Proceedings of The Fourth International Conference on Informatics & Applications, Takamatsu, Japan, 2015

An Efficient Ant Colony Optimization for Two-dimensional Bin Packing

Problem with Defect Issue
Hao-Chun Lu, Yi-Siao Wan, and Yu-Ting Shueh
Department of Information Management, Fu Jen Catholic University, Taiwan.
076135@mail.fju.edu.tw

ABSTRACT
In various industries, manufacturers must cut
rectangular items from stock sheets. In the actual
setting, stock sheets inevitably contain some defects,
and manufacturers have to avoid cutting the
products which contain the defects (i.e., TFTLCD
industry, e-paper); therefore, the two-dimensional
bin packing problems with defect (2DBPPWD)
emerges. However, no effective method for solving
2DBPPWD in the current literature has been
developed. To address the research gaps and
industrial requirements, this study proposes an
effective approach that integrates ant colony
optimization with the flawless corner space
placement algorithm to solve 2DBPPWD. Finally,
we compare the proposed approach with the
traditional approach (genetic algorithm integrated
with bottomleft placement) to solve C21 (Hopper
&Turton, 2001) and N13 (Burke et al., 2004)
problems. Computational results demonstrate that
our proposed approach can solve 2DBPPWD
effectively and significantly outperform those of
current approaches.

KEYWORDS
Ant colony optimization, two-dimensional bin
packing problems, defect, flawless corner
space placement algorithm.

1. INTRODUCTION
Bin packing problems are a classic optimal
problem in operations research. These
problems have a similar logical structure that
can be considered as a set of rectangular items
being arranged on a set of bins. It has been
studied in various applications in varying
industries, i.e., TFT-LCD substrates cutting
problem[1], thin-film transistor liquid crystal
ISBN: 978-1-941968-16-1 2015 SDIWC

panel cutting problem [2], steel coil cutting

problem [3], steel bridge construction [4],
crosscutting rectangular products from wood
stocks [5], reel and sheet cutting at a paper mill
[6], the relative algorithms [7], etc. These goals
are producing items from the stock sheets in
order to maximize material utilization and
minimize wastage. This optimal problem is
known as a two-dimensional bin packing
problem (2DBPP) in literature [8], and there
are many methods can solve 2DBPP well. For
example, Jakobs [9] designed the well-known
genetic algorithm (GA) integrated bottom-left
placement (BLP) algorithm, Leung et al. [10]
proposed a mixed simulated annealing (SA)
integrated with GA for improving the executing
time, Liu et al. [11] proposed particle swarm
optimization (PSO) integrated bottom-left fill
placement (BLFP) algorithm to solve 2DBPP,
etc.
In fact, the stock sheets inevitably contain
some defects and manufactures have to avoid
the defects for cutting the products in the real
world. As a result, it arise the two-dimensional
bin packing problems with defect (2DBPPWD)
shown in Figure 1. The current methods in
literates are qualified for solving the 2DBPP;
however, there is no method has the ability to
solve the 2DBPPWD effectively. For the
research gaps and industrial requirements, this
study proposes an effective approach that
integrates ant colony optimization (ACO) with
flawless corner space placement (FCSP)
algorithm to solve the 2DBPPWD.

83

Proceedings of The Fourth International Conference on Informatics & Applications, Takamatsu, Japan, 2015

Figure 1. 2DBPP with defect issue

The remainder of this study is organized

as follows. Section 2, classical placement
algorithms are reviewing. Section 3 develops
the FCSP algorithm for conquering the bin with
defect issue. Section 4, we present how to
integrate ACO with FCSP to solve the
2DBPPWD. In Section 5, various numerical
experiments demonstrate the advantages of our
proposed method. Section 6 contains some
concluding remarks

the sliding movements of item as downward or

leftward.
At first, item A will be placed in the top
right of bin, and it move downward until
blocked by the defect 1; therefore, item A move
leftward until blocked by the defect 2. Item A
cannot move downward or leftward anymore
now; then, item B will be placed sequentially.
As showing in Figure 2, item B cannot be
placed in this bin by BLP algorithm. Figure 3
shows the intuitive way of placing items into
same bin shown in Figure 2. Base on the
packing result by BLP algorithm, it is clearly
show that BLP algorithm is inefficient while
dealing with defective case.
3

2
1

B
3

3
2

A
1

3
2

3
2

A
2

A
1

A
1

Figure 2. Place items into bin with defects by BLP

2. PLACEMENT ALGORITHM REVIEW

In this section, we use existing popular
placement algorithms to pack items into the
bins with defects to demonstrate the
inappropriateness and inefficiencies of these
algorithms. In applying the current popular
placement algorithms, we randomly generate
some defects on the bin and treat the defects as
rectangular items fixed on the bin. Then, it is
easy to understand how to place the items
which never overlap with any other items
existing on the bin.

2.1. Bottom Left Placement Algorithm

Bottom Left placement (BLP) algorithm is a
well-known algorithm that takes as input a list
of rectangular items and places each item in
tune into bin. It firstly places the item in the
top-right location and makes successive moves
of sliding it as far down and left as possible.
Figure 2 indicates the inefficiencies of BLP
algorithm in the defective case. The black
squares are representing the defects on the bin.
However, BLP algorithm makes successive
moves of sliding while the defects will block

ISBN: 978-1-941968-16-1 2015 SDIWC

Figure 3. The intuitive way to place items into bin

with defects

2.2. Bottom Left Fill Placement Algorithm

The Bottom Left Fill placement (BLFP)
algorithm is a popular well-known algorithm
that packs the set of items into the bin with
another placing strategy. In the traditional
2DBPP, the performance of BLFP algorithm is
superior to BLP algorithm. The BLFP
maintains a list of location points in a
bottom-left ordering to indicate where the
empty areas may be placed items. When
placing an item, the BLFP algorithm starts with
the lowest and leftmost point for checking
whether the item can be placed with left-bottom
justifying without overlapping any placed items.
If the item can be placed without overlapping,
the item is placed and the point list is updated
to indicate any new points; otherwise, the next
point in the point list is selected until the item
can be placed without overlap occurring.

84

Proceedings of The Fourth International Conference on Informatics & Applications, Takamatsu, Japan, 2015

Hopper and Turton[12] found that BLFP

algorithm outperforms BLP algorithm by up to
25%.
Figure 4shows the items placed into the
bin with defects by BLFP algorithm. According
to the placing rules of BLFP algorithm, it only
considers the item can be placed with the insert
point without overlapping any placed items, but
it is no idea to treat how to place the item with
justifying insert point without overlapping any
defect. Because BLFP algorithm cannot deal
with the defects on the bin, it clearly
demonstrates that BLFP algorithm is
inappropriate for dealing with 2DBPP with
defects. Maybe BLFP algorithm can shift the
item upward or rightward for avoiding the
overlapping areas with defects; however, those
shifting steps will generate many and complex
strategies for considering the situations of
many defects spreading on the bin and the
follow-up related problems.
Combining BLFP with shifting strategy is
not the original BLFP algorithm; therefore, it
can be treated as a best-fit strategy and is
complex enough to develop the specially
algorithm, flawless corner space placement
(FCSP)
algorithm,
for
dealing
with
2DBPPWD.

remaining space of the placed CS into several

new CSs that can contain other small
rectangular items. FCSP algorithm also merges
small CSs to form a larger CS if they can be
merged. This merging step improves the utility
of any unused space on the bin and reduces the
number of fragmented CS generated by the
splitting process of FCSP. The above process
continues until no rectangular items needed be
placed. The main processes of FCSP are shown
in Figure 5 and describe as follows.

Figure 5. Processes of FCSP

(1) Choose: First, wesearch all CSs that can

fit the target item and choose the fittest CS
from CSs to be placed the target item for
minimal waste and reducing unused
fragmented spaces. For example, there are
3 CSs (C1, C2, and C3) that can fit the
item A in Figure6. After evaluating the
choose process, FCSP algorithm will place
item A in C1.If the placed area in C1
contains defects, FCSP algorithm will
continuously move the item upward or
rightward until item can avoid the defects.

Figure 4. Place items into bin with defects by BLFP

3. PROPOSES FLAWLESS CORNER

SPACE PLACEMENT ALGORITHM
This section proposes an FCSP algorithm to
deal with 2DBPPWD. When try to place a
rectangular item into a bin, the FCSP algorithm
only considers the available rectangular spaces
which are tracked by a set of corner spaces
(CSs). Each CS contains the bottom-left
coordinate, width, length of unused rectangular
space, and CS can be overlapped with other CS
existing on the same bin. CSs are core
components in FCSP algorithm. If FCSP
algorithm chooses an appropriate CS for
placing the rectangular item, it will split the

ISBN: 978-1-941968-16-1 2015 SDIWC

Figure 6. Example of choose

(2) Split: After placing item into a CS, the

remaining unused space in placed CS will
be split into multiple new CSs. According
to the location of item placing in CS, a
placed CS can be split at most four CSs.
Figure 7 describes one of the splitting
situations.

85

Proceedings of The Fourth International Conference on Informatics & Applications, Takamatsu, Japan, 2015

Figure 7. One of the splitting situations

In Figure 7, the grey area denotes the placing

item, Item A, and the white area is the
remaining unused space in placed CS. FCSP
algorithm considers all four edges of placed
item to decide how to split the remaining
unused space in placed CS.
Step 1: In Figure 8, if the top edge of Item A, a,
is inside the placed CS, FCSP
algorithm splits the topCS off from
placed CS where topCS is type 1 CS
and represents the CS of topCS.
Step 2: In Figure 9, if the right edge of Item A,
d, is inside the placed CS, FCSP
algorithm splits the rightCS off from
placed CS where rightCS is type 2 CS
and represents the CS of rightCS.
Step 3: Because the bottom edge of Item A, b,
is not inside the placed CS, FCSP
algorithm dose not splits any CS.
Step 4: Because the left edge of Item A, c, is
not inside the placed CS, FCSP
algorithm dose not splits any CS.
After considering all edges of Item A,
Figure 10 demonstrates that the original placed
CS is removed and the new CSs,topCS and
rightCS, will be inserted to the set of CSs.

(3) Adjust: There are two kinds of adjusting,

that is, type 1 adjusting and type 2 adjusting.
Type1 adjusting is horizontally extending for
type1 CS, and type2 adjusting is vertically
extending for type2 CS. Figure 11 shows the
adjusting of type 1 for the two type1 CSs, C1
and C2. Because C2completely contains C1,
the size of C1 should be extended to the same
as size of C2.

Figure 11. Type1adjust

(4) Merge: After finishing all adjustings, there

maybe exists some CSs which have the
same coordinate,width , and height. In this
process, FCSP algorithm only keeps one
of the CSs which have the same
properties.

4. INTEGRATING ACO WITH FCSP

In traditional ACO, ants represent traveller to
visit each city one by one until all cities have
been visited. In this study of 2DBPPWD, we
use item ID represent the city and the
reciprocal of items height as distance.
Therefore, we will get a sequence of item ID
when ant completely visited all cities. After the
sequence of item ID placed by FCSP, FCSP
algorithm can evaluate the fitness of the
sequence for updating the pheromones on the
trail.

5. NUMERICAL EXPERIMENTS

Figure 8. Split the

topCS off

Figure 9. Split the

rightCS off

Figure 10. All CSs

ISBN: 978-1-941968-16-1 2015 SDIWC

We use the C21 problems [13] and the N13

problems [14] as benchmark to compare the
performances between ACO-FCSP and
GA-BLP. Because the C21 and N13 do not
have defects, we add some defects into bin in
this experiment. The executing program is
coded by C# with Visual Studio 2012, and all
experiments were run on PC with Intel
Core2 Duo E6320Processor and 4GB RAM.
Both of GA-BLP and ACO-FCSP cannot
obtain any result of N13 within 10 hours;
therefore, N13 does not compare in this study.
Table 1 demonstrates that the proposed

86

Proceedings of The Fourth International Conference on Informatics & Applications, Takamatsu, Japan, 2015

ACO-FCSP can obtain better solution in every

case except N13, it is clear to show that

ACO-FCSP is more effective and efficient than

GA-BLP.

Table 1. Experiment Result

24

N1

10

20

60

64

C1P2

17

34

38

24

N2

20

40

78

54

C1P3

16

32

57

23

N3

30

60

63

55

C2P1

25

50

41

17

N4

40

80

95

84

C2P2

25

50

19

17

N5

50

100

179

110

C2P3

25

50

43

17

N6

60

120

275

106

C3P1

28

56

52

33

N7

70

140

140

107

C3P2

29

58

58

35

N8

80

160

129

86

C3P3

28

56

39

33

N9

100

200

270

161

C4P1

49

98

79

67

N10

200

400

427

235

C4P2

49

98

83

67

N11

300

600

299

157

C4P3

49

98

85

67

N12

500

2000

470

450

C5P1

73

146

159

99

N13

3152

36000

N/A

N/A

C5P2

73

146

163

100

C5P3

73

146

158

100

C6P1

97

194

159

131

C6P2

97

194

170

127

C6P3

97

194

167

128

C7P1

196

392

316

276

C7P2

197

394

366

255

C7P3

176

352

327

263

Best

6. CONCLUSIONS
This study proposes an efficient ACO that
incorporates a novel placement algorithm,
FCSP, to solve 2DBPPWD effectively. The
results
of
computational
experiments
demonstrate that our proposed approach can
solve 2DBPPWD effectively and significantly
outperforms the current approaches.
REFERENCES

ISBN: 978-1-941968-16-1 2015 SDIWC

Executing

No of item

25

Best

time(s)

Problem

32

Executing

No of item
16

time(s)

Problem
C1P1

GA-BLP ACO-FCSP

GA-BLP ACO-FCSP
Best

Best

[1] J. F. Tsai, P. L. Hsieh, and Y. H. Huang, An

optimization algorithm for cutting stock problems
in the TFT-LCD industry, Computers & Industrial
Engineering, vol. 57, pp. 913919, 2009.
[2] H. C. Lu, Y. H. Huang, and K. A. Tseng, An
integrated algorithm for cutting stock problems in
the thin-film transistor liquid crystal display
industry, Computers & Industrial Engineering, vol.
64(4), pp. 10841092, 2013.
[3] W. Zheng ,P. Ren, P. Ge, Y. Qiu, and Z. Liu,
Hybrid heuristic algorithm for two-dimensional
steel coil cutting problem, Computers & Industrial
Engineering, vol. 62(3), pp. 829838, 2011.

87

Proceedings of The Fourth International Conference on Informatics & Applications, Takamatsu, Japan, 2015

[4] Y. Cui and Y. Lu, Heuristic algorithm for a cutting

stock problem in the steel bridge construction,
Computers & Operations research, vol. 36(2), pp.
612622, 2009.
[5] M. P. Reinders, Cutting stock optimization and
integral production planning for centralized wood
processing, Mathematical and Computer Modeling,
vol. 16(1), pp. 3755, 1992.
[6] M. H. Correia, J. F. Oliveira, and J. S. Ferreira,
Reel and sheet cutting at a paper mill, Computers
& Operations Research, vol. 21(8), pp. 12231243,
2004.
[7] T. A. De Queiroz, F. K. Miyazawa, Y.
Wakabayashi, and E. C. Xavier, Algorithms for
3D guillotine cutting problems: Unbounded
knapsack, cutting stock and strip packing,
Computers & Operations research, vol. 39(2), pp.
200212, 2012.
[8] A. Lodi, S. Martello, and M. Monaci,
Two-dimensional packing problems: A survey,
European Journal of Operational Research, vol. 141,
pp. 241252, 2002.
[9] S. Jakobs, On genetic algorithms for the packing
of polygons, European Journal of Operational
Research, vol. 88, pp. 165181, 1996.
[10] T. W. Leung, C. K. Chan, and M. D. Troutt,
Application
of
a
mixed
simulated
annealing-genetic algorithm for a 2D packing
problem, European Journal of Operational
Research, vol. 145(3), pp. 530542, 2003.
[11] D. S. Liu, K. C. Tan, C. K. Goh, and W. K. Ho,
On solving multi objective bin packing problems
using particle swarm optimization, presented at
IEEE Congress on Evolutionary Computation,
Vancouver, BC, July 1621, 2006, pp.2095 2102.
[12] E. Hopper and B. C. H. Turton, A review of the
application of meta-heuristic algorithms to 2D Strip
Packing Problems, Artificial Intelligence Review,
vol. 16, pp. 257300, 2001.
[13] E. Hopper and B. C. H. Turton, An empirical
investigation of meta heuristic and heuristic
algorithms for a 2D packing problem, European
Journal of Operational Research, vol. 128, pp.
3457, 2001.
[14] E. K. Burke, G. Kendall, and G. Whitwell, A new
placement heuristic for the orthogonal stock cutting
problem, Operational Research, vol. 52 (4), pp.
655671, 2004.

ISBN: 978-1-941968-16-1 2015 SDIWC

88