Beruflich Dokumente
Kultur Dokumente
ABSTRACT
In the present study, multi-objective optimization of Forward-Curved (FC) blade centrifugal
fans is performed in three steps. In the first step, Head rise (HR) and the Head loss (HL) in a set
of FC centrifugal fan is numerically investigated using commercial software NUMECA. Two
meta-models based on the evolved group method of data handling (GMDH) type neural
networks are obtained, in the second step, for modeling of HR and HL with respect to
geometrical design variables. Finally, using the obtained polynomial neural networks, multiobjective genetic algorithms are used for Pareto based optimization of FC centrifugal fans
considering two conflicting objectives, HR and HL.
Keywords: forward-curved blade centrifugal fan; multi-objective optimization; CFD; GMDH;
genetic algorithms.
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
63
1. INTRODUCTION
Forward-Curved (FC) blade centrifugal fans or squirrel cage fans are the group of turbo
machines which occur in industry in large scales. Nowadays, increasing demands and competition
require the use of good models to describe the operation of FC centrifugal fans. Kim and Seo [1]
presented a response surface method using three dimensional Navier-Stokes analyses to optimize
the shape of a forward-curved blade centrifugal fan and finally improved the efficiency of the fan.
Lu et al. [2] numerically investigated the internal flow field of centrifugal fans. They could
decrease the head loss and increase the total pressure using splitter blades in centrifugal fans.
Optimization of FC fans is indeed a multi-objective optimization problem rather than a single
objective optimization problem that has been considered so far in the literature. Sugimura et al. [3]
investigated a multi-objective optimization process on centrifugal fans using multi-objective
robust design exploration method (MORDE). They tried to determine the design variables which
have the minimum turbulent noise level and the maximum efficiency. Besides applications to fan
optimization, in recent years there have been many efforts to increase the performance of different
types of turbo machines. Safikhani and Nourbakhsh [4] investigated a multi-objective
optimization approach on centrifugal pumps. They finally presented the Pareto front for
centrifugal pumps and defined five optimum points which had the best efficiency and cavitation
behavior. Derakhshan et al. [5, 6] optimized a pump as a turbine machine (PAT) for increasing the
efficiency using the genetic algorithms (GAs) and incomplete sensitivities method.
In centrifugal fans there are some objective functions which are not independent of each other,
like efficiency, head rise and input shaft power, so these parameters are not suitable for multiobjective optimization process. Head rise and the head loss are important and independent
objective functions which can be used in a multi-objective optimization process. These objective
functions are either obtained from experiments or computed using very timely and high-cost
computational fluid dynamic (CFD) approaches, which cannot be used in an iterative
optimization task unless a simple but effective meta-model is constructed over the response
surface from the numerical or experimental data. Therefore, modeling and optimization of the
parameters is investigated in the present study, by using GMDH-type neural networks and multiobjective genetic algorithms in order to maximize the head rise and minimize the head loss.
System identification and modeling of complex processes using input-output data have
always attracted many research efforts. System identification techniques are applied in many
fields in order to model and predict the behavior of unknown and/or very complex systems
based on given input-output data [7]. In this way, soft-computing methods [8], which concern
computation in an imprecise environment, have gained significant attention. The main
components of soft computing, namely, fuzzy logic, neural network, and evolutionary
algorithms have shown great ability in solving complex non-linear system identification and
control problems. Many research efforts have been developed that make use of evolutionary
methods as effective tools for system identification [9]. Among these methodologies, Group
Method of Data Handling (GMDH) algorithm is a self-organizing approach by which
gradually complicated models are generated based on the evaluation of their performance on a
set of multi-input-single-output data pairs Xi , yi (i51, 2, , M). The GMDH was first
developed by Ivakhnenko [10] as a multivariate analysis method for complex systems modeling
and identification, which can be used to model complex systems without having specific
knowledge of the systems. The main idea of GMDH is to build an analytical function in a feed
forward network based on a quadratic node transfer function [11] whose coefficients are
obtained using regression techniques.
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
64
In recent years, however, the use of such self-organizing networks leads to successful
application of the GMDH-type algorithm in a broad range of areas in engineering, science, and
economics [12]. Moreover, there have been many efforts in recent years to deploy GAs to design
artificial neural networks since such evolutionary algorithms are particularly useful for dealing
with complex problems having large search spaces with many local optima [13]. In this way, Gas
have been used in a feed forward GMDH-type neural network for each neuron searching its
optimal set of connections with the preceding layer [14]. In the former reference, authors have
proposed a hybrid genetic algorithm for a simplified GMDH-type neural network in which the
connection of neurons are restricted to adjacent layers. Moreover a multi-objective genetic
algorithm has also been recently used by some of authors to design GMDH-type neural
networks considering some conflicting objectives [15, 16].
In this paper, the head rise and the head loss in a set of forward-curved blade centrifugal fans
are numerically investigated using NUMECA. Genetically optimized GMDH type neural
networks are then used to obtain polynomial models for the effects of geometrical parameters of
the FC fans on both HR and HL. This approach of meta-modeling of those CFD results allows
the use of iterative optimization techniques. The obtained simple polynomial models are then
used in a Pareto based multi-objective optimization approach to find the best possible
combinations of HR and HL, known as the Pareto front. The corresponding variations of design
variables, namely, geometrical parameters, known as the Pareto set, constitute some important
and informative design principles.
Standard ke model
k2
Lk
LVi
Ck zn
{ui uj
Lxi
e
Lxj
De
L
k2
Le
e
LVi
e2
~
Ck zn
{Ce2
{Ce1 ui uj
Dt Lxj
Lxj
k
e
Lxj
k
Dk
L
~
Dt Lxj
The dimensions of the case study in the present paper and some operating conditions for the
simulations are shown in Table 1. The simulations are performed using Numeca software.
Firstly one blade is modeled in Auto Blade 3.6 and then the Design 3D environment of Numeca
automatically generates the database with different design variables.
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
65
Value
333
210
150
0.34
634
5
30000
1
To parameterize the camber line curve, the simple Bezier method is used [17]. A schematic
definition of simple Bezier method is shown in Fig. 1. The design variables in this method are
leading edge angle (b1), trailing edge angle (b2) and the stagger angle (c). In the present paper
two sections are defined in the blades, one on hub and one on shroud as shown in Fig. 2. It is
supposed that b1 , b2 and c are equal at hub and shroud section due to the 2D nature of FC
blades centrifugal fan, which can mathematically be given by
b1Hub ~ b1Shroud ~Design Variable
The design variables and their range of variations are shown in Table 2. By changing the
geometrical independent parameters according to the Table 2, various designs will be generated
66
and evaluated using CFD. Consequently, some meta-models can be optimally constructed using
GMDH-type neural networks, which will be further used for multi-objective Pareto based
design of such centrifugal fans. In this way, 132 various CFD analyses have been performed due
to those different design geometrics.
For CFD grid generation, the Auto Grid environment of Numeca is coupled with the Auto
Blade environment. To test for grid independency, three grid types (named A, B, and C) with
increasing grid density are studied and their details are listed in Table 3. The computational
results of three grid types for different mass flow rates are compared in Table 4. As can be seen,
the maximum difference between the results is less than 6 % so the grid type (A) is used for all
computations in the present study. Figure 3 shows the details of the computational grid for the
centrifugal fans. The physical model used in the solver is the Reynolds-Averaged NavierStokes
equations and the k-e turbulence model. Mass flow, k and e are imposed at the fan inlet. A
static pressure outlet boundary condition is used at the outlet and finally periodic boundary
condition is applied between two blades. The computation is continued until the solution
converged with a total residual of less than 25. Samples of numerical results, using CFD are
shown in Table 5. A typical pressure contour in one of the simulations is shown in Fig. 4.
The results obtained in such CFD analysis can now be used to build the response surface of
both the head rise and the head loss for those different 132 geometries using GMDH-type
polynomial neural networks. Such meta-models will, in turn, be used for the Pareto-based
multi-objective optimization of the FC fans. A post analysis using the CFD software
NUMECA is also performed to verify the optimum results using the meta-modeling approach.
Finally, the solutions obtained by the approach of this paper exhibit some important trade-offs
From
To
c (deg)
b1 (deg)
b2(deg)
N (no. of blades)
7
12
25
25
25
47
60
40
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
67
Inlet
Outlet
Periodic
Meridional
20
24
29
25
27
33
79
86
104
37
43
55
444690
588632
762820
A
B
C
among those objective functions which can be simply used by a designer to optimally
compromise among the obtained solutions.
It is now possible to train a GMDH-type neural network to predict the output values ^yi for any
given input vector X ~xi1 ,xi2 ,xi3 , . . . ,xin
That is
^yi ~f^xi1 ,xi2 ,xi3 , . . . ,xin
i~1,2 . . . M ,
The problem is now to determine a GMDH-type neural network so that the square of difference
between the actual output and the predicted one is minimized, that is
M h
X
i2
f^xi1 ,xi2 ,xi3 , . . . ,xin {yi ?min
i~1
HL (m)
Max
Diff (%)
Max
Diff (%)
14.82
12.10
10.86
14.05
12.65
10.77
14.78
12.08
10.93
5.19
4.50
1.46
1.22
0.99
0.88
1.21
0.98
0.88
1.20
0.98
0.86
1.63
1.00
2.27
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
68
General connection between inputs and output variables can be expressed by a complicated
discrete form of the Volterra functional series in the form of
y~a0 z
n
X
ai xi z
i~1
n X
n
X
aij xi xj z
i~1 j~1
n X
n X
n
X
10
aijk xi xj xk z . . .
Which is known as the Kolmogorov-Gabor polynomial [18]. This full form of mathematical
description can be represented by a system of partial quadratic polynomials consisting of only
two variables (neurons) in the form of
Input Data
Num
c(deg)
b1(deg)
b2(deg)
HR(m)
HL(m)
1
2
3
4
5
6
7
8
9
130
131
132
7.45
7.45
9.54
24.45
7.45
24.45
7.45
24.4
7.45
12.14
46.14
46.14
46.14
12.14
29.14
29.14
46.14
12.14
42.77
59.77
59.77
59.77
42.77
25.77
25.77
25.77
25.77
40
40
40
35
35
35
30
30
30
14.236
18.777
15.587
18.89
13.951
12.530
10.841
11.360
10.373
1.559
2.056
1.707
2.008
1.481
1.331
1.168
1.220
1.113
9.54
7.45
9.54
12.14
46.14
12.14
59.77
42.77
42.77
25
25
25
9.488
15.427
5.768
0.745
1.729
0.645
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
69
^y~G xi ,xj ~a0 za1 xi za2 xj za3 xi xj za4 xi 2 za5 xj 2
11
There are two main concepts involved within GMDH-type neural networks design, namely,
the parametric and the structural identification problems. In this way, some authors presented a
hybrid GA and singular value decomposition (SVD) method to optimally design such
polynomial neural networks. The methodology in these references has been successfully used in
this paper to obtain the polynomial models of HR and HL. The obtained GMDH-type
polynomial models have shown very good prediction ability of unforeseen data pairs during the
training process which will be presented in the following sections.
The inputoutput data pairs used in such modeling involve two different data tables obtained
from the CFD simulation discussed in Section 2. Both of the tables consist of four variables as
inputs, namely, the geometrical parameters of the FC fans c, b1, b2 (Fig. 1) and N (number of
blades) and outputs, which are HR and HL. The tables consist of a total of 132 patterns, which
have been obtained from the numerical solutions to train and test such GMDH type neural
networks.
However, in order to demonstrate the prediction ability of the evolved GMDH type neural
networks, the data in both inputoutput data tables have been divided into two different sets,
namely, training and testing sets. The training set, which consists of 112 out of the 132 input
output data pairs for HR and HL, is used for training the neural network models. The testing
set, which consists of 20 unforeseen inputoutput data samples for HR and HL during the
training process, is merely used for testing to show the prediction ability of such evolved
GMDH type neural network models.
The GMDH type neural networks are now used for such inputoutput data to find the
polynomial models of head rise and head loss with respect to their effective input parameters. In
order to design, genetically, such GMDH type neural networks described in the previous
section, a population of 10 individuals with a crossover probability (Pc) of 0.7 and mutation
probability (Pm) 0.07 has been used in 500 generations for HR and HL. The corresponding
polynomial representation for Head Rise (HR) is as follows:
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
70
12a
12b
12c
12d
12e
12g
Similarly, the corresponding polynomial representation of the model for Head Loss (HL) is in
the form of
Y 1 ~1:2905{0:0799c z0:01905b1 z0:00382c2 z3:383e{005b1 2 {0:000769c b1
13a
13b
13c
The very good behavior of such GMDH type neural network model for head rise is also
depicted in Fig. 5, both for the training and testing data. Such behavior has also been shown for
the training and testing data of head loss in Fig. 6. It is evident that the evolved GMDH type
neural network in terms of simple polynomial equations successfully model and predict the
outputs of the testing data that have not been used during the training process. The models
obtained in this section can now be utilized in a Pareto multi-objective optimization of the FC
centrifugal fans considering both HR and HL as conflicting objectives. Such a study may unveil
some interesting and important optimal design principles that would not have been obtained
without the use of a multi-objective optimization approach.
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
71
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
72
F X ~f1 X , f2 X , . . . , fk X T ,
14
15
16
Where X [<n is the vector of decision or design variables, and F X [<k is the vector of
objective functions, which must each be either minimized or maximized. However, without loss
of generality, it is assumed that all objective functions are to be minimized.
In order to investigate the optimal performance of the FC centrifugal fan in different
conditions, the polynomial neural network models obtained in section 3 are now employed in a
multi-objective optimization procedure. The two conflicting objectives in this study are Head
Rise (HR) and Head Loss (HL) that is to be simultaneously optimized with respect to the design
variables c, b1, b2 (Fig. 1) and N (Number of blades). The multi-objective optimization problem
can be formulated in the following form:
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
73
8
Maximize
>
>
>
>
>
>
< Minimize
Subject to
>
>
>
>
>
>
:
17
74
Table 6. The values of objective functions and their associated design variables of the optimum
points.
Point
A
B
C
D
c (deg)
b1(deg)
b2(deg)
HR (m)
13.48
12.54
13.33
17.38
12.36
12.14
12.01
42.19
25.24
57.82
57.57
58.57
25
25
33
38
6.5036
10.695
14.376
18.981
HL (m)
.52068
.8093
1.156
1.8272
function in the Pareto front would cause a worse value for another objective. The
corresponding decision variables of the Pareto front shown in Fig. 7 are the best possible
design points so that if any other set of decision variables is chosen, the corresponding values of
the pair of objectives will locate a point inferior to this Pareto front. Such inferior area in the
space of the two objectives is in fact bottom/right side of Fig. 7.
In Fig. 7, the design points A and D stand for the best head loss and the best head rise.
Moreover, the other optimum design point, B can be simply recognized from Fig. 7. The design
point, B exhibit important optimal design concepts. In fact, optimum design point B obtained in
this paper exhibits an increase in head loss (about 23.58 %) in comparison with that of point A
whilst its head rise improves about 30.6 % in comparison with that of A.
It is now desired to find a trade-off optimum design point compromising both objective
functions. This can be achieved by the method employed in this paper, namely, the mapping
method. In this method, the values of objective functions of all non-dominated points are
mapped into interval 0 and 1. Using the sum of these values for each non-dominated point, the
trade-off point simply is one having the minimum sum of those values. Consequently, optimum
design point C is the trade-off point which has been obtained from the mapping method.
There are some interesting design facts which can be used in the design of such FC fans. It is
clear from Figs. 9 and 10 that from point A to B, design variables c, b1 and N are nearly
constant whereas b2 varies almost linearly. Similarly, from point B to point C, the geometrical
75
Fig. 9. Variations of design variables corresponding to the Pareto front of head rise.
Fig. 10. Variations of design variables corresponding to the Pareto front of head loss.
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
76
Fig. 11. Overlay graph of the obtained optimal Pareto front with the numerical data.
design parameters c, b1 and b2 are nearly constant and the number of blades (N) varies almost
linearly. Finally from point C to point D, design parameters b2 and N are nearly constant
whereas c and b1 varies almost linearly. These useful relationships that indefeasible between the
optimum design variables of FC fans cannot be discovered without the use of multi-objective
Pareto optimization process presented in this paper.
The Pareto front obtained from the GMDH-type neural network model (Fig. 7) has been
superimposed with the corresponding CFD simulation results in Fig. 11. It can be clearly seen
from this figure that such obtained Pareto front lies on the best possible combination of the
objective values of CFD data, which demonstrate the effectiveness of this paper, both in
deriving the model and in obtaining the Pareto front.
In a post numerical study, the design points of the obtained Pareto front have been reevaluated using CFD. It should be noted that the optimum design points of the Pareto set are
not included in the training and testing sets utilized meta-modeling using GMDH-type neural
network which makes such re-evaluation sensible. The results of such CFD analysis reevaluations have been compared with those of numerical results using the GMDH model in
Table 7. As seen the CFD data agree well with the GMDH data.
5. CONCLUSION
Genetic algorithms have been successfully used both for optimal design of generalized
GMDH type neural network models of Head Rise and Head Loss in FC centrifugal fans and
for multi-objective Pareto based optimization of such processes. Two different polynomial
relations for HR and HL have been found by evolved GS-GMDH type neural networks using
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
77
HL(m)
GMDH
CFD
Error (%)
GMDH
CFD
Error (%)
6.5036
10.695
14.376
18.981
6.351
10.09
14.11
18.25
2.31
5.99
2.87
4.01
.52068
.8093
1.156
1.8272
.501
.778
1.10
1.765
3.79
3.98
5.09
3.23
some CFD simulations for inputoutput data of the fans. The derived polynomial models have
been then used in an evolutionary multi-objective Pareto based optimization process so that
some interesting and informative optimum design aspects have been revealed for fans with
respect to the design variables such as geometrical parameters of c, b1, b2 (Fig. 1) and number of
blades (N). Consequently, some very important tradeoffs in the optimum design of FC
centrifugal fans have been obtained and proposed based on the Pareto front of two conflicting
objective functions. Such combined application of GMDH type neural network modeling of
inputoutput data and subsequent non-dominated Pareto optimization process of the obtained
models is a very promising technique for discovering useful and interesting design relationships.
REFERENCES
1. Kim, K. and Seo, S., Application of numerical optimization technique to design of forwardcurved blades centrifugal fan, Japan Society of Mechanical Engineers International Journal
Series B, Vol. 49, No. 1, pp. 152158, 2006.
2. Lu, Y., Lu, Q., Wang, J. and Zhang, J., 3D Numerical simulation of centrifugal fan inside flow
field and modification investigation, Proceedings of the International Conference on Power
Engineering, pp. 429433, 2007.
3. Sugimura, K., Jeong, S., Obayashi, S. and Kimura, T., Kriging-model-based multi-objective
robust optimization and trade-off rule mining of a centrifugal fan with dimensional
uncertainty, Journal of computational science and technology, Vol. 3, No. 1, 2009.
4. Safikhani, H. and Nourbakhsh, A., Modeling and multi-objective optimization of centrifugal
pumps using CFD and genetic algorithms, Engineering Applications of Computational Fluid
Mechanics, in press, 2010.
5. Derakhshan, S., Mohammadi, B. and Nourbakhsh, A., Incomplete sensitivities for 3D radial
turbo machinery blade optimization, Computers & Fluids, Vo. 37, No. 10, pp. 13541363, 2008.
6. Derakhshan, S., Mohammadi, B. and Nourbakhsh, A., The comparison of incomplete
sensitivities and genetic algorithms application in 3D radial turbo machinery blade
optimization, Computers & Fluids, Vo. 39, No. 10, pp. 20222029, 2010.
7. Astrom, K.J. and Eykhoff, P., System identification: A survey, Automatica, Vol. 7, pp. 123162,
1971.
8. Sanchez, E., Shibata, T. and Zadeh, L.A., Genetic Algorithms and Fuzzy Logic Systems, World
Scientific, Riveredge, NJ, 1997.
9. Kristinson, K. and Dumont, G., System identification and control using genetic algorithms,
IEEE Transactions on Systems, Man and Cybernetics, Vol. 22, No. 5, pp. 10331046, 1992.
10. Ivakhnenko, A.G., Polynomial theory of complex systems, IEEE Transactions on Systems,
Man and Cybernetics, SMC-1, pp. 364378, 1971.
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
78
11. Farlow, S.J., Self-Organizing Method in Modeling: GMDH Type Algorithm, Marcel Dekker Inc.,
1984.
12. Porto, V.W., Evolutionary computation approaches to solving problems in neural computation, In Handbook of Evolutionary Computation Back, T., Fogel, D. B. and Michalewicz, Z.
(Eds), Institute of Physics Publishing and New York: Oxford University Press, pp. D1.2:1
D1.2:6, 1997.
13. Amanifard, N., Nariman-Zadeh, N., Borji, M., Khalkhali, A. and Habibdoust, A., Modeling and
Pareto optimization of heat transfer and flow coefficients in micro channels using GMDH type
neural networks and genetic algorithms, Energy Converse Management, Vol. 49, pp. 311325,
2008.
14. Atashkari, K., Nariman-Zadeh, N., Golcu, M., Khalkhali, A. and Jamali, A., Modeling and
multi-objective optimization of a variable valve timing spark-ignition engine using polynomial
neural networks and evolutionary algorithms, Energy Converse Management; Vol. 48, No. 3,
pp. 10291041, 2007.
15. Safikhani, H., Hajiloo, A. and Ranjbar, M.A., Modeling and multi-objective optimization of
cyclone separators using CFD and genetic algorithms, Computers & Chemical Engineering, in
press, Chemical Engineering Research and Design, 2010.
16. Safikhani, H., Akhavan-Behabadi, M.A., Nariman-Zadeh, N. and Mahmoodabadi, M.J.,
Modeling and multi-objective optimization of square cyclones using CFD and neural
networks, in press, 2010.
17. Auto Blade v.7 user guide. Numeca, 2006.
18. Coello Coello, C.A. and Christiansen, A.D., Multi-objective optimization of trusses using
genetic algorithms, Computers & Structures, Vol. 75, pp. 647660, 2000.
19. Pareto, V., Cours deconomie politique, Lausanne, Switzerland, Rouge, 1896.
Transactions of the Canadian Society for Mechanical Engineering, Vol. 35, No. 1, 2011
79