ISSN 1715-7862 [PRINT]

Advances in Natural Science ISSN 1715-7870 [ONLINE]

Vol. 8, No. 3, 2015, pp. 20-26 www.cscanada.net
DOI:10.3968/7342 www.cscanada.org

Improved Adaptive Genetic Algorithm in Optimal Layout of Leather Rectangular

Parts

LIU Hao[a],*; QIU Jianxin[a]

[a]
School of Mechanical Engineering, Shanghai University of optimization technology is essential in the leather industry
Engineering Science, Shanghai, China. and an important means to save raw materials and make
*
Corresponding author.
full use of the resources. The merits of nesting cutting
Received 9 April 2015; accepted 25 August 2015 directly impact on the leather production efficiency and
Published online 26 September 2015 economic benefits. Artificial leather nesting cutting is
generally based on two-dimensional rectangular sample
Abstract rule nesting. For irregular leather nesting, complementary
In the mass customization of Leather products (such sample makes the fight against and filling, cluster
as sofa), the intelligent layout is the key technology to analysis, take the smallest envelope rectangle and
improve material utilization. The paper faces artificial other Pretreatment, irregular sample transformed into a
leather fabric cutting problem, most can be converted into rectangular sample, finally, in the rules of artificial leather
a rectangle packing layout problem. This paper proposes fabrics to optimize nesting. Such as artificial leather sofa
a new improved adaptive genetic algorithm. Crossover nesting cutting is a typical two-dimensional rectangular
and mutation probability of genetic algorithm adaptively Nesting engineering problems. It requires a set of leather
adjust on the basis of logistic curve equation and the rectangle samples which are distributed over a given
shortcomings of traditional adaptive genetic algorithm specification leather fabric with an optimal arrangement,
solved well. The remaining rectangle algorithm as the various leather samples do not overlap each other and
decoding algorithm and adopting New cross-ways, the not more than leather fabric border, making full use of
niche technology controlled whether the child individual leather fabrics, improving their utilization. Algorithms
replacement the parent individual or not accelerating have a reliable fast convergence and nesting calculation is
convergence rate. Examples show that the algorithm efficient.
of leather fabrics nesting is effective and a substantial Optimal layout of rectangular pieces is NP-hard
increase in the utilization of leather fabric. problem, there is still more literature that dedicated to
finding efficient algorithms to solve a wide variety of
Key words: Optimal layout of rectangular parts;
two-dimensional rectangular pieces of packing problem.
Adaptive genetic algorithm; Niche technology; The
Research work in this area as early as linear programming
remaining rectangle algorithm
and the knapsack problem (Reda & Abd, 1994; Cao &
Zhou, 1994) thinking applied to the problem of nesting,
Liu, H., & Qiu, J. X. (2015). Improved Adaptive Genetic
Algorithm in Optimal Layout of Leather Rectangular Parts.
but it has a great computational complexity, suitable
Advances in Natural Science, 8 (3), 20-26. Available from: for small-scale problem solving, not very practical in
h t t p : / / w w w. c s c a n a d a . n e t / i n d e x . p h p / a n s / a r t i c l e / v i e w / 7 3 4 2 engineering.
DOI: http://dx.doi.org/10.3968/7342 With the development of intelligent algorithms
and successfully applied to TSP and space allocation,
literature (Huang, Qi, & Tan, 2012; Thiruvady, Meyer, &
Ernst, 2008; Jiang & Lu, 2008) make the particle swarm
INTRODUCTION optimization, ant colony algorithm, genetic algorithms,
Leather fabric used in clothing, automobile, home, simulated annealing algorithm apply to the Rectangular
airplane seats, luggage, and other industries. Nesting Strip Packing Problem. However, due to the limitations of

Copyright © Canadian Research & Development Center of Sciences and Cultures 20

 LIU Hao; QIU Jianxin (2015).
Advances in Natural Science, 8 (3), 20-26

the algorithm itself, the nesting effect is not too satisfied. motherboard. After completion of the maximum height
In Literature (Tang & Tang, 2005; Jia, Yin, & Luo, 2002). of nesting (that is the highest leather rectangular
GA combined with layout algorithm of BL, BLF to sample boundary of figure nesting, also known as
optimize the layout, however, the algorithm is not very the highest nesting contour) H high ≦H (referring to
good stability and convergence rate. Figure 3 Hhigh). While meeting certain cutting process
After a lot of genetic iterative, genetic algorithm requirements.
eventually evolved to the optimal layout in probability Optimization objectives: The minimum gap of
theory from the random search method. It has the between Leather rectangular sample to make leather fabric
advantage of strong global search capability, but has maximum utilization (η).
the disadvantage of premature and local convergence. Since guarantee the lowest Hhigh of the highest contour
Adaptive genetic algorithm (Lee, Lin, & Chen et al., nesting figure, between leather rectangular sample gap
2011) solve the problem that the crossover rate is too is the minimum in nesting figure, so the optimization
large, the best individual structures destroyed, and the modeling in this paper can be simplified to only consider
crossover probability is too small to make evolutionary the impact of the highest contour nesting map. Due to
stagnation. It also solves the question that the mutation the width of Motherboard leather and the total area of all
probability value is too small to difficult produce a new leather Rectangular samples is constant, the utilization
individual, mutation probability value is too large, GA has of leather fabric η only relevant with total area of all
no evolutionary effect and other issues. However, since rectangles and area of using motherboard sample, the
the value of the adaptive genetic algorithm crossover optimization objective function is:
and mutation probability close to zero in later evolution, η = h/Hhigh , (1)
evolution will occur stagnation. In addition, adaptive where: n
genetic algorithm only consider the average value of
h = ∑ w i × hi / W . (2)
group fitness and optimal fitness value and it does not i =1

consider the impact of its evolution from the entire group where, h is theoretical optimal height, which is the
setting produces. resulting all the leather samples rectangular area divided
In this paper, an improved adaptive genetic algorithm by the width of leather motherboard. h is constant, Hhigh is
is employed. First, obtain Optimal into the row order of variable. The objective function value η the closer 1, the
a set of rectangular leather samples. In the process of closer optimal nesting.
Leather rectangular sample layout, automatically adjust
the crossover and mutation probability depending on
the circumstances, it well overcomes the shortcomings 2. THE KEY TECHNOLOGY AND BASIC
of traditional adaptive genetic algorithm. In order to F L O W A L G O R I T H M O F L E AT H E R
make the evolutionary process does not fall into local
optimum, this paper adopted a new way of cross. Finally,
RECTANGULAR SAMPLE LAYOUT
the remaining rectangle decoding algorithm to determine In this paper, using the improved adaptive genetic
the rules of the row of rectangular samples of leather, algorithm to obtain the optimal order of the row of
achieving automatic nesting. rectangular leather sample group, and then use the
remaining rectangle algorithm to determine the rules
of the row of rectangular samples of leather, automatic
1. LEATHER RECTANGLE PACKING nesting. In this paper, using leather rectangular optimal
layout sample problem describes the improved adaptive
PROBLEM MATHEMATICAL MODEL
genetic algorithm basic flow.
Leather rectangular nesting optimization problem is
described as follows: In the wide W, height H of the 2.1 Genetic Encoding
rectangular artificial leather motherboard emissions Optimization leather rectangular sample nesting use
without overlapping rectangular leather samples {P1, P2 decimal integer coding. To a group of n rectangular
.... Pi .... Pn}, anyone leather rectangular samples Pi (i = leather samples P 1, P 2, ... P n with a natural number 1
1,2, ..., n) can be expressed using a dimensional binary to n correspond to numbered，the number of leather
array: rectangular samples constitute a random integer string
Pi = (wi , hi ) , X={x 1, x 2,...x i.....x n} (x i represents a rectangle leather
where wi, hi are the leather rectangular sample Pi’s width sample), thus forming an individual, that is, a group
and height. of leather rectangle row sample group which has been
Optimization constraints: Each rectangle leather determined to be the order of the rows. Coding sequence
sample does not exceed the edges of the rectangular of the individual elements in the sample is the order of
artificial leather motherboard and its edges must be leather rectangle. x i has positive and negative points,
parallel to the edges of the rectangular artificial leather representing nesting manner of leather rectangular

21 Copyright © Canadian Research & Development Center of Sciences and Cultures

Improved Adaptive Genetic Algorithm in
Optimal Layout of Leather Rectangular Parts

sample, that is horizontal or vertical. When xi> 0, leather

rectangular sample does not rotate, horizontal nesting.
When x i <0, leather rectangular sample rotation 90°,
vertical nesting. For example, a set of random samples
of rectangular leather{2, 7，-3，-5, 1, 8，-6, 4},
representing the discharge order is 2 →7 → 3 → 5 → 1
→ 8 → 6 → 4, the number -3,-5, -6 of leather rectangular
sample rotation 90°, others is horizontal. Each individual
sample corresponds to one kind of leather rectangular
nesting diagram, m numbers of aforementioned
individuals constitute a population. Figure 2
2.2 The Size of the Initial Population Remaining Rectangular Nesting Algorithm
The size of the population has a large impact on 2.4 Fitness Function
optimization of final results and the convergence rate. In Genetic Algorithm evolutionary process, individuals
The larger the population size, the more mode of leather decoding nesting, with the fitness F to evaluate the quality
rectangular sample nesting processing, the more the rich of nesting map (Nesting Results). Individual fitness are
diversity of the population, the greater the opportunity associated with utilization, H value of the formula (1) is a
to evolve into optimal layout map, but the convergence fixed value, so fitness function simplifies to:
slows down. Population size is too small, it will affect the   F = 1/Hhigh (3)
final result of nesting. In this paper, using the adaptive 2.5 Choose Operator
population size m, whichever m twice the number n of According to the principle of survival of the fittest,
rectangle leather sample, m = 2n. choosing large fitness individuals from the population.
2.3 Leather Rectangular Layout Sample Decoding Before selecting the population, to accelerate the search
Algorithm speed of optimal solution, to prevent the destruction of
In order to evaluate the fitness of the individual, the preferred individual in the crossover and mutation
individuals need to decode operation, the population operations processes, using elite selection strategy.
of individuals transformed into rows comp image. Selecting (1-GGAP) × m of the best individual directly
Nesting algorithm determines the rules of the row of inherited to the next generation, not through crossover and
leather rectangular sample. In this paper, the remaining mutation operations. In this paper, using roulette selection
rectangular nesting algorithm as a decoding method. method, setting the gap to GGAP, the population size
The remaining rectangular nesting algorithm (Jaya is m, selecting GGAP × m (if not an integer rounding)
Thomas·Narendra, 2014) with a rectangle Data set to individuals do genetic manipulation.
represent currently remaining of leather fabric, any 2.6 Crossover Operator
unused space emissions (including isolation gap) are Cross operations mainly generate a new order of the row
included in the remaining rectangle collection; Before of rectangular leather samples. In order to prevent the
rectangular pieces of leather into the row, The most emergence of local optimal solution, using a new cross-
reasonable position emission based on the remaining way. The selection of the individuals was divided into two
rectangle centralized data to overcome the lowest level groups in accordance with the fitness value from small to
algorithm can not enter the hollow area of the defect row. large order: Large fitness value into one group, the smaller
Individuals with a coding example, its the sort listed (1, 2, fitness value is divided into one group. During cross-
3, 4, 5). Using different algorithms for nesting, the results operation, from each of the two groups were selected to
shown in Figure 1, 2 do a pair crossover, so making optimum individual and
poor individuals have a greater probability of mating, so
that the whole population is close to the optimal solution,
to prevent local optima the emergence. In this paper, using
two-point crossover mode, from two groups each selected
an individual pair, producing a total of n / 2 (rounded)
groups. Choosing a pair of cross matched group of X1,
X2 regard as a parents. In [1, n] random generate two
integers r1, r2 (0≤ r1≤ r2≤ n), rand randomly generated
in [0,1], If crossover probability Pc <rand, not cross
operation, otherwise cross operation. In the crossover
Figure 1             operation, X 2 genes of located between r 1 and r 2 are
“Minimum Level” Algorithm Nesting copied to Offspring1 in same position, the rest position

Copyright © Canadian Research & Development Center of Sciences and Cultures 22

 LIU Hao; QIU Jianxin (2015).
Advances in Natural Science, 8 (3), 20-26

by X1 gene duplication. In order to avoid the generation the corresponding position of Offspring1 by the order of
of invalid individuals, if gene copied from X1 conflict appearing in X2. Using the same method of generating
with a cross section, using some method map copied to Offspring2. For example:
ParentX1
3 -5 1 10 -2 9 -8 6 4 7
                 ↑             ↑
                  r1            r2
ParentX2              ↓              ↓
6 -9 3 -7 -4 2 1 8 10 5

Progeny Offspring1
3 -5 -9 -7 -4 2 1 6 8 10

Progeny Offspring2
6 1 3 10 -2 9 -8 4 7 5

2.7 Mutation Operator constant population size.

Variation of generating a new order of the rows of leather 2.10 Crossover and Mutation Probability of
rectangular sample is auxiliary operations. Because Adaptive Genetic Algorithm
the individual is an integer encoding with a symbol, In crossover and mutation operation of GA, the adaptive
so variation in two ways. One is the rotation variation, crossover probability Pc and mutation probability Pm
namely leather rectangular sample 90 0 after the row significant impact on the performance of the algorithm,
rotation; the other is the position variation. In this paper, the genetic algorithm convergence performance depends
the use of two variants of a combination of ways. For the primarily on two core operations of the crossover and
individual X mutation , Firstly, rand randomly generated mutation operator. Srinivas proposed adaptive genetic
in [0,1]. If the mutation probability Pm≥rand, as follows: algorithm whose a dynamic adjustment of crossover
In [0, n] randomly generated r1, r2, r3 (r2≥r1), the r1th gen and mutation probability based on adaptation values
and r2th gen are exchanged, the first r3 alleles reverse (Patnailklm, 1994). Adaptive crossover probability Pc
signs. As X = {- 2,7, -3,5,1, -4,6}, when r1 = 2, r2 = 6, r3 and mutation probability Pm can automatically change
= 4, After mutation, X ‘= {- 2, -4, -3, -5,1,7,6}; When Pm over the fitness value, but individual fitness value close
<rand, do not do mutation. to or equal to the maximum fitness value, Pc and Pm
2.8 Based on the Niche Technology of Primary close or equal to zero, the algorithm is easy to fall into
Election System local optimum. Ren Ziwu put forward an improved
Using the niche technology of primary election system adaptive genetic algorithm (Ren & San, 2006) to solve
(Dong & Yan, 2013) decided to whether offspring replace the above problems. But poor individuals have poor
parent individuals after crossover and mutation operations ability to crossover and mutation, prone to stagnation.
to maintain the diversity of population. Furthermore, do not consider setting Pc and Pm from
the entire population evolution. Pc and Pm of adaptive
Cross operation: for X1 parent and offspring Offspring1,
genetic algorithm of adjustment formula based on
if F (Offspring1) ≥F
literature (Chen, 2011): Individuals of poor adaptation,
(X1), the offspring substitute parent, otherwise leave
should have a larger Pc and smaller Pm. For individuals
the parent X1; for X2 parent and offspring Offspring2, if
with high fitness, should impart corresponding values
F (Offspring2) ≥F (X2), the offspring substitute parent,
of Pc and Pm according to the degree of evolution and
otherwise leave the parent X2.
adaptation of times. For individual with fitness value
Crossover and mutation operation select the best
which approaching best individuals, should have a larger
individuals into the next generation in between the parent
Pc and a smaller Pm. This is mainly basis makes a strong
and offspring, the offspring is always better than or equal
global search ability and weak local search capability in
to the parent, evolutionary direction is always towards the
early evolution. As evolution progresses, global search
optimal direction of development.
ability dwindling, local search capabilities continue to
2.9 Re-Insert Operation strengthen, which prevent the individuals of high fitness
Re-insert operation: After the operations of crossover, destruction, to facilitate the evolution to the optimal
mutation and niche technology of primary election system, solution, to prevent premature phenomenon appearing.
the offspring population and elite individuals together to In this paper, on the basis of logistic curve equation,
form a new generation of populations, in order to maintain using an improved adaptive genetic algorithm Pc and Pm

23 Copyright © Canadian Research & Development Center of Sciences and Cultures

 EX + 1
ω=
DX
Improved Adaptive Genetic Algorithm in 1
Optimal Layout of Leather Rectangular Parts Pc = − k1
− 0.1
1+ e ω

dynamically adjusting the formula, adopting a new Pc k2

and Pm, well meeting the above requirements.  Pm = 1 . (8)
Reflecting fitness value of the population’s average 5(1 + e ω )
expectation EX: Where, F avg represents the average fitness of the
F +F +…Fm population of individuals, m represents the size of the
   EX = Favg = 1 2m . (4)
population, F i (i = 1, 2 ... n) represents the fitness of
Reflecting the value of individual fitness level discrete individual i, k1 and k2 are two constants, k1 of the range (0,
DX: +∞), k2 of the range (0,1).
F 2+F 2+…Fm2
   DX = 1 2m - favg2. (5) 2.11 Convergence Criteria
The degree of similarity between individuals of the In this paper, the convergence criteria set: if utilization of
optimal nesting solution greater than the set value ηk, stop
population representedEX + 1 similarity coefficient ω:
by the
ω = EX + 1 evolution; otherwise hereditary algebra greater than the set
  ω = DX , (6) maximum number of evolutionary Gmax stopped. Finally,
DX
1 exporting the optimal solution and the corresponding
Pc P= = −1k1 − 0−.01,.1 (7) output nesting map. Rectangular Nesting algorithm with
c − k1
1 +1e+ ωe ω main flow chart shown in Figure 3.

k
Pm P= = 2 k 21
m 1
5(15+(1e+ e)ω )
ω

Figure 3
Rectangular Nesting Algorithm With Main Flow Chart
Gmax=100, ηk=0.98, Generation Gap GGAP=0.9. Artificial
3. EXPERIMENTAL ANALYSIS
leather fabric width is 2 m, assuming infinite length. The
In order to verify the validity of firefly simulated annealing size and number of rectangular leather sofa fabric as
algorithm, a set of selected samples of rectangular leather shown in Table 1:
sofa from a furniture plant for nesting computing. Set the Using matlab software for simulation，nesting results
parameters as follows: k1=10, k2=0.1, Convergence criteria shown in Figure 5 , nesting height of the evolutionary

Copyright © Canadian Research & Development Center of Sciences and Cultures 24

 LIU Hao; QIU Jianxin (2015).
Advances in Natural Science, 8 (3), 20-26

process diagram in Figure 6. The results can be seen from In order to verify the efficiency of the algorithm,
Figure , nesting height Hhigh = 4,395 mm, leather fabric above a set of leather sofas rectangular sample
utilization of 94.49%, obtaining the number of iterations respectively use GA (Standard genetic algorithm)
of the final layout diagram for the first 62 times, the entire and AGA (Adaptive Genetic Algorithm) algorithm
nesting process run time of 159.8s. nesting, the results were compared, remaining rectangle
Table 1 algorithm as the decoding algorithm; Based on this
The Size and Number of Rectangular Leather Fabric algorithm (improved adaptive genetic algorithm in this
Kind of The length of the The width of the paper), using the lowest level algorithm as the decoding
rectangular rectangular pieces rectangle pieces Quantity algorithm nesting. The results obtained for the above-
pieces long (mm) wide (mm) described algorithm are compared, as shown in Table
1 700 410 6 2. In order to ensure comparability of the algorithm,
parameter settings should be the same. GA parameters
2 650 300 6
are set as follows: Population size m=72, GGAP=0.9,
3 650 150 6 Pc=0.9, Pm=0.1, Gmax=100, ηk=98%; AGA’s parameters
4 650 650 6 are as follows: Population size m=72, GGAP=0.9, Initial
crossover probability Pc1=0.9, Pc2=0.6, Initial mutation
5 500 400 2 probabilityPm1=0.5, Pm2=0.1, Gmax=100, ηk=98%; In this
6 500 250 4 algorithm basis, using the lowest level algorithm as the
decoding algorithm nesting algorithm uses the same
7 700 500 2
parameters of this algorithm. For the four algorithms,
8 550 265 2 respectively, running 10 times, obtaining the maximum
9 550 400 2 height H max, the minimum height H min, average height
H avg . After each run, the maximum number K max of
iterations of the final solution, minimum Kmin, average
Kavg, average utilization ηavg.
Table 2
Compares the Four Algorithms Results

This paper This paper

algorithm(the algorithm(the lowest
GA AGA remaining rectangle level algorithm
as the decoding as the decoding
algorithm) algorithm)

Hmax 4575 4510 4470 4480

Hmin 4400 4360 4300 4320

Havg 4459 4429.5 4410 4418.5

Kmax 96 72 98 96
Figure 4        
Nesting Results      Kmin 3 22 44 47

Kavg 61,5 55.1 66 65.4

ηavg(%) 93.56 94.18 94.59 94.42

Can be drawn from Table 2, for leather rectangular

sample layout, the results obtained using this algorithm
regardless of maximum height Hmax, the minimum height
H min, average height H avg, average η avg are better than
other algorithms. Each time, the final solution with
maximum iterations Kmax, minimum Kmin, average Kavg
generally greater than other algorithms. This further
validates that the adaptive genetic algorithm used in
the later stage of evolution is still a strong ability to
mutate and local search ability, to avoid late algorithm
Figure 5 stagnation and “premature” appearance.
Nesting Height Evolution

25 Copyright © Canadian Research & Development Center of Sciences and Cultures

Improved Adaptive Genetic Algorithm in
Optimal Layout of Leather Rectangular Parts

CONCLUSION Huang, L., Qi, J., & Tan, Y. (2012). A genetic-discrete paticle
swarm optimization algorithm for rectangular packing.
(a) Improvement of Algorithms: In this paper, Based Chinese Journal of Electronics, 6(40), 1103-1107.
on the adaptive genetic algorithm, crossover and adaptive Jaya Thomas·Narendra, S. C. (2014). A new metaheuristic
crossover probability, mutation probability are improved genetic-based placement algorithm for 2D strip packing.
, using a mechanism based on a pre-selected niche Springerlink, 47(9), 1-16
technology, obtaining the best leather rectangular layout Jia, Z. X., Yin, G. F., & Luo, Y. (2002). Two-dimensional
sample order, and then using the remaining rectangle irregular parts packing with genetic algorithm. Journal of
algorithm to achieve automatic nesting. Computer-Aided Design & Computer Graphics, 5(14),
(b) Effect of the algorithm: By comparison of the data 467-470.
with the GA and AGA experiments prove the algorithm is Jiang, X. B., & Lu, X. Q. (2008). Optional packing of rectangles
better and superior performance. For leather rectangular with an adaptive simulated annealing genetic algorithm.
layout sample, using this algorithm obtains maximum Journal of Computer-Aided Design & Computer Graphics,
utilization of material, later evolution remains the 11(20), 1426-1431.
effective evolution. Improved adaptive genetic algorithm Lee, C. P., Lin, W. S., & Chen, Y. M., et al. (2011). Gene
is proposed to solve the problem that production of good selection and sample classification on microarray data based
utilization and efficiency can not take into account in the on adaptive genetic algorithm/k-nearest neighbor method.
leather rectangle sample layout. The algorithm not only Expert Systems With Applications, 38(5), 4661-4667.
improves utilization of leather sample material, but also Patnailklm, S. M. (1994). Adaptive probabilities of crossover
has fast convergence rate. In the garment leather business and mutation in genetic algorithms. IEEE Transaction on
applications, this algorithm significantly improves the System, Man and Cybernetics, 24(4), 656-667.
economic efficiency of enterprises. Reda, M. S., & Abd, E. A. (1994). An interactive technique
for cutting stock problem with multiple objects. European
Journal of Operational Research, 78(3), 304-317.
REFERENCES Ren, Z. W., & San, Y. (2006). Improved adaptive genetic
Cao, J., & Zhou, J. (1994). Rectangular nesting of backpack algorithm and its application research in parameter
algorithm. Chinese Mechanical Engineering, 5(2), 11-12. identification. Journal of System Simulation, 1(18), 41-44.
Chen, C. (2011). Improved adaptive genetic algorithm and its Tang, K. W., & Tang, W. K. S. (2005). Metal cutting with hybrid
application. Guangzhou, South China University of genetic algorithmic (pp. 735-739). Proceedings of the 3rd
Technology. IEEE International Conference Industrial Informatics, Perth.
Dong, D. W., & Yan, Y. H. (2013). Adaptive genetic simulated Thiruvady, D. R., Meyer, B., & Ernst, A. T. (2008). Strip packing
annealing algorithm in optimal layout of rectangular parts. with hybrid ACO: Placement order is learnable. IEEE
China Mechanical Engineering, 18(24), 2499-2504. Congress on Evolutionary Computation, 1207-1213.

Copyright © Canadian Research & Development Center of Sciences and Cultures 26