You are on page 1of 14

International Journal of Heat and Fluid Flow 96 (2022) 108994

Contents lists available at ScienceDirect

International Journal of Heat and Fluid Flow


journal homepage: www.elsevier.com/locate/ijhff

Understanding loss generation mechanisms in a centrifugal pump using


large eddy simulation
Esra Sorguven a, *, Sevil Incir b, Jonathan Highgate a
a
Thermo Fluid Mechanics Research Centre (TFMRC), University of Sussex, Falmer BN1 9RS, UK
b
Research and Development – Heating, Alarko Carrier Sanayi ve Ticaret AS, GOSB 41480, Gebze, Kocaeli, Turkey

A R T I C L E I N F O A B S T R A C T

Keywords: Water pumps are amongst the most frequently used turbomachines, which consume about 9% of the global
Decarbonisation energy production. This paper provides insight on the loss generation mechanisms in inline centrifugal pumps
Circulation operating under realistic conditions through a detailed analysis based on large eddy simulation (LES). The
Centrifugal pump
equations for the resolved, sub-grid and wall entropy generation terms are derived using the LES filtering and
Large eddy simulation
Entropy generation
Smagorinsky-Lilly formalism. Specific exergy is calculated at the grid points. Entropy generation and exergy
Exergy transfer within the flow domain are visualised to highlight the loss generation mechanisms within the pump.
Results show that there is a strong component interaction. The double-bended inlet pipe initiates Dean-vortices at
the eye of the impeller. This asymmetric inflow coupled with the disturbances at the cut-off cause a non-uniform
flow through the blade passages. Other drivers of the asymmetrical loading of the blade passages are high
leakage and unsteady and asymmetrical mass and exergy transfer between the dead zone. The paper also in­
vestigates the effect of the grid resolution on the first and second order statistics of the flow field and its turbulent
characteristics. A thorough evaluation of the discretisation error is presented to provide a measure for the
required grid density and demonstrate the extent and impact of the temporal and spatial discretisation error in
centrifugal turbomachinery.

were operating far from their best efficiency point (BEP) and therefore
1. Introduction consuming more energy than necessary to run the heating system. It was
estimated that more than 30 TWh per year can be saved in EU just by
Pumps and pumping systems play an important role in global energy eliminating the use of oversized or inefficient pumps (Products et al.,
consumption and greenhouse gas emissions. Comprehensive data ana­ 2011).
lyses show that electric motors are the largest contributor to the global The European Commission set out ecodesign requirements for water
electricity demand and consume 46% of the electricity generated in the pumps in 2012 to avoid energy inefficiencies (“Commission Regulation
world (Waide and Brunner, 2011). About 70% of the energy supplied by (EU) No 547/, 2012). Ecodesign considers pump characteristics such as
electric motors is used by turbomachinery, with a breakdown of com­ rotational speed, size and hydraulic performance, to define a limit for
pressors 32%, fans 19% and pumps 19% (Waide and Brunner, 2011). the energy efficiency index (EEI). The extended ecodesign pump review
This means that pumps consume about 9% of the electricity generated in report published in 2018 shows that the total energy consumption of all
the world. pump systems was 225 TWh/year by 2015, and majority of the pumps
Over 90% of the electric motor systems are small sized, such as investigated were already in scope of the ecodesign regulation 547/
heating circulation pumps. Building heating systems in EU, employ over 2012 (Ecodesign Pump Review Extended report, 2014). This study
100 million circulation pumps, which consumed about 50 TWh of predicts that with the current regulations, the total annual energy con­
electricity in 2011, (i.e. 2% of the overall electricity consumption in EU) sumption of water pump units will rise to 261 TWh/year by 2030;
and emitted over 30 million tons of carbon dioxide (Products et al., however, by applying an extended product approach (EPA) at least 43
2011). Until recent years, the main energy inefficiencies were caused TWh/year of electricity can be saved by 2030. EPA takes into account
due to installing oversized circulator pumps in building heating systems, the “extended product”, i.e. pump, electric motor and the variable speed
with an attempt to reduce customer complaints. These oversized pumps drive, as well as the load profiles and the control algorithms (Arun

* Corresponding author.
E-mail address: e.sorguven@sussex.ac.uk (E. Sorguven).

https://doi.org/10.1016/j.ijheatfluidflow.2022.108994
Received 11 March 2022; Received in revised form 26 April 2022; Accepted 3 May 2022
Available online 2 June 2022
0142-727X/© 2022 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

Nomenclature r radius, m
s Specific entropy, J/(kgK)
A Surface area, m2 T Temperature, K
cp Pressure coefficient u Specific internal energy, J/kg
ex Specific exergy, J/kg ui Flow velocity in i-direction, m/s
Ex
˙ destr Exergy destruction rate, W U Blade velocity, m/s
E(k) Turbulent kinetic energy, J/kg v Velocity magnitude, m/s
g Gravitational accelaration, m/s2 Ẇ Work transfer rate, W
G LES filter kernel z elevation, m
h Specific enthalpy, J/kg
Greek symbols
H Pump head, m
β Blade angle
k wavenumber, 1/m
δij Dirac delta
kSGS Subgrid scale turbulent kinetic energy, J/kg
Δ LES filter length, m
kresolved Resolved turbulent kinetic energy, J/kg
η Efficiency
n Rotating speed, rpm
μ Viscosity, Pa • s
N Data size for statistical sampling
ρ Density, kg/m3
ns Specific speed
φ Arbitrary flow variable
p pressure, Pa
ϕ Flow coefficient
P Power, W
ψ Load coefficient
Q Flow rate, m3 /s
Q̇ Heat transfer rate, W

Shankar et al., 2016). By including the load profiles into the calculation,
total energy consumption can be estimated more realistically. As a result Table 1
of these comprehensive surveys and policy changes, considerable effort Details of the geometry and operating conditions of the investigated centrifugal
has been put in to operate pump systems at or very near to their BEP, pump.
especially by using variable speed drives and intelligent control Impeller inlet radius on hub and on shroud r1,hub = 0.008m, r1,shroud =
algorithms. 0.018m
As regulations force and intelligent technologies enable pumps to Impeller outlet radius (r2 ) 0.025m
Blade inlet angle (β1 ) 17.5o
operate at or near their BEP for the most of their life time, it becomes
Blade outlet angle (β2 ) 49.5o
more important to understand the loss generation mechanisms at the Blade number (N) 8
BEP. A vast literature has been dedicated to investigate the flow phe­ Nominal flow rate (QBEP ) 6.64m3 /h
nomena under off-design conditions. Recent examples of such studies Nominal head (HBEP ) 3.44m
carried out using LES include investigation of rotating stall (Zhou et al., Nominal rotating speed 3, 300rpm (345.56rad/s)
Blade passing frequency (BPF) 440 Hz
2019), instabilities at low flow rates (Hagiya et al., 2019), unsteady
Reynolds number based on the inlet velocity and 46, 744
features (Cui et al., 2020) and pressure fluctuations (Zhang et al., 2019; impeller outlet diameter (Re = ρcinlet D2 /μ)
Li et al., 2022; Posa, 2021). These studies shed light on how and why Q 0.31
Flow coefficient (ϕ = ,)
losses increase dramatically as centrifugal pumps move away from their A2 u2
gH 0.45
BEP. However, there is a gap in the literature on using advanced tech­ Load coefficient (ψ = 2 )
u
niques to analyse pump hydrodynamics at the BEP. This study aims to √2 ̅̅̅̅̅̅̅̅̅̅
56
Q/fq
fill this gap by performing LES of a typical pump used in building heating Specific speed (ns = n 0.75 )
H
systems: a small-sized, inline, medium specific speed, centrifugal pump.
A commercially available pump is chosen for the study, and the flow
domain is kept in high-fidelity with the 3D CAD model to mimic the real details of the geometrical setup and numerical methodology, respec­
hydrodynamics within the pump. Exergy and entropy generation tively. Section 4.1, introduces a thorough investigation of the grid
equations are implemented into the CFD code to visualise the loss gen­ independency with respect to the first and second statistics of the flow
eration mechanisms. variables, and the turbulent characteristics of the flow field. In section
This paper is organised as follows: sections 2 and 3 are devoted to the 4.2, numerical results are evaluated to calculate the exergy transfer and

Fig. 1. (a) 3D model of the investigated inline pump, (b) numerical mesh on the surface of the blades and hub of the coarsest grid level (i.e., 33⋅106 CVs).

2
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

Fig. 2. Schematic view of the experimental test setup (1: valve, 2: inlet pressure transmitter, 3: circulation pump, 4: outlet pressure pump, 5: water tank, 6:
electromagnetic flowmeter, 7: pneumatic control valve).

entropy generation. Flow domain is divided into seven sub-regions


τij = ui uj − ui uj (5)
(control volumes), and the exergy destruction in each control volume
is calculated. Section 4.3 provides a closer look to the time-dependent The subgrid scale stresses resulting from the filtering are unknown,
flow structures within each control volume and a discussion of the and needs to be modelled. There are numerous SGS models available in
interaction between the main flow structures. The paper ends with the the literature, varying in levels of complexity and CPU-time re­
conclusion section. quirements. An overview can be found in (Shang, 2019; Zhiyin, 2015).
In the following simulations, dynamic Smagorisky-Lilly subgrid scale
2. Geometrical setup model is employed. Dynamic Smagorinsky-Lilly model relies on the idea
of defining a second test filter Δ,
̃ which is larger than the original filter
An inline, centrifugal, circulation pump (Alarko Optima 5/8), which
(in the following simulations Δ ̃ = 2Δ) in order to extract additional
is used in domestic heating and cooling systems, is investigated in this
information about the turbulence characteristics. Filtering the subgrid
study (Fig. 1). Table 1 summarizes the geometrical parameters of the
scale stress (Eq. (4)) with the test filter (Δ)
̃ results in:
pump, as well as the operating conditions at the BEP. Pump character­
istics are measured with an experimental setup that is designed and τ̃ij = ũ ̃
i uj − ui uj (6)
operated according to EN 16297-1 and 2 (Fig. 2). Pressure difference
between the pump inlet and outlet, flow rate, rotational speed and Applying both filtering operations to the incompressible Navier-
power consumption are simultaneously measured, and repeatability of Stokes equations result in the subgrid scale tensor of the velocity field ũi :
the experimental data is assured. BEP is defined as the maxima of the
overall efficiency, which is defined as: τij = ũi ũj − ũ
i uj (7)

gρQH Now, consider the resolved flow field ui , evaluating the resolved
η= (1) turbulent stress with the test filter results in:
P
= ũi ũj − ũ
i uj (8)
3. Methodology
L ij

Substituting equations (7) and (8) into equation (6) results in the
The governing equations employed for LES are obtained by filtering Germano’s identity:
the time-dependent Navier-Stokes equations. The filtering process
effectively filters out the eddies whose scales are smaller than the filter L ij = τij − τ̃ij (9)
size, Δ, which is defined as the cube root of the grid volume (Δ = In this equation L ij can be calculated based on the resolved flow
1
(dx⋅dy⋅dz)3 ). The resulting equations thus govern the dynamics of large field; however the terms in the right hand side of the equation (τij and τ̃ij )
eddies. The filtering operation of any flow variable φ is denoted with an need to be modelled. Assuming scale similarity, and using eddy viscosity
overbar in Eqn. (2), and defined as the convolution integral of the field modelling, these terms can be modelled as:
with a filter kernel G. Filtering substantially reduces the amplitude of the
1
high-frequency spatial Fourier components (Sagaut, 2006). τ̃ij − 2 ̃
kk δij = 2CΔ |S|Sij
τ̃ (10)
3

(2) ⃒ ⃒
′ ′ ′
φ(x) = φ(x )G(x, x )dx 1 ⃒ ⃒
D τij − τkk δij = 2C(2Δ)2 ⃒̃S ⃒S̃ij (11)
3
here, D is the domain of concern. The filtered form of the incompress­ Substituting these equations into the Germano’s identity, and solving
ible, unsteady continuity and momentum equations are: for the variable C results in:
∂ui )
(3) ( 1
=0 Lij − Lkk δij
∂xi 3 (12)
C=
( ) ( ) Mij Mij
∂ui ∂ ui uj ∂ ∂u 1 ∂p ∂τij
+ = μ i − + (4) ( ⃒ ⃒ )
∂t ∂xj ∂xj ∂xj ρ ∂xi ∂xj
where Mij = − 2 (2Δ)2 ⃒̃
⃒ ⃒̃
S⃒Sij − Δ2 |S|S
̃ . Simulations are performed
ij

where τij is the subgrid scale stress, which is defined as: with the commercial software Fluent, and the dynamic Smagorinsky-
√̅̅̅̅
Lilly constant CDS = C is limited between zero and 0.23 to avoid nu

3
­
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

Fig. 4. Non-dimensional experimental pump characteristic curves and numer­


ical predictions of the load coefficient.

Fig. 3. Comparison of the temporal and spatial discretisation of LES studies for
low specific speed pumps (Zhang et al., 2019; Zhou et al., 2019; Cui et al., 2020; Table 2
Hagiya et al., 2019; Kye et al., 2018, and Posa, 2021). Load coefficient predicted by different levels of grid resolution.
Load coefficient, ψ , at the BEP Error relative to the
merical instability. fine grid experiment
Sliding interface is defined between the rotating zone (surrounding
coarse grid 0.469 3.3% 6.8 ± 2%
the impeller) and the stationary zone, which consists of the inlet, outlet medium grid 0.465 2.4% 5.8 ± 2%
pipes and the volute. SIMPLE pressure–velocity coupling scheme with fine grid 0.454 – 3.3 ± 2%
second order pressure and bounded central difference discretization.
Implicit second order time discretisation is employed and a fixed time
step size of 1.25 10-5 s is used for all grid levels. Inlet boundary condition
is set as constant volumetric flow rate, and outlet boundary condition is
set as constant pressure. CPU times required for one revolution is 100,
240 and 408 h for the 33, 48 and 80 million control volume grids
respectively, using 128 cores on the high performance computing sys­
tem at the University of Sussex.

4. Results and discussion

4.1. Grid independency study

There is a large variation in the grid resolution of centrifugal pump


LES in the literature. Studies within the last 5 years have used numerical
grids ranging between 2.5 and 512 million control volumes, and number
of time steps per revolution ranging between 360 and 5120 (Fig. 3). All
of these LES studies involve a grid refinement study, and report that they
have achieved grid independent numerical solutions. This large Fig. 5. Time averaged and root mean square values of the pressure coefficient
discrepancy in the number of control volumes employed in LES pump around the volute surface on the mid plane.
simulations, make it worth to investigate the effect of grid resolution in
detail. In the following, a comparison of first and second order statistics In turbomachinery simulations, grid independency is usually
and turbulence characteristics is given for the different grid levels. Nu­ assessed with the accuracy in predicting the pump characteristic curve
merical data is also compared with experimental data, where available, (Fig. 4). Here, LES predicts the pump load coefficient with less than 7%
to achieve a fair indication of discretisation error. error even with the coarsest mesh, and the error decreases to below 4%
Here, we have 3 levels of grid resolution: coarse (33 million CVs), with the finest mesh. Load coefficient prediction with the finest grid is
medium (48 million CVs) and fine (80 million CVs) meshes. Simulations well within the experimental measurement error range (Table 2).
for all mesh levels have been performed with the same time step size These findings indicate that even the coarsest mesh does a satisfac­
(1.25e-5s). Simulations have been performed until statistically steady tory job in predicting time-averaged flow variables. We observe an
state is achieved (approximately 30 revolutions), and afterwards data improvement with increasing grid resolution, however, the impact of
has been recorded for 10 revolutions to perform post processing and increasing the grid resolution is limited. This conclusion is well aligned
statistical analysis. with other published pump LES results in the literature (Posa, 2021a;
a) Effect of grid resolution on the time-averaged flow variables. Posa, 2021b; Posa and Lippolis, 2018).
The fundamental method to test grid independency of CFD simula­ b) Effect of grid resolution on the error root mean square of flow
tions is to compare time-averaged flow variables directly solved by the variables.
CFD code (such as pressure and velocity) from different levels of grid One way of investigating the effect of the grid resolution on the
resolution. If the error to the finest grid level is reasonably small, results temporal characteristics of the flow field is to calculate the root mean
are considered to be grid independent. In cases where experimental data square (rms), which serves as a measure of the amplitudes of time-
is available, a validation can be performed by calculating the relative dependent fluctuations. The time-averaged pressure coefficient is
error with respect to the experimental data. defined as:

4
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

Table 3
Flow parameters to test the grid independency.
coarse grid medium grid fine grid coarse grid medium grid fine grid

Polar angle cP Error cP Error cP cP,rms Error cP,rms Error cP,rms

45o 0.898 4% 0.884 3% 0.860 0.013 10% 0.015 0% 0.015


67.5o 0.888 3% 0.881 2% 0.863 0.012 29% 0.015 8% 0.016
90o 0.879 4% 0.866 2% 0.848 0.010 33% 0.014 8% 0.016
112.5o 0.888 4% 0.883 3% 0.854 0.010 23% 0.013 5% 0.013
135o 0.890 3% 0.886 3% 0.861 0.010 15% 0.011 5% 0.012
225o 0.907 3% 0.901 2% 0.883 0.006 28% 0.008 2% 0.008
247.5o 0.914 3% 0.905 2% 0.884 0.008 14% 0.007 1% 0.007
270o 0.922 4% 0.916 3% 0.890 0.010 24% 0.008 4% 0.007
292.5o 0.929 4% 0.924 3% 0.896 0.006 16% 0.007 7% 0.007
315o 0.921 3% 0.913 2% 0.893 0.005 9% 0.005 2% 0.005

1 ∑N p(x, y, z, ti ) − pref
cP (x, y, z) =
N i=1 1 2 (13)
ρU
2 2

here pref is a reference pressure, which is chosen as the time and surface-
averaged inlet pressure predicted with the fine grid. Rms of the pressure
coefficient is calculated using the following definition:
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
⎛ ⎞2̅


√ 1 ∑N ⎜p(x, y, z, ti ) − pref
cP,rms (x, y, z) = √
√N ⎝ − cP ⎠
⎟ (14)
i=1 1
2
ρU 22

Fig. 5 and Table 3 compare the cP , and cP,rms extracted from the
simulations with the three grid levels, at distinct points on the volute
circumference. The points are selected on the intersection of the volute
surface and a plane normal to x-axis that goes through the mid-height of
the volute. The angle between the radius of the data extraction point and
z-axis is defined as θ, and θ is positive in the direction of impeller
rotation.
Fig. 6. Fourier transformation of the resolved turbulent kinetic energy 15 mm
The time-averaged values, cP , are highly-correlated and exhibit the downstream the volute exit.
same increasing trend for all grid levels. cp,rms values are highest near the
cut-off, where the interaction between the periodic pressure waves
induced by the blade passing and the cut-off surface triggers a highly
turbulent flow field. The unsteady pressure fluctuations remain high in
the first quarter of the volute (between 0◦ and 90◦ ), where the cross-
sectional area of the volute is still relatively small and constraining
the flow field. Afterwards, cp,rms values decrease in the rotational di­
rection, and achieve the lowest values towards the exit of the volute.
Table 3 further demonstrates that the error relative to the fine mesh
is 4% or less for the coarsest, and 3% or less for the medium mesh.
Aligned with the previous findings on the time averaged load coefficient,
local cP values are also well predicted with even the coarsest mesh. The
rms values however, do exhibit a different pattern. In the coarse mesh
error percentages can get as high as 33%, and cP,rms values for the coarse
mesh is lower compared to the other two grid levels. This is an indication
that 33 million elements is not enough to resolve the short-life-span flow
structures, which substantially influence the instantaneous turbulent
flow field, but do not contribute to the time averaged flow field. As grid
resolution increases smaller scales become visible. This leads to an in­
crease in the cP,rms , i.e. a higher portion of the turbulent fluctuations can
Fig. 7. Ratio of the resolved turbulent kinetic energy to the total turbulent
be resolved, and error levels lower than 8% are achieved. One can
kinetic energy on the h50 plane in the impeller.
conclude from these results that rms values have a high dependence on
the grid resolution.
Fig. 6 shows the spatial Fourier transformation of the resolved tur­
c) Effect of grid resolution on turbulent characteristics.
bulent kinetic energy at 15 mm downstream of the volute exit. Since the
An important criteria to assess the quality of LES is the ratio of
data is extracted downstream the volute exit, within the outlet pipe,
resolved turbulent kinetic energy to the overall turbulent kinetic energy.
peaks at the blade passing wavelength are not visible. Periodic pro­
Resolved turbulent kinetic energy, kresolved , is defined based on the
duction of the energy-carrying large eddies in the wake of blades, are
resolved velocity fluctuations, as:
visible within the impeller and volute, and spatial FFTs in these regions
1 [ ′2 ′ ′ ]
(15) show clear peaks at the blade passing wavelength and it harmonics.
kresolved (t, x) = u +v2 +w2
2 However, towards the exit from the volute, a mixing of the blade wake
and the cut-off wake occurs. This mixing causes a shift from larger into

5
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

99.0% for the medium, and 99.3% for the fine mesh.
kresolved
Instantaneous contours of the resolved spectrum (kresolved +kSGS ) on the
mid span (h50) plane for the fine mesh, shows that 99% of the turbulent
kinetic energy spectrum is resolved in the majority of this plane. There
are limited areas near the blade leading edges on the suction side and
mid-chord length on the pressure side, where resolved spectrum ratio is
less than 99%. But, investigating the time history of the minimum value
shows that instantaneous values do not fall below 85% at any point on
this surface.

4.2. Flow field and loss generation mechanisms within the pump

To gain insight on the complex physics within the centrifugal pump,


the flow domain is divided into seven control volumes as shown in the
figure below. This approach allows a systematic analysis of the mass,
momentum and energy transfer and draw conclusions on the in­
Fig. 8. Control volumes selected within the pump for mass and exergy balances teractions of various flow phenomena.
(CV1: inlet pipe, CV2: mixing area upstream of the eye of the impeller, CV3: Leakage flow rate is hard to predict either numerically or experi­
impeller, CV4: leakage, CV5: dead zone, CV6: mixing region upstream of the mentally, mainly due to the geometrical constraints. Leakage cross-
volute, CV7: volute and exit pipe). sectional area is too small for measurement techniques such as hot
wire anemometer (HWA) and involves non-planar surfaces which makes
smaller scales, and additional entropy generation. As the figure shows, optical access difficult for techniques such as laser Doppler anemometry
the numerically predicted turbulent kinetic energy cascade is parallel to (LDA) or PIV. Numerically, discretising such a small surface adequately
the theoretical k− 3 curve between the k values 500 and 5000 1/m.Fig. 7..
5
to predict flux reliably increases the computational demand. Thus, a
Subgrid scale turbulent kinetic energy can be represented as a common simplification in CFD studies is to neglect the leakage area,
function of the sub-grid scale viscosity and filter length: assuming that there is less than 5% leakage flow if specific speed is
( ) higher than 20 (Kye et al., Aug. 2018). Here, leakage area is resolved
μSGS 2 with dense grids, and grid independency is checked to assure that the
kSGS (t, x) = (16)
ρCDS Δ modelling and discretisation errors are negligible. Results show that
Another indicator of the LES grid quality is the ratio of the resolved about 10% of the inlet mass flux flows through the leakage. This leakage
kresolved flow mixes with the inlet mass flux in zone 2, and flows through the
turbulent kinetic energy to the total turbulent kinetic energy (kresolved ).
+kSGS
impeller. Our results indicate that the leakage flux is non-negligible even
kresolved
Celik et al. (2005) suggested to employ kresolved +kSGS
≥ 0.85 as a limit for in pumps with specific speeds higher than 20.
LES quality. Such a universal measure is especially important, when The second point to highlight in Fig. 9 is flux through the dead zone.
detailed experimental data is not available. A further advantage is that There is an asymmetric and time-dependent flow in and out of the dead
this ratio gives a distribution within the flow domain. This allows to zone underneath the impeller. However, the surface-averaged mass flux
locate coarsely-discretised areas within the flow domain and use local into the dead zone remains zero at every time step, as expected for
grid refinement to improve the overall grid quality. incompressible flows.
Here, volume integral of the resolved spectrum is evaluated for the Specific exergy of a fluid is defined as the useful work that can be
impeller (CV3) and the volute (CV6 + CV7) regions, to compare the extracted when the fluid is brought to equilibrium with its environment
different grid resolutions quantitatively. Results show that impeller is at the dead state through reversible processes. Thermodynamic and fluid
well resolved in all grid levels withV1 ∭ kresolved
dV > 99%. For the dynamic properties at the dead state is denoted with the subscript 0.
kresolved +kSGS
CV3
1( 2 )
volute, the value of V1 ∭ CV6+CV7 kresolved
dV is 98.3% for the coarse, ex = (h − h0 ) + T0 (s − s0 ) + v − v20 + g(z − z0 ) (17)
kresolved +kSGS 2

Fig. 9. Mass flux within the pump (not to scale).

6
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

Fig. 10. Exergy flow within the pump (not to scale).

Table 4 Table 5
Time averaged and root mean square values of the mass and exergy flux within Exergy destruction rate in each control volume within the pump.
the pump. Exergy destruction rate
Mass Normalised Exergy flux Normalised Surface
CV1 0.28 W 1.3%
flux rms of mass ṁe = ∬ rms of and time
CV2 0.13 W 0.6%
ṁ(kg/s) flux ρexvn dA(W) exergy flux averaged
CV3 9.13 W 41.1%
ṁrms ṁerms specific
CV4 1.65 W 7.4%
ṁ ṁe exergy
CV5 1.34 W 6.1%
1
∬ exdA CV6 3.42 W 15.4%
A
CV7 6.25 W 28.1%
inlet 1.84 0.00% 46.52 0.18% 25.30 Total 22.20 W 100%
a 1.84 0.00% 46.23 0.18% 25.15
b 2.02 0.03% 54.27 0.15% 25.08
c 0.18 0.90% 8.16 0.28% 42.35 is defined as the time and surface averaged pressure on the inlet surface
d 0.18 0.89% 9.81 0.33% 57.45
and v0 is taken as zero. Note that the absolute value of the specific exergy
e 2.02 0.03% 132.54 0.05% 67.95
f 0.00 39.6% 1.34 8.12% 59.20 at a point does change as a function of the thermo-fluid dynamic prop­
g 1.84 0.05% 117.96 0.05% 64.12 erties at the chosen dead state (i.e. here P0 and v0 ). However, the dif­
outlet 1.84 0.00% 111.72 0.00% 60.77 ference of specific exergy between two points within the system (Δex),
and exergy destruction (Ex˙ destr ) are independent of the choice of the dead

To calculate the specific exergy of the water flowing inside the state.
investigated pump, let us rewrite this equation in terms of flow pa­ The general form of the exergy balance equation written for a control
rameters that can be extracted from the incompressible LES. The first volume is:

d
∫∫∫ ∑n ∑m ∑k ( T0
)
(ρex)dV = (ṁex) in − (ṁex) out + 1 − ˙ destr
Q̇j − Ẇ net − Ex (19)
dt in=1 out=1 j=1 Tj

( ( P0
))
term on the right hand side can be replaced by (u − u0 ) + Pρ − ρ . The
here, the term describing the exergy transfer due to heat is zero, since
first Gibbs equation states that ds = duT
+ P d(1/
T
ρ)
. Since water flow within ( )
the pump is incompressible, i.e. dρ = 0, second term in the Gibbs Q̇ = 0. All terms except the exergy destruction Ex˙ destr are calculated
equation drops, and internal energy change can be replaced by du =
by post processing the LES results at each time step. Volume integrals are
cdT. Furthermore, there is no observable temperature change in the ( )
pump flow, therefore dT = 0. Thus, for an incompressible fluid under­ calculated to obtain the accumulation term dtd ∭ (ρex)dV , which might
going an isothermal process, both specific internal energy and specific be non-zero instantaneously, but its time-averaged value has to
entropy remain constant (du = u − u0 = 0, ds = s − s0 = 0). When the converge to zero. Exergy flux through the control surfaces are calculated
( )
last term g(hL − hL0 ) is ignored, since it is negligibly small compared to via surface integrals, as ṁex = ∬ ρexvn dA , where vn is the normal ve­
the other two terms, the equation becomes: locity on the control surface dA. Ẇnet is only non-zero on the blade
surfaces, where work is transferred from the rotating surfaces of the
1 1( )
ex = (P − P0 ) + v2 − v20 (18) impeller to the fluid. Instantaneous torque values are extracted from the
ρ 2
simulation to calculate the instantaneous work transfer to the fluid as
here P0 , and v0 are the dead state pressure and velocity, respectively. P0 Ẇ = τθ̇. Fig. 10 and tables 4 and 5 are summarizing the results of these

7
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

Fig. 11. Velocity and exergy contours on the exit of the inlet pipe (surface a in Fig. 8) a: time averaged axial velocity component, b: time averaged magnitude and
vectors of the secondary velocity, c: time averaged specific exergy, d: instantaneous specific exergy.

calculations. impeller (see Fig. 1a, and Fig. 8). Bend radii and Reynolds number are
Time-averaged results show that the impeller transfers 87.4 W of the main parameters that influence the boundary layer thickness dis­
work to the fluid, and that 22.2 W of this work is lost due to the hy­ tribution along the pipe inner surface, and thus, the strength of the Dean
drodynamic irreversibilities within the pump. Investigating the time type vortices in the secondary flow. Eqs. (20) and (21) define the mag­
history of mass and exergy flux values show that there is very limited nitudes of the time-averaged secondary (csec ) and axial (cx ) velocities,
fluctuation with respect to time; thus, rms values remain low. respectively.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅
uy + uz
4.3. Local loss generation mechanisms csec =
Q (20)
Ain
4.3.1. CV1: Inlet pipe
√̅̅̅̅̅
Impellers are designed for a rotationally symmetric inflow, which is ux
very difficult to achieve in real applications. 90◦ elbows and bends in the cx =
Q (21)
inlet pipe are very common in realistic systems. In an inflow type pump, Ain
the inlet pipe first curves upwards, goes through an inflexion point and
Fig. 11 shows kidney-like Dean type vortices appearing on the outlet
maximum height, and then curves downwards sharply into the eye of the

Fig. 12. Exergy contours on the eye of the impeller (surface b in Fig. 8) a: instantaneous, b: time averaged in the relative frame.

8
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

Fig. 13. Contours of rms and time averaged pressure coefficient at h25, h50 and h75 surfaces of the impeller.

Fig. 14. Contours of instantaneous vorticity magnitude.

Fig. 15. Contours of instantaneous wall entropy generation (a), and time averaged wall shear stress (b).

of the inlet pipe (surface a in Fig. 8). Exergy contours demonstrate that The losses within the inlet pipe is only 1.3% of the total loss; how­
the effect of the blades on the flow field is visible in time-dependent ever, the interaction between this rotationally asymmetric flow and
snapshots (Fig. 11 d), but are filtered out through time-averaging impeller influences the entire system. These findings support the exist­
(Fig. 11 c). The sharp turn downwards at the end of the inlet pipe cau­ ing studies in literature: rotationally asymmetric flow in the eye of the
ses a flow separation on the lower surface. The effect of this flow sep­ impeller results in reduction in the efficiency and pressure head, an
aration is visible in Fig. 11 a, where in the left hand side, a very low axial increased risk of local cavitation (Mittag and Gabi, 2015) and strong
velocity region appears. Most of the entropy generation occurs in this non-uniformity in the impeller channels (Li et al., 2020).
zone, where also the wall shear stress values are nearly 5 times larger
than the rest of the inlet pipe. Not only velocity, but also pressure is low 4.3.2. CV2: Mixing upstream of the eye of the impeller
in this region, resulting in low exergy values (Fig. 11 c and d), as well as Flow through the inlet pipe and the leakage (surface c in Fig. 8) are
an increased risk of local cavitation. mixed in CV2, and flow out through the eye of the impeller. The neck

( )
Fig. 16. Contours of the entropy generation sgen,D + sgen,D′ .

9
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

Fig. 17. Time averaged, rms and instantaneous x-velocity (U) and exergy on the surface f (in Fig. 8).

ring causes some blockage in the circumference. to the cut-off (blades 1 and 2). Especially at h50 and h75 we observe a
Figure below shows the instantaneous and time-averaged exergy larger low-pressure region stretching up to about 80% of the chord
contours in the eye of the impeller. The Dean vortices and the low- length of blade 1 on the suction side. The cP,rms contours give further
exergy zone are clearly visible. Time averaging is performed in the insight, showing a large temporal fluctuation in pressure values between
relative frame; thus, the distorted inflow is filtered out. In both figures, these blades next to the cut-off, compared to other blade passages.
high exergy values are visible in the circumference. This is due to the The instantaneous vorticity contours (Fig. 14) reveal that the
relatively high exergetic fluid flowing in from the leakage. The average asymmetric inlet flow feeds in vortices into the passage between blades
specific exergy on the surface c is 42.35 J/kg, which is substantially 1 and 2. To a smaller extent passage between blades 1 and 8 is also
larger than the average specific exergy on the surfaces a and b, which are affected. In the in-line type pumps, the inlet and exit pipes are in the
25.15 J/kg and 25.08 J/kg, respectively. Flow through the surface c is same axis (here z-axis). Therefore, both the position of the cut-off and
axisymmetric, demonstrating only negligible variation in velocity or the inlet high vorticity area are affecting the same blade passage. Hence,
exergy with respect to θ. we observe a larger region of high vorticity around the blades in this
Fig. 12 shows that there is a non-uniform, asymmetric flow in the eye area.
of the impeller. The total exergy flowing through an impeller passage Highest vorticity levels are visible in h75. Passage vortex occupies
( ) (
θ+2π
∬ θ 8 ρexu⋅dA at a given instant has a high variation. Lowest ∬ θ 8
θ+2π nearly half of the blade passage, indicating high mixing losses. The
) strength of the asymmetric-inlet-vortex is decreasing in the direction
ρexu⋅dA values are experienced by the blades passing through the low- from shroud to hub. On the h25 surface only notable vortices occur on
exergy zone on the left hand side. Furthermore, there is a non-zero dex the suction side of the blade trailing edges. This is due to the migration
dr
of the hub boundary layer towards the suction side corner.
gradient influencing each blade passage, which shifts high exergy flow
To identify the loss generation mechanisms, local entropy generation
towards the shroud.
is calculated. Similar to all flow variables in LES, entropy generation is
also filtered, and includes resolved and modelled parts. Following the
4.3.3. CV3: Impeller
derivation of local entropy generation by Kock and Herwig (Kock and
In this section flow and loss generation within the impeller is
Herwig, 2004), and introducing the LES-specific filtered variables as
explained using the contour plots of the pressure coefficient (Fig. 13),
derived in Higgins et al. (xxxx), Geurts and Fröhlich (2002), we obtain
vorticity magnitude (Fig. 14), exergy (Fig. 16) and entropy generation
the resolved entropy generation term as:
(Fig. 17) on three meridional surfaces, i.e. h25, h50 and h75 (h0 is the

( [( ) ( )2 ( )2 ] ( ) ( ) ( ) )
μ ∂u 2 ∂v ∂w ∂u ∂v 2 ∂u ∂w 2 ∂v ∂w 2
sgen,D = 2 + + + + + + + + (22)
T ∂x ∂y ∂z ∂y ∂x ∂z ∂x ∂z ∂y

hub surface and h100 is the shroud surface).


Flow remains attached within the blade passages, and there is a
smooth increase in the pressure within blade passages (Fig. 13). There is The entropy generation due to the subgrid scale fluctuations is
some difference between the cp distributions in the blade passage nearest derived using the Smagorinsky-Lilly subgrid scale model:

10
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

Fig. 18. Time averaged and instantaneous exergy contours.

( [( ′ ) (∫ ∫ )
μ ∂u 2
( ′ )2
∂v scales CV sgen,D dV < CV sgen,D
′ dV . This is expected because the sub-grid
sgen,D′ = 2 +
T ∂x ∂y structures are small and dissipative. The resolved scales, on the other
( ′ )2 ] ( ′ ′) ( ′ ′) ( ′ ′)
) hand, involve the large, energy-carrying eddies. Majority of dissipation
∂w ∂u ∂v 2 ∂u ∂w 2 ∂v ∂w 2 occurs in the smallest scales of turbulence. But still, resolved compo­
+ + + + + + +
∂z ∂y ∂x ∂z ∂x ∂z ∂y nents make a non-negligible contribution, because here the grid reso­
ρε ρ (μSGS /ρ)3 lution is fine enough to resolve a considerable range of wavelengths.
=
T
=
T (CDS Δ)4 This observation is supported by the results in section 4.1.c, where it is
shown that the resolved spectrum accounts for over 90% of the total
(23)
turbulent kinetic energy.
where Δ is the LES filter length, CDS is the sub-grid dynamic constant, Similar to the vorticity, entropy generation is high on h75 surface
( ′) and decreasing towards the hub. This result, namely that the losses are
∂v ∂v

and ε is the turbulent kinetic energy dissipation rate ε = ν∂xij ∂xji . higher towards the shroud, is also observed by Posa et al. in (Posa and
Equations (22) and (23) are defined for a point in the flow domain. On Lippolis, 2019; Posa et al., 2016). Interfaces between CV3, CV6 and CV7
the wall surface, entropy is generated mainly due to the wall velocity are shown in the plot to provide a measure of the entropy generation in
gradient appearing due to the no-slip condition. different zones. In the impeller (CV3) entropy is generated on the blade
√̅̅̅̅̅̅̅̅ surfaces, and in the mid passage near the shroud. The interaction be­
τwall uP τwall + τwall tween the distorted inflow, cut-water and blades cause a strongly non-
sgen,wall = = y (24)
T T ρ uniform flow within the blade passages.

here uP is the velocity at the center of the first grid above the wall, and 4.3.4. CV4: Leakage
τwall is the local wall shear stress. As seen in equation (24), and revealed Leakage path is substantially limited by the neck ring. Here, the flow
in Fig. 15 wall shear stress (WSS) and wall entropy generation are domain is reconstructed based on the 3D CAD geometries. Even small
correlated. In Fig. 15 the effect of the horseshoe vortex at the inlet and manufacturing variances are expected to have a large impact in this
impeller blade tip are visible. In the time-averaged WSS, this is seen as region. The mass flow rate through the leakage will be influenced by
two nearly-continuous streaks; whereas in the instantaneous sgen,wall von- these manufacturing variances. For this ideal case, leakage flux accounts
Karman type vortex shedding structures are visible. There are two sets for about 10% of the inlet mass flux. There is limited and highly scat­
structures: one parallel to the shroud, the other beginning at the inlet tered data on the leakage flux percentage for low and medium specific
blade tip and mitigating towards the hub. High entropy generation is speed pumps. One reason might be that manufacturing variances are
observed on the leading edges. There is some entropy generation on the usually not measured in the experimental investigations, and are not yet
hub surface, indicating the boundary layer migration towards the suc­ studied numerically.
tion side corner (effects visible in Fig. 14 on the surface h25). Water with an average exergy of 57.45 J/kg flows through surface d,
( )
Fig. 16 shows the total entropy generation sgen,D + sgen,D′ at a given into the CV4, and flows out through the surface c with an average exergy
instant. The entropy generation occurs mostly in small of 42.35 J/kg. Losses through this area is mainly due to wall friction, and

11
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

Fig. 19. Space – time plots of entropy generation and exergy at three positions in the blade wake: r2 + 1mm, r2 + 5mm,.r2 + 9mm.

accounts for 7.4% of the total exergy destruction. 4.3.6. CV6: Mixing downstream the impeller
There is mass and exergy flux to the CV6 from four surfaces: e
4.3.5. CV5: Dead zone (impeller outlet), d (leakage inlet), f (dead zone) and g (inlet to the
The annular surface separating volute from the dead zone (surface f volute). 2.02 kg/s of high-exergy water is flowing into the CV6 through
in Fig. 8) is plotted in Fig. 17 to visualise the time-averaged, rms, and the impeller outlet (surface e). As visible in Fig. 18, blade wakes carry
instantaneous x-velocity component and specific exergy. There is a the highest exergy levels. Instantaneous contours show that this high-
highly complicated flow through this surface with simultaneous in and exergy wakes are spiralled out into the volute. At positions on h25
out flow, and high gradients in time, r, and θ. Time-averaged U values downstream the cut-off, they almost reach the volute surface. High-
show that upstream the cut-off, at high-radii, there is a net inflow into exergy wakes are also mixed with the medium-exergy flow from the
the dead zone (positive time-averaged U values between blade passage within the CV6 and CV7.
270o < θ < 360o ). The net time-averaged outflow downstream the cut- The total exergy flowing into CV6 from surface e is 132.54 W, and
off between 0o < θ < 120o , balances this out, resulting in net-zero- there is nearly no temporal variation in this value (Table 4). 9.81 W of
flow through the surface. Instantaneous U contours demonstrate that this exergy flows into the leakage (surface d) almost steadily with
there are fluctuating vortex structures, which change the location of in- negligible fluctuations. A net exergy of 1.34 W is flowing into the dead
and out-flow from time step to time step. High rms values in the up­ zone (surface f) and destroyed. Contrary to the exergy flow distribution
stream of the cut-off and between 180o < θ < 270o , highlight the re­ on the other surfaces, this exergy flux demonstrates high temporal and
gions with the highest fluctuations. spatial variation as discussed in the previous section. 117.96 W of exergy
Time averaged exergy contours show a high-exergy flow domain in is flowing through the imaginary interface between the CV6 and CV7
the upstream of the cut-water. That coincides with the inflow zone. (surface g).
Therefore, high exergy water is flowing into the dead zone, loses its work
potential as it is rotated by the lower surface of the hub, to friction and 4.3.7. CV7: Volute and the exit pipe
viscous forces, and flows out of the dead zone with a lower exergy. CV7 is the region where the high-exergy blade wakes continue to
Instantaneous and rms contours of specific exergy show limited fluctu­ spiral out, and interactions with the cut-water become more apparent.
ation in time. Thus, mainly the same exergy profile is observed at every There is a low-exergy flow downstream the cut-off, especially visible on
time step. However, since there is a high temporal variation in mass flux, the h50 and h75 surfaces. That is interacting with the blade wakes and
rms of the exergy flux through this surface (8.12%, see Table 4) is suppressing the convection of the high-exergy wake especially for blades
substantially higher than the rest of the surfaces within the pump. About 1 and 8.
6% of the total exergy destruction occurs in CV5. To gain further insight on the wake propagation, space–time plots of
entropy generation and exergy are plotted Fig. 19. These space–time

12
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

plots are very useful to interpret the temporal evolution of wake struc­ and computed at each grid point. Results are post-processed to calculate
tures passing through a fixed 1D region; but, unfortunately, rarely the mass and exergy flux within the pump. Results indicate that 41% of
employed in centrifugal pump analysis. Data is extracted from circles the total exergy destruction occurs within the impeller, whereas 28.1%
with three different radii (r2 + 1mm, r2 + 5mm, and r2 + 9mm) on the is destructed within the volute, and 15.4% is destructed within the
h50 plane, at each time step. First two radii are within CV6, and the last mixing zone downstream the impeller.
one is within CV7 close to the cut-off radius. The plots show the time step Investigation of the flow structures within the pump shows that, even
in x-axis, and location of the data points in θ-direction in y-axis. though the exergy destruction within the inlet pipe only accounts for
At the immediate wake of the blade trailing edges (r2 + 1mm) both 1.3% of the total exergy destruction, the flow distortion generated there
entropy generation and exergy plots demonstrate clear streaks. These causes additional loss mechanisms within the entire domain. The Dean
streaks are convected with a rate of 45o per 1 blade passing period (BPP). vortices in the eye of the impeller causes a highly non-uniform blade
That rate is equal to the blade rotational speed. Hot spots begin at the passage flow. This is observed in the high entropy generation regions
high-pressure corner of the trailing edge with a sudden increase in the appearing mid blade passage. The fact that the positions of the Dean
entropy generation and specific exergy. High values continue to appear vortices and the cut-off coincides in inline pumps, amplifies the asym­
throughout the blunt trailing edge. Towards the suction-side corner of metry within the impeller. The high loss generation is associated with
the trailing edge, mixing with the medium-exergy blade passage flow the coupled interactions between the bended inlet pipe, cut-off region,
begins. Therefore, the contours following the suction side of the blade high leakage flow and the highly-fluctuating asymmetrical flow towards
are not as sharp as in the pressure-side. the dead zone.
At r2 + 5mm, convection rate of the streaks exhibit a larger variation.
45o
Entropy generation streaks have nearly a rate of BPP for θ ≈ 45o . This CRediT authorship contribution statement
o
10
rate decreases in the rotational direction and approaches to BPP for
o o o
θ ≈ 340 . In the range 340 < θ < 45 , streaks with different rates are Esra Sorguven: Conceptualization, Methodology, Supervision,
superposed. The specific exergy space–time plot demonstrates parallel Writing – original draft, Writing – review & editing. Sevil Incir: Vali­
45o
streaks with a rate of approximately BPP between 45o < θ < 300o , and a dation, Investigation, Writing – review & editing. Jonathan Highgate:
0 o Software, Investigation, Writing – review & editing.
nearly horizontal streak with a rate of approx. BPP , (i.e. no effect of the
impeller rotation, high-exergy region remains in the same angular po­
sition). This and the very-low-exergy region in 315o < θ < 360o , are the Declaration of Competing Interest
direct consequences of the cut-off.
Data extracted at r2 +9mm is within the CV7, and comes very close to The authors declare that they have no known competing financial
the cut-off location at θ = 0o . Here, the entropy generation streaks are interests or personal relationships that could have appeared to influence
0o 15o the work reported in this paper.
nearly horizontal, with rates ranging between BPP and BPP . The interac­
tion with the cut-off is especially visible in the range 340o < θ < 45o ,
References
where large entropy generation values appear, and the high-entropy-
( 0o )
generation regions remain steadily fixed in space BPP . Specific exergy Waide, P., Brunner, C.U., 2011. Energy-efficiency policy opportunities for electric motor-
does not reach the high values seen at lower radii, except for a few driven systems international energy agency energy efficiency series.
Products, B., The, S., Index, E.E., Intelligent, T., E, Europe, Pumps, E.P., Circulation
limited hot-spots. At this higher radius within the mid-volute region, pumps : recommendations, no. March 2011, pp. 1–4, 2015.
flow is more diffused and a more uniform distribution in the flow exergy Commission Regulation (EU) No 547/2012 of 25 June 2012 implementing Directive
is observed. 2009/125/EC of the European Parliament and of the Council with regard to
ecodesign requirements for water pumps, Jun. 2012.
[4] Ecodesign Pump Review Extended report (final version), 2018. [Online]. Available:
5. Conclusion www.ecopumpreview.eu.
Arun Shankar, V.K., Umashankar, S., Paramasivam, S., Hanigovszki, N., 2016.
A comprehensive review on energy efficiency enhancement initiatives in centrifugal
One component in decarbonising heating is to reduce the energy pumping system. Appl. Energy 181, 495–513.
consumption of the circulation pumps. Therefore, it is essential to have Zhou, P., Dai, J., Yan, C., Zheng, S., Ye, C., Zhang, X., 2019. Effect of stall cells on
an understanding on the loss generation mechanisms in centrifugal pressure fluctuations characteristics in a centrifugal pump. Symmetry 11 (9).
https://doi.org/10.3390/sym11091116.
pumps operating in realistic systems and conditions. This paper aims to
Hagiya, I., Kato, C., Yamade, Y., Fukaya, M., Nagahara, T., 2019. Analysis of blade-
serve this purpose by investigating an inline pump using large eddy passage flow of a mixed-flow pump at performance-curve instability. in: IOP
simulation. Conference Series: Earth and Environmental Science, 2019, 240 (3). doi: 10.1088/1755-
The existing centrifugal pump LES studies in the literature exhibit a 1315/240/3/032038.
Cui, B., Zhang, C., Zhang, Y., Zhu, Z., 2020. Influence of cutting angle of blade trailing
large variation in both spatial and temporal discretisation level. Nu­ edge on unsteady flow in a centrifugal pump under off-design conditions. Appl. Sci.
merical grids range between a few million to a few hundred million grid (Switzerland) 10 (2). https://doi.org/10.3390/app10020580.
elements. The thorough evaluation of the discretisation error presented Zhang, N., Liu, X., Gao, B., Wang, X., Xia, B., 2019. Effects of modifying the blade trailing
edge profile on unsteady pressure pulsations and flow structures in a centrifugal
in section 4.1. aims to address this issue in the literature and demon­ pump. Int. J. Heat Fluid Flow 75, 227–238. https://doi.org/10.1016/j.
strate the merits and limits of LES numerical grids in predicting first and ijheatfluidflow.2019.01.009.
second order statistics and turbulent characteristics. Even lower grid Li, D., Qin, Y., Wang, J., Zhu, Y., Wang, H., Wei, X., 2022. Optimization of blade high-
pressure edge to reduce pressure fluctuations in pump-turbine hump region.
densities are capable of predicting time-averaged flow variables reli­ Renewable Energy 181, 24–38. https://doi.org/10.1016/j.renene.2021.09.013.
ably. However, resolving temporal fluctuations (rms), and turbulent Posa, A., 2021a. LES study on the influence of the diffuser inlet angle of a centrifugal
characteristics require a finer resolution of the turbulent scales, i.e. pump on pressure fluctuations. Int. J. Heat Fluid Flow 89, 108804. https://doi.org/
10.1016/j.ijheatfluidflow.2021.108804.
higher numbers of control volumes. It is shown that the grid with 80 Sagaut, P., 2006. Large Eddy Simulation for Incompressible Flows. Springer-Verlag,
million control volumes is satisfactory for a reliable prediction of the Berlin/Heidelberg, 10.1007/b137536.
time-averaged and root mean square of the flow variables, as well as the Yeru Shang, 2019. A numerical analysis of a corrugated channel flow by Large Eddy
Simulation.
turbulent characteristics of the flow field.
Zhiyin, Y., 2015. Large-eddy simulation: past, present and the future. Chin. J. Aeronaut.
Loss generation mechanisms can be best visualised with the entropy 28 (1), 11–24.
generation. The equations for resolved, sub-grid and wall entropy gen­ Kye, B., Park, K., Choi, H., Lee, M., Kim, J.-H., Aug. 2018. Flow characteristics in a
eration terms, as well as the specific exergy are derived using the LES volute-type centrifugal pump using large eddy simulation. Int. J. Heat Fluid Flow 72,
52–60. https://doi.org/10.1016/J.IJHEATFLUIDFLOW.2018.04.016.
filtering and applying the Smagorinsky-Lilly formalism in the subgrid Posa, A., 2021b. LES investigation on the dependence of the flow through a centrifugal
scales. These additional equations are coded into the main flow solver pump on the diffuser geometry. Int. J. Heat Fluid Flow 87, 108750.

13
E. Sorguven et al. International Journal of Heat and Fluid Flow 96 (2022) 108994

Posa, A., Lippolis, A., 2018. A LES investigation of off-design performance of a Kock, F., Herwig, H., 2004. Local entropy production in turbulent shear flows: a high-
centrifugal pump with variable-geometry diffuser. Int. J. Heat Fluid Flow 70 Reynolds number model with wall functions. Int. J. Heat Mass Transf. 47 (10–11),
(February), 299–314. https://doi.org/10.1016/j.ijheatfluidflow.2018.02.011. 2205–2215. https://doi.org/10.1016/J.IJHEATMASSTRANSFER.2003.11.025.
Celik, I.B., Cehreli, Z.N., Yavuz, I., 2005. Index of resolution quality for large Eddy Higgins, C., Parlange, M.B., Meneveau, C., Energy dissipation in large-eddy simulation:
simulations. J. Fluids Eng. 127 (5), 949–958. https://doi.org/10.1115/1.1990201. dependence on flow structure and effects of eigenvector alignments.
Mittag, S., Gabi, M., 2015. Experimental and numerical investigation of centrifugal Geurts, B.J., Fröhlich, J., 2002. A framework for predicting accuracy limitations in large-
pumps with asymmetric inflow conditions. J. Therm. Sci. 24 (6), 516–525. https:// eddy simulation. Phys. Fluids 14 (6). https://doi.org/10.1063/1.1480830.
doi.org/10.1007/s11630-015-0817-8. Posa, A., Lippolis, A., 2019. Effect of working conditions and diffuser setting angle on
Li, Y.-B., Fan, Z.-J., Guo, D.-S., Li, X.-B., 2020. Dynamic flow behavior and performance pressure fluctuations within a centrifugal pump. Int. J. Heat Fluid Flow 75, 44–60.
of a reactor coolant pump with distorted inflow. Eng. Appl. Comput. Fluid Mech. 14 Posa, A., Lippolis, A., Balaras, E., 2016. Investigation of separation phenomena in a
(1), 683–699. radial pump at reduced flow rate by large eddy simulation. J. Fluids Eng. 138 (12),
FE-15-1318.

14

You might also like