Journal of Mechanical Science and Technology 32 (2) (2018) 947~958

www.springerlink.com/content/1738-494x(Print)/1976-3824(Online)
DOI 10.1007/s12206-018-0145-x

Design optimization of HRSG inlet duct geometry for

improving flow uniformity using meta-heuristic algorithm†
Hyun-Kyoo So1, Tae-Hyun Jo1, Yong-Han Lee1, Bon-Chan Koo2 and Do-Hyung Lee1,*
1
Department of Mechanical Design Engineering, Hanyang University, Ansan, Korea
2
Department of Mechanical Engineering, Hanyang University, Seoul, Korea

(Manuscript Received July 7, 2017; Revised November 9, 2017; Accepted November 27, 2017)

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Abstract

The HRSG extensively affects all performance of CCPPs. The inlet duct geometry of an HRSG is the most essential part for determin-
ing heat exchange in the main body, in terms of flow uniformity. In the present study, numerical analysis of the HRSG flow characteris-
tics and design optimization of inlet duct geometry for improving flow uniformity at the front section of the main body were performed
to meet the trend requirements. A new inlet duct geometry, which has maximum flow uniformity, was proposed through design optimi-
zation procedures using a genetic algorithm. Specifically, the actual operating condition of the D-top model HRSG was applied and the
pressure recovery coefficient and diffuser efficiency were considered. In the optimized design, a recirculation area was formed at the top
internal wall of the second expansion stage. Results indicate that the forming of the recirculation area improves flow uniformity by rotat-
ing movement and spreading the high-speed flow.
Keywords: Design optimization; Flow uniformity; Genetic algorithm; Heat recovery steam generator
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

CCPPs continues to increase and currently approaches about

1. Introduction
60 % [4]. Thus, searching for methods for drawing high ther-
Growth in industry and world population leads to an incre- mal efficiency of CCPPs and the HRSG is becoming more
mental increase of energy demand. The growing quantity of important.
carbon emissions due to this increase becomes the main cul- The HRSG is the core facility of CCPPs for generating sup-
prit in environmental problems and accelerates global warm- plementary electricity by recovering heat energy from the
ing. Furthermore, the application possibility of renewable exhaust gas of thermal power plants [5]. Specifically, combus-
energy is economically far from certain and fuel prices con- tion gas exhausted from thermal power plants that have high
tinue to rise depending on the decrement of fossil fuel re- temperature and high pressure flows into the steam turbine
sources [1]. For these reasons, Combined cycle power plants cycle, such that the amount of thermal energy from the ex-
(CCPPs), which are more efficient than the conventional hausted gas can be recovered in the HRSG once again. Con-
thermal power plants, are practical and essential at this mo- sequently, because the thermal efficiency of the HRSG has a
ment in time for overcoming the environmental crisis [2]. strong influence on the overall performance of CCPPs, many
CCPPs are a way of generating electricity by combining two studies to improve the thermal efficiency of HRSG have been
types of thermodynamic cycles. These are the Rankine cycle conducted.
of conventional thermal power plants and Brayton cycle of the Major parameters related to the improvement of HRSG
Heat recovery steam generator (HRSG). In terms of thermal thermal efficiency are inlet duct geometry, exhaust gas tem-
efficiency, CCPPs surpass conventional thermal power plants perature, the composition of heat exchangers, and pinch point
by more than 10 %. Furthermore, CCPPs have several other temperature differences [6]. Over the last several decades, the
advantages, such as low pollutants, lower greenhouse gas above mentioned major parameters have been analyzed and
emission, and greater operational flexibility [3]. Because of thermodynamic, thermoeconomic optimizations have been
these advantages, CCPPs have attracted more attention during performed for increasing the thermal efficiency of the HRSG.
several decades in energy field. The thermal efficiency of Valdes et al. proposed a methodology for finding design pa-
*
rameters that affect thermal efficiency from the ther-
Corresponding author. Tel.: +82 31 400 4742, Fax.: +82 31 436 8146
E-mail address: [email protected] moeconomic perspective [7]. Franco and Russo proposed a
†
Recommended by Associate Editor Seong Hyuk Lee thermodynamic and thermoeconomic optimization method to
© KSME & Springer 2018
948 H.-K. So et al. / Journal of Mechanical Science and Technology 32 (2) (2018) 947~958

improve the thermal efficiency of the HRSG, up to 60 % in- caused by thermal stress concentration. Thus, it is demanding
stead of the conventional pinch point method for the HRSG to lengthen the heat exchange tubes while maintaining high
operation [8]. Kaviri et al. analyzed the effect of inlet gas tem- flow uniformity in the inlet duct. For this reason, it is neces-
perature on HRSG thermal efficiency and found the optimized sary to design the shape of the inlet duct to improve the inlet
inlet gas temperature through exergoenvironmental optimiza- flow uniformity.
tion [9]. Hajabdollahi et al. performed thermodynamic, exer- In this respect, Lee et al. conducted a computational and ex-
goeconomic analyses and multi-objective optimization using a perimental study of the effect of swirl flow from the rear sec-
Genetic algorithm (GA). Moreover, they also predicted the tion of the gas turbine to the HRSG inlet duct [13]. Hegde et al.
behavior of the objective function using regression analysis proposed an inlet duct geometry with improved flow uniform-
for optimum curve fitting [2]. Recently, the HRSG developed ity through Computational fluid dynamics (CFD) simulation
from a single-pressure to multi-pressure stage in order to re- and confirmed that flow uniformity is the dominant factor of
cover heat energy from high temperature exhaust gases ac- HRSG thermal efficiency [12]. Shin et al. analyzed the inlet
cording to the characteristics of CCPPs gas turbines, which are flow pattern by numerical simulation [14]. Ameri and Dor-
being improved for thermal efficiency and the rising tempera- cheh modeled the inlet duct geometry of the horizontal drum
ture of the exhaust gases [5]. Kim et al. proposed a methodol- type HRSG and analyzed the effect of the component located
ogy for minimizing steam drum thermal stress during the start- behind the inlet, on the flow uniformity [15]. Hanafizadeh et
up procedure by predicting and analyzing the transient behav- al. simulated the inlet flow of the horizontal drum type HRSG
ior of the inlet at the HRSG high-pressure section of HRSG and proposed an optimized inlet duct geometry [16]. The
[10]. Casarosa et al. performed thermodynamic and ther- above studies on the optimization of inlet duct geometry and
moeconomic optimization for the two-pressure stage and the analysis of flow characteristics, have provided theoretical
three-pressure stage HRSG, respectively, and found that the guidelines for the improvement of flow uniformity. However,
thermal efficiency increased by nearly 60 % [11]. Recently, it is not applicable to the industrial field because the informa-
Alobaid et al. modeled the actual three-pressure stage HRSG tion of the structure and flow condition of the rear section of
and analyzed the behavior of the warm start-up procedure by the gas turbine is not taken into account.
simulation [1]. If the exhaust gas composition of the gas turbine, internal
Most of these studies, which analyzed and optimized for structure between the gas turbine and HRSG inlet, and flow
thermal efficiency and thermal economic efficiency of the condition is not accurately reflected, the analysis of the inlet
HRSG, include common design parameters, such as drum effect and design optimization of inlet duct geometry cannot
pressure, pinch point temperature differences, steam mass be effective in the actual HRSG design. Within the path from
flow, and inlet gas temperature. However, HRSG steam pro- the rear section of the gas turbine to the HRSG inlet, there are
duction is more directly related to the thermal efficiency than several components, such as guide vanes, and these compo-
the above design parameters. The steam production depends nents have a decisive effect on the HRSG inlet flow. The
on the amount of heat transfer between the exhaust gas and HRSG inlet has a high turbulent flow that has a strong swirl
heat exchangers. Therefore, sufficient improvement of thermal and shows a large flow characteristic difference from the con-
efficiency can be expected by considering the flow uniformity ventional flow through the components between the gas tur-
of the exhaust gas that inflows to the HRSG inlet. The flow bine and HRSG inlet. Moreover, the gas turbine exhaust gas is
uniformity of the exhaust gas is directly related to the inlet composed of syn-gas and is dissimilar to general air in com-
duct geometry. Moreover, it is the most important parameter position. The analysis using general air composition and using
to determine the temperature uniformity at the HRSG inlet, the exhaust gas composition, clearly show different results in
and amount of heat transfer and exchange in the pressure stage. the behavior of the HRSG inlet flow.
The temperature uniformity and amount of heat transfer of the In this study, the design optimization for a full-scale analy-
exhaust gas increases as the flow uniformity increases at the sis and inlet duct geometry was performed to improve the
inlet, which also increases the thermal efficiency of the HRSG flow uniformity of the inlet after selecting the HRSG, which is
[12]. In recent times, CCPPs adopted a method of lengthening actually operating in the industrial field as an analytical model.
the tube length rather than increasing the number of HRSG The information of exhaust gas composition, HRSG inlet flow
heat exchange tubes due to construction site constraints. As condition, and the components of the connection section of the
the heat exchange tube lengths are increased, the height of the D-Top HRSG model, which is in actual operation, have been
HRSG is increased, such that the inlet duct angle is increased reflected in order to improve the accuracy of the analysis un-
more precipitously. Increasing the inlet duct angle causes an like previous studies. In addition, a more accurate numerical
escalating velocity deviation of the HRSG inlet gas. This ad- analysis was made possible by grasping the composition of
versely affects the thermal efficiency and heat transfer of the HRSG inlet gas. Design optimization was performed using
HRSG [6]. In addition, non-uniform bottom-focused high GA after extracting 30 design points by setting the x, y-
temperature flow, which was caused by increased inlet height, coordinates of the inlet duct deflection angles θ1, θ2 as design
has a decisive effect on the HRSG durability. Securing high parameters. Thus, the optimized inlet duct geometry of the
flow uniformity can prevent durability degradation, which is HRSG was determined and compared to the conventional
 H.-K. So et al. / Journal of Mechanical Science and Technology 32 (2) (2018) 947~958 949

¶r ¶
+ ( rui ) = 0 (1)
¶t ¶xi
¶ ¶
( rui ) + ( r u j ui ) =
¶t ¶x j
(2)
¶p ¶ é æ ¶u ¶u ö 2 ù
- + ê( m + mt ) ç i + j ÷ - kd ij ú + Si
¶xi ¶x j ê ç ÷
ë è ¶x j ¶xi ø 3 úû

¶ ¶ ¶ éæ m ö ¶k ù
¶t
( r e) +
¶x j
(
ru j k = )
¶x j
êçç m + t
s
÷÷ ú + G - re (3)
ëêè k ø ¶x j ûú

where ui, uj are velocity components, k is turbulent kinetic

energy, ε is the turbulent kinetic energy dissipation rate, and G
is the turbulent production term. Moreover, source term, Si , is
Fig. 1. Composition of heat recovery steam generator. heat exchanger pressure loss, which should be porous medium
at momentum, Eq. (2).
design, and optimized design based on the area rate, which The realizable k-ε turbulence model, which is a two-
meets from the 50 % to 150 % range of average flow direction equation turbulence model among RANS models, was adopt-
velocity and Root mean square (RMS) to determine the flow ed for reasonably predicting the turbulent flow characteristics
uniformity. of a sudden cross-sectional change in the HRSG inlet [15].
The realizable k-ε model, Eqs. (4) and (5), is represented be-
low.
2. Numerical analysis
In this study, a numerical analysis was performed to inves- ¶ ¶ ¶ éæ m ö ¶k ù
tigate the flow characteristics in an HRSG inlet duct. The inlet ¶t
( rk ) +
¶x j
(
r ku j = )
¶x j
êçç m + t
s
÷÷ ú + Pk + Pb - re
ëêè k ø ¶x j ûú
duct consists of a circular duct extending from the gas turbine
to the HRSG inlet, followed by a rectangular two-stage expan- (4)
¶ ¶
sion duct. The high temperature exhaust gas entering the
¶t
( re ) +
¶x j
(
re u j = )
HRSG passes through the inlet duct, and is expelled from the
(5)
stack through heat exchangers arranged perpendicular to the ¶ éæ m ö ¶e ù e2 e
flow direction and damper. êçç m + t ÷÷ ú + r C1Se - r C2 + C1e C3e Pb
¶x j ëêè s e ¶ x
ø j ûú k + ne k
Fig. 1 shows the composition of the entire HRSG. The
HRSG is comprised of an inlet duct, five module sections,
which are composed of three-pressure stages, an SCR layer, where μ is laminar viscosity coefficient and μt is turbulent
damper, and stack. In the analysis domain selection, Patil et eddy viscosity coefficient. The model constants C1ε is 1.44, C2
al. [4], Hanafizadeh et al. [6] omitted the rear section of the is 1.9, C2ε is 1.92, C3ε is -0.33 and σk is 1.0, σε is 1.2.
HRSG inlet, which includes the pressure stages, although the
pressure stages and damper at the downstream of the inlet 2.2 Boundary condition
duct affect the overall flow behavior. The inlet flow entering
from the rear section of the gas turbine is interrupted by heat This work reflects the geometry, dimension, and inflow gas
exchange tubes in the main body and changed the flow con- condition of the 152.5 MW three-pressure stage D-Top model
dition due to the temperature difference by heat exchange, HRSG. Figs. 2(a) and (b) show the actual operating profile of
while passing through the main body, stack and damper. the flow direction velocity and temperature of the analytical
Consequently, in this study, the analysis domain covered the model inlet. Table 1 provides the inlet flow of the operating
whole area from the HRSG inlet to the three-pressure stage, condition at 100 % output of the HRSG. The HRSG inlet di-
and the damper and set as porous zones, in order to consider ameter is 6400 mm and a square obstacle with a length of
the influence of the pressure stages and damper on the inlet 3640 mm is installed at the center of the inlet. Caused by the
flow behavior. A full-scale three-dimensional flow analysis square obstacle and high velocity of circular edge, recircula-
was performed using ANSYS 15.0 Fluent as a tool. tion area was generated at the rear section of the obstacle, thus
negative velocity value was represented in Fig. 2(a) contour.
In addition, three-singularity point, which has extremely high
2.1 Governing equation
velocity over 100 m/s, was generated due to the three-blade
The flow governing equations are expressed as continuity, guide vane. The HRSG inlet flow characteristic is affected by
Eq. (1), momentum, Eq. (2), and energy transport, Eq. (3), for the square obstacle and guide vane, located between the rear
three-dimensional compressible turbulent flow. section of the gas turbine and HRSG inlet. The radial and
950 H.-K. So et al. / Journal of Mechanical Science and Technology 32 (2) (2018) 947~958

Table 1. Inlet flow condition at 100 % output.

Velocity (m/s) 50
Mass flow rate (kg/s) 615.306
Density (kg/m3) 0.37
Temperature (°C) 638.5
Temperature difference (°C) 557.3

Table 2. Inlet condition of syn-gas at each temperature.

T (K) 373 423 473 573 673 773 873

3
ρ (kg/m ) 0.926 0.817 0.739 0.603 0.513 0.447 0.388
(a)
Cp (103J/kg·K) 1.098 1.105 1.117 1.144 1.174 1.204 1.235
μ (10-5N·s/m2) 2.045 2.266 2.479 2.883 3.262 3.622 3.780
-5 2
ν (10 m /s) 2.208 2.775 3.353 4.782 6.355 8.105 9.722

(b)

Fig. 2. (a) Inlet flow direction velocity profile; (b) inlet temperature
profile.

Fig. 3. Generating geometry of entire HRSG in computational space.

tangential velocities were calculated from the above profile
and given data. The radial and tangential velocities are the in computational space. The geometry consists of the inlet
mean velocity values of each point, which are -3.008 m/s and duct into which the gas is introduced, HRSG pressure section
11.188 m/s, respectively. Also, since the calculated Reynolds where the heat exchange occurred, and stack through which
number is higher than 106, the HRSG inlet flow can be judged the heat exchanged gas is discharged. The three-pressure stag-
as a strong turbulent flow. es of the HRSG consist of five modules with heat exchangers
Moreover, the temperature difference of approximately 557 and the damper installed in the stack. Inside the pressure stage,
K, which is caused by the high temperature flow through the restrainers, which support the long heat exchangers, are ar-
HRSG heat exchange tubes, raises the density difference. The ranged equally at 2.8 m apart. The HRSG inlet diameter is 6.4
total density variation rate of the outlet in comparison with the m, the total length from the inlet to the rear section of the main
inlet was 61.95 %, thus, compressibility was considered in body, which includes pressure stages, damper and stack, is
numerical analysis. In most flow analysis, compressibility 40.49 m, height is 22.2 m, and width is 10.4 m.
occurs due to the high flow rate. However, in this numerical The total number of meshes is 3.5 million and the mesh
analysis, compressibility occurs only by temperature differ- type is hexahedron and tetrahedron. The inlet domain is com-
ence. Thus, the compressible flow model was applied to accu- posed of a tetrahedral type mesh and all domains, except for
rately calculate the compressibility turbulence flow caused by inlet region, are composed of a hexahedral type mesh. The
the temperature difference. orthogonal quality of the mesh was evaluated to determine the
Table 2 lists the density of inlet gas condition, specific heat quality of the lattice configuration. The overall average of
at constant pressure, viscosity, and kinematic viscosity, ac- orthogonal quality is 0.95, and therefore, the grid system is
cording to temperature. Inlet gas is a syn-gas and the syn-gas considered to be well structured.
property is determined by the composition of N2, O2, H2O,
CO2 and Ar. The composition of Ar included in syn-gas was 2.3 Numerical method
obtained by using the HYSYS tool, and calculation was per-
formed by a piecewise linear interpolation of density, Cp, con- Modeling the complex HRSG geometry, such as the heat
ductivity, and viscosity. exchanger tubes within the pressure stages, in practice is time
Fig. 3 shows the geometry composition of the entire HRSG consuming and virtually impossible. Choi et al. treated the
 H.-K. So et al. / Journal of Mechanical Science and Technology 32 (2) (2018) 947~958 951

pressure stage module with a porous zone in the heat flow Table 3. Numerical and actual performance of temperature difference.
analysis of the vertical drum type HRSG [17]. Therefore, in
Temperature
this study, the module, including the inner heat exchanger Actual performance Numerical result
difference (K)
tubes, is assumed to be a porous zone in the flow analysis. The Module 1 113.1 114.7
momentum source term, which is the sum of viscous resis-
Module 2 161.2 160.8
tance and inertial resistance, was used to calculate the porous
SCR 0 0.2
zone. In the case of HRSG heat exchange tubes, only the iner-
tial resistance term is considered because the viscous resis- Module 3 107 106.5
tance term is very small compared to the inertial resistance Module 4 86.1 85.9
term. The inertial resistance term is shown in Eq. (6) below. Module 5 89.9 89.3
The vertical direction resistance and transverse direction resis- Damper 0 0.6
tance are 20 % and 50 %, respectively, and the heat sink con-
dition is considered to simulate heat transfer at each porous
zone. diffuser stall reduces pressure recovery and diffuser efficiency
[18]. In the measurement of pressure recovery and diffuser
1 efficiency, the inlet was set to the HRSG inlet and outlet was
Dp = C r nU 2 (6)
2 set to the front section of module 1. The pressure recovery
coefficients and the diffuser efficiency are shown in the fol-
where C is the inertial resistance factor, Δn is the porous zone lowing Eqs. (10) and (11), respectively.
thickness, and U is the average flow direction velocity.
The numerical analysis was based on 100 % HRSG output P2 - P1
and the Finite volume method (FVM) was adopted to analyze Cp = (10)
1
rU12
the entire area of the HRSG. A no-slip condition was applied 2
to the entire outer wall of the HRSG, and the calculation was Cp
repeated about 4000 times. h= 2
(11)
æA ö
In the contemporary horizontal type HRSG, which has been 1- ç 1 ÷
adopting the method of lengthening the heat exchange tubes è A2 ø
due to the construction site constraints, the inlet flow uniform-
ity is significant parameter with respect to the durability and where Cp is the pressure recovery coefficient and η is diffuser
thermal efficiency. Thus, in this study, the uniformity of flow efficiency.
direction velocity was set as main parameter. As parameters to
determine the flow uniformity at the front of module 1 in the 2.4 Verification
HRSG inlet, the RMS deviation and cross-sectional area ratio,
which meet the range from 50 % to 150 % of average flow In order to verify the validity of the numerical analysis re-
direction velocity compared to the total cross-sectional area, sults, the actual operating performance and numerical analysis
was set. The RMS deviation and cross-sectional area ratio are results of the 152.5 MW three-pressure stage D-Top model
stated as Eqs. (7)-(9), respectively. Eqs. (7) and (8) indicate HRSG were compared. The results are presented in Table 3.
RMS deviation and Eq. (9) indicates effective area ratio. The comparison contents are set to the temperature difference
of each module composed of heat exchangers. The inlet and
U RMS outlet temperatures of each module were measured to calcu-
RMS deviation = ´ 100 (%) (7)
U late the temperature difference. The temperature difference
2 values, which were obtained by numerical simulation at each
S (u - U )
U RMS = (8) heat exchanger module treated as a porous zone, are within
N 2 K variation when compared with actual operating perform-
AREA±50% ance data. Consequently, it is considered that the numerical
P±50% = ´ 100 (%) (9)
AREAtotal simulation is properly processed in simulating the actual flow
behavior of the HRSG.
where u is local velocity, U is average flow direction velocity
and P±50% is effective area.
3. Design optimization
Further, the pressure recovery coefficient and diffuser effi-
ciency were calculated in the HRSG inlet duct. The HRSG Design optimization was performed on the HRSG inlet duct
inlet duct can be treated as a diffuser having different upper geometry in order to determine the advanced geometry that
and lower side angles with two expansion stages. An occur- would improve the flow uniformity. PIAnO was used as a tool
rence of diffuser stall in the diffuser greatly affects pressure to accomplish design optimization. PIAnO basically integrates
recovery and diffuser efficiency. In general, the occurrence of and automates analysis procedures related to computer-aided
952 H.-K. So et al. / Journal of Mechanical Science and Technology 32 (2) (2018) 947~958

Fig. 5. Design points distribution in design range.

where Pc is the deflection point of the double-expansion inlet

diffuser of HRSG and U is average flow direction velocity.

3.2 Optimization method

Fig. 4. Design parameters and boundary condition of design domain.
In this procedure, the Optimal Latin hypercube design
engineering, and applies the latest design techniques, such as (OLHD) technique among DOE was applied to acquire 30
Design of experiments (DOE), optimization, and reliability design points. Next, the Kriging method of meta-model meth-
analysis, to such automated analysis procedures to provide a ods was adopted for effectively managing the CPU execution
multidisciplinary integrated design solution. Consequently, time and dealing with nonlinear design parameters. Also, the
design optimization was performed using an appropriate DOE, GA technique was used to determine the global optimum val-
meta-model method, and optimization algorithm to success- ue for improving flow uniformity.
fully determine the HRSG inlet duct geometry with maximum The DOE was performed based on the above parametric
flow uniformity. study. The number of design points to assign the DOE chart
and method is significant for clear design optimization and
system analysis. Thus, the OLHD technique, which is a DOE
3.1 Design formulation
space filling sampling method, was used to extract the 30 de-
The x, y-coordinates of point Pc are set as design parameters sign points. In general, OLHD offers an acceptable accuracy,
in order to perform design optimization. Fig. 4 shows point Pc even for a small sample size. However, 30 design points were
and the limitations of design domain. By setting the x, y- judged to be sufficient for securing robustness in this proce-
coordinates of point Pc to the design parameters, the optimum dure due to the high non-linearity and unpredictable correla-
inlet duct angles θ1, θ2 were efficiently determined. Prior to tion of design parameters. Consequently, OLHD offers 30
performing design optimization, the definition and formula- design points properly with respect to two design parameters
tion of the optimization problem were executed. In general, among the DOE methods. This technique can uniformly di-
definitions and formulations of optimization problems take up vide the ranges of each parameter. Accordingly, this technique
a large part in the overall process. The design input variable was judged to be appropriate for analyzing the main effect of
and object function Eq. (12) are defined in the followings. each design parameter. OLHD, which has large samples, is
Finding point Pc, which has maximum flow uniformity was also better suited for GA, which is applied in the next step [19].
formulated by design formulation procedure. In this optimiza- Fig. 5 shows the distribution of 30 cases of x, y-coordinates of
tion problem, the constraint conditions were not considered design points selected using the OLHD technique in the de-
and uniformity of the flow direction velocity was shown be- sign range.
low as an object function. The object function data for x, y-coordinates and overall
predicted tendency were shown in Figs. 6(a) and (b). The
Find Location of Pc overall tendency of object function was expected to present a
Maximize Velocity RMS Normal ± 50 % Area (%) quadratic form with respect to x-coordinate location. On the
Object function =
other hand, non-linear behavior was expected with respect to
y-coordinate location and roughly represented a cubic form.
{Area Area = ( 0.5U £ U £ 1.5U )} ´ 100 (%) (12) The results of OLHD show a non-linear relationship, as ex-
Totalarea pected, between design parameters and object function. In
 H.-K. So et al. / Journal of Mechanical Science and Technology 32 (2) (2018) 947~958 953

(a)

Fig. 7. Meta-surface of object function with respect to x, y-coordinate.

(b)

Fig. 6. (a) Object function data and predicted tendency of x-coordinate;

(b) object function data and predicted tendency of y-coordinate. (a)

these design optimization tendencies, maximum flow uni-

formity can be predicted at 25 %, 0 % of x-coordinate location
and y-coordinate location, respectively.
In order to perform a clearer analysis, the meta-surface was
generated using the Kriging method. The Kriging method is
suitable for generating a meta-surface with non-linear rela-
tionships. The meta-surface was generated with respect to
design parameters x-coordinate location and y-coordinate
location. The design domain of conventional geometry was
(b)
mapped to the orthogonal domain for definite analysis of the
correlation between the input parameters and object function. Fig. 8. (a) Flow uniformity rate with respect to x-coordinate; (b) flow
Fig. 7 represents the meta-surface, and Figs. 8(a) and (b) rep- uniformity rate with respect to y-coordinate.
resent the object function with respect to x-coordinate location
and y-coordinate location, respectively. In this result, x- the possibility of being trapped in a local optimum is de-
coordinate location is judged to be much sensitive comparing creased. After defining the above object function, the first
with y-coordinate location. Optimum value, which was pre- generation, which is composed of a series of genes, is gener-
dicted by the meta-surface, was featured at about 84.9 % flow ated. If the requirement of object function is not satisfied with
uniformity. this generation, a new generation is generated through certain
In sequence, the GA method was adopted to determine op- procedures, such as selection, mating, and mutation, towards
timum value. The above meta-surface result was expressed as the optimum value of object function. Consequently, the glob-
a non-linear surface, such that the global optimum finding al optimum value, which maximizes the HRSG inlet flow
algorithms were required. The GA, which is a stochastic op- uniformity, is determined by undergoing the above process.
timization algorithm based on the principles of evolution in
nature, is the proper method for accurately finding a global
4. Results and discussion
optimum value compared with other gradient based methods
in non-linear problems [20]. In the GA, the global optimum In this study, the numerical analysis of flow characteristics
value is searched exhaustively in solution space. Moreover, at the inlet duct of D-Top model HRSG and design optimiza-
954 H.-K. So et al. / Journal of Mechanical Science and Technology 32 (2) (2018) 947~958

Table 4. Velocity and standard deviation of optimized design.

Optimized design Conventional design

Average flow
7.078 7.031
direction velocity (m/s)
50 % / 150 % of
3.539 / 10.618 3.535 / 10.546
average velocity (m/s)
Standard deviation 3.742 3.894

Table 5. Temperature conditions and uniformity of optimized design.

Optimized design Conventional design

Average
909.21 909.22
temperature (K)
Min / max
904.05 / 913.27 902.28 / 913.26
temperature (K)
Temperature
(a) 27.85 23.35
uniformity (%)

Fig. 10. Velocity distribution histogram.

The average velocity, 50 % and 150 % of velocity magni-

(b) tude and standard deviation were represented in Table 4, and
Fig. 9. (a) Optimized inlet duct geometry; (b) conventional inlet duct the histogram of velocity distribution was shown in Fig. 10.
geometry. The black graph represents the velocity distribution of the
optimized design and white graph represents the velocity dis-
tion for inlet duct geometry using GA were performed. Thus, tribution of the conventional design, respectively. As shown in
in the optimization procedure, the x, y-coordinates of the op- Table 4, the mean velocities of the conventional design and
timized design parameters were 2850.886 mm and 1000.506 optimized design are almost similar. Whereas, the standard
mm, respectively, and the flow uniformity of the optimized deviation of the optimized design was 3.90 % lower than the
design, which was probed at the front section of module 1 was conventional design. The standard deviation of the optimized
84.835 %. This value is 3.4 % higher than the conventional design is 3.741795 and standard deviation of the conventional
design. The optimum value calculated through the GA indi- design is 3.893759. As shown in the histogram of Fig. 10, the
cates the approximate value of 84.900 % and has a 0.077 % velocity distributions of the optimized and conventional de-
difference compared with the predicted value by meta-surface. signs similarly appear as normal distribution, respectively,
Thus, the HRSG inlet duct geometry was optimized as shown whereas the normal distribution peak point of the optimized
in Fig. 9(a), and the conventional geometry was represented in design was more adjacent to mean velocity than conventional
Fig. 9(b). In the optimized design, the deflection point Pc is design. Consequently, the peak point of the velocity normal
located further ahead and lower compared with the conven- distribution is located closer to the mean velocity and the
tional design. The deflection angles, θ1, θ2, of the optimized standard deviation is decreased, such that the flow uniformity
geometry were 19.338° and 71.276°, respectively, and the is improved. Thus, it can be judged that the optimization has
slope of the second-extension stage of the inlet duct became been performed reasonably.
more gradual and lengthened, whereas the slope of first- In addition, the temperature condition and temperature uni-
extension stage became more precipitous and shortened. formity of optimized design and conventional design was
 H.-K. So et al. / Journal of Mechanical Science and Technology 32 (2) (2018) 947~958 955

(a)
(a)

(b)

Fig. 11. (a) Inlet flow streamline of optimized design; (b) inlet flow (b)
streamline of conventional design. Fig. 12. (a) Flow direction velocity contour of optimized design; (b)
flow direction velocity contour of conventional design.

analyzed. Temperature conditions and temperature uniformity

was represented in Table 5. Whereas the average temperature Figs. 13(a) and (b) represent the area distribution where re-
and maximum temperature of optimized and conventional circulation exists in the HRSG inlet duct of the optimized
design are almost equal to 909.21 K and 913.27 K, the mini- design and conventional design, respectively. The recircula-
mum temperature of the optimized design increased about 1.8 tion area was measured through volume-rendering method
K in comparison with the conventional design. Since the tem- with respect to certain cells, which have negative flow direc-
perature difference between maximum and minimum tem- tion velocity. In comparison to the recirculation area, in the
perature was narrow, the area, which meet the range of ±1 K conventional design, the recirculation area was observed at the
of average temperature, was calculated to compare the tem- lower middle of the front section of module 1. On the other
perature uniformity. In the analysis results, the temperature hand, in the optimized design, the recirculation area located at
uniformity of optimized design was 4.5 % increased as com- the lower middle section was considerably reduced and the
pared with the conventional design. Consequently, the tem- new recirculation area was formed at the upper internal wall
perature difference decreases and the temperature uniformity inlet duct. This phenomenon can also be confirmed by the
increases evidently through the optimization procedure. streamlines in Fig. 11(a). The volume of the recirculation area
Figs. 11(a), (b), 12(a) and (b) represent HRSG inlet duct was increased about 31.133 m3 in the optimized design com-
streamlines and velocity contours at the front section of mod- pared to the conventional design in the inlet duct.
ule 1 of the optimized design and conventional design, respec- Subsequently, the pressure recovery coefficient and diffuser
tively. In the above velocity contour, the contour legends rep- efficiency of the optimized design and conventional design
resent the velocity, which is from 50 % to 150 % of the mean were calculated at the HRSG inlet duct. The high-speed rotat-
velocity, with respect to the flow direction and the white areas ing flow which, is separated from the boundary layer of the
represent the remaining velocities. The velocity singularity first expansion stage upper wall, is occurred diffuser stall,
generated by the internal compositions and strong swirling which is also called recirculation area. In general, generating a
flow cannot spread up and is concentrated at the bottom of the recirculation area necessarily reduces the pressure recovery
HRSG due to the energetic flow direction velocity component. coefficient and diffuser efficiency of the inlet duct. In the con-
The unevenly broad white area of the conventional design is ventional design, the pressure recovery coefficient and dif-
noticeably reduced in the optimized design, and converging to fuser efficiency were calculated 1.67 and 1.72, respectively. In
the bottom of the HRSG is observed. the optimized design, the pressure recovery coefficient and
956 H.-K. So et al. / Journal of Mechanical Science and Technology 32 (2) (2018) 947~958

velocity and strong swirling components, is the main cause of

the remarkable flow uniformity enhancement, compared with
the minor reduction of the diffuser efficiency. As shown in
Figs. 11(a) and 12(a), the high-speed flow at the bottom center
of the inlet is dispersed by the rotating movement of the recir-
culation area generated at the top internal wall of the HRSG
inlet duct, compared with Figs. 11(b) and 12(b), undergoing
the design optimization procedure.

5. Conclusion
In this study, numerical analysis and design optimization
were performed to improve the flow uniformity of an HRSG
(a)
inlet duct, which has two-expansion stages. The actual operat-
ing conditions of a 152.5 MW three-pressure stage D-Top
model HRSG were applied for analyzing a full-scale 3-
dimensional flow. Consequently, the optimized geometry of
the HRSG inlet duct was determined using a GA. The per-
formance of the optimized design and conventional design
was evaluated based on the flow uniformity, pressure recovery
coefficient, and diffuser efficiency.
The optimized inlet duct geometry has 84.835 % flow uni-
formity at x, y-coordinates 2850.886 mm, 1000.506 mm, re-
spectively, and this result is the maximum value of 30 cases.
The recirculation area generated at the top internal wall dis-
perses the high velocity singularity, which is leaned too much
toward to the bottom of HRSG inlet duct. Thus, the diffuser
(b) efficiency of the optimized design reduced by 1.16 %, how-
ever, the flow uniformity was enhanced up to 3.4 % compared
Fig. 13. (a) Domain of recirculation area in optimized design; (b) do-
main of recirculation area in conventional design. with the conventional design, which is not considered the
systematic correlation of non-linear design parameters.
The results of this paper indicate that the adoption of opti-
diffuser efficiency were calculated 1.65 and 1.70, respectively. mized inlet duct geometry is a favorable choice for enhancing
Diffuser efficiency can exceed one, relatively depending on flow uniformity and, by extension, thermal efficiency. Apply-
the flow uniformity at the inlet and outlet of the diffuser [18]. ing this new optimized geometry to the installed HRSG,
In a general diffuser, the flow uniformity at the inlet is more henceforth, is an outstanding and non-costing way to accept.
uniform than the flow uniformity at the outlet, however, in the Improvement of flow uniformity enhances the durability of
HRSG inlet duct selected in this paper, the exhaust flow from the HRSG in terms of heat stress as well as thermal efficiency.
the gas turbine, which is the inlet flow of the HRSG, becomes In order to achieve a more precise analysis, flow vibration by
more uniform passing through the inlet duct. Thus, in this fluctuation, and swirl effect by guide vane should be consid-
research, the calculated value of the diffuser efficiency could ered. Gas turbine combustion, which causes strong fluctuation
be higher than one. As a result, the optimized design has re- due to the explosion, effectuates flow vibration at HRSG inlet.
duced the pressure recovery coefficient and diffuser efficiency Additionally, to implement the quantitative evaluation of the
about 1.19 % and 1.16 %, respectively, compared to the con- durability of heat exchange groups and the HRSG thermal
ventional design. efficiency in accordance with the improvement of flow uni-
Consequently, as the size of the recirculation area, which formity, precise modeling of heat exchange groups, which
adversely affects the diffuser efficiency, is expanded, the dif- were treated as porous zones in this study, and reflecting exact
fuser efficiency is reduced only 1.16 %. However, the flow material properties are required to future studies. In this sense,
uniformity has been remarkably enhanced about 3.4 %. More- this paper is expected to be an adequate guideline for these
over, as the pressure recovery coefficient and diffuser effi- future studies.
ciency were reduced, the flow direction velocity decreased
slightly. However, the standard deviation of the flow direction
Acknowledgment
velocity declined about 3.9 %. The diminution of the standard
deviation leads to improve the flow uniformity at the HRSG This work was supported by the Human Resources Devel-
inlet. The dispersion of velocity singularity, which has high opment of the Korea Institute of Energy Technology Evalua-
 H.-K. So et al. / Journal of Mechanical Science and Technology 32 (2) (2018) 947~958 957

tion and Planning (KETEP) grant funded by the Korea gov- [7] M. Valdes, A. Rovira and M. D. Duran, Influence of the heat
ernment Ministry of Knowledge Economy (No. KETEP2017 recovery steam generator design parameters on the ther-
4010201310). moeconomic performances of combined cycle gas turbine
power plants, International Journal of Energy Research, 28
(14) (2004) 1243-1254.
Nomenclature------------------------------------------------------------------------
[8] A. Franco and A. Russo, Combined cycle plant efficiency
C : Inertial resistance factor increase based on the optimization of the heat recovery
Cp : Pressure recovery coefficient steam generator operation parameters, International Journal
G : Turbulent production term of Thermal Sciences, 41 (9) (2002) 843-859.
k : Turbulent kinetic energy [9] A. G. Kaviri, M. N. M. Jaafar and T. M. Lazim, Exergoenvi-
n : Thickness of porous zone ronmental optimization of heat recovery steam generators in
P : Pressure combined cycle power plant through energy and exergy
Pc : Inlet deflection point analysis, Energy Conversion and Management, 67 (2013)
P±50% : Effective area 27-33.
Si : Pressure loss of heat exchangers [10] T. S. Kim, D. K. Lee and S. T. Ro, Analysis of thermal
T : Temperature stress evolution in the steam drum during start-up of a heat
U : Average flow direction velocity recovery steam generator, Applied Thermal Engineering, 20
u : Local flow direction velocity (2000) 977-992.
X1 : x-coordinate of deflection point [11] C. Casarosa, F. Donatini and A. Fanco, Thermoeconomic
X2 : y-coordinate of deflection point optimization of heat recovery steam generators operating pa-
Y1 : (Object function) Flow uniformity rate rameters for combined plants, Energy, 29 (2004) 389-414.
ε : Turbulent kinetic energy dissipation rate [12] N. Hegde, I. Han, T. W. Lee and R. P. Roy, Flow and heat
η : Diffuser efficiency transfer in heat recovery steam generators, Journal of Energy
θ1 : First expansion angle Resources Technology, 129 (3) (2007) 232-242.
θ2 : Second expansion angle [13] B. E. Lee, S. B. Kwon and C. S. Lee, On the effect of swirl
μ : Laminar viscosity coefficient flow of gas turbine exhaust gas in an inlet duct of heat re-
μt : Turbulent eddy viscosity coefficient covery steam generator, Transactions-American Society of
ν : Kinetic viscosity coefficient Mechanical Engineers Journal of Engineering for Gas Tur-
bines and Power, 124 (3) (2002) 496-502.
[14] H. S. Shin, D. H. Kim, H. J. Ahn, S. M. Choi and G. C.
References
Myung, Investigation of the flow pattern in a complex inlet
[1] F. Alobaid, K. Karner, J. Belz, B. Epple and H. G. Kim, duct of a heat recovery steam generator, Energy and Power,
Numerical and experimental study of a heat recovery steam 2 (1) (2012) 1-8.
generator during start-up procedure, Energy, 64 (2014) [15] M. Ameri and F. J. Dorcheh, The CFD modeling of heat
1057-1070. recovery steam generator inlet duct, International Journal of
[2] H. Hajabdollahi, P. Ahmadi and I. Dincer, An exergy-based Energy Engineering, 3 (3) (2013) 74.
multi-objective optimization of a Heat recovery steam gen- [16] P. Hanafizadeh, M. M. Siahkalroudi and P. Ahmadi, Ex-
erator (HRSG) in a Combined cycle power plant (CCPP) us- perimental and numerical investigation of optimum design
ing evolutionary algorithm, International Journal of Green of semi industrial heat recovery steam generator inlet duct,
Energy, 8 (1) (2011) 44-64. Applied Thermal Engineering, 104 (2016) 375-385.
[3] P. Ahmadi, I. Dincer and M. A. Rosen, Exergoeconomic and [17] S. M. Choi, H. Moon, S. H. Kim, J. S. Park and H. H. Cho,
environmental analyses and evolutionary algorithm based The effects of thermal spreaders on reducing thermal cracks
multi-objective optimization of combined cycle power plants, in heat recovery steam generators, Applied Thermal Engi-
Energy, 36 (10) (2011) 5886-5898. neering, 108 (2016) 1251-1260.
[4] N. Patil, M. Kavade and A. Patil, Study of gas flow behavior [18] Blevins and D. Robert, Applied fluid dynamics hand-
in HRSG inlet duct with CFD tools, International Journal of book, Van Nostrand Reinhold Co. (1984) 144-153.
Mechanical Engineering applications Research, 3 (01) [19] R. Stocki, A method to improve design reliability using
(2012) 146-151. optimal Latin hypercube sampling, Computer Assisted Me-
[5] V. Ganapathy, Heat-recovery steam generators: Understand chanics and Engineering Sciences, 12 (4) (2005) 393.
the basics, Chemical Engineering Progress, 92 (8) (1996) [20] O. Yeniay, A comparative study on optimization methods
32-45. for the constrained nonlinear programming problems,
[6] P. Hanafizadeh, S. Falahatkar, P. Ahmadi and M. M. Si- Mathematical Problems in Engineering, 2 (2005) 165-173.
ahkalroudi, A novel method for inlet duct geometry im- [21] H. Walter, C. Dobias, F. Holzleithner and R. Hofmann,
provement of heat recovery steam generators, Applied Ther- Numerical analysis of the fluid flow in a channel between
mal Engineering, 89 (2015) 125-133. gas turbines and heat recovery steam generator, 2nd Interna-
958 H.-K. So et al. / Journal of Mechanical Science and Technology 32 (2) (2018) 947~958

tional Conference on Fluid Mechanics and Heat and Mass Do-Hyung Lee received his Ph.D. de-
Transfer (2011). gree in aerospace engineering from Uni-
versity of Michigan, United States of
America, in 1996. He was Postdoctoral
Hyun-Kyoo So is in M.D. course at Fellow in NASA Ames Research Center,
Applied Computational Design & Fluid in 2000. He is a Professor of Mechanical
Dynamics laboratory, Department of Design Engineering in University of
Mechanical Design Engineering, Univer- Hanyang, Ansan, Korea.
sity of Hanyang, Ansan, Korea. His
interests include computational fluid
dynamics, design optimization, and flow
experimental study.