Professional Documents
Culture Documents
304980
304980
A THESIS SUBMITTED TO
THE GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCES
OF
MIDDLE EAST TECHNICAL UNIVERSITY
BY
H. SİNEM GÜNSEL
FEBRUARY 2012
i
Approval of the thesis:
Date: 09.02.2012
ii
I hereby declare that all information in this document has been
obtained and presented in accordance with academic rules and ethical
conduct. I also declare that, as required by these rules and conduct, I
have fully cited and referenced all material and results that are not
original to this work.
Signature :
iii
ABSTRACT
Günsel, H. Sinem
iv
ÖZ
Günsel, H.Sinem
v
To my family and my love
vi
ACKNOWLEDGEMENTS
Finally, I must express my great thanks to Emrah Günsel, for his love,
friendship, patience, support and holding me whenever I fall down. Without
his understanding and care, such a work will not exist.
vii
TABLE OF CONTENTS
ABSTRACT .................................................................................................iv
ÖZ ................................................................................................................ v
ACKNOWLEDGEMENTS ...........................................................................vii
TABLE OF CONTENTS.............................................................................viii
CHAPTERS
1 INTRODUCTION ....................................................................................... 1
viii
4.1 PROBLEM DEFINITION .............................................................19
4.1.1 Assumptions ........................................................................20
4.1.2 Formulation ..........................................................................21
4.1.3 Parameters ..........................................................................21
4.1.4 Decision Variables ...............................................................23
4.1.5 Model ...................................................................................24
4.2 SAMPLE PROBLEM ..................................................................26
7 CONCLUSION .........................................................................................48
REFERENCES ............................................................................................50
APPENDICES
A GLOSSARY ............................................................................................52
ix
D.1 Test1: 2 Magazines – N1 – Mixed – 2_D1..............................57
D.2 Test2: 2 Magazines – N1 – Mixed – 2_D2..............................58
D.3 Test3: 2 Magazines – N1 – Successive – 2_D1 .....................59
D.4 Test4: 2 Magazines – N1 – Successive – 2_D2 .....................60
D.5 Test5: 2 Magazines – N2 – Mixed – 2_D1..............................61
D.6 Test6: 2 Magazines – N2 – Mixed – 2_D2..............................62
D.7 Test7: 2 Magazines – N2 – Successive – 2_D1 .....................63
D.8 Test8: 2 Magazines – N2 – Successive – 2_D2 .....................64
D.9 Test9: 4 Magazines – N1 – Mixed – 4_D1..............................65
D.10 Test10: 4 Magazines – N1 – Mixed – 4_D2..........................66
D.11 Test11: 4 Magazines – N1 – Successive – 4_D1 .................67
D.12 Test12: 4 Magazines – N1 – Successive – 4_D2 .................68
D.13 Test13: 4 Magazines – N2 – Mixed – 4_D1..........................69
D.14 Test14: 4 Magazines – N2 – Mixed – 4_D2..........................70
D.15 Test15: 4 Magazines – N2 – Successive – 4_D1 .................71
D.16 Test16: 4 Magazines – N2 – Successive – 4_D2 .................72
x
LIST OF TABLES
TABLES
Table 4.2.1: Presence of projectiles in magazine 1 .....................................26
Table 4.2.2: Presence of projectiles in magazine 2 .....................................26
Table 4.2.3: Travel time from node i to node j in magazine 1 ......................28
Table 4.2.4: Travel time from node i to node j in magazine 2 ......................28
Table 4.2.5: Unloading projectile order for sample problem ........................29
Table 6.4.1.1: Test Sets for Test Set1 .........................................................39
Table 6.4.1.2: Demands for projectile types used in Test Sets ...................40
Table 6.4.1.3: N1 Magazine Situation .........................................................40
Table 6.4.1.4: N2 Magazine Situation .........................................................40
Table 6.4.2.1: Test Sets for Test Set 2 ........................................................41
Table 6.4.2.2: Demands for projectile types used in Test Sets ...................42
Table 6.5.1.1: Comparison for Mixed Unloading Scenario ..........................44
Table 6.5.1.2: Comparison for Successive Unloading Scenario..................46
Table 6.5.2.1: Comparison for Test Set 2....................................................47
Table B-1: Calculation of distances between i and j in magazines ..............53
Table D1-1: Test 1 GAMS Solution .............................................................57
Table D1-2: Test 1 Heuristics’ Solution (Mixed) ..........................................57
Table D2-1: Test 2 GAMS Solution .............................................................58
Table D2-2: Test 2 Heuristics’ Solution (Mixed) ..........................................58
Table D3-1: Test 3 GAMS Solution .............................................................59
Table D3-2: Test 3 Heuristics’ Solution (Successive – Magazine) ..............59
Table D3-3: Test 3 Heuristics’ Solution (Successive – Overall) ..................59
Table D4-1: Test 4 GAMS Solution .............................................................60
Table D4-2: Test 4 Heuristics’ Solution (Successive – Magazine) ..............60
Table D4-3: Test 4 Heuristics’ Solution (Successive – Overall) ..................60
xi
Table D5-1: Test 5 GAMS Solution .............................................................61
Table D5-2: Test 5 Heuristics’ Solution (Mixed) ..........................................61
Table D6-1: Test 6 GAMS Solution .............................................................62
Table D6-2: Test 6 Heuristics’ Solution (Mixed) ..........................................62
Table D7-1: Test 7 GAMS Solution .............................................................63
Table D7-2: Test 7 Heuristics’ Solution (Successive – Magazine) ..............63
Table D7-3: Test 7 Heuristics’ Solution (Successive – Overall) ..................63
Table D8-1: Test 8 GAMS Solution .............................................................64
Table D8-2: Test 8 Heuristics’ Solution (Successive – Magazine) ..............64
Table D8-3: Test 8 Heuristics’ Solution (Successive – Overall) ..................64
Table D9-1: Test 9 GAMS Solution .............................................................65
Table D9-2: Test 9 Heuristics’ Solution (Mixed) ..........................................65
Table D10-1: Test 10 GAMS Solution .........................................................66
Table D10-2: Test 10 Heuristics’ Solution (Mixed) ......................................66
Table D11-1: Test 11 GAMS Solution .........................................................67
Table D11-2: Test 11 Heuristics’ Solution (Successive – Magazine) ..........67
Table D11-3: Test 11 Heuristics’ Solution (Successive – Overall) ..............67
Table D12-1: Test 12 GAMS Solution .........................................................68
Table D12-2: Test 12 Heuristics’ Solution (Successive – Magazine) ..........68
Table D12-3: Test 12 Heuristics’ Solution (Successive – Overall) ..............68
Table D13-1: Test 13 GAMS Solution .........................................................69
Table D13-2: Test 13 Heuristics’ Solution (Mixed) ......................................69
Table D14-1: Test 14 GAMS Solution .........................................................70
Table D14-2: Test 14 Heuristics’ Solution (Mixed) ......................................70
Table D15-1: Test 15 GAMS Solution .........................................................71
Table D15-2: Test 15 Heuristics’ Solution (Successive – Magazine) ..........71
Table D15-3: Test 15 Heuristics’ Solution (Successive – Overall) ..............71
Table D16-1: Test 16 GAMS Solution .........................................................72
Table D16-2: Test 16 Heuristics’ Solution (Successive – Magazine) ..........72
Table D16-3: Test 16 Heuristics’ Solution (Successive – Overall) ..............72
xii
LIST OF FIGURES
FIGURES
Figure 1.1 - T-155 mm self-propelled Fırtına (Storm) howitzer..................... 1
Figure 2.1.1 - T-155 mm Self-Propelled Fırtına (Storm) Howitzer ................ 5
Figure 2.1.2 - Ammunition Resupply Vehicle (ARV) ..................................... 5
Figure 2.1.3 - Projectile Magazine Units....................................................... 6
Figure 2.1.4 - Horizontal Projectile Transfer Unit ......................................... 6
Figure 2.1.5 - Vertical Projectile Transfer Unit.............................................. 7
Figure 2.1.6 - External Conveyor Unit .......................................................... 7
Figure 2.2.1 - The snapshot of the screen when a requirement occurs ......10
Figure 2.2.2 - The snapshot of the screen when a plan is done ..................10
Figure 3.1.1 – A Hamiltonian tour................................................................12
Figure 3.2.1.1 – A nearest neighbor tour for 100 nodes ..............................16
Figure 3.2.2.1 - 2-opt exchange ..................................................................18
Figure 4.1.3.1 – Calculation of t12.222 ..........................................................23
Figure 4.2.1 - Presence of projectiles in magazines ....................................27
Figure 6.1.1 - The snapshot of GAMS code user interface .........................36
Figure 6.1.2 - The snapshot of GAMS output user interface .......................36
Figure 6.2.1 – The graphical user interface of the heuristic algorithm .........37
xiii
LIST OF SYMBOLS AND ABBREVIATIONS
xiv
CHAPTER 1
INTRODUCTION
As the technology has developed, Turkish Army get advanced in the field of
arms and started to make their own weapons. The first artillery weapon
system produced in Turkey is T-155 mm self-propelled Fırtına (Storm)
howitzer. This howitzer has the capability of rapid movement in rough
terrains and can open fire very quickly. Each howitzer can carry 48
ammunitions with itself and can spend these ammunitions in 6 minutes with
a maximum rate of fire. So supplying ammunitions to these howitzers
becomes very crucial in warfare. To complete this necessity, Ammunition
Resupply Vehicle (ARV) is also being developed in Turkey. The glossary for
the military terminology can be seen in Appendix A.
1
Ammunition Resupply Vehicle (ARV) will be used to meet Fırtına howitzers’
ammunition demand for tactical requirements of higher firing rate (firing
frequency) by off-road mobility and survivability. New ARV’s aim is to
improve the current capabilities of Fırtına’s by operating in groups with
howitzers. The transfer of ammunitions from ARV’s to Fırtına is to be
optimized for an effective improvement of firing rate. In this thesis the
transferring order of carried ammunitions is being optimized to minimize the
total ammunition transferring time. By minimizing the total transferring time,
total movement of the mechanisms in ARV’s will also be minimized.
Minimum movement will also make the ARV long-lived.
Four magazines, each carrying twenty four ammunitions, are similar to the
ninety six nodes to be visited by the gripper. Similar problems are also
solved in literature. Vehicle routing problem (VRP), multiple travelling
salesman problems (m-TSP) are some of them. For example, in TSP the
aim is to find the cheapest tour that contains all cities starting and returning
the origin. However, in our problem the aim is to find the cheapest tour
starting from the origin of the grippers but not all nodes are necessary. So
the problem is modeled as a variant of TSP in this thesis. For ARV, four
magazines are assumed to be independent from each other. The demand of
Fırtına howitzers from different types and quantities of ammunitions are
given. Here, four salesmen are supposed to visit four magazines’ nodes to
satisfy a total number of demanded ammunitions.
2
In this thesis, the problem is modeled as TSP and solved with GAMS, a
small version of the problem with 2 magazines is solved optimally. On the
other hand, the problem with 4 magazines can only be solved partially and a
local optimum could be found after an out of memory error. The obtained
solutions show the order of ammunitions to be transferred from the ARV.
Not only the mixed ammunition type ordering considered, but also the
successive unloading scenarios are included in this thesis.
3
CHAPTER 2
2.1 BACKGROUND
4
Figure 2.1.1 - T-155
T mm Self-Propelled
elled Fırtına (Storm) Howitzer
5
ammunitions automatically with a computerized control system. The
Automatic Ammunition Transfer System (AATS) which has been developed
by ASELSAN Inc. consists of four projectile magazine units (each of
capacity is 24 projectiles), a horizontal projectile transfer unit with a gripper
on it, a vertical projectile transfer unit, an external conveyor unit as well as
electronic units, which will not be mentioned in this thesis.
6
Figure 2.1.5 - Vertical Projectile Transfer Unit
7
The ARV is capable of transferring 96 projectiles and 96 charges to T-155
Fırtına howitzers. In this thesis only projectiles will be focused since they are
loaded and transferred by the automation system.
The ammunitions are loaded to the ARVs from Ammunition Supply Points
(ASP). The external conveyor unit extends over the ASP and the projectiles
are put on it manually. The conveyor starts to move through the ARV. At the
end of the conveyor, the projectile is given to the Vertical Projectile Transfer
Unit and then to the Horizontal Projectile Transfer Unit. The Horizontal
Projectile Transfer Unit gives it to the planned Projectile Magazine Unit’s
planned node. While new projectiles are being loaded, the previous ones
move to the backsides as Projectile Magazine Unit turns right or left.
8
Projectile Transfer Unit when the magazine brings it to the front. The
Horizontal Projectile Transfer Unit gives it to the vertical one and the vertical
one gives it to the External Conveyor Unit. External Conveyor Unit extends
over the front of the ARV and lines up with the rear of the howitzer.
2.2 MOTIVATION
With this current system which can be replaced by this thesis algorithm, the
projectile transferring times to howitzer will be more than the howitzer firing
rate. As explained earlier, Fırtına howitzers have really high firing rates by
off-road mobility and survivability. In the warfare, supplying the ammunitions
within the required time and location is very crucial. Since at the transferring
moment the howitzers cannot fire to targets, the transferring should be as
quick as possible. In this thesis, main aim is to speed up this transferring
time to provide capability for rapid deployment without being exposed to
hostile units counter-firing thus to enable higher firing rate.
The snapshots of the commander screen in the current system can be seen
in Figure 2.2.1 and Figure 2.2.2.
9
Figure 2.2.1 - The snapshot of the screen when a requirement occurs
10
CHAPTER 3
LITERATURE SURVEY
11
Figure 3.1.1 – A Hamiltonian tour
Subject to:
= 1, ∀ Є {1, 2, . . . , }
= 1 , ∀Є {1, 2, . . . , }
≤ ││ − 1, ∀ ⊆ Є {2, . . . , }
Є {0, 1}
12
TSP has many applications like;
• Drilling of printed circuit boards: the holes that are drilled are the
nodes to be visited. The distances between nodes are calculated as
the time to move the head from one node to another.
13
application of mTSP can be applied by changing the number of salesmen.
The number of salesmen can be bounded variable or fixed. Also without a
boundary, each salesman can increase the cost and salesman cost can be
also included in the problem. If the cities to be visited have a time limitation
and if the salesman should obey a time boundary to visit each city, the
problem is called mTSP with time windows in the literature. In fact cargo
distribution for food stuff can be an example. Another application of mTSP
can be applied by bounding the number of cities each salesman should visit.
Minimum or maximum bound for number of cities or minimum or maximum
distance bound to be visited can be applied [10].
14
companies, personnel/school bus routing, etc. Many exact and approximate
algorithms are developed in the literature to solve these problems [11].
3.2 HEURISTICS
The TSP is known to play an important role for many of the most widely
used clever algorithms, heuristics that run quicker and give better solutions
for the given problems. Since TSP is a basic model, different heuristics are
developed and tested for better tours [13]. The heuristics can be
categorized by two; construction and improvement.
To start a search in the space for optimization, an initial and feasible point is
needed. By adding nodes or arcs to a starting set, a tour is constructed.
Here are some of them:
15
In this thesis, time is the main concern. Due to the structure of the problem,
nearest neighbor heuristic satisfies the requirements.
16
Node: k=3 NNs
1 2-7-8
2 1-3-6
3 4-1-5
4 2-1-5
Y Y
• Lin-Kernighan Heuristic
17
finding a smaller cost, the tour is said to be 2-optimal [14]. Figure 3.2.2.1
illustrates the 2-opt mechanism [17].
18
CHAPTER 4
19
This problem is modeled as a version of TSP. The gripper of the horizontal
projectile transfer unit is seen as the traveling salesman where the nodes of
the projectile magazine units are seen as the cities to be visited. However
there is a few striking difference that everyone can notice at first look.
• In classical TSP; the aim is to find a tour which visits all the cities with
minimum cost. On the other hand, in our problem; the aim is to find a
tour which visits cities until all demands (needed number of
projectiles) are satisfied again with minimum cost.
• In classical TSP; cities are static and the traveling salesman goes to
the cities. However in our problem, cities are not static as expected.
In reality going to a city in our problem means; gripper is going to the
front side of the selected projectile magazine unit and then projectile
which is presented at selected node is coming towards the front side
of its magazine unit.
4.1.1 Assumptions
• The howitzer assignments are done by FDC one by one since during
the transferring time for one howitzer, the rest demanding howitzers
generally change their locations. For every howitzer visit, the problem
is resolved.
20
is received, the selection of the Projectile Magazine Units
appropriately will improve the situation.
4.1.2 Formulation
Indices are;
4.1.3 Parameters
21
• Npjm = binary parameter that shows the presence of projectile p in
node j of magazine m,
tijm = 2.hm + Ѳ . min {│j - i│, 24 - │j – i│}. (For all mixed scenario and
successive scenario where i and j have same projectile types.)
• t0jm gives us the first movement time of grippers, since we know the
front nodes of magazines, we can easily determine the value of t0jm.
22
the origin in h2 seconds. The time from node 12 to node 22 in
magazine 2 is illustrated in Figure 4.1.3.1.
Gripper
The decision made in this thesis is the order of unloading the projectiles
demanded. Thus, decision variables show the order of these projectiles.
Here is the definition of the binary decision variables:
v
23
4.1.5 Model
X . t
(1)
" !
Subject to:
"'
%& = 1, ∀( (2)
"!
%& = 0 , ∀( (3)
!)*
"!
%& = 1, ∀( (4)
!
"!
% ≤ 1, ∀ , ∀( (5)
!
"!
% ≤ 1, ∀ , ∀( (6)
!
"! −
"!/'
= 0, ∀ ≥ 1 , ∀1 ≥ 2, ∀( (7)
"!
% = 78 , ∀9 (8)
" ! 6
Є345
"!
% ≤ ;<=7" , ∀( (9)
!
"! "
% ≤ ?8 , ∀ , ∀( (10)
! 8
"!
% Є {0, 1} (11)
24
• Objective function (1) minimizes the total time.
• Constraint set (2) ensures that for every magazine the grippers leave
their origins (the starting point, node 0), at state 1.
• Constraint set (3) restricts the leaving node 0 at only state 1. After
state 2, the grippers cannot leave the starting point for each
magazine.
• Constraint set (4) returns the grippers to origin exactly once, i.e.
returning node 0 is compulsory.
• Constraint set (5) ensures that departure from node i to any node in
that magazine cannot be greater than 1 for each magazine.
• Constraint set (6) ensures that arrival to node j from any node in that
magazine cannot be greater than 1 for each magazine.
• Constraint set (7) balances the incoming and outgoing nodes. If there
exists a coming arc to a node at state s-1 then there should be a
departuring arc also at state s. This constraint excludes node 0 since
coming to node completes the tour for that magazine.
This model is solved in GAMS. The GAMS formulation of this model can be
seen in Appendix C.
25
4.2 SAMPLE PROBLEM
• i, j Є {0, 1, 2,Y,6}
• m Є {1, 2}
• s Є {1, 2, 3,Y,7}
• p Є {a, b}
• da = 6, db = 5
Npj1 0 1 2 3 4 5 6
a 0 0 0 1 0 1 1
b 0 1 1 0 1 0 0
N pj2 0 1 2 3 4 5 6
a 0 1 0 0 1 0 1
b 0 0 1 1 0 0 0
26
• The presence of projectile types can be seen in Figure 4.2.1:
4
Figure 4.2.1
4 - Presence of projectiles in magazines
• Load1 = 6, load2 = 5
• h1 = 1, h2 = 2,
2 Ѳ=1
27
Table 4.2.3: Travel time from node i to node j in magazine 1
tij1 0 1 2 3 4 5 6
0 0 4 3 2 3 4 5
1 0 100 3 4 5 4 3
2 0 3 100 3 4 5 4
3 0 4 3 100 3 4 5
4 0 5 4 3 100 3 4
5 0 4 5 4 3 100 3
6 0 3 4 5 4 3 100
tij2 0 1 2 3 4 5 6
0 0 6 5 4 5 6 7
1 0 100 5 6 7 6 5
2 0 5 100 5 6 7 6
3 0 6 5 100 5 6 7
4 0 7 6 5 100 5 6
5 0 6 7 6 5 100 5
6 0 5 6 7 6 5 100
28
Table 4.2.5: Unloading projectile order for sample problem
s 0 1 2 3 4 5 6 7
m
Magazine1 0 4 3 2 1 6 5 0
Magazine2 0 4 3 2 1 6 0
The unloading can be done starting from magazine 2 and continuing from
magazine 1 or starting from magazine 1 and continuing from magazine 2.
One projectile from magazine 1 and one projectile from magazine 2 can
also be applicable in the problem.
29
CHAPTER 5
HEURISTIC ALGORITHM
30
In this chapter; nearest neighbor method as construction and 2-opt method
as improvement algorithm is explained in detail.
For the mixed unloading scenario, steps of the algorithm are explained
below:
Step 5: Choose the projectile with minimum distance to the origin of the
gripper and add it to the ‘unloading order’ list. Add it’s time to the
‘unloading time’.
31
For the successive unloading scenario, steps of the algorithm are explained
below:
Step 4.1: Construct lists for each demanded projectile type with the
information of in which magazine and node they present.
Step 4.2: Select the order of the projectile types. (for ex. ABC)
Step 4.3: Choose the necessary projectile from the list with minimum
distance to the origin of the gripper and add it to the ‘unloading
order’ list. Add it’s time to the ‘unloading time’.
Step 4.4: Update the initial position of the magazine by the chosen
projectile’s node.
Step 4.6: If demand of all projectiles becomes 0, continue with Step 4.1
for the next permutation of the demanded projectile types.
To compare the outputs with GAMS results, a penalty should also be added
in successive loading scenario. To be consistent with the GAMS,
successive-magazine method is applied. In this method penalty is added at
Step 4.5 for each projectile type change in the magazine orders. On the
32
other hand, to reflect the real life, successive-overall method which
generates penalty once only for each projectile type changes overall is
applied at Step 4.6.
5.2 2- OPT
Steps of the algorithm for both mixed and successive scenarios are like as
follows:
Step 2: Create pairs with two consecutive projectiles, remove their edges,
invert and combine them to the tour again. Check the ‘unloading
time’, if it is better than the best obtained time, add to the solution
set. Repeat Step 2 for all duos.
33
Step 5: From the list that constructed in Step 2, new lists are formed as
List1, List2 and List3. List1 is the same with the list in Step 4 (b, b,
b, c, a, a, Y). List2 contains two projectile pairs and their
combinations from the list in Step 4 (bb, bb, bc, ca, ac, aa, Y). List
3 contains three projectile pairs and their combinations. (bbb, bbc,
bba, bab, caa, Y)
Step 6: Remove pairs from the constructed tour and try all pairs in List1,
List2 and List3. Take their solutions one by one, if it is better than
the best obtained time, add to the solution set.
Step 7: Show the minimum of the solution set as the best solution.
34
CHAPTER 6
COMPUTATIONAL EXPERIMENTS
6.1 GAMS
35
Figure 6.1.1 - The snapshot of GAMS code user interface
36
6.2 HEURISTIC ALGORTIHM
37
6.3 TEST ENVIRONMENT
As a test environment, at first step different machines are used for GAMS.
To cope with the GAMS’ high usage of memory, best hardware is chosen
and all the computational experiments are run on it. The machine is an
Intel(R) Xeon(R) CPU X5650, 2.67 GHz 2 processors desktop computer
with 4 GB RAM and 64-bit operating system. The algorithm is also run on
the same machine.
Two different test sets are constructed in this thesis. In test set 1; the
comparison of GAMS and heuristic for mixed and successive unloading
scenario is done. In test set 2, the behavior of the heuristic for different
demands for mixed and successive unloading scenario is analyzed. For
both of test sets, same magazine situations; N1 and N2 are used. In N1
type, approximately %50 and in N2 type, approximately %85 of the
magazines are full.
In Test Set 1 the experiments are done using the following data sets:
• 2 magazines
• 4 magazines
38
The output method;
Different demands
Test no M N T D
1 2_Demand1
Mixed
2 2_Demand2
N1
3 2_Demand1
Successive
4 2_Demand2
2 magazine
5 2_Demand1
Mixed
6 2_Demand2
N2
7 2_Demand1
Successive
8 2_Demand2
9 4_Demand1
Mixed
10 4_Demand2
N1
11 4_Demand1
Successive
12 4_Demand2
4 magazine
13 4_Demand1
Mixed
14 4_Demand2
N2
15 4_Demand1
Successive
16 4_Demand2
39
Table 6.4.1.2: Demands for projectile types used in Test Sets
Demands a b c d
2_Demand1 3 6 5 4
2_Demand2 2 7 4 4
4_Demand1 8 7 3 5
4_Demand2 9 10 8 9
The initial positions are taken for all test sets are; 3, 5, 3, 8 respectively for
magazines 1, 2, 3, 4.
These tests are run in GAMS and heuristic for mixed and successive (both
for successive magazine and successive overall) unloading scenarios. The
40
results for Test Set 1 can be seen in tables of Appendix D. The results
analysis is done in part 6.5.1.
In Test Set 2 the experiments are done using the following data sets:
• 4 magazines
Different demands
Test no M N T D
1 4N1_Demand 1-5
Mixed
2 4N1_Demand 6-10
N1
3 4N1_Demand 1-5
Successive
4 4N1_Demand 6-10
4 magazine
5 4N2_Demand 1-5
Mixed
6 4N2_Demand 6-10
N2
7 4N2_Demand 1-5
Successive
8 4N2_Demand 6-10
41
Demands are taken as below:
Demands a b c d Total
4N1_Demand 1 5 5 5 5 20
4N1_Demand 2 3 8 8 1 20
4N1_Demand 3 8 2 1 9 20
4N1_Demand 4 7 7 3 3 20
4N1_Demand 5 2 3 3 12 20
4N1_Demand 6 10 10 11 9 40
4N1_Demand 7 5 12 11 12 40
4N1_Demand 8 10 12 8 10 40
4N1_Demand 9 9 9 11 11 40
4N1_Demand 10 8 12 10 10 40
4N2_Demand 1 9 9 9 9 36
4N2_Demand 2 15 15 3 3 36
4N2_Demand 3 4 4 16 12 36
4N2_Demand 4 9 9 16 2 36
4N2_Demand 5 3 3 15 15 36
4N2_Demand 6 18 18 18 18 72
4N2_Demand 7 16 20 20 16 72
4N2_Demand 8 13 21 20 18 72
4N2_Demand 9 18 21 16 17 72
4N2_Demand 10 18 21 21 12 72
The initial positions are taken for all test sets are; 3, 5, 3, 8 respectively for
magazines 1, 2, 3, 4.
Magazine situations and movement times are same with Test Set 1.
The tests in Test Set 2 are run only in heuristic for mixed and successive-
overall unloading scenarios. Successive-magazine is omitted since GAMS
is not used in this set. The results analysis is done in part 6.5.2.
42
6.5 TESTS ANALYSIS
For the mixed unloading scenarios in Test Set 1, GAMS generally finds
global optimals. For test 13 and 14, since the number of variables increases
to 57.500, GAMS cannot find global optimal. The program gives out of
memory error approximately 1 hour later.
The heuristic also finds very good solutions in a short time when compared
to GAMS. At test 14, GAMS finds a local optimal as 172, on the other hand
the heuristic finds 152 in the construction part and improves it to 149 less
than half minute. In fact, this result is expected at the beginning. Due to the
polyhedral structure of the magazines, the nearest neighbor heuristic should
definitely give good results. Instead of a metaheuristic like Genetic
Algorithms, the heuristic algorithm developed for this thesis gives better
results although its simplicity.
In table 6.5.1.1 the costs and CPU times can be seen for construction
heuristic, improvement heuristic and GAMS one by one. As expected, after
construction, 5 of 8 tests could not be improved. The other 3 tests; test 1,
test 10 and test 14 improves 1,46%, 0,68%, 1,97% respectively. This shows
us that nearest neighbor heuristic works well for this problem type.
Another point that attracts attention about the tests is; for the cases with
same number of demands when the load of the magazines increases, the
costs decreases. To illustrate pair test 1 - test 5, test 9 - test 13 can be
analyzed. Test 1 and test 5, test 9 - test 13 have the same demand with
43
different magazine loads. Test 5 and test 13 give lower best costs than test
1 and test 9.
Final analysis about the tests is about the main reason of the thesis. The
main aim of the thesis is to decrease the transferring time of the current
system. As example, test 9 and test 13 are taken. (Since the software is not
completed the current system’s results are obtained manually.) For test 9
and 13, the current system gives 231 and 233 respectively. This means the
thesis reach its objective by improving the current system for test 9 and 13
by 31% and 57%.
Contrary to the mixed unloading scenario, the GAMS get difficulty for finding
global optimal (only 3 of 8 tests are global optimal). The program again
gives out of memory error. Although the program finds local optimum, it
generally gives the unloading order except test 12. It gives 589 as cost
however the results are not shown. This test is run on different machines to
see the order, however all machines are unsuccessful to show the results.
44
The heuristic again finds very good solutions in a short time when compared
to GAMS. To compare with GAMS, successive-magazine (SM) column
should be used. To see the gap between GAMS column and successive-
overall (SO) column, results can be examined. Successive-overall results
are really different from SM and GAMS.
From the table 6.5.1.2 it is easily recognized, after construction, again tests
could not be improved. This justifies us that nearest neighbor heuristic
works well for this problem type, also for successive unloading scenario.
Another point that attracts attention about the tests of successive unloading
is; for the cases with same number of demands the successive gives higher
costs than mixed ones. It is obviously normal that some punishments cannot
be inevitable.
Final analysis about the tests is again about the current system. As
example, test 11 and test 15 are taken. (Since the software is not completed
the current system’s results are obtained manually.) For test 11 and 15, the
current system gives 365 and 376 respectively. This justifies the thesis
reach its objective by improving the current system for test 11 and 15 by
28% and 47%.
45
Table 6.5.1.2: Comparison for Successive Unloading Scenario
In test set 2, the effect of different demands is put forward. Tests are run in
the heuristic algorithm. The results can be seen in Table 6.5.2.1. In the
table, best values for each test and averages of them are given. For the
mixed and successive overall unloading scenarios, evaluation times do not
differ more than 1 second. For 4N1_Demand1-5 (20 of 45) and
4N2_Demand1-5 (36 of 82) best cost variations are about %20 where
4N1_Demand6-10 (40 of 45) and 4N2_Demand6-10 (72 of 82) best cost
variations are about %0.5. This is because of the demanded / loaded
ammunition quantity ratio. When this ratio gets smaller, the variation
between different demand sets increases as expected.
46
Table 6.5.2.1: Comparison for Test Set 2
47
CHAPTER 7
CONCLUSION
Our problem assumes the mechanical parts are distinct from each other and
the total time of transferring all demanded projectiles is evaluated as the
cost of problem. So, the unloading time of the ARV is minimized in this
study. Also the unloading order of the projectiles is studied in two scenarios;
the mixed and the successive unloading scenario.
Our experimental results (Test Set 1) show that for small size problems,
GAMS and heuristic algorithm give meaningful results and the heuristic is
48
better for larger problems in terms of computational time and solution
quality. Also it can be said in any way, GAMS or heuristic, the current
system can be improved. Again in tests (Test Set 2), the demanded / loaded
ammunition quantity ratio attracts the attention. It is recognized that when
this ratio gets smaller, the variation between different demand sets
increases as expected.
The aim of this study is to define the mathematical model of ATS and
minimize the unloading time by GAMS and heuristic .The extension of this
study may include the following issues:
49
REFERENCES
[1] Murat Faik Yaren, “Fırtına Howitzer Fire Control System and Ammunition
Resupply Vehicle”, ASELSAN Magazine, 2009
[2] Cor A.J. Hurkens, Gerhard J. Woeginger, “On the nearest neighbor rule
for the traveling salesman problem”, Operations Research Letters 32, 1-4,
2004
[3] E.L. Lawler, J.K. Lenstra, A.HG. Rinnooy Kan, D.B. Shmoys (Eds.) “The
Traveling Salesman Problem: A Guided Tour of Combinatorial
Optimization”, Wiley, Chichester, 1985
50
[10] Tolga Bektaş, “The multiple traveling salesman problem: an overview of
formulations and solution procedures”, Omega 34, 209-219, 2006
[11] Jeroen Wouter Boer, "Approximate models and solution approaches for
the vehicle routing problem with multiple use of vehicles and time windows."
METU M.Sc. thesis, 2008
51
APPENDIX A
GLOSSARY
Artillery: Artillery is the general name of the weapons and soldiers that
throws a projectile with the help of an explosion of change/powder.
Howitzer: Howitzer is the weapon that can shoot a projectile with both low
angle of elevation (like guns) and high angle of elevation (like mortars).
Ammunition: Ammunition is the general name of the material that is used for
each round of weapons shot. So, ammunition can be said to be the
combination of charge/powder (the chemical material which explores to
throw the projectile away), fuse (the mechanical compound that starts the
explosion of the projectile at the target) and projectile (the main part which is
thrown and explores at the target area to harm the target).
Fire Direction Center (FDC): FDC is the commander unit center which is
also an armored vehicle. The decisions of missions and necessary
precautions for the next missions are controlled by FDC. For example,
which Fırtına will shoot to which target and how many projectiles will be shot
is decided by FDC. So, FDC should also decide the demand of each Fırtına
and FDC should order the ARV to transfer the necessary ammunition to
these Fırtına's.
52
APPENDIX B
CALCULATION OF TIJM
m
tij 0 1 K 4 K 11 K 22 K 24
1.0 0 2h1 + 3Ѳ Y 2h1 Y 2h1 + 7Ѳ Y 2h1 + 6Ѳ Y 2h1 + 4Ѳ
1.1 0 M Y 2h1 + 3Ѳ Y 2h1 + 10Ѳ Y 2h1 + 3Ѳ Y 2h1 + 1Ѳ
K 0 Y Y Y Y Y Y Y Y Y
1.4 0 2h1 + 3Ѳ Y M Y 2h1 + 7Ѳ Y 2h1 + 6Ѳ Y 2h1 + 4Ѳ
K 0 Y Y Y Y Y Y Y Y Y
1.24 0 2h1 + 1Ѳ Y 2h1 + 4Ѳ Y 2h1 + 11Ѳ Y 2h1 + 2Ѳ Y M
2.0 0 2h2 + 3Ѳ Y 2h2 Y 2h2 + 7Ѳ Y 2h2 + 6Ѳ Y 2h2 + 4Ѳ
2.1 0 M Y 2h2 + 3Ѳ Y 2h2 + 10Ѳ Y 2h2 + 3Ѳ Y 2h2 + 1Ѳ
K 0 Y Y Y Y Y Y Y Y Y
2.4 0 2h2 + 3Ѳ Y M Y 2h2 + 7Ѳ Y 2h2 + 6Ѳ Y 2h2 + 4Ѳ
K 0 Y Y Y Y Y Y Y Y Y
2.24 0 2h2 + 1Ѳ Y 2h2 + 4Ѳ Y 2h2 + 11Ѳ Y 2h2 + 2Ѳ Y M
3.0 0 2h3 + 3Ѳ Y 2h3 + 6Ѳ Y 2h3 + 11Ѳ Y 2h3 Y 2h3 + 2Ѳ
3.1 0 M Y 2h3 + 3Ѳ Y 2h3 + 10Ѳ Y 2h3 + 3Ѳ Y 2h3 + 1Ѳ
K 0 Y Y Y Y Y Y Y Y Y
3.22 0 2h3 + 3Ѳ Y 2h3 + 6Ѳ Y 2h3 + 11Ѳ Y M Y 2h3 + 2Ѳ
K 0 Y Y Y Y Y Y Y Y Y
3.24 0 2h3 + 1Ѳ Y 2h3 + 4Ѳ Y 2h3 + 11Ѳ Y 2h3 + 2Ѳ Y M
4.0 0 2h4 + 10Ѳ Y 2h4 + 7Ѳ Y 2h4 Y 2h4 + 11Ѳ Y 2h4 + 11Ѳ
4.1 0 M Y 2h4 + 3Ѳ Y 2h4 + 10Ѳ Y 2h4 + 3Ѳ Y 2h4 + 1Ѳ
K 0 Y Y Y Y Y Y Y Y Y
4.11 0 2h4 + 10Ѳ Y 2h4 + 7Ѳ Y M Y 2h4 + 11Ѳ Y 2h4 + 11Ѳ
K 0 Y Y Y Y Y Y Y Y Y
4.24 0 2h4 + 1Ѳ Y 2h4 + 4Ѳ Y 2h4 + 11Ѳ Y 2h4 + 2Ѳ Y M
53
APPENDIX C
GAMS FORMULATION
alias (i,j) ;
$include "D:\Sinem\Tez\model\Test_Sets\TEST13\TEST13_N2.txt" ;
parameters
/a8
b7
c3
d 5/
/ 1 17
2 22
3 21
4 22/;
54
table t(m,i,j) travel time in magazines
$include "D:\Sinem\Tez\model\Test_Sets\TEST13\TEST13_T.txt" ;
binary variables x(i,j,m,s) 1 if arc from node i to node j in magazine m at state s is used
otherwise 0
variable z objective_function
equations
constraint2(m) leaving node 0 cannot happen after state 2 for each magazine
constraint6(j,m,s) if there exists a coming arc to a node then there should be a departuring
arc also(excluding node 0)
objective objective_function ;
constraint4(m,i).. sum((s,j),x(i,j,m,s))=l= 1;
constraint5(m,j).. sum((s,i),x(i,j,m,s))=l= 1;
55
constraint9(m,j)$(ord(j) ge 2).. sum((s,i),x(i,j,m,s)) =l= sum(p,N(m,p,j));
tez.WORKSPACE=3500;
put opt;
'Workmem 30000'/ ;
putclose opt;
56
APPENDIX D
57
D.2 Test2: 2 Magazines – N1 – Mixed – 2_D2
58
D.3 Test3: 2 Magazines – N1 – Successive – 2_D1
59
D.4 Test4: 2 Magazines – N1 – Successive – 2_D2
60
D.5 Test5: 2 Magazines – N2 – Mixed – 2_D1
61
D.6 Test6: 2 Magazines – N2 – Mixed – 2_D2
62
D.7 Test7: 2 Magazines – N2 – Successive – 2_D1
63
D.8 Test8: 2 Magazines – N2 – Successive – 2_D2
64
D.9 Test9: 4 Magazines – N1 – Mixed – 4_D1
65
D.10 Test10: 4 Magazines – N1 – Mixed – 4_D2
66
D.11 Test11: 4 Magazines – N1 – Successive – 4_D1
67
D.12 Test12: 4 Magazines – N1 – Successive – 4_D2
68
D.13 Test13: 4 Magazines – N2 – Mixed – 4_D1
69
D.14 Test14: 4 Magazines – N2 – Mixed – 4_D2
70
D.15 Test15: 4 Magazines – N2 – Successive – 4_D1
71
D.16 Test16: 4 Magazines – N2 – Successive – 4_D2
72