You are on page 1of 12

Original Article

Proc IMechE Part E:


J Process Mechanical Engineering
Thermohydraulic optimization of 2019, Vol. 233(3) 589–600
! IMechE 2018

triple concentric-tube heat exchanger: Article reuse guidelines:


sagepub.com/journals-permissions

A multi-objective approach DOI: 10.1177/0954408918779232


journals.sagepub.com/home/pie

Tarikayehu Amanuel and Manish Mishra

Abstract
In the present study, optimization of heat transfer and pressure drop characteristics in a triple concentric tube heat
exchanger has been done using the results of numerical simulation. A commercial CFD software ANSYS Fluent v17.0 has
been employed for simulating the flow and heat transfer, while optimization has been done by Response surface meth-
odology (RSM) and Genetic algorithm (GA). The effective parameters in the study are Reynolds number
(2500 4 Re 4 10,000) and Length to hydraulic diameter ratio (100 4 L/Dh 4 220). The optimum values, as well as
the functional relationship between the design factors (Re and L/Dh) and response variables (Nu and f), have also been
developed. It has been found that both the design factors (Re and L/Dh) have a strong influence on the response variables
(Nu and f). With the increase in Re (flow rate), a large growth in Nusselt number and decline in friction factor has been
observed. However, with the increase in L/Dh, an enormous decrease in both Nusselt number and friction factor has
been found.

Keywords
Triple concentric-tube, numerical simulation, optimization, thermohydraulic performance, genetic algorithm

Date received: 4 March 2018; accepted: 29 April 2018

possible without making a major compromise in the


Introduction desired output. This is when multi-objective optimiza-
In an industrial sector, the need to transfer heat tion comes into play.
from one fluid to another makes the use of heat Optimization of heat exchangers has been the focus
transfer devices inevitable. Energy, being a scarce of many researchers as it aims to develop the optimum
resource, is nowadays demanding tremendous effort setting for the desired variable under specified con-
to conserve it for longer future. On one end, the straints. Mishra et al.1 applied GA based optimization
need for saving energy by properly handling critical technique for crossflow plate-fin heat exchanger using
heat flux conditions is a big challenge. On the other offset-strip fins. They have considered a specific prob-
end, the availability of limited space in industries lem and developed an optimum solution for geomet-
adds up another difficulty in the field. To develop rical parameters as well as fin specifications that
an acceptable compromise on these two limitations, would result in minimum entropy generation for a
a number of works have been conducted and are specified heat duty. Several studies of multi-objective
still in progress. Several mechanisms, such as by alter- optimization for different configurations of heat
ing the geometry, using extended surfaces and exchangers have also been performed in the past dec-
developing a better heat transfer fluids have all been ades. Multi-objective optimization of plate heat
applied to obtain an enhanced heat transfer. exchanger using genetic algorithm has been done for
However, most of the times, the acceptability of thermal and hydraulic characteristics.2 They have
these efforts faces a challenge as there is always a
possibility of unwanted results being involved in the Department of Mechanical and Industrial Engineering, Indian Institute of
process. Since the inherent nature of these problems is Technology Roorkee, Roorkee, India
unlikely to fully avoid, there comes a point where
Corresponding author:
choice should be made to develop a reasonable bal-
Tarikayehu Amanuel, Department of Mechanical and Industrial
ance between these conflicting objectives. In a more Engineering, Indian Institute of Technology Roorkee, Roorkee,
elaborate way, one needs to develop a mechanism to Uttarakhand 247667, India.
get maximum benefit keeping the penalty as low as Email: castrofdl@gmail.com
590 Proc IMechE Part E: J Process Mechanical Engineering 233(3)

conducted sensitivity analysis to study the effect of


Numerical procedure
geometrical parameters on thermal and hydraulic per- In this section, the preliminary numerical simulation
formance. The results show that the port center dis- of heat transfer and pressure drop of triple concentric-
tance, plate spacing and a number of thermal plates tube heat exchanger has been discussed. Initially, the
have a strong influence on the performance of the heat simulation has been conducted for a fixed geometry of
exchanger. Wang et al.3 implemented a combination the heat exchanger (length to hydraulic diameter
of CFD, RSM, and GA to conduct multi-objective ratio, L/Dh ¼ 140) and Re: 2500–10,000. The heat
optimization of shell-and-tube heat exchanger with transfer and pressure drop results of this simulation
fold helical baffles. They studied the effect of the were documented. Later the simulation has been
design parameters (helical angle, overlapped degree, extended for varying length to hydraulic diameter
and shell-side inlet velocity) on the Nusselt number ratio: L/Dh ¼ 100,180,220. Two possibilities of vary-
and pressure drop. The results of their study revealed ing L/Dh can be considered here; varying either the
that both the Nusselt number and pressure drop length or hydraulic diameter one at a time. In
increase with the decrease of helical angle and shell- the present work, the length of the tube was allowed
side inlet velocity, unlike with the increase in the over- to vary in such a way that the non-dimensional factor
lapped degree, which also caused an increase of the is maintained at the above-mentioned values.
same. They also developed a functional relationship This offers additional advantage of studying the indi-
between the objective functions (Nu and f) and the vidual effects of each parameter as well as uniformity
design parameters. Wen et al.4 have also performed of the analysis. In the subsequent simulations, the
multi-parameter optimization of shell-and-tube heat heat transfer values of the previous simulation
exchanger with helical baffles. They considered ther- have been used. In doing so, the effect of varying
mal resistance as the objective function to study the geometry and flow rate on the thermohydraulic per-
effect of helix angle, overlapped degree and shell side formance of triple concentric-tube heat exchanger can
velocity. The results indicated that the entransy dissi- be studied.
pation thermal resistance increased with the helix
angle but decreased with the shell-side velocity while
the effect of overlapped degree was insignificant.
Computational domain
Multi-objective shape optimization of double pipe A Triple concentric-tube heat exchanger and its com-
heat exchanger with inner corrugated tube has been putational domain are shown in Figure 1. The
done by Han et al.5 RSM has been applied consider- exchanger consists of three tubes such that the hot
ing four design parameters (dimensionless corruga- fluid flows through the inner annulus whereas the
tion pitch, dimensionless corrugation height, cold fluid and the intermediate temperature fluids
dimensionless corrugation radius and Reynolds flow through the inner tube and outer annulus,
number) while overall heat transfer performance has respectively. Pure water with the flowrate ranging
been the objective function. They finally developed from laminar to turbulent has been used in all the
the optimum set of design parameters for which the three passages. The detail geometrical description of
overall heat transfer performance was found the max- the heat exchanger is given in Table 1.
imum. Additional works dealing with multi-objective
optimization of the thermohydraulic performance of
Governing equations and boundary conditions
heat exchangers are available in the published
literature.6–14 The flow and heat transfer for the hot fluid flowing in
Despite so many important works being performed the inner annulus were defined by conservation equa-
on the optimization of heat exchangers with /without tions of mass, momentum and energy.
nanofluids for the desired outputs, the focus of those Heat transfer occurs among the fluids through the
studies is mainly on conventional two-fluid heat walls separating them. The governing equations in a
exchangers. Therefore, conducting an investigation generic form including continuity, momentum and
on three-fluid heat exchangers would be a great energy equations as well as turbulence model can be
effort. In this study, optimization of the heat transfer expressed in tensor notation as follows.
and pressure drop characteristics of triple concentric-  
tube heat exchanger has been performed based on r: V~ ¼ r:ð  rÞ þ S ð1Þ
CFD results. Response surface methodology (RSM)
and Genetic algorithm (GA) approaches have been where the terms  and S represent the appropriate
applied to develop an optimum response as well as diffusion and source terms, respectively. The details of
a functional relationship based on the studied the governing equations are available in published
range of design/control parameters. Nusselt number work.15
and friction factor have been chosen as response For modelling turbulent flow, k-" turbulence model
variables while Reynolds number and length to has been adopted. The k-" turbulence model com-
hydraulic diameter ratio have been used as design prises two additional equations; namely, turbulent
variables. kinetic energy (k) and rate of dissipation (").
Amanuel and Mishra 591

Intermediate temperature fluid

Hot fluid

Cold fluid

Figure 1. Schematic representation of Triple Concentric-tube Heat exchanger and its Computational domain.

Table 1. Geometrical and boundary conditions.

Geometry Inner tube Inner annulus Outer annulus

Inner diameter[m] D1i ¼ 0.01 D2i ¼ 0.022 D3i ¼ 0.033


Outer diameter[m] D1o ¼ 0.012 D2o ¼ 0.024 D3o ¼ 0.036
Thickness, t[m] 0.001 0.001 0.0015
Hydraulic diameter, Dh[m] 0.01 0.01 0.009
Length, L [m] 1.41 1.41 1.41

Boundary conditions
Fluid stream Inlet Outlet
Cold fluid Velocity inlet Pressure outlet
Tin ¼ 30 C (303K)
Hot fluid Velocity inlet Pressure outlet
Tin ¼ 60 C (333K)
Intermediate temperature fluid Velocity inlet Pressure outlet
Tin ¼ 40 C (313K)
Fluid-wall interface Coupled /conjugate heat transfer
Outer wall No-slip condition and Insulated

Turbulent kinetic energy The constants in equations 3 and 4 are defined as


 
t,m
r:ðVkÞnf ¼ r: rk þðGm  m "Þnf ð2Þ C1 ¼ 1:44, C2 ¼ 1:92, C ¼ 0:09, k ¼ 1:0, " ¼ 1:3
k nf
Turbulent rate of dissipation
  In the present work, RNG k-" ‘‘renormalization
t,m "
r:ðV"Þnf ¼ r: r" þ ðC1 Gm  C2 m "Þnf group’’ turbulence model is chosen. The RNG
" nf k model is more accurate and reliable for a wider class
ð3Þ of flow than the standard k-" model.16 These govern-
ing and turbulence equations with the corresponding
where Gm represents the generation of turbulent kin- boundary conditions have been solved to obtain the
etic energy due to mean velocity gradients; sk and s" performance parameters: Nusselt number and friction
are effective Prandtl numbers for turbulent kinetic factor.
energy and rate of dissipation, respectively; C1 and At the hot fluid inlet, inlet velocities corresponding
C2 are constants and t,m is the eddy viscosity and to the values of operating Re and the inlet tempera-
is modeled as follows. tures of each stream were specified. The complete
description of the boundary conditions is given in
k2 Table 1. For a particular simulation, the inlet veloci-
t,m ¼ m C ð4Þ
" ties of cold and intermediate temperature fluids were
592 Proc IMechE Part E: J Process Mechanical Engineering 233(3)

fixed while the inlet velocity of hot fluid was allowed Reynolds number
to vary.
vDh
Re ¼ ð12Þ

Solution procedure
The commercial CFD package ANSYS Fluent v17.0 Hydraulic diameter
has been used to solve the present problem. Second-
order upwind differencing scheme with SIMPLE algo- 4A
Dh ¼ ð13Þ
rithm was applied to solve the convection terms. P
Convergence criteria for all equations were set at
106. RNG k-" model is adopted for turbulence mod- The Darcy friction factor for fully developed flow
eling as it offers better capturing near critical zones. in a circular tube
A point implicit (Gauss-Seidel) linear equation solver
along with an Algebraic multi-grid (AMG) has been 2 PDh
f¼ ð14Þ
applied to solve the resulting linear systems. v2 L

Data reduction
The quantitative analysis of thermohydraulic per-
Grid independence test and code validation
formance is performed based on the approach fol- Unstructured mesh with quadrilateral elements was
lowed in literature.17 generated for simulation. Grid independence study
The heat transferred from the hot fluid through has been conducted using six different grid sizes
inner annulus with 81,040, 106,580, 131,160,189,346, 286,741 and
  301,238 elements. Nusselt number and friction
Qh ¼ mCp h ðTi  To Þh ð5Þ factor have been determined for the above varying
grid sizes at a certain value of Reynolds number
The heat transferred to the cold fluid in the inner (Re ¼ 2500) and length to hydraulic diameter ratio
tube (L/Dh ¼ 140). Figure 2(a) and (b) shows the plot of
  the same. It has been found that both the Nusselt
Qc ¼ mCp c ðTo  Ti Þc ð6Þ number and friction factor have not shown significant
changes after a grid size of 131,160. Thus, the same
The heat transferred to the intermediate tempera- grid size has been considered in the simulation for the
ture fluid in the outer annulus present study as it offers an optimum solution in terms
  of computational time and accuracy of the results.
Qi ¼ mCp i ðTo  Ti Þi ð7Þ To gain confidence in the numerical approach, the
Thus, the average heat transfer among the three results of the present work has been validated with
fluids can easily be given as published experimental work in the literature.18
CFD simulation has been done according to the
½Qh þ Qc þ Qi  actual operating conditions of the experiment.
Qav ¼ ð8Þ
2 Performance factors (Nusselt number and friction
factor) obtained from CFD simulation have been
Further, the average heat transfer coefficient of the compared to those obtained from the experiment as
hot fluid in the inner annulus is given by well as standard correlations; i.e., Dittus-Boelter,
Gnielinski and Blasius empirical correlations as
Qav
hh ¼ ð9Þ shown in Figure 3(a) and (b).
As LMTDav It has been shown that the result from present ana-
where the average log-mean temperature difference lysis satisfactorily agreed with those from correl-
among the three fluid streams can be calculated as ations. Therefore, the present numerical model can
adequately be considered for further optimization
½LMTDhc þ LMTDhi  study.
LMTDav ¼ ð10Þ
2
Multi-objective optimization
The dimensionless performance parameters have
been summarized as below. In this section, the simulation results of thermohy-
Nusselt number draulic performance have been used to determine
the optimum condition based on the considered
hDh range of the design parameters. The primary aim is
Nu ¼ ð11Þ
k to develop a functional correlation (relationship)
between ‘‘input variables’’ (Reynolds number and
Amanuel and Mishra 593

(a) (b)
22.62 0.0573

22.61 0.0572

22.61 0.0571
Re=2500,L/Dh=140
Re=2500, L/Dh=140
Nu

22.60 0.057

f
22.60 0.0569

22.59 0.0568

22.59 0.0567
80000 120000 160000 200000 240000 280000 320000 80000 120000 160000 200000 240000 280000 320000
Grid size Grid size

Figure 2. Grid independence: (a) Nusselt number and (b) Friction factor.

(a) (b)
55
0.065
50 Blassius
0.06
45 Gnielinski
0.055
40 Simulation
35 0.05 Experimental[18]
Nu

30 Dittus-Boelter 0.045
25 Gnielinski 0.04
20 Simulation
0.035
15 Experimental[18]
10 0.03
2000 4000 6000 8000 10000 2000 4000 6000 8000 10000

Re Re

Figure 3. Validation of (a) Nusselt number and (b) friction factor.

length to hydraulic diameter ratio) and ‘‘the of design parameters on the response variables. Its
response’’ (Nusselt number and Friction factor) importance is immense when the cost of generating
using Response surface methodology (RSM). The too many quantitative data is very expensive.
Nusselt number and friction factor results obtained Successful implementation of RSM helps in develop-
from CFD has been used for the same. These results ing a reasonable correlation between the design par-
have been used for optimization using the commercial ameters and response variables through statistical
software MINITAB 14.0. The range of design param- analysis of experimental data.
eters considered is given as: In RSM, the generic form of the most common
polynomial model is expressed as
. Reynolds number:(Re:2500–10,000)
. Length to hydraulic diameter ratio: (L/Dh:100–220) X
n X
n X
n
Y ¼ a0 þ ai xi þ aii x2i þ aij xi xj þ "
i¼1 i¼1 i,j¼1
First, the relationship between the design factors
and response variables has been obtained. Later, the ð15Þ
optimal solution has been developed by multi-objec-
tive optimization within Genetic algorithm toolbox in where Y ¼ response variable, a0 ¼ intercept, ai ¼ lin-
another commercial software MATLAB R2016a. ear regression coefficient of ith factor, aii ¼ quadratic
regression coefficients of the ith factor, aij ¼ inter-
action coefficient of the ith and jth factors, " ¼ statis-
Response surface methodology tical/random error.
RSM is one of the most productive tools to develop a Once the regression fit of the polynomial model
functional relationship between simultaneous effects was done between the design factors and the response
594 Proc IMechE Part E: J Process Mechanical Engineering 233(3)

variables obtained, analysis of variance (ANOVA) The optimization technique generally begins with an
has been performed to statistically analyze the results. initial set of random solution (population) and
ANOVA predicts the robustness of the approach advances toward better solutions through successive
through the determination of certain statistical iterations (generations). In multi-objective optimiza-
terms. In ANOVA, P-values less than 0.05 represent tion, there is no single (unique) solution for optimiz-
that the model terms are ‘‘statistically significant’’, ing multiple objectives simultaneously. Thus, the
meaning that the corresponding factor is considered target is to discover a reasonable set of the optimal
to have a significant impact on the response. While P- solutions called as Pareto-front. A more detailed
values greater than 0.05 means that the model terms review of the implementation of GA in various heat
are ‘‘insignificant’’ and they have been neglected in exchanger applications can be found in literature.21–25
further analysis. Further, the large F-value also indi- In the present study, multi-objective optimization
cates the great significance of the regression model. approach based on a non-dominated sorting genetic
R2, the coefficient of multiple determination, predicts algorithm (NSGA-II)26 has been implemented. The
the quality of fit. Generally, the value of R2 ranges optimal parameters (Reynolds number and Length
from 0 to 1, with values closer to 1 showing a higher to hydraulic diameter ratio) that result in the min-
quality of the response model. imum pressure drop (friction factor) and maximum
heat transfer (Nusselt number) under certain con-
Statistical analysis. A proper choice of design of experi- straints (working limits of the design parameters)
ments (DOE) is important for effective implementa- have been determined. The advantage of this
tion of RSM. The matrix contains two factors with approach is that a uniformly distributed Pareto
values on the uncoded basis, Re with five levels and front and an optimal solution can be obtained imme-
L/Dh with four levels, forming 20 runs. The corres- diately in a single run.
ponding responses (Nu and f), as predicted by CFD
have also been presented. Later, a central composite
design (CCD) with the face-centered approach has
Results and discussion
been applied to analyze with less number of data, In this section, results of the effect of design variables
without compromising the adequacy of prediction. on Nu and f have been presented. The validity of the
The second-order CCD is one of the most preva- results has also been investigated by studying various
lent and suitable DOE approaches among the kinds of outputs from the analysis.
RSM.19 To develop the best fit for the polynomial In Table 4, it can clearly be seen that the F and
model in Eq.15, face-centered central composite P-values of the model terms are highly significant. The
design (CCD) with the modified values of the design Large F-values of Nusselt number (825.35) and fric-
variables has been conducted. Each of the design vari- tion factor (464.87) indicates that the proposed model
ables take on new values on three levels (1 ¼ low, 0 is adequate to predict the response variables accur-
¼ midpoint, 1 ¼ high). Based on this setting, the ately. Further, linear, quadratic and interaction
quadratic polynomial regression has been performed terms are also highly significant. For all these terms,
by Minitab14.0. The summary of the design factors the associated P-values are less than 0.05 for both the
and their uncoded values along CFD results of Nu and f, confirming that all the model terms are
response variables are summarized in Tables 2 and highly significant. Thus, all the model terms should
3. Further, a genetic algorithm has been applied to be included in developing the functional relationship.
these results to get the optimum values.

Genetic algorithm Table 3. Levels of design factors and CFD results of response
variables.
A genetic algorithm is a very powerful tool in hand-
Design factors Response variables
ling optimization of various engineering problems. It
(uncoded values) (uncoded values)
is a search based technique that is fundamentally
based on Darwin’s theory of natural selection.20 Run Order Re L/Dh Nu f

1 2500 140 20.5885 0.0662134


2 10,000 140 36.8984 0.0398602
Table 2. Design variables and levels of their values. 3 2500 220 13.1995 0.0424368
Coded values 4 10,000 220 23.6792 0.0255468
5 2500 180 16.1273 0.0518672
Design factors symbol 1 0 1 6 10,000 180 28.9031 0.0312238
Reynolds number Re 2500 6250 10,000 7 6250 140 30.9013 0.0475710
Length to hydraulic L/Dh 140 180 220 8 6250 220 19.8372 0.0304887
diameter ratio 9 6250 180 24.2055 0.0372640
Amanuel and Mishra 595

Table 4. ANOVA table for Nusselt number and friction factor.

Degree of
Source Freedom(DF) Seq SS Adj SS Adj MS F P

Nusselt number
Regression 5 445.149 445.149 89.030 825.35 0.000
Linear 2 428.092 428.092 214.046 1984.31 0.000
Square 2 8.559 8.559 4.280 39.67 0.007
Interaction 1 8.498 8.498 8.498 78.78 0.003
Residual Error 3 0.324 0.324 0.108
Total 8 445.473
Friction factor
Regression 5 0.001256 0.001256 0.000251 464.87 0.000
Linear 2 0.001188 0.001188 0.000594 1098.69 0.000
Square 2 0.000046 0.000046 0.000023 42.78 0.006
Interaction 1 0.000022 0.000022 0.000022 41.42 0.008
Residual Error 3 0.000002 0.000022 0.000001
Total 8 0.001258

Table 5. Estimated Regression Coefficients for Nusselt model. Hence, it can be concluded that the model
number and friction factor. reflects the response variables quite accurately.
Figure 4(a) and (b) depicts the plots of residuals for
Term Coef SE Coef T P
the Nusselt number and friction factor respectively.
Nusselt number As it can be observed from the normal probability
Constant 43.6027 0.2448 99.037 0.000 plots in both the figures, the residuals are evenly dis-
Re 0.005062 0.1341 49.180 0.000 tributed on both sides of the straight line indicating
L/Dh 0.32 0.1341 39.369 0.000 that the distribution of residuals for the Nusselt
Re 2
1.2434E-07 0.2322 7.530 0.005 number and friction factor are normal and thus, pro-
(L/Dh)2 0.000691 0.2322 4.760 0.018 vide a decent fit to the data.
Re* L/Dh 9.717E-06 0.1642 8.876 0.003 Figure 5(a) and (b) displays the additional plots in
S ¼ 0.3284 R-Sq ¼ 99.9% R-Sq(adj) ¼ 99.8%
examining the quality of fit for Nu and f respectively.
In both the figures, comparison of actual (CFD pre-
Friction factor diction) against the RSM prediction is illustrated. It
Constant 0.16,481 0.000548 67.841 0.000 can be observed that all the data points precisely fall
Re 9.60,438E-06 0.000300 35.478 0.000 on the straight line, indicating adequacy of the model
L/Dh 7.5622E-04 0.000300 30.638 0.000 to successfully predict the response variables. In other
Re2 3.1408E-10 0.000520 8.497 0.003 words, the predicted values of the response variables
(L/Dh)2 1.18,822E-06 0.000520 3.657 0.035 are in close agreement with numerically obtained
Re* L/Dh 1.57,720E-08 0.000368 6.436 0.008 values of the same. Therefore, it can be anticipated
S ¼ 0.0007352 R-Sq ¼ 99.9% R-Sq(adj) ¼ 99.7% that the developed RSM model has comprehensive
accuracy.
Based on this setting, the functional relationships
between the design variables (Re and L/Dh) and
Table 5 displays the estimated regression coeffi- response variables (Nu and f) have been established
cients of each term for the response variables. As it as follows.
can be seen, the linear terms of Reynolds number and
length to hydraulic diameter ratio (Re, L/Dh) are the Nu ¼ 43:6027 þ 0:005062Re  0:32ðL=Dh Þ
most important terms for a Nusselt number as well as
 1:2435  107 Re2 þ 0:000691ðL=Dh Þ2
friction factor. This simply means that both Reynolds
number and length to hydraulic diameter ratio mainly  9:7171  106 ðRe  L=Dh Þ ð16Þ
affect the response variables. All the other terms with
their respective order of importance are presented in f ¼ 0:16481  9:60438  106 Re
the table. In addition, the quality of the fit is given  7:5622  104 ðL=Dh Þ þ 3:1408  1010 Re2
by the coefficient of multiple determination, R2. Its
value for Nusselt number is computed as (R-Sq þ 1:18822  106 ðL=Dh Þ2
(adj) ¼ 99.8%), meaning that only 0.2% of the total þ 1:5772  108 ðRe  L=Dh Þ ð17Þ
variation in Nu cannot be reflected by the regression
596 Proc IMechE Part E: J Process Mechanical Engineering 233(3)

(a)

(b)

Figure 4. Residual plots for (a) Nusselt number and (b) friction factor.

In the developed model, the insignificant terms


Response surface analysis
have been eliminated and only the significant ones
are considered. Eq. (16) and (17) have further been In this section, the effect of design variables on the
used as objective functions in multi-objective opti- response is discussed. Figure 6(a) shows the variation
mization tool of genetic algorithm in MATLAB of Nusselt number with Reynolds number and length
R2016a and the optimal results are obtained. The to hydraulic diameter ratio. From the figure, it can be
same has been discussed later in this session. seen that Nu is strongly dependent on both Re as well
Amanuel and Mishra 597

(a) (b) 0.07


40

35 0.06

30
0.05

CFD predicted
CFD predicted

25

0.04
S 0.215011
20
S 0 .0 0 0 4 8 1 3
R -S q 99.9%
R-S q 9 9 .9 %
R - S q ( a d j) 99.9%
0.03 R - S q ( ad j) 9 9 .9 %
15

10 0.02
10 15 20 25 30 35 40 0.02 0.03 0.04 0.05 0.06 0.07

RSM Predicted RSM Predicted

Figure 5. Comparison of (a) Nusselt number and (b) friction factor.

(a) (b)
Re L /D h Re L /D h
3 0 .0 0.055

2 7 .5 0.050

2 5 .0
0.045
Nu

2 2 .5
0.040
2 0 .0

0.035
1 7 .5

1 5 .0 0.030
2500 6250 10000 140 180 220 2500 6250 10000 140 180 220

Figure 6. Main effects plot: (a) Nusselt number and (b) friction factor.

(a) (b)
0.09
50 Re
Re
2500 0.08 2500
4500
4500
6500 0.07 6500
40
8500 8500
10000 10000
0.06
Nu

30
0.05

0.04
20
0.03

10 0.02
100 140 180 220 100 140 180 220
L /D h L /D h

Figure 7. Interaction plot: (a) Nusselt number and (b) friction factor.
598 Proc IMechE Part E: J Process Mechanical Engineering 233(3)

as L/Dh. With the increase in Re, Nu strongly factor with Reynolds number and length to hydraulic
increases, while it decreases with the increase in diameter ratio. Similar to the above case, a strong
L/Dh. This is an indication that both the Reynolds dependence of f on both the Re and L/Dh is observed.
number and length to hydraulic diameter ratio are With the increase in Re and L/Dh, f highly decreases.
very dominant in affecting the response variables. Thus, both Re and L/Dh are strongly dominant in
Similarly, Figure 6(b) shows the variation of friction affecting the friction factor.
Figure 7(a) and (b) shows the combined effect of
Reynolds number and length to hydraulic diameter
ratio on the Nusselt number and friction factor,
Table 6. Function values and decision variables. respectively. It can easily be seen that, for a certain
value of one of the design factors, the response vari-
Run order Re L/Dh Nu f
able strongly varies. For example, at Re ¼ 2500, the
1 9232.708 219.9997 23.04581 0.026091 Nusselt number decreases from 20.58 at L/Dh ¼ 140
2 8126.233 219.6131 22.23544 0.026887 to 13.2 at L/Dh ¼ 220 i.e. by 35.8%. Similarly, for
3 3356.834 219.9537 15.06431 0.038911 the same Re, the friction factor decreases from 0.047
4 5338.557 219.3147 18.76112 0.03226 at L/Dh ¼ 140 to 0.0305 at L/Dh ¼ 220 i.e. by 36%.
5 6860.083 219.9149 20.86265 0.028664 Likewise, at a certain value of L/Dh¼140, the Nusselt
number increases from 20.58 at Re¼2500 to 36.89 at
6 4169.681 219.3031 16.71795 0.035955
Re ¼ 10,000 i.e. by 44.2%. However, for the same
7 7836.433 219.789 21.94586 0.027194
value of L/Dh, the friction factor decreases from
8 4916.234 219.6154 18.04249 0.033448
0.0662 at Re ¼ 2500 to 0.0398 at Re ¼ 10,000 i.e. by
9 3524.511 219.0561 15.45692 0.038405 39.8%.
10 8673.61 219.8874 22.66706 0.026387
11 3691.548 219.1221 15.79352 0.037745
12 8884.578 219.9554 22.81663 0.026249
Optimization
13 7559.151 219.6024 21.68206 0.027576 Table 6 shows the functional values and the decision
14 6169.735 218.5219 20.06912 0.030267 values corresponding to the optimal situation. The
15 4717.608 219.2892 17.71928 0.034119 values finally lead to Figure 8 depicting the Pareto
16 3830.836 219.9532 16.02699 0.037073 front, i.e., the set of design variables that give the
17 6555.51 218.9134 20.56044 0.029381 optimum values of Nusselt number and friction
factor. As all the points in the (Re and L/Dh)
18 3356.584 219.9537 15.06379 0.038912
Pareto front are optimal points, they exhibit no

0.044

0.042

0.040

0.038 Pareto front

0.036
Friction factor

0.034

0.032

0.030

0.028

0.026

0.024
12 14 16 18 20 22 24
Nusselt number

Figure 8. Pareto front for objective functions.


Amanuel and Mishra 599

dominance over one another. Moving from one opti- has helped in reducing the size of the required data
mal point to another, one could easily observe an as the RSM provides useful interaction between the
improvement in one objective function while the various parameters of the system.
other objective function deteriorates and vice-versa. . The optimum values of the response variables in
Thus, based on the desired objective, which may the form of Pareto front have also been determined
vary depending on the situation/application, the opti- by GA in the range considered for design factors.
mum choice of the design variables can easily be made . Hence, the present analysis could be very helpful
from the available results within the considered range for engineers in making the optimal choice of oper-
of parameters. ating parameters based on the desired objective for
Thus, for the range of design factors considered in a concentric-tube three-fluid heat exchanger.
the present study, the results obtained from the Pareto
front have can be accepted as optimum values of the
Declaration of conflicting interests
response variables, Nu and f.
The author(s) declared no potential conflicts of interest with
respect to the research, authorship, and/or publication of
Conclusion this article.
Numerical simulation for thermohydraulic perform-
ance of triple concentric tube heat exchanger using Funding
CFD and its optimization using RSM and GA have The author(s) disclosed receipt of the following financial
been conducted. After validation of the numerical support for the research, authorship, and/or publication
approach, optimization has been performed to of this article: The authors acknowledge the financial sup-
port provided by the Council of Scientific and Industrial
develop a functional relationship between design fac-
Research (CSIR) for the Extra-Mural Research (EMR) at
tors and response variables as well as to determine the Department of Mechanical and Industrial Engineering,
optimum values for the range of design variables. The Indian Institute of Technology Roorkee.
design factors considered are Reynolds number (Re:
2500–10,000) and length to hydraulic diameter ratio
ORCID iD
(L/Dh: 100–220). Nusselt number (Nu) and friction
factor (f) have been chosen as the response variables. Tarikayehu Amanuel http://orcid.org/0000-0001-6226-
Based on the results of numerical simulation and 2098.
multi-objective optimization, the basic findings of
this study have been summarized as follows. References
1. Mishra M, Das PK and Sarangi S. Second law based
. Analysis of variance (ANOVA) showed that the optimisation of crossflow plate-fin heat exchanger
developed model is adequate to successfully predict design using genetic algorithm. Appl Therm Eng 2009;
the response variables. 29: 2983–2989.
. The functional relationship developed between the 2. Imran M, Pambudi NA and Farooq M. Thermal and
design factors and the response variables can be hydraulic optimization of plate heat exchanger using
multi objective genetic algorithm. Case Stud Therm Eng
used to predict the thermohydraulic characteristics
2017; 10: 570–578.
at any condition.
3. Wang S, Xiao J, Wang J, et al. Application of response
. Both the design factors (Re and L/Dh) have a surface method and multi-objective genetic algorithm to
strong influence on the Nusselt number and fric- configuration optimization of Shell-and-tube heat
tion factor. With the increase in Re (flow rate), a exchanger with fold helical baffles. Appl Therm Eng
large growth in Nusselt number and decline in fric- 2018; 129: 512–520.
tion factor has been observed. However, with the 4. wen J, Gu X, Wang M, et al. Multi-parameter optimiza-
increase in L/Dh, a large decrease in both Nusselt tion of shell-and-tube heat exchanger with helical baffles
number and friction factor have been found. based on entransy theory. Appl Therm Eng 2018; 130:
. It was found that, for a certain value of one of the 804–813.
design factors, the response variable strongly 5. Han H-Z, Li B-X, Wu H, et al. Multi-objective shape
varies. optimization of double pipe heat exchanger with inner
corrugated tube using RSM method. Int J Therm Sci
 At Re ¼ 2500, the Nusselt number and friction
2015; 90: 173–186.
factor drop by 35.8% and 36% respectively for
6. Gosselin L, Tye-Gingras M and Mathieu-Potvin F.
an increase in L/Dh from 140 to 220.
Review of utilization of genetic algorithms in heat trans-
 For a certain value of L/Dh¼140, the Nusselt fer problems. Int J Heat Mass Transf 2009; 52:
number increases by 44.2% for an increase in 2169–2188.
Re from 2500 to 10,000. However, for the 7. Razelos P. A critical review of extended surface heat
same value of L/Dh, the friction factor decreases transfer. Heat Transf Eng 2003; 24: 11–28.
by 39.8%. 8. Iqbal Z, Syed KS and Ishaq M. Optimal convective heat
. The predicted value of the response variables is pre- transfer in double pipe with parabolic fins. Int J Heat
cisely closer to the numerical results and hence, it Mass Transf 2011; 54: 5415–5426.
600 Proc IMechE Part E: J Process Mechanical Engineering 233(3)

9. Milani Shirvan K, Mirzakhanlari S, Kalogirou SA, 24. Darvish Damavandi M, Forouzanmehr M and
et al. Heat transfer and sensitivity analysis in a double Safikhani H. Modeling and Pareto based multi-objec-
pipe heat exchanger filled with porous medium. Int J tive optimization of wavy fin-and-elliptical tube heat
Therm Sci 2017; 121: 124–137. exchangers using CFD and NSGA-II algorithm. Appl
10. Zheng N, Liu P, Liu Z, et al. Numerical simulation and Therm Eng 2017; 111: 325–339.
sensitivity analysis of heat transfer enhancement in a 25. Tayal MC, Fu Y and Diwekar UM. Optimal design of
flat heat exchanger tube with discrete inclined ribs. Int heat exchangers: a genetic algorithm framework. Ind
J Heat Mass Transf 2017; 112: 509–520. Eng Chem Res 1999; 38: 456–467.
11. Alzoubi MA and Sasmito AP. Thermal performance 26. Liu C, Bu W and Xu D. Multi-objective shape opti-
optimization of a bayonet tube heat exchanger. Appl mization of a plate-fin heat exchanger using CFD and
Therm Eng 2017; 111: 232–247. multi-objective genetic algorithm. Int J Heat Mass
12. Tgarguifa A, Abderafi S and Bounahmidi T. Energetic Transf 2017; 111: 65–82.
optimization of Moroccan distillery using simulation
and response surface methodology. Renew Sustain
Energy Rev 2017; 75: 415–425.
13. Kaushal M, Dhiman P, Singh S, et al. Finite volume Appendix
and response surface methodology based performance
prediction and optimization of a hybrid earth to air Notation
tunnel heat exchanger. Energy Build 2015; 104: 25–35.
14. Sun L and Zhang C-L. Evaluation of elliptical finned- A cross-sectional area [m2]
tube heat exchanger performance using CFD and AMG algebraic multi-grid
response surface methodology. Int J Therm Sci 2014; As the contact surface area between hot
75: 45–53. fluid and the other fluid streams [m2]
15. Moraveji MK and Ardehali RM. CFD modeling (com- Cp specific heat [J/Kg-K]
paring single and two-phase approaches) on thermal D diameter of tube [m]
performance of Al2o3/water nanofluid in mini-channel h heat transfer coefficient [W/m2K]
heat sink. Int Commun Heat Mass Transf 2013; 44: k thermal conductivity [W/mK]
157–164. L length of the exchanger [m]
16. Mohammed HA, Abbas AK and Sheriff JM. Influence
LMTD log-mean temperature difference [K]
of geometrical parameters and forced convective heat
m the mass flow rate of fluid streams
transfer in transversely corrugated circular tubes. Int
Commun Heat Mass Transf 2013; 44: 116–126. [Kg/s]
17. Gomaa A, Halim MA and Elsaid AM. Experimental Nu Nusselt number
and numerical investigations of a triple concentric- P wetted perimeter of the cross-section
tube heat exchanger. Appl Therm Eng 2016; 99: [m]
1303–1315. ST source term
18. Singh SK, Mishra M and Jha P. Experimental investi- v average fluid velocity [m/s]
gations on thermo-hydraulic behaviour of triple con- P pressure drop [Pa]
centric-tube heat exchanger. Proc IMechE, Part E: J
Process Mechanical Engineering 2015; 229: 299–308.  dynamic viscosity [Pa-s]
19. Hemmat Esfe M, Hajmohammad H, Toghraie D, et al. r density of the fluid [Kg/m3]
Multi-objective optimization of nanofluid flow in D generic diffusion term
double tube heat exchangers for applications in energy  dependent variable
systems. Energy 2017; 137: 160–171.
20. Goldberg DE and Holland JH. Genetic algorithms and
machine learning. Mach Learn 1988; 3: 95–99. Subscripts and superscripts
21. Sajedi R, Taheri M and Taghilou M. On the multi-
objective optimization of finned air-cooling heat av average
exchanger: Nano-fluid effects. J Taiwan Inst Chem c cold fluid
Eng 2016; 68: 360–371. h hot fluid, hydraulic diameter
22. Vahdat Azad A and Vahdat Azad N. Application of i intermediate temperature fluid, inner
nanofluids for the optimal design of shell and tube diameter, inlet
heat exchangers using genetic algorithm. Case Stud
o outer diameter, outlet
Therm Eng 2016; 8: 198–206.
23. Sanaye S and Hajabdollahi H. Thermal-economic
1 inner tube
multi-objective optimization of plate fin heat exchanger 2 inner annulus
using genetic algorithm. Appl Energy 2010; 87: 3 outer annulus
1893–1902.

You might also like