See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/282449315

A Comparative Study of Simulated Annealing with different Cooling Schedules

for Geometric Optimization of a Heat Transfer Problem According to
Constructal Design

Article in Scientia Plena · August 2015

DOI: 10.14808/sci.plena.2015.081321

CITATIONS READS

0 590

6 authors, including:

Gill Velleda Gonzales Elizaldo dos Santos

Instituto Federal de Educação, Ciência e Tecnologia Sul-Rio-Grandense (IFSUL) Universidade Federal do Rio Grande (FURG)
10 PUBLICATIONS 36 CITATIONS 302 PUBLICATIONS 1,614 CITATIONS

SEE PROFILE SEE PROFILE

Leonardo Emmendorfer Liercio Isoldi

Universidade Federal do Rio Grande (FURG) Universidade Federal do Rio Grande (FURG)
41 PUBLICATIONS 123 CITATIONS 289 PUBLICATIONS 1,415 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Estudo Numérico e Otimização Geométrica do Dispositivo de Placa Horizontal Submersa View project

Modelling Social Systems View project

All content following this page was uploaded by Liercio Isoldi on 06 January 2016.

The user has requested enhancement of the downloaded file.

 VOL. 11, NUM. 08 2015
www.scientiaplena.org.br doi: 10.14808/sci.plena.2015.081321

A Comparative Study of Simulated Annealing with Different

Cooling Schedules for Geometric Optimization of a Heat
Transfer Problem According to Constructal Design
Um Estudo Comparativo do Simulated Annealing com Diferentes Cooling Schedules para a
Otimização Geométrica de um Problema de Transferência de Calor de Acordo com o
Constructal Design

G. V. Gonzales1,*; L. R. Emmendorfer1; L. A. Isoldi1; E. da S. D. Estrada2; L. A.

O. Rocha2, E. D. dos Santos1
1
Programa de Pós­Graduação em Modelagem Computacional (PPGMC), Universidade Federal do Rio Grande
(FURG),96021­900, Rio Grande, Brasil
2
Department of Mechanical Engineering (DEMEC), Universidade Federal do Rio Grande do Sul (UFRGS),90050­
170, Porto Alegre, Brasil
*[email protected]
(Recebido em 22 de setembro de 2014; aceito em 29 de dezembro de 2014)
The Simulated Annealing is a widely used meta­heuristic to combinatorial and optimization problems
studied since 1983, when it was first proposed. This algorithm has a particular parameter, the cooling
schedule (specifically the decrease of temperature) which is still defined for each problem empirically.
This study has the aim to find the best cooling schedule(s) for geometric optimization of a heat transfer
problem. More specifically it is performed the geometrical evaluation of an isothermal Y­shaped cavity
intruded into conducting solid wall with internal heat generation. In this research, Constructal Design is
employed to determine the constraints and objectives for geometric evaluation of this problem. The
cooling schedule(s) recommended here will contribute for future applications of Simulated Annealing
heuristic in association with Constructal Design to seek for the optimal shapes of complex geometries in
heat transfer problems, as well as, evaluate the influence of geometric parameters over thermal
performance of the problem. Six different methods of decrease of temperature are evaluated in twenty
samples of thirty simulations for each cooling schedule. From these simulations, it was extracted the
percentage of convergence to the global optimal point. Results showed that the Fast cooling schedule led
to the worst performance in the seek for the global optimal shapes, while the three hybrid cooling
schedules proposed here are the most recommended to find the optimal geometrical configurations in heat
transfer problems in association with Constructal Design (BoltzExp, ConstExp1 and ConstExp2).
Keywords: Constructal Design, Simulated Annealing, Geometric Optimization

O Simulated Annealing é uma meta­heurística amplamente utilizada para problemas combinatórios e de

otimização estudada desde 1983, quando foi proposta. Este algoritmo possui um parâmetro particular, o
cooling schedule (especificamente o decaimento de sua temperatura) que ainda é definido de forma
empírica para cada tipo de problema. Este estudo tem como objetivo encontrar o melhor cooling
schedules(s) para a otimização geométrica de um problema de transferência de calor. Mais
especificamente, é realizada a avaliação geométrica de uma cavidade isotérmica em forma de Y inserida
em um sólido com geração de calor interna. Nesta pesquisa, o método Constructal Design é empregado
na determinação de restrições e objetivos para a avaliação geométrica deste problema. Os cooling
schedules recomendados aqui, contribuirão para futuras aplicações da heurística Simulated Annealing em
associação ao Constructal Design na busca pelas formas ótimas de geometrias complexas em problemas
de transferência de calor, assim como, na avaliação da influência dos parâmetros geométricos sobre o
desempenho térmico do problema. Seis diferentes métodos de decaimento de temperatura são avaliados
em vinte amostras de trinta simulações para cada cooling schedule. A partir destas simulações, foi
extraída a porcentagem de convergência ao ponto de ótimo global. Os resultados mostram que o cooling
schedule Fast obteve a pior performance na busca pelas geometrias ótimas globais, enquanto que três
cooling schedules híbridos propostos aqui são os mais recomendados na busca pelas configurações
geométricas ótimas em problemas de transferência de calor em associação ao Constructal Design
(BoltzExp, ConstExp1 e ConstExp2).
Palavras­chave: Constructal Design, Simulated Annealing, Otimização Geométrica

081321 –1
 G.V. Gonzales et al., Scientia Plena 11, 081321 (2015) 2

1. INTRODUCTION

The algorithm Simulated Annealing (SA) is a meta­heuristic proposed by Kirkpatrick [1]. It

is an implementation of Metropolis et al. [2] method in a sequence of iteration controlled by a
temperature parameter, based on physic annealing process. According to Kirkpatrick [1] the
temperature parameter and the behavior of temperature decay is defined empirically. This
parameter is responsible for the control of temperature in the algorithm and is called cooling
schedule. This parameter has the highest influence over the search for the optimal results [3,4].
There are in literature several studies about the best cooling schedules or variants of classical
SA, see Refs. [5­13]. However, there is no universally recommended cooling schedule to seek
for the best geometries in heat transfer problems.
Constructal Design derives from Constructal Theory, which is the view that geometry
patterns of any animate or inanimate flux systems, which are seen in nature is not result of
chance, but emerged by a physical principle, the Constructal Law [14,15]. Constructal Law
states that “For a finite­size flow system to persist in time (to live), its configuration must
evolve in such a way that provides easier access to the currents that flow through it”. The form
to employ Constructal Law for geometric evaluation of flux systems is the Constructal Design
[14], which is a method based on objectives and restrictions used to minimize the imperfections
of the investigated systems. This method has also been applied for improvement of performance
in engineering problems. Recently, several C, T, Y­shaped cavities, as well as, complex and
multiple cavities has been studied by means of Constructal Design, see Refs. [16­24].
The problem studied here is a Y­shaped cavity that is intruded into a conducting solid wall
with internal heat generation. The heat is just only removed by Y­shaped cavity and the
objective is to optimize its geometry to minimize the thermal resistance of the solid. In general,
as more complex is the cavity geometry to be evaluated more degrees of freedom (DOF) are
required to define. The optimization of this kind of problem requires a large computational
effort due to the high number of required simulations for complete optimization. In this
framework, recent studies [21, 22] employed genetic algorithms (GA) to seek for the best
shapes. The achieved results for few degrees of freedom were compared with those achieved
with Exhaustive Search (ES) which consists on the simulation of all geometric possibilities.
Afterwards, new results were proposed using only GA for the whole degrees of freedom of the
problem.
The employment of Simulated Annealing (SA) in association with Constructal Design for
geometrical optimization of cavities has not been studied in literature at the authors knowledge,
representing a novel important study. Then, the study of appropriate annealing cooling schedule
is also an important subject. The application of algorithm without adjusted configurations can
lead to errors in the achievement of global optimal shapes, as well as, in the evaluation of
influence of each geometric parameter (DOF) over thermal performance of the problem, mainly
in cases where it is not possible to compare the results with those achieved with ES. In this
sense, this study intends to reach to a recommendation about the best cooling schedules for a
reliable optimization with SA meta­heuristic for future optimization of complex cavities into
Constructal Design framework.

2. MATHEMATICAL MODELING

Figure 1 shows the computational domain of the studied problem. The Y­shaped cavity acts
removing the heat generated by solid domain (gray region). The seek for the optimal geometry
of the cavity is given by variation of the degrees of freedom: H/L, t1/t0, L1/L0 and α. The
temperature field inside the solid domain is solved for every studied cases and the maximum
temperature is collected. The best thermal performance of the system is obtained when the
maximum temperature inside the solid domain is minimized.
 G.V. Gonzales et al., Scientia Plena 11, 081321 (2015) 3

Figure 1: Computational Domain of Y­shaped cavity into a solid with internal heat generation

More details of mathematical model can see in [17]. The main objective is to minimize the
dimensionless maximum excess of temperature (θmax) in the solid domain, given by
dimensionless equation:

Tmax  Tmin
θmax =
q''' A
k (1)

3. NUMERICAL MODEL

The function represented by Equation (1) is determined numerically by solving the

temperature field in every assumed configurations (H/L, t1/t0, L1/L0, α) and calculating θmax to see
whether θmax can be minimized by varying the configuration. The numerical solution is
performed with the Finite Element Method (FEM), see Refs. [24, 25]. The domain is discretized
with linear triangular elements. The solution is performed in MATLAB environment, precisely
the PDE (partial­differential­equations) toolbox [26]. The grid was non­uniform in both x and y
directions, and varied from one geometry to the next. The appropriate mesh size was determined
by successive refinements (h­adaptively), increasing the number of elements four times from the
current mesh size to the next mesh size, until the following criterion is satisfied:

T max
i
 θmax
i+1 i
/ θmax |< 1 u 10 4 (2)

where θi represents the maximum value of the temperature evaluated with current mesh, θi+1
represents the temperature evaluated with next refined mesh (which has four elements more
than the previous mesh).
Table 1 illustrates the refinement mesh process with the aim to find an independent mesh
that does not influence results of temperature field. The following results were performed by
using a range between 2240 and 143360 triangular elements. The independent mesh is achieved
for the third refinement (35840 elements). The grid independence test is performed for a fixed
geometry with the following parameters: ϕ = 0.05, ψ = 0.5, H/L = 1.0, t1/t0 = 2.0, L1/L0 = 0.5, α =
0.94.
 G.V. Gonzales et al., Scientia Plena 11, 081321 (2015) 4

Table 1: Grid independence test

Number of Elements θmax |(θimax­ θi+1max)/ θimax|
2240 0.079046 5.6600×10­4
8960 0.079001 1.8978×10­4
35840 * 0.078986 5.6971×10­5
143360 0.078982 ­­­­­­­­­­­­­
* Independent mesh

The results obtained in grid independence test were compared to previous results found in
Refs.[18] and [21]. The comparison is depicted in Table 2.

Table 2: Comparison between the results achieved with the present model and those showed in literature.
Reference θmax
Present Work 0.0762
Lorenzini et al. [18] 0.0762
Lorenzini et al. [21] 0.0762

4. SIMULATED ANNEALING

The combinatorial optimization algorithm Simulated Annealing (SA) was proposed firstly in
Kirkpatrick et al. [1] and it has a heuristic based in production of resistant materials, like glass
or metal, submitted to annealing process. Initially the material is heated at high level
temperatures and then cooled slowly to obtain the minimization of energy between the atoms of
the material. The algorithm of Kirkpatrick et al. [1] applies the optimization proposed by
Metropolis et al. [2] for various temperature levels.
Based on the heuristic presented, the algorithm SA starts the search for the possible solutions
in the search space and accept new solutions according with a probability given by Eq. (3),
which depends on current iteration temperature of algorithm. According to Eglese [3], the initial
temperature, the method as the temperature decrease, the number of neighbor proposes at each
level of temperature, the reannealing parameter and the stop criterion, are named cooling
schedule. For high algorithm temperatures of the annealing process the algorithm proposes
distant new geometric neighbor point and has more probability to accept bad solutions, i.e.,
distant from the optimal shape. As the algorithm temperature decreases the new proposed
neighbor points are near from the current point and the probability to accept bad solutions
decreases considerably.

1
P=
1+exp
(maxΔ(T )) (3)

In Eq. (3) ∆ is the difference between the value returned by Objective Function of the new
configuration in relation to Objective Function value for the best evaluated point. More
precisely, when a new geometry neighbor is proposed in comparison with the current solution,
the value of Objective Function, which it is intended to be minimized, is calculated for this new
geometry (neighbor). Afterwards, it is calculated Δ subtracting the value of Objective Function
for the neighbor solution and Objective Function for the current solution. The acceptance
probability of solution also depends on the actual algorithm iteration. Once the numerical model
used here led to an achievement of different temperatures for each degree of freedom, for
calculation with Eq. (3) it is used the highest temperature of domain [26].
To apply the SA algorithm in the geometric optimization of the Y­shaped cavity, it was
needed the same discretization of the space search realized in the study of Lorenzini et al. [21].
The implementation of SA was run in MATLAB environment. More precisely, it is used the
Global Optimization Toolbox with a specific function to SA execution named simulannealbnd.
 G.V. Gonzales et al., Scientia Plena 11, 081321 (2015) 5

The SA solver of the MATLAB is extremely adaptable, i.e., it is possible to develop new
algorithms for several functions as cooling schedule, annealing function or the acceptance
function.
The present work applied the SA algorithm for the same problem previously studied in Refs.
[16, 21] with the Exhaustive Search (ES) and Genetic Algorithm (GA), which were employed to
seek for the optimal shapes. In the present study, the main intention is to find the best
configurations for SA (cooling schedule) that combined with Constructal Design can be
employed to seek for the best geometries of complex cavity shapes. Twenty samples with 30
executions of the SA is realized trying to found the once optimized ratio of L1/L0, (L1/L0)o, and
the twice optimized angle α, αoo, for three different values of the ratio t1/t0: t1/t0 = 10.0, 11.0 and
12.0, which represents the optimal region for the ratio t1/t0 and with the fixed parameters studied
here, see Lorenzini et al. [21]. In this analysis it was extracted the mean percentage of
achievement of the global optimal geometries with each cooling schedule.
For optimization of two DOF of the problem, the search space formed by L1/L0 and α is
discretized in the same solution space employed for GA algorithm used in Lorenzini et al. [21].
The SA is evaluated for 15 different values of DOF L1/L0, as can be seen in the vector
represented in Eq. (4), while the DOF α is varied in the interval 1.00 ≤ α ≤ 1.57 with a step of
Δα = 0.01. The results present in Lorenzini et al. [21] were used to validate the optimal
configuration found here with SA and it is also used as stop criterion for the algorithm.

||
0.001
0.002
0.003
0.005
0.007
0.01
0.02
L1 / L 0= 0.03
0.05
0.07
0.1
0.3
0.5
0.7
1
(4)

The SA algorithm requires a function that must propose the neighbor solutions from the
current point during the optimization process, this function is called annealing function, which
is developed in the present study for optimization with two DOFs (L1/L0, α). The proposal of
new neighbor for the possible values of DOFs must respect the discretized space, i.e., generating
in a random way the values inside the search space. The random number function used is the
default function of MATLAB together with an adaptation calculus for generation of valid values
in search space and inside of the discretization established for each degree of freedom.
The cooling schedules observed in this work has the same initial parameters of SA, defined
in the start of algorithm, only difference is the temperature decrease that is the investigation
goal. Six functions of cooling was studied, the default functions of MATLAB Optimization
Toolbox (Exponential, Fast and Boltz) and more three hybrid and customized function which
will be called here: BoltzExp, ConstExp1 and ConstExp2. These functions are respectively
represented by the following equations:

k sa
Ti = T0 u 0.95 (5)
 G.V. Gonzales et al., Scientia Plena 11, 081321 (2015) 6

T0
T i=
k sa (6)
T0
T i=
ln (k sa ) (7)
­ T0
° : k sa d 80 for T0 = 100
Ti = ® ln k sa
° T0 u 0.95
k sa T0 / 2
: k sa > 80 for T0 = 100
¯ (8)
­ T0
°° T0 : k sa d
2
Ti = ®
° k sa T0 / 2 T
°¯ T0 u 0.95 : k sa > 0
2 (9)
­ T
°° T0 : k sa d 0
2
Ti = ®
° k sa T0 / 2 T
°¯ T0 u 0.9 : k sa > 0
2 (10)

where ksa is the annealing parameter, which is the same value for iteration number until
reannealing, i.e., the number of iterations before the recalculation of temperature. Reannealing
raises the temperature after the algorithm accepts a certain number of new points, and start the
search again at the higher temperature. The number of iterations after reannealing is configured
in MATLAB Optimization Toolbox, which by default is considered 100. Ti in above equations
represents the temperature in the current iteration and T0 represents the initial temperature of the
algorithm, that is assumed T0 = 100 by default.
In this work the proposed cooling temperature, named Boltzexp, is a hybrid form between the
models described in Equations (5) and (6). Basically, the function consists on the evaluation of
the temperature obtained with the two equations and the employment of the lowest temperature
obtained with both methods in the current iteration. The value ksa =80 in Eq. (8), represents the
number of iteration when the function changes the model of decrease temperature, i.e., from
Boltz to Exponential.
Two more cooling schemes are proposed, the ConstExp1 and ConstExp2, represented by
Eqs. (9) and (10), respectively. These models are very similar, the only difference between them
is the constant of decrease in the cooling temperature, which are respectively given by 0.95 and
0.9. Both schedules maintain the temperature fixed in the initial steps until the number of
iterations is equal to the half of initial temperature (T0). In first steps of those models, the
algorithm runs freely at initial temperature and with full probability to accept new points with
the purpose to find a better initial point until the temperature decrease. In Figure 2 it is possible
to observe the behavior of temperature decay with the number of iterations for the several
cooling schedules evaluated here.
 G.V. Gonzales et al., Scientia Plena 11, 081321 (2015) 7

Figure 2: The behaviour of algorithm temperature as a function of the number of iterations for the six
different cooling schedule of SA (T0 = 100 for all models).

5. RESULTS AND DISCUSSION

In this study, it is evaluated only the SA parameter named decrease of temperature. The other
parameters are maintained fixed with the sole interest to evaluate the influence of cooling
schedule function in the search for the optimal shapes of the studied Y­shaped cavity.
As above mentioned, the study focused in the optimal region for t1/t0 reported in Lorenzini et
al. [21]. For all geometries evaluated, the parameter H/L is considered constant H/L = 1.0 and
the areas fractions are ϕ = 0.05 and ψ = 0.3. The results of optimal configuration in Lorenzini et
al. [21] were used as stop criterion for the algorithm SA.
Table 3 shows for each cooling schedule the percentage of achievement of the optimal
geometry, i.e., which minimizes the dimensionless maximum excess of temperature (θmax). From
the 30 executions for each sample (in a total of 20 samples) for each cooling schedule and for
each evaluated ratio of t1/t0, it was possible to reach to statistics measures of the achievement of
global optimal geometry. Results of Table 3 reveal that SA algorithm is not deterministic and
they show the importance of statistical analysis to understand the behavior and influence of the
different parameters in optimization algorithms. For example, Fast cooling schedule reaches to
the optimal shape only 12.94 % of the evaluated samples. On opposite, the ConstExp2 leads to
an achievement of 62.44 %, in average, of the optimal shapes, showing that the new proposed
cooling schedules are more effective in the search for the global optimal shapes than the default
models presented in MATLAB Optimization Toolbox. Among the evaluated cooling schedules,
the worst performance was obtained with the Fast method, which clearly is not recommended
for this kind of problem. Even for a specific sample, where the Fast cooling schedule reaches to
the optimal in 36.67 % of the simulations, the performance was inferior to the average
performance of other functions as Boltz, BoltzExp, ConstExp1 and ConstExp2. It can also be
noticed that the method Exponential also has a lower percentage to find the global optimal
shape, the mean percentage is only 27.50 % with a standard deviation of 16.86 %. The hybrid
cooling schedules allowed the increase of the number of times where the global optimal
geometry for Y­shaped cavity with two DOFs is find, being in this primary analysis more
reliable for combination with Constructal Design for optimization of complex cavities in heat
transfer realm. However, these methods (BoltzExp, ConstExp1 and ConstExp2) also increase the
magnitude of standard deviation, which can be negative for representation of local effects of
degrees of freedom over the thermal resistance of the problem. The ideal case is that one which
 G.V. Gonzales et al., Scientia Plena 11, 081321 (2015) 8

in average reaches to a high number of cases where the geometry is optimized and with a few
variations (standard deviation) when the geometrical parameters are varied.

Table 3: Percentage distribution of statistics in the achievement of optimal shapes for each cooling
schedule.
Statistical Measures / Fast Exponential Boltz BoltzExp ConstExp1 ConstExp2
Function
Mean 12.94 27.50 42.72 54.39 54.67 62.44
Absolute Deviation 8.50 14.75 14.71 17.98 19.80 22.68
Standard Deviation 10.11 16.86 17.09 20.93 23.57 25.63
Variance 102.10 284.23 292.10 438.23 555.63 656.82
Mode 3.33 6.667 33.33 30.00 20.00 83.33
Median 11.67 30.00 36.66 53.33 56.67 71.67
Smaller 0.00 0.00 13.33 16.67 10.00 10.00
Largest 36.67 60.00 73.33 90.00 93.33 96.67

The Table 4, ANOVA was used for proof the difference between the cooling schedules
analyzed. In this analysis was employed the one­way ANOVA, because the percentage sample
for each treatment, or cooling schedule, are independents [27]. Also was observed twenty
rounds of thirty runs of the SA for each value of t1/t0 (10,11 and 12) in optimal region for this
case. So, were evaluated six different treatments, each one of them with sixty samples of the
percentage in find the optimal shapes at thirty optimization processes. Therefore the model
employed for ANOVA is balanced, i.e., the same number of samples for each treatment. The P­
Value, shown in Table 4, is very minor than the significance level 0.001, this indicates that at
least one distribution of the percentages is different [27].

Table 4: Analysis of variance between all cooling schedules.

Source of Degrees of Sum of Mean Squares F­Value P­Value
Variation Freedom Square
Schedules 5 107143 21428.6 55.2 2.56401e­42
Residuals 354 137417.1 388.2 ­­­ ­­
Total 359 244560 ­­­ ­­­ ­­­

Figure 3 depicts the percentage of global optimization for each cooling schedule. It can be
seen a similarity between three last cooling schedules: BoltzExp, ConstExp1 and ConstExp2. In
average these cooling schedules are better than the other three studied methods (Fast,
Exponential and Boltz). Results also showed that in local conditions (one specific sample) the
methods ConstExp1 and ConstExp2 reach to the global optimal shape only in 10 % of the cases
and for other specific samples they reached to magnitudes of achievement higher than 90 %,
which can be seen in black traces. It is also important to mention that for these cooling
schedules local optimal shapes with a performance similar to that obtained for the global
optimal shapes were reached. In this sense, in general, the hybrid cooling schedules showed for
the present study the most recommended functions for application of SA and Constructal
Design for geometrical optimization of heat transfer problems.
 G.V. Gonzales et al., Scientia Plena 11, 081321 (2015) 9

Figure 3: The comparison and classification between analysed cooling schedules (b_exp=
BoltzExp;c_exp1 = CosntExp1; c_exp2 = ConstExp2).

To corroborate the previous findings, Figure 3 also shows the classification of all studied
cooling schedules is performed based on pairwise comparison and the mean value of percentage
distribution of achievement of global optimal shape. Figure 3 was made with data generated by
Software­R, a free software environment for statistical computing. Moreover, the Tukey test was
employed for generated this graph. The level of significance employed in the Tukey test was
0.05, what resulted in a confidence rate of 95%. Results clearly revealed the inferiority of Fast
and Exponential cooling schedules and the superiority of new schedules to find the optimal
geometries in the present studied cases.

6. CONCLUSION

In this work, six functions of the cooling schedule were compared and classified. This
analysis employed the SA algorithm combined with Constructal Design for geometrical
optimization of complex Y­shaped cavities. Results showed that SA is a very useful tool for
optimization in the framework of Constructal Design, but this meta­heuristic requires an
adequate selection of its main parameters, mainly the cooling schedule. For the case with more
degrees of freedom when ES is not applicable, the reliable of meta­heuristics used is very
important for geometric optimization, as well as, the evaluation of effects of each degree of
freedom over the thermal performance of the system.
In the present study it was noticed that the Fast cooling schedule led to the worst
performance, reporting several local optimal geometries for the Y­shaped cavity intruded into
conductive solid wall with internal heat generation. Even when the best results obtained with
Fast cooling schedule are taken into account, the performance is inferior to those reached with
hybrid cooling schedules (BoltzExp, ConstExp1 and ConstExp2). The other two functions for
cooling schedule (Exponential and Boltz) led to intermediate performance.
The best results were obtained with the new cooling schedules proposed (BoltzExp,
ConstExp1 and ConstExp2). The increase in the achievement of global optimal shapes for the
 G.V. Gonzales et al., Scientia Plena 11, 081321 (2015) 10

hybrid cooling schedules are concerned with the increase of freedom given by the algorithm to
explore the search space in the initial iterations. When the algorithm starts with bad initial point,
he has more iteration to change to a better point until the beginning of temperature decay. It is
important to mention that the new cooling schedules were the only functions that reach to a
maximum percentage of more than 90 % in the achievement of optimal shapes.
In the future, the results generated in this paper will be employed in a same class of problem
but with a new geometry, the double­T shaped cavity. This geometry has a more degrees of
freedom and you complete optimization would consume a large computational effort with
Exhaustive Search and will need the employment of SA for geometric optimization.

7. AGRADECIMENTOS

Professors L. A. Isoldi, L. A. O. Rocha and E. D. dos Santos thank CNPq for research grant.

6. REFERENCES

1. Kirkpatrick S, Gelatt CD, Vecchi MP. Optimization by Simulated Annealing. Science. New Series.
1983; 220 (4598): 671­680.
2. Metropolis N, Rosenbluth AW, Rosenbluth MN, Teller AH. Equation of State Calculations by Fast
Computing Machines. J. Chem. Phys. 1953; 21: 1088­1092.
3. Eglese RW. Simulated Annealing: A tool for Operational Research. Eur. J. Oper. Res. 1990; 46:.271­
281.
4. Bertsimas D, Tsitsiklis J. Simulated Annealing, Statistical Science 1993; 8 (1): 10­15.
5. Szu H. Hartley R. Fast Simulated Annealing, Phys. Lett. A 1987; 122: 157­162.
6. Hajek B. Cooling Schedules for Optimal Annealing. Math. Oper. Res. 1988; 13 (2): 111­129.
7. Ingber L.Very Fast Simulated Re­annealing, Math. Comput. Model. 1989; 12 (8): 967­973.
8. Ingber L. Simulated Annealing: Patrice versus Theory, Math. Comput. Model. 1993; 18 (11): 29­57.
9. Nourani Y, Andresen B. A comparison of simulated annealing cooling strategies, J. Phys. A: Math.
Gen. 1998; 31: 8373–8385.
10. El­Bouri A, Azizi N, Zolfaghari S. A comparative study of a new heuristic based on adaptive memory
programming and simulated annealing: The case of job shop scheduling, Eur. J. Oper. Res. 2007; 177:
1894–1910, doi: 10.1016/j.ejor.2005.12.013.
11. Schneider JJ, Puchta M. Investigation of acceptance simulated annealing — A simplified approach to
adaptive cooling schedules, Physica A 2010; 389 (24): 5822–5831, 10.1016/j.physa.2010.08.045.
12. Kastanya D. Evaluating the effects of cooling schedule on the performance of ADORE, Ann. Nucl.
Energy 2013; 62: 483–486, doi: 10.1016/j.anucene.2013.06.019.
13. Brusco MJ. A comparison of simulated annealing algorithms for variable selection in principal
component analysis and discriminant analysis, Comput. Stat. Data An. 2014; 77: 38–53, doi:
10.1016/j.csda.2014.03.001.
14. Bejan A. Shape and Structure, from Engineering to Nature. Cambridge University Press, Cambridge,
UK; 2000.
15. Biserni C, Rocha LAO, Bejan A. Inverted Fins: Geometric Optimization of the Intrusion into a
Conducting Wall, Int. J. Heat Mass Tran. 2004; 47: 2577­2586.
16. Estrada E, Dos Santos ED, Isoldi LA, Rocha LAO. Genetic Algorithm and Constructal Design for
Geometric Optimization of Y‐Shaped. In: V MCSUL, Rio Grande ­ RS. V Conferência Sul de
Modelagem Computacional, 2012. v. 1. p. 45­50.
17. Lorenzini G, Biserni C, Isoldi LA, Dos Santos ED, Rocha LAO. Constructal Design Applied to the
Geometric Optimization of Y­shaped Cavities Embedded in a Conducting Medium. J. Electronic
Packaging. 2011; 133: 041008­1 – 041008­8, doi: 10.1115/1.4005296.
18. Lorenzini G, Garcia FL, Dos Santos ED, Biserni C, Rocha LAO. Constructal design applied to the
optimization of complex geometries: T­Y­shaped cavities with two additional lateral intrusions cooled
by convection, Int. J. Heat Mass Tran. 2012; 55 (5 – 6): 1505­1512, doi:
10.1016/j.ijheatmasstransfer.2011.10.057.
19. Lorenzini G, Biserni C, Link FB, Isoldi LA, Dos Santos ED, Rocha LAO. Constructal Design of T­
Shaped Cavity for Several Convective Fluxes Imposed at the Cavity Surfaces. J. Eng. Thermophys.
2013; 22 (4): 309­321, doi: 10.1134/S1810232813040048.
View publication stats

G.V. Gonzales et al., Scientia Plena 11, 081321 (2015) 11

20. Hajmohammadi MR, Poozesh S, Campo A, Nourazar SS. Valuable Reconsideration in the Constructal
Design of Cavities. Energy Convers. Manage. 2013; 66: 33–40, doi:
10.1016/j.enconman.2012.09.031.
21. Lorenzini G, Biserni C, Estrada ESD, Dos Santos ED, Isoldi LA, Rocha LAO. Genetic Algorithm
Applied to Geometric Optimization of Isothermal Y­Shaped Cavities. J. Electronic Packaging. 2014;
136 (3): 031011­1­031011­8, doi: 10.1115/1.4027421.
22. Lorenzini G, Biserni C, Estrada ED, Isoldi LA, Dos Santos ED, Rocha LA.O. Constructal Design of
Convective Y­Shaped Cavities by Means of Genetic Algorithm. J. Heat Trans­T ASME. 2014; 136
(7): 071702­071702­10, doi: 10.1115/1.4027195.
23. Xie Z, Chen L, Sun F. Geometry Optimization of T­Shaped Cavities According to Constructal
Theory. Math. Comput. Model. 2010; 52 (9 – 10):1538–1546, doi: 10.1016/j.mcm.2010.06.017.
24. Reddy JN, Gartling DK. The Finite Element Method in Heat Transfer and Fluid Dynamics. CRC,
Boca Raton, Florida,USA;1994.
25. Zienkiewicz OC, Taylor RL. The finite element method: The Basis. Volume 1, Butterworth­
Heinemann, Oxford,.Ed 5; 2000.
26. MATLAB User’s Guide, Version 6.0.088, Release 12, The Mathworks Inc., Natick, MA, 2000.
27. Hines WW, Montgomery DC, Goldsman DM, Borror CM. ,2006. Tradução: VR Lima, Probabilidade
e Estatística na Engenharia. Cap.12, Rio de Janeiro, LTC.