the past 10 weeks.

25
Week Demand (crates)
1 2000
2 2025
3 1950
4 2000
5 2100
6 2050
7 2000
8 1975
9 1900
10 2000

Total Demand 20 000

Average per week 2000

Holding Costs
These are costs associated with maintaining or carrying a given level of inventory and
they depend on the size of the inventory.
First there is the cost of financing the inventory investment. This cost of capital is
usually expressed as a % of the amount invested. Netbrew estimates its cost of
capital at an annual rate of 18%.
Other holding costs includes, insurance, taxes, breakages and warehouse overhead. Net
brew estimates these other costs at an annual rate of approximately 7% of the value of its
inventory.
Total holding costs for the beer inventory is 25% (18% + 7%) of the value of the
inventory.
Assume the cost of 1 crate of castle beer is $8. The cost of holding one crate of beer in
inventory for 1 year is 0,25 ($8) = $2

Ordering Costs
This costs is considered fixed regardless of the order quantity. It covers the preparation
of the voucher, the processing of the order including payment, postage, telephone,
transportation, invoice verification, salaries of the purchasers etc. The manager estimates
its ordering costs are $32 per order regardless of the quantity requested in the order.

3.2.2 Formulation of the EOQ Model

The holding costs, ordering costs and demand information are the 3 data items that must
be known before the use of the EOQ model.
Let Q be the order quantity, the how much to order decision involves finding the value of
Q that will minimize the sum of holding and ordering costs.
Assume that Natbrew opens 5 days each week and has a constant demand rate of 2000
crates per week or 400 crates per day.
The graph of the level of beer over one order cycle of length T is as shown below.

26
 Q

1
Q Average Inventory level
2

T
Time

Inventory level over many cycles is as below.

Average Inventory Level

Time

Let I = annual holding cost rate.

C = Unit cost of the inventory item

C h = annual cost of holding one unit in inventory

Annual cost of holding one unit in inventory is

C h  IC

27
Cost of one crate of castle is C = $8 with I = 225%

Annual holding cost for one crate is

C h  IC = 0,25 (8)
= $2

Annual holding cost = (Average inventory level ) (annual holding cost per unit)

1
= QC h
2

Let D be the annual demand for the product.

for Natbrew D = 52 (2000 crates per week)

= 104,000 crates per year.

D
By ordering Q units every time we order we have to place orders per year.
Q

If C o is the cost of placing one order,

Annual ordering cost = (Number of orders per year) (cost per order)
D
= Co
Q

Total annual costs TC = Annual holding costs + Annual ordering cost

1 D
ie TC  QC h  C o
2 Q

For castle beer ( C h = $2, C o = $32 D = 104,000

1 104 000
TC  Q(2)  (32)
2 Q
3 328 000
= Q
Q

28
3.3 The how much to order decision

Find Q that will minimize the total annual costs (TC).

3 328 000
For the castle beer TC  Q 
Q

Using Q = 8000, TC = 8000 + 3,328 00/8000 = $ 8 416

When Q = 5000, TC = 5000 + 3,328,00/500 = $5 666

Several other values of Q gives the graph below

Diagram
600 TC

Annual holding cost

400
Cost ($)

Annual ordering cost

200

Q*

Order Quantity (Q)

3.3.1 Derivative of Q* Using Calculus

1 D
Since TC  QC h  C o
2 Q
Differentiating the above equation with respect to Q gives

d (TC ) 1 DC
 Ch  2o
dQ 2 Q

d (TC )
at minimum 0
dQ

29
 1 DC 2 DC 0
Ch  20  0  Q2 
2 Q Ch
ie
2 DC 0
 Q* 
Ch

For the castle beer;

2(104000)32
Q* 
2
 1824 cases
*
NB Q will always balance the holding costs and the ordering costs

3.4 When to –Order Decisions

The inventory position for an item is defined as the amount of inventory on hand plus the
amount of inventory on order. The when to order decision is expressed in terms of a
reorder point, which is the inventory position at which a new order should be placed.
The manufacturer of castle beer guarantees a 2 – day delivery on any order placed.
Assuming a constant demand rate of 400 crates per day we expect 2( 400) = 800 crates to
be sold during the 2 days it takes a new order to reach the company.
The two day period is called the lead time and the 800 crates demand during this period
is called lead – time demand.
So they had to order when the inventory level reaches 800 crates. The reorder point is the
same as the lead –time demand.
The reorder point r is given by
r  dm where

r = reorder point
d = demand per day
m = lead time for a new order in days

Period between orders is called cycle time(T) and is given by:

Q*
T ( in years)
D
12Q *
or T  ( in months)
D
250Q *
or T  (in days)
D

30
3.5 Economic Production lot size model
In this model, instead of assuming that goods arrive in a shipment of size Q as in the
EOQ model, we will assume that units are supplied to inventory at a constant rate over
several days or week.
The constant supply rate assumption implies that the same number of units is supplied to
inventory over a period of time eg 10 units every day. Once an order is placed production
begins and a constant number of units is added to inventory each day until the production
run has been completed.
In this model production rate should be greater than demand rate.

The inventory pattern is as below

Production phase Non production phase

Inventory
Level Average inventory level

Time

This model assume two cost; holding costs and production set up costs.
Production set up costs include labor, material etc and is a fixed cost that occurs for every
production run regardless of the production lot size.

The total cost model

Let Q be the production lot size ie the total number of units produced during the
production period.
Let d = daily demand rate for the product
p = daily production rate for the product
t = number of days for a production run.
Since p d, the daily inventory build up rate during production phase is p - d. After t
days the level of inventory will be ( p  d )t .

Max inventory level = ( p  d )t

Since Q is the production lot size it means

31
 Q  pt
Q
t  days (length of a production run)
p

Maximun inventorylevel  ( p  d )t
Q
 ( p  d)
p
d
 (1  )Q
p

1
Average inventorylevel  max imum inventorylevel
2
1 d
 (1  )Q
2 p

Therefore
Annual holding costs   Ave. Inventory level Annual holding cos ts 
1 d 
 1  QC h
2  p

If D is the annual demand for the product and C o is the set up cost for a
production run then,

 No of production runs Set up cos t 

Annual set up cos t    
 per year  per run 

D
 Co
Q

1 D D
 TC  1  QC h  Co
2 P Q

P is the annual production for the year.

d D
NB 
p P

32
 d (TC ) 1 D DC
 (1  )C h  2 o
dQ 2 P Q

2 DC 0
 Q* 
D
(1  )C h
P

Example 3.2

The New Green Key Bar Soap is produced on a production line that has an annual
capacity of 60 000 bars. The annual demand is estimated at 26 000 bars with the demand
rate essentially constant throughout the year. The cleaning, preparation and set up of the
production line cost approximately $135.00. The manufacturing cost per bar is $4.50 and
the annual holding cost is figured at a 24% rate.
(a)What is the recommended production lot size.
(b) Find
(i)Total cost
(ii) the reorder point and cycle time between production runs given that
the lead time is 1 week.

Solution
C h  IC  0.24($4.50)  $1.08
(a)

2(26000)135
Q* 
26000
(1  )1.08
60000

 3 387

(b) (i)
1 D D
TC  1  QC h  C o
2 P Q

 $2 073

33
 (ii)
r  dm
 ( weekly demand)(1 week lead time)
26000
 (1)
52
 500 bars

and
250Q
T
D
250(3387)

26000
 33 working days

So a production of 3387 bars had to be planned every 33 days

3.6 Quantity Discounts for the EOQ Model

In this model the cost per unit varies depending on the quantity ordered.

Example 3.3

The data and cost analyses below show an annual holding cost of 20%, an ordering cost
of $49 per order, and an annual demand of 5000 units. What quantity should be ordered?

Discount Order size Discount % Unit Cost

Category
1 0-999 0 $5
2 1000-2499 3 $4,85
3 2500 + 5 $4,75

34
Solution

Let Q1 , Q2 and Q3 represent the order quantity for discount category 1,2,3 respectively.

2 DC o
Since Q *  and C h  IC  (0.20)C
Ch

The following steps should be followed

Solution

Step 1 Calculate Q* based on the unit cost associated with the discount category 1, 2, 3
respectively.
2(5000)49
Q1*   700
0.2(5)

2(5000)49
Q2*   711
0.2(4.85)

2(5000)49
Q3*   718
0.2(4.75)

Step 2 For the Q* that is too small to qualify for the assumed discount price adjust the
order quantity upward to the nearest order quantity that will allow the product to be
purchased at the assumed price.

Purchase cost = annual demand x unit cost (which was constant and never affected by
the inventory order b4)
Q D
So the Total Cost TC  C h  C o  DC (i)
2 Q
N.B if a calculated Q for a given discount price is large enough to qualify for a bigger
discount, that value of Q cannot lead to an optimal solution.

Step 3 For each of the order quantities resulting from steps 1& 2, compute the total
annual cost using the unit price from the appropriate discount category and the order
quantity yielding the minimum total annual costs is the optimal order quantity.

The results after step 3 are summarized below

35
Discount Unit cost Order Holding Ordering Purchase Total Cost
category quantity Cost Cost Cost TC
1 5 700 350 350 25000 25 700
2 4.85 1000 485 245 24.250 24 980
3 4.75 2500 1188 98 23.700 25 036

So a decision to order 1000 units at the 3% discount rate yields the minimum cost
solution Q* = 1000

3.7 A Single – Period Inventory Model with Probabilistic Demand

This single- period inventory model treat demand as probabilistic and best described by
probability distribution.
In situations where the demand rate is not deterministic models have been developed that
treat demand as probabilistic and best described by a probability distribution.
The single- period inventory model refers to inventory situations in which one order is
placed for the product; at the end of the period, the product has either sold out, or there is
a surplus of unsold items that will be sold for a salvage value.
The single- period inventory model is applicable in situations involving seasonal or
perishable items that cannot be carried in inventory and sold in future periods. Eg
newspapers, winter coats and the single period inventory problem is sometimes referred
to as the news boy problem.
Incremental analysis is a method that can be used to determine the optimal order quantity
for a single- period inventory model.
Incremental analysis addresses the how- much – to –order question by comparing the cost
or loss of ordering one additional unit with the cost or loss of not ordering one additional
unit.

Co = cost per unit of overestimating demand (this cost represent the loss of ordering
one additional unit and finding it cannot be sold).

Cu = cost per unit of underestimating demand (this cost represents the opportunity
loss of not ordering one additional and finding that it could have been sold

The general expressions for the expected loss for overestimating and underestimating
order –quantity are:

EL(Q * 1)  Co P(demand  Q*) (1)

EL(Q*)  Cu P( Demand  Q*) (2)

But P(demand  Q*)  P(demand  Q*)  1 (3)

 P (demand  Q*)  1  P (demand  Q*) (4)

Substituting (4) into (2) yields
EL (Q * 1)  Cu [1  P(demand  Q * (5)

36
 When EL (Q * 1)  EL (Q*) (6)
there is no economic advantage to increasing the order quantity by one additional
unit. Substituting (1) and (5) into (6);

Co P(demand  Q*)  Cu [1  P(demand  Q*) (7 )

Solving for P(demand ≤ Q*) we have

Cu
P(demand  Q*)  (8)
C o  Cu
and this states the general condition for the optimal order quantity Q in the single –
period inventory model.
Thus to establish an optimal order quantity for single- period models we must identify the
probability distribution that describes the demand for the item and the costs of
overestimation and underestimation.

EXAMPLE 3.4
The J & B shop sells calendars with different colonial picture for each month.
The once – a – year order for each year’s calendar arrives in September.
From past experience, the Sept- Jul demand for the calendars can be approximated by
a N (500,1202 ).
The calendars cost $1 500 each and J&B sells them for $3 000 each.
(a) If J &B throws out all unsold calendars at the end of July (ie salvage value is
zero) how many calendars should be ordered?
(b) If J & B reduces the calendar price to $ 1000 at the end of July and can sell all
surplus calendars at this price, how many calendars should be ordered?

solution

(a) Q ~ N (500,1202 ) C 0  1500 Cu  3000  1500  1500

1500
P (demand  Q*) 
3000

Normalising this equation yields

 Q * 500  1
P Z  
 120  2

Q * 500
 0 ( from normal tables)
120

 Q*  500

37
 Pdemand  Q * 
1500
2000

 Q * 500 
P Z    0.75
(b)  120 

Q * 500
 0.675
120
Q*  581

Example 3.
An Air- Conditioning Company is considering the purchase of a special shipment of
portable air conditioners manufactured in Japan. Each unit will cost the company
$80 000 and it will be sold for $ 125 000.
The company does not want to carry surplus air conditioners over to next year.
Thus all surplus air conditioners will be sold to a wholesaler for $50 000 per unit.
Assume that the air conditioners demand follows a N (20, 82)
(a) what is the recommended order quantity?
(b) What is the probability that the company will sell all the units it orders?

solutions

(a) Cu  45 000 C 0  30 000

Cu
Pdemand  Q * 
45

C 0  C u 75

 Q * 20 
P Z    0.6
 8 

Q * 20
 0.253
8

Q*  22

PSell all   Pdemand  Q *

 1  Pdemand  Q *
(b)
 1  0.60
 0.40

38
3.8 An Order –Quantity, Re – order Point Model with Probabilistic Demand
(Multi – Period)

Reorder point

Order placed

Time Stock out

The inventory level decreases at a non-constant rate based on probabilistic demand.

A new order is placed whenever the reorder point is reached.
At times, the order quantity of Q units will arrive before inventory reaches zero.
However at times, higher demand will cause a stock out before a new order is
received. As with other order –quantity, reorder point models, the manager must
determine the order quantity Q and the re- order point for the inventory system.

3.8.1 How much to order decision

The how much to order question is answered by the following equation.

2 DC 0
Q* 
Ch

3.8.2 The when – to – order decision

If a normal distribution is used for lead-time demand, the general equation for r is
r  t  Z where Z is the number of std deviation necessary to obtain the acceptable
stockout probability.

39
EXAMPLES 3.
Barons Inc provides a variety of auto parts to small local garages. Barons purchase parts
from manufacturers according to the EOQ mode and then ship the parts from a regional
warehouse directly to its customers. For a particular type of springs, Baron’s EOQ
analysis recommends orders with Q* = 25 to satisfy an annual demand of 200 mufflers.
There are 250 working days and the lead-time averages 15 days.
(a) What is the re –order point if Barons assumes a constant demand rate?
(b) Suppose that an analysis shows that the lead –time demand follow N (12, 2,52). If
the management can tolerate one stock out per year, what is the revised re- order
point?
(c) What is the safety stock for part (b) if Cn = $5 / year. What is the extra cost due to
the uncertainty demand?

solution

 200 
(a) r  dm   15  12
 250 

D 200
(b)   8 orders per year
Q 25

P1 stockout per cycle 

1
 0.125
8

P(stockout) = 0.125

40
For area in tail = 0.125, z = 1.15

r  12
Z  1.15
2 .5

r  12  1.15(2.5)  14.875  15

(c) Safety stock = 15 – 13 = 3

Added cost = 3($5) = $15 per year

41
 Chapter 4

DECISION ANALYSIS
4.1 Introduction

Decision analysis is used to determine optimal strategies when a decision maker

is faced with several decisions alternatives of future events.
For example a manufacturer of seasonal clothing would like to manufacture more
if consumer acceptance and demand for the product is high and manufacture less
if demand is low.
Selection of the best production- volume decision from among several alternatives
when the decision maker is faced with the uncertainty of future demand is a
problem suited for decision analysis.

4.2 STRUCTURING THE DECISIN PROBLEM

EXAMPLES 4.1

Alfa computer systems is a newly formed computer service firm that specializes in
information services such as surveys and data analysis.
The company is in the final stages of selecting a computer system for one of its
branches.
The company is currently attempting to determine the size of the computer system
that would be most economical, between one of three computer systems which differ
in size and capacity.
The three decision alternatives, denoted by d1, d2, d3, are as follows:
d1 = large computer system
d2 = medium computer system
d3 = small computer system

The second step is to identify the future events that might occur, and these events, which
are not under the control of the decision maker, are referred to as the state of nature.
The states of nature for the company are as follows:

S1 = high customer acceptance of the company’s services

S2 = low customer acceptance of the company’s services

Given the 3 decision alternatives and two states of nature, which computer system should
the company select.

42
4.2.1 Payoff Tables

In decision analysis terminology, the outcome resulting from a certain decision and the
occurrence of a particular state of nature is a referred to as the payoff. The profit and loss
that the company will make for taking a certain decision and the occurrence of a state of
nature is shown below.

STATES OF NATURE

Decision Alternatives High Acceptance (s1) Low Acceptance (s2)

Large system $200 000 - $20,000
Medium system $150 000 $20,000
Small system $100 000 $60,000

Payoff (profit in $) of Alfa C.S

The notation used in the payoff table is V (di, Sj) which is the payoff associated with
decision alternative di and state of nature Sj.

eg V (d1, S2) = $ - 20 000

DECISION TREES

A decision tree provides a graphical representatives of the decision making process. For
ACS

High (S1) $200 000

2
Large(d1)

Low(S2) $-20 000

High(S1) $150 000

1 3
Medium (d2)

Low(S2) $20 000

High(S1) $100 000

Small (d3)
4

Low(S1) $60 000

43
Note node1 (which is a Square) is called a decision node whereas 2,3 and 4 are state of
nature nodes.

4.3 DECISION MAKING WITHOUT PROBABILITIES

4.3.1 Optimistic Approach

The optimistic evaluates each decision alternatives in term of the best payoff that can
occur.
In the above example, the optimistic approach would lead the decision maker to choose
the alternative corresponding to the largest profit.
For problems involving minimization, this approach leads to choosing the alternative
with the smallest payoff.
For the above problem the decision is to buy the large systems.
For maximization the approach is referred to as the maximax approach and for a
minimization it is referred to as the minimin approach.

Decision alternative Maximum payoff

d1 200 000
d2 150 000
d3 100 000

4.3.2 Conservative Approach

This evaluates each decision alternative in terms of the worst payoff that can occur.
The decision alternative recommended is the one that provides the best of the worst
possible payoff.
For a maximizations problem this approach is referred to as the maximin and for
minimization problem it is referred to as the minmax.
For A.C.S.

Decision alternatives Minimum payoff

d1 - 20 000
d2 20 000
d2 60 000

4.3.3 MINIMAX REGRET APPROACH

This approach is neither purely optimistic nor purely conservative.

Suppose that d3 was chosen and it happens that service is high, the resulting profit would
be $100 000.
Since S1 occurred we see that d1 yielding a profit of $20 000 would have the best
decision.

44
 Opportunity loss or regret = difference between best payoff (200 000) and payoff
experienced 100 000

= 200 000 – 100 000

= $100 000
If we choose d2 and state S1 occurred

Opportunity loss or regret = 200 000 – 150 000 = 50 000

In general for a maximization problem.

Opportunity loss or regret R(di, Sj) = V* (Sj) – V (di, Sj) where;

R (di, Sj) = regret associated with decision alternative with decision alternative di and
state of nature sj
V* (sj) = best payoff value under state of nature sj
V(di, Sj) = payoff associated with decision alternative di and state of nature sj

In minimization problems V* (sj) will be the smallest entry in column j, and

R (di,Sj) = V(di, Sj ) – V*(Sj)

Regret or opportunity loss of A.C.S. is as below,

State of nature

Decision alternatives S1 S2
d1 0 80 000
d2 50 000 40 000
d3 100 000 0

Decision alternatives Maximum regret or opportunity loss

d1 80 000
d2 50 000minimum of the maxim regret
d3 100 000

NB Although the 3 approach have provided different recommendations,

ultimately, the decision maker will have to choose the most appropriate opproach and
then make the final decision accordingly.

45
4.4 DECISION MAKING WITH PROBABILITIES

In many decision making situation, it is possible to obtain probability estimates for each
of the states of nature.
When these probabilities are available, the expected value approach can be used to
identify the best decision alternatives, after evaluates each decision alternative in terms of
its expected value.
Consider the Alfa computer system (A.C.S)

Let N = the number of states of nature. P(Sj) = probability of state of nature Sj

Conditions
P (sj)  0 for all states of nature.

 P( S
j 1
j )  P( S1 )  P( S 2 )  ......... P( S N )  1

Expected Value (EV) of decision alternative di is;

N
EV (d i )   P( S j )V (d i , S j )
j 1

For A.C.S. Let P(S1) = 0.3 and P(S2) = 0.7

EV (d1 )  0.3(200 000)  0.7(20 000)  $46 000

EV (d 2 )  0.3(150 000)  0.7(20 000)  $59 000
EV (d 3 )  0.3(100 000)  0.7(60 000)  $72 000
P (S1) = 0.3 $200 000

2
Large(d1)

P(S2) = 0.7 $-20 000

P(S1) = 0.3 $150 000

1 3
Medium (d2)

P(S2) = 0.7 $20 000

P(S1) = 0.3 $100 000

Small (d3)
4
P(S1) = 0.7 $60 000

46
 2
Large(d1) EV = $46 000

1 3
Medium (d2) EV = $59 000

Small (d3)
4 EV = $72 000

4.4.1 SENSITIVITY ANALYSIS

For A.C.S P(S1) = 0.3 and P(S2) = 0.7 these probabilities might change and such changes
is referred to as sensitivity analysis.
For example Let P(S1) = 0.6 and P(S2) = 0.4

EV (d1 )  0.6(200 000)  0.4(20 000)  $112 000

EV (d 2 )  0.6(150 000)  0.4(20 000)  $98 000
EV (d 3 )  0.6(100 000)  0.4(60 000)  $84 000

For cases
To get p equal any two equation

220 000p – 20000 = 40 000 + 60 000

180 000p = 80 000

p = 0,44

ie when p = 0,44 all the decision alternatives will provide the same EV for p  0,44 d3
provides the largest EV.
sensitivity analysis can assist management in deciding when it is worth investing more
time and money to obtain better probability estimates .

47
NB Similar sensitivity analysis with 3 d a and two S on should not be expected to result
in the same type of graph.
In cases where s o n are more than 3 a computer software package can be used to assist
with the computations

4.5 EXPECTED VALUE OF PERFECT INFORMATION (EV PT)

A.C.S. should conduct a market research study that would evaluate consumer needs and
such a study could help by improving the probability assessments for the states of
nature.
If the cost of obtaining the market research information exceeds its value then the
company should not conduct the research study.
If perfect information were available the decision strategy will be as follows:
If s1 occurs then select d1
‘’ s2 ‘’ ‘’ ‘’ ‘’ ‘’ select d3

since P (s1) = 0,3 & P (s2) = 0,7, we see that there is a 0,3 probability that ACS will
make $200 000 and 0,7 probability that ACS will make $60 000.
So the expected value of the decision strategy that uses perfect information is
0,3 ( 200 000) + 0,7 (60 0000) = $102 000.
$102 000 is referred to as the expected value with perfect information (EV w PT)
When perfect information was not available, the expected value of approach resulted in
d3 with an expected value of $72 000
$72 00 is the expected value without perfect information (EV W 0 P1)
The expected value of perfect information (EVP1) – EV wo P1
= 102 000 – 72000
= $30 000
In other words $30 000 represents the additional expected value that can be obtained of
perfect information were available about the states of nature.
So ACS should not pay more than $30 000 for any information no matter how good.

4.6 DECISION ANALYSIS WITH SAMPLE INFORMATION

To make the best possible decision, the decision maker may want to seek additional
information about the states of nature.
This new information can be used to revise or update the prior probabilities so that the
final decision is based on better probability estimates for the states if nature.

In the ACS example, P (s1) = 0,3 and P (S2) = 0,7 and these probabilities are referred to
as prior probabilities for the states of nature.
Suppose ACS has decided to carry a market research, which will provide new
information that can be combined with the prior probabilities by using Bayes theorem.
The revised probabilities are called posterior probabilities and the new information
obtained through research or experiments is referred to as indicator.

48
The marketing research study outcome was as follows.
I1 = favorable market research report (ie the individually contacted express interest in
ACS services)
I1 = favorable market research report ( ie the individually contacted express little interest
in ACS)
Insert a diagram

The end result of the Bayesian revision process depicted above is a set of posterior
probabilities of the form P(sj/ Ik) ie sj will occur given what market research study was
indicator Ik
Assume that in the ACS problem the following are available

MARKET RESEARCH REPORT

Insert a table

Developing a Decision Strategy

Insert diagram

4.6.1 Computing Branch Probabilities

Prior probabilities in the ACS problem were given as P (S1) = 0,3 & P (S2) = 0,7

Also insert the solutions

To calculate the probabilities P (Ik) and P (sj / Ik) , Bayesian procedure is applied.

First note that:

The market research report is favorable (I1) and the state of nature turns to be high
acceptance (S1), is written as (I1 S1)
Also I1 and son turns to be S2 is written (I1 S2)
The probabilities of these two are written P (I1 S2) and P (I1  S2)

Insert the formulae

49
The indicator probabilities are:

Insert the solutions

4.6.2 COMPUTING BRANCH PROBABILITIES A TABULAR

PROCEDURE

First for each indicator Ik we form a table consisting of the following 5 columns
1 States of nature Sj
2 Prior probabilities P(sj)
3 Conditional probabilities
4 Joint probabilities
5 Posterior probabilities

Then given any indicator Ik the following procedure can be used to calculate
Insert the solution
Step 1
In column 1, the states onto the problem being analyzed.

Step 2
In column 2, enter the prior probability to each of state of nature.

Step 3
In column 3 enter the appropriate value of ( I k sj ) for each s – o – n specified in column
1

Step 4
To compute each entry in column 4 multiply column 2 x corresponding c 3

Step 5

Add the entries in column 4. The sum is the value of P (I k)

Step 6

To compute each entry in column 5, divide the corresponding entry in column 4 by P (Ik)
Do for P (T2)

4.6.3 An Optimal Strategy

After computing the branch probabilities, we can use the expected value approach to
determine the decision strategy.

Working backward through the decision tree in use to calculate the expected value at
each s-o – n mode.

50
Enter formulae’s

So since d1 leads to an expected value of $150,324 the optimal decision at node 2 is d1.
Similarly at node 3 optimal decisions at this branch is d3. ($63, 480)

Finally EV (node 1) inserts the formulae

the value $90,402 is viewed as the expected value of the optimal decision strategy when
the market research study is used.
We will need to know the results of the market research study before deciding on any
system.
The results of the decision analysis at this point, however, have provided the following
optimal strategy.

IF Then
Report favorable Select large system (d1)
Report unfavorable Select small system (d3)

4.7 EXPECTED VALUE OF SAMPLE INFORMATION

Since the additional information provided by the market research firm will result in an
added cost, in terms of the fee paid to the research firm, management may question the
value of this market research information.
The value of sample information is often measured by calculating what is referred to as
the expected value of sample information
For maximization problems insert the format and diagram.
So ACS should be willing to pay up to $18,402 for the market research.
Insert format

NB Higher efficiency ratings indicate that the sample information is almost as good as
perfect information, and additional sources of information should not be worthwhile.
Lower efficiency ratings for sample information lead the decision maker to look for other
types of information

51
52
 Chapter 5

WAITING LINE MODELS (QUEUING THEORY)

5.1 Introduction

The study of waiting lines is referred to as the queuing theory.

Waiting line models consist of Mathematics formulas and relationships that can be used
to determine the operating characteristics (performance measures) for a waiting line.
Some of the operating characteristics that are of interest are:

(1) The probability that there are no units in the system.

(2) The average number of units in the waiting line.
(3) The average number of units in the system (the number of units in the waiting
line + number of units being served).
(4) The average time a unit spends in the waiting line.
(5) The average time a unit spends in the system. (line plus service)
(6) The probability that an arriving unit has to wait for service.
(7) The probability that of n in the system ( n  0 ).

5.2 THE STRUCTURE OF A WAITING LINE SYSTEM

Example 5.1
Hefilis fast- food restaurant, sells hamburgers, cheeseburgers, French fries, soft drinks,
milk shakes, pizas and Chinese food and a lot more dessert selections.
Although Hefilis would like to serve each customer immediately at times more customers
arrive than can be handled by the food- service staff. Thus customers wait in the line to
place and receive their orders.
Hefilis is concerned that the methods they are currently using to serve customers are
resulting in excessive waiting times. Management has asked that a waiting line study be
performed to help determine the best approach to reducing waiting times and improving
services.

SINGLE CHANNEL WAITING LINE

Order taking
and order
filling

53
5.2.1 The Process of Arrivals

For many waiting line situation, the arrivals occur in a random fashion, ie each arrival is
independent of other arrivals, and we cannot predict when the next arrival will occur.
It was found out that the Poisson probability distribution provides a good description of
the arrival pattern.

The probability of x arrivals in a specified time period is defined as;

x e 
P( x)  for x = 0, 1, 2 ,………
x!

where x = the number of arrivals in the period.

 = the average or mean number of arrivals per period.

Suppose that Hefilis has analyzed data on customer arrivals and has concluded that mean
arrival rate is 45 customers per hour ie
45
  45 arrivals per hour or   arrivals per minuate
60

Probability of x arrivals per minute is;

x e  (0.75) x e 0.75
P( x)  
x! x!

Probability of 0, 1, 2 arrivals during a in minute period is

(0.75) 0 e 0.75
P(0)   0.4724
0!

(0.75)1 e 0.75
P(1)   0.3543
1!

5.2.2 The Distribution of Service Times

The service time is the time the customer spends at the service facility once the service
has started.
It has been determined that the exponential probability distribution often provides a good
approximation of service times in waiting line situations.

54
 P[ service time  t ]  1  e  t
where
μ = the average or mean number of units that can be served per time period.

Hefilis has found out that a single food server can process an average of 60 customer
orders per hour.
Or 6060 = 1 customer per minute.

So μ = 1 customer minute

P[ service time  0.5]  1  e 1( 0.5)

 1  0.6065

5.2.3 Queue Discipline

Most waiting line system assume a FIFO Basis (first in first out)
Other are FILO basis eg in an elevator
LILO
LIFO

5.3 The Single – Channel Waiting Line Model with Poisson Arrivals and
Exponential Service Times

The following Assumptions regarding the waiting line are applicable.

(1) the waiting line has a single channel.

(2) The pattern of arrivals follows a Poisson probability distribution
(3) The service times arrivals follows an exponential probability distribution
(4) The queue discipline is FIFO.

Operating Characteristic

 = the mean or average number of arrivals per time period (mean Arrival Rate)

 = the mean or average number of services per time period (mean service rate)

(1) the probability that there are no units in the system is


P ( 0)  1 


55
 (2) the average number of units in the waiting line

2
Lq 
 (   )

(3) The average number of units in the system


L  Lq 

(4) The average time a unit spends in the waiting line

Lq
Wq 

(5) The average time a unit spends in the system

1
W  Wq 


(6) Probability that an arriving unit has to wait for service


Pw 

(7) Probability of n units in the system

n

Pn    P0


The above formulas are applicable only when the mean service rate μ is greater than

the mean arrival rate λ. ie when 1


For Hefilis problem since λ = 0,75 customers per minute and μ = 1 customer per
minute.

56
 
P0  1   0.25

2
Lq   2.25
 (   )
L  3 customers
Wq  3 min
W  4 min
Pw  0.75

Manager’s Use Of Waiting

Looking at the results of the single- channel waiting line for Hefilis
Customers wait for 3 minutes before service.
75% of the arriving customers have to wait.
There is probability of 0,1335 that seven or more customers are in the system at one
7

time P7    P0

Hefilis should consider alternative designs or plans for improving the waiting line
operation.

Improving the Waiting Line Operation

Generally service improvements are made along the following lines .

(1) Increase the mean service rate m by making a creative design charge or by
employing new technology.
(2) Add parallel service channels so that more units can be serviced at a time.

Assume Hefilis had considered employing an order filler who will assist the order taker at
the cash register.With this new deign management has estimated the mean service rate to
increase from 60 customers per hour to 75 customer per hour.
Or from 1 customer per min to 1,25 customer/min

Using λ = 0,75 and μ = 1,25 the following results were obtained.

Probability of no units in the system = 0,400
Average number of units in the waiting line = 0,900
Average number of units in the system = 1,500
Average time in the waiting line = 1,2 minuets
Average time in the system = 2.00
Probability on arriving unit has to wait = 0,600
Probability that 7 more are in system = 0,028

57
 Example5.
The reference desk at NUST library receives requests for assistance.Assume that a
poisons probability distribution with a mean of 10 requests per hour can be used to
describe the arrival pattern and that service times follow the exponential probability
distribution with mean service rate of 12 requests per hour.

(1) What is the probability that there are no requests for assistance in the system?
(2) What is the average number of request that will be waiting for service.
(3) What is the probability waiting time in minutes before services begins.
(4) What is the average time at the reference deist in minutes (wait t service).
(5) What is the probability that a new arrival has to wait for service?

5.4 THE MULTIPLE- CHANNEL WAITING LINE MODEL WITH

POISSON ARRIVALS AND EXPONENTIAL

This model consists of two or more channels or service locations that are assumed to
be identical in terms of service capability.

Formulas used in this model use the following assumptions;

(1) The waiting line has two or more service channels.
(2) Arrivals follows a Poisson probability distribution.
(3) Service time for each channel follows an experimental probability distribution.
(4)  (mean service rate) is the same for each channel.
(5) Arrivals wait in a single line and then move to the first open channel for services.
(6) Queue discipline is FIFO

Operating Characteristics
 = the mean arrival rate for the system
 = the mean service rate for each channel
k = No of channels
k = mean service rate for the multiple- channel system.

The formulas following apply if k >

1 The probability that there are no units in the system.

1
P0 
  n
   k
k 1      
      k 

n 0  
 
n! k!  k   
 
 

58
2 The average number of units in the waiting line.

k
   

Lq   P
k  1!k   2 0

(3) The average number of units in the system


L  Lq 

(4) The average time a unit spends in the waiting line

Lq
Wq 

(5) The average time a unit spends in the system

1
W  Wq 


(6) Probability that an arriving unit has to wait for service

1    k  k 
Pw      P0
k!     k   
(7) Probability of n units in the system

n
  

Pn   P0 for n  k
n!

n
  

Pn   n k  P0 for n > k
k! k

Assume that Hefilis wishes to open another cash register and order processing station
so that two customers can be served simultaneously.

We will assume that k= 2,  = 0,75 and  = 1

59
 
P0  1   0.4545

2
Lq   0.1227
 (   )
L  0.8727customers
Wq  0.16 min
W  1.16 min
Pw  0.2045

It can be observed that;

1. The average time required between when you customer enters, the waiting line
and when the customer receivers the order (waiting time + service time) is
reduced from w = 4 min to w = 1,16 min.
2. Length of waiting line is reduced from Lq = 2,25 customer to Lq = 0,1227
customer.
3. Wq from Wq = 3 min to Wq = 0,16
4. The % of customers having to wait for service is reduced from Pw = 0,75 or 75%
to Pw = 0,2045 or 20,45%.

ECONOMIC ANALYSIS OF WAITING LINES

To develop a total-cost model for awaiting line, we begin by defining the following
notation.

Cw = the waiting cost per time period for each unit.

L = the average number of units in the system
Cs = the service cost per time period for each channel.
k = No of channels.
TC = total cost per time period.

TC = Cw + Cs k

In the Hefilis problem , the waiting cost per period would be the cost per minute for a
customer waiting for services.
Although the cost is not a direct cost to Hefilis but of Hefilis ignores this cost and allows
long waitng lines customers will ultimately take their business elsewhere, thus
experiencing lost sale and incur a cost.
Service cost include servers’s wages, benefits, and any other direct costs associated with
operating the service channel.

Lets suppose that Hefilis is willing to assign a cost of $10 per hour for customer waiting
time and service cost is estimated to be $7 per hour.

60
For single –channel system where L = 3 customers

TC Cw L + Cs K
= $10 (3) + $7 (1)
= $37.00 per hour

for Two – channel system (where L = 0,8727 customer

TC = 10(0,8727 ) + 7(2)
= $22,73 per hour

Diagram
600 TC

Service cost

400
Cost ($)

Waiting cost
200

Number of service channels

61