Академический Документы
Профессиональный Документы
Культура Документы
Wen-Guang LI
Corresponding author:
Dr Weng-Guang Li
E-mail: liwg38@yahoo.com.cn
1
ABSTRACT
The hydraulic performance of a centrifugal pump handling water and viscous oils was
investigated numerically by using a CFD code based on a steady or unsteady, three-dimensional,
incompressible turbulent flow, where the turbulence effect is involved with the standard k −ε
turbulence model combined with the non-equilibrium wall function in this work. The flow passages of
the pump are subject to a particular surface roughness. The effect of liquid viscosity on the impeller
and pump performance was indicated by examining the hydraulic parameters, such as theoretical
head, hydraulic efficiency, slip factor, hydraulic loss coefficient etc in terms of flow rate. The
unsteady flow, wall function and turbulence flow models effects on the performance are also
indicated. The computed pump performance has been compared with the experimental data. The
results showed that the influence of the relative position between volute tongue tip and impeller blade
trailing edge on the performance is less significant. The volute offers increasing influence on the flow
at the impeller exit at low flow rate, causing reduced impeller theoretical head at much low flow rate.
The “sudden-rising head effect” is confirmed to exist and high viscosity and particular large surface
roughness are responsible for this effect.
Key words: centrifugal pump impeller volute viscosity performance roughness CFD
2
1 Introduction
For recent years, CFD (computation fluid dynamics) has become one vital tool of solving
complicated flow problems in research and engineering communities. As a result, it has found
substantially increasing applications in fluid dynamics of rotor-dynamic pumps. The steady 3D
incompressible turbulent flows in isolated centrifugal impellers have been studied numerically in [1-
5]. The turbulent flows in centrifugal diffuser pumps in [6, 7] and those in centrifugal volute pumps in
[8-15] have been conducted by existing CFD codes, accordingly the performance curves against flow
rate were also established based on the CFD results. In some work, for example, in [10, 11, 13-15],
the computed performance curves have been compared with those measured. Nevertheless, these
contributions shed light on the understanding of fluid dynamics and the performance optimization of
rotor-dynamic pumps. It is very interesting to note that in these numerical investigations, water is
usually used as a working liquid and the roughness of wetted surface of flow passages is specified to
be zero. However, in the practical applications of pump, the liquids handled are in a wide range.
Further, the wetted surface roughness perhaps is rough enough sometimes. Therefore, the
computations of flow in centrifugal pumps, where liquids other than water and rough surface are
involved, should be targeted.
On one hand, for many years, significant contributions have been denoted to the investigation
into the performance and flow in centrifugal pumps handling viscous oils in [16-23]. In more recent
years, the hydraulic characteristics of the centrifugal pumps with respective two different impellers
having 20o and 44 o blade exit angles were measured when water and highly viscous oils were pumped
[24]. The time-averaged turbulent flows of water and viscous oils in the impellers and volute of a
centrifugal pump were investigated experimentally by using LDV in [25, 26].
On the other hand, a laminar steady flow caused from high-viscosity of fluid in an isolated
impeller was examined numerically in [27]. A time-averaged steady turbulent flow of viscous oil in a
centrifugal pump impeller was calculated by means of a CFD code-PHOENICS in [28], the computed
results were compared with those given by using LDV measurements. Unfortunately, because of the
limitation in both computer hardware and code itself, the calculated results were poor. The highly
viscous effect of fluid on pump affinity laws was addressed experimentally and numerically by a
small centrifugal blood pump in [29]. The numerical simulations indicated the Reynolds number is
necessary to insure accurate scaling in the small pump. It should be noticed that these numerical
investigations fail to examine detailed variations of hydraulic performance and internal flow field of
centrifugal pumps in terms of liquid viscosity, causing the relations of hydraulic loss in the impeller
and volute with liquid viscosity are unavailable. Moreover, an effective comparison of performance
between experiment and numerical outcomes isn’t made. So that the numerical computations of
viscous oil flow and corresponding hydraulic performance of centrifugal pumps need to be
investigated in detail. In this paper, we tend to estimate numerically the performance in a centrifugal
pump with an impeller blade exit angle of 20o and a volute when the pump transports water and
3
mechanical oils with various kinematical viscosities and densities by using a CFD code-FLUENT to
investigate effects of the viscosity on the performance on a theoretical basis.
For a centrifugal volute pump, an interaction between impeller blades and volute tongue exists.
Nevertheless, this interaction should be taken into account in the CFD computations. In FLUENT, the
interaction can be handled with both MRF (multiple reference frame) method and sliding-mesh
technique. Currently, two routines can be applied to deal with the interaction: one is steady MRF
method (frozen rotor method in other codes) by which steady flow computations are carried out at a
series of relative position between blade and volute tongue; one is unsteady MRF method by which
unsteady flow calculations are launched with the aid of the sliding-mesh technique. For the
unsteadiness of flow is ignored in the steady MRF method, it is less accurate. In the unsteady MRF
method, however, both the rotational effect of impeller and the steadiness of flow are involved, the
method is more accurate than the former. Further, as a reward, it is able to offer time-dependent flow
variables and pump performance. In general, the time consumed to get a periodical solution with the
unsteady MRF method is as long as about 50 times the steady one [11]. Consequently, the unsteady
MRF method perhaps is applicable in the flow and performance computation of a pump at just a few
operating points. If we are just interested in steady flow and steady performance of a pump, it should
be practical to make use of the steady MRF method in the corresponding computations.
In the paper, we mainly tend to have theoretical effects of liquid viscosity on the performance of
a centrifugal pump with volute in a steady manner. However, we also attempt to a little bit involve
some critical factors, such as flow steadiness, wall function and turbulence model effects to make sure
the viscosity effects have been identified correctly.
2 Computational Models
2.1 Flow Domains
An end-suction, single-stage, centrifugal pump was severed as the computational model, whose
both performance and flow field were investigated experimentally already in [24-26]. The pump
specifications at the design point are: flow rate Q =25m3/h, head H =8m, rotational speed n
=1450r/min, specific speed ns =1317 (USA) =93 (China1). The impeller geometrical parameters are
as follows: eye diameter De =62mm, discharge diameter D2 =180mm, number of blades Z =4,
blade discharge angle β2 b =20°, blade warp angle φ=140°, blade outlet width b2 =18mm. Outside
the impeller, a volute with 40mm wide rectangular cross-section and a discharge nozzle of 50mm
diameter is installed. The volute tongue tip is located at a circle of 190mm diameter (Fig. 1).
As mentioned above, the impeller-volute interaction can be dealt with by either unsteady or
steady MRF method. Since the steady rather than unsteady performance of centrifugal pump is
interesting for us at the fist phase of research, the steady MRF method is applied in this contribution.
1
the specific speed of a pump is defined to be n s = 3.65 nQ 0.5 H 0.75 ( rpm , m 3 /s, m)
4
Subsequently, the flow domain geometry of the pump has been built shown Fig. 1(a). The flow
domain consists of a suction pipe, impeller and volute. The domain is created just based on the
dimensions of the solid body presented in the design drawings of the pump, and the variations in
dimension and geometrical shape due to casting process were not taken into account. It should be
noticed that the two spaces between the pump casing and both walls of the impeller shroud and hub
were neglected in the domains.
(a) (b)
Fig. 1 Flow domain of the model pump (a) and the relative position between a blade and the
volute tongue (b)
In the steady MRF method, the pump performances corresponding to a series of the relative
position between impeller blade and volute tongue need to be investigated. Because of the symmetry
of the impeller, even the four blades 1 through 4 illustrated in Fig. 1(b) pass through the tongue
sequentially, it is enough to conduct numerical computations of flow at a series of relative position
between blade 1 and the tongue. The relative position is specified by the circumferential angle θ
measured from the tip of the tongue to the terminal edge of the pressure surface of blade 1. In the
computations, θ =90o, 75o, 60o, 45o, 30o and 15o, respectively.
5
fields in the impeller and volute of the model pump have been investigated experimentally at both low
flow rate and the best efficiency points in [25, 26] by means of LDV when tap water and oil 2 were
handled by the pump. To compare the performances computed with those measured, these four liquids
were exactly applied in the numerical computations.
Table 1 Physical properties of liquids
Liquid Water Oil 1 Oil 2 Oil 3
6
pipe sliding mesh plane is generated at the impeller eye inlet, where is located at 35mm upstream the
datum plane. Likewise, an impeller-volute sliding mesh plane is specified as a cylindrical surface with
a diameter of 92.5mm.
The suction pipe fluid domain was descritized with hexahedron cell, but both the impeller and
volute flow domains were done with tetrahedron cell due to their complicated geometry. It was
confirmed that when the total number of cells is increased up to 910,000 (10,000cells in the suction
pipe, 210,000 cells in the impeller and 520,000 cells in the volute) the pump performance will be
independent on the cell-size. Unless otherwise stated, the computational results were under this
number of cells. The effect of cell size on the pump performance can be found out in the Appendix.
The under-relaxation factors of pressure, velocity, turbulence kinetic energy and its dissipation
rate are given to be 0.3, 0.5, 0.8 and 0.8 respectively. The convergent criterion for all residuals of
pressure, velocity, turbulence kinetic energy and its dissipation rate is chosen to be 1.5×10-4.
Consequently, the difference of mass flow rate between the pump inlet and outlet is reduced as small
as 8.02×10-5kg/s, so that the continuity equation is much fairly satisfied between flow domains.
V 1 y τw ρ
= ln +B (1)
τw ρ κ ν
where κ is Von Karman constant, κ =0.41; B the constant determined by experiments, B =5.5;
y the normal distance between a node of cell adjacent to the wall and the wall; ν the kinematical
viscosity of fluid; ρ the density of fluid; τw the shear tress on the wall. Since the fluid velocity V
at the point of distance y from the wall has been available, the determination of τw is fairly
7
straightforward.
For a rough wall, the roughness Reynolds number k s+ (= k s τw ρ ν ) is defined to indicate
the effects of roughness on the velocity profile and in turn the shear stress on the wall. The roughness
k s is known as the equivalent sand roughness and is linked to the arithmetic average of absolute
values of the actual roughness of the wall with particular finish by k s =6 Ra , where Ra =12.5 μm
-50 μm for a cast wall. If Ra =50 μm , then k s =300 μm . This sand equivalent roughness
estimate is comparable to 250 μm , a sand equivalent roughness of natural surface of cast iron
recommended in [30].
Based on the data of hydraulic experiments made by Nikuradse in 1933 with straight circular
cross-section pipes lined inside with uniform sand grains, Ioselevich and Pilipenko (1974) clarified
that: if k s+ ≤ 2.25 , the flow in the boundary layer is in a hydraulic smooth regime and the velocity
profile and shear stress on the wall is unaffected by roughness but by viscosity ; If 2.25 < k s+ ≤ 90 ,
the flow is a transition regime and the profile and stress are impacted by both roughness and viscosity;
if k s+ > 90 , then the flow is in a hydraulic rough regime and the flow is influenced by roughness
only. The velocity profile involved roughness effects is written as [30]
V 1 y τw ρ
= ln + B − ∆B (2)
τw ρ κ ν
where ∆B is a function in terms of roughness. As a result of the curve fitting to Nikuradse’s data,
Ioselevich and Pilipenko got an analytical expression of ∆B as follows:
0
k s+ ≤ 2.25
1 k s −2.25
[ )]
+
∆B = ln
κ 87 .75 +k s+ +
sin 0.4285 ln k s −0.811 ( 2.25 < k s+ ≤ 90 (3)
1 k s+ > 90
ln k s+
κ
In FLUENT, (3) is applied to involve the effect of surface roughness of a wall on the boundary
layer flow with an adoption by using a roughness coefficient C ks to indicate the uneven property of
rough elements of a practical surface finish and eventually is expressed with the following formula
0
k s+ ≤ 2.25
1 k s −2.25
[ )]
+
∆B = ln
κ 87 .75
+C ks k s+ +
(
sin 0.4285 ln k s −0.811 2.25 < k s+ ≤ 90 (4)
1 k s+ > 90
( +
ln 1 +C ks k s )
κ
where C ks =0.5-1.0 is usual choice. Since the shape and distribution of rough elements of cast
8
surface more resembled to those of uniform sand grains in hydraulic experiments available, it was to
let C ks =0.75 in the paper. It has been shown that a large C ks results into a low pump head
computed. As C ks is increased from 0.5 to 0.75, the estimated head at the best efficiency point is
dropped as low as 2% only. Therefore, C ks seems to have minor influence on the computational
results.
Because the treatment of surface roughness effect on boundary layer flows was conducted in
FLUENT by using existing data gained under ideally experimental circumstances on a fundamental
basis of fluid mechanics but nothing new, if a difference of computational flow parameters from
experiment occurs for a flow over rough walls, we shouldn’t be surprised, but consider the case to be
normal as well.
3 Results
3.1 Effect of Relative Position between Blade and Tongue
13
To illustrate the effect of the relative position between
0.9
impeller blade and volute tongue on the
pump 12hydraulic performance, the numerical computations upper
of margin
flow have been carried out at six
upper margin 0.8
positions of blade 1 related to volute tongue, i.e. θ =90o, 75o, 60o, 45o, 30o and 15o. The working fluid
11 mean
mean 0.7
µm
is water and the surface roughness of both impeller and volute is Ra =50lower margin
. Fig. 2 indicates the
impeller theoretical head H i and hydraulic efficiency ηhi , pump head H and hydraulic efficiency
10 0.6
H (m)
ηhi
o o
data1
θ=90 data1
θ=90
i
ηh in9terms of theθ=75
flow
o
data2 lower Q at various θ . The impeller
ratemargin
0.5
theoretical head H
θ=75 is defined as the
data2
i
o
o
o
data3 θ=60 data3 θ=60
o
o
data4
θ=45 data4
θ=45
mass-averaged
8 total energy head rise of fluid from the impeller
o
data5
θ=30
0.4
inlet to outlet. The impeller hydraulic o
data5
θ=30
o
o
data6
θ=15 data6
θ=15
efficiency is defined
data7as the ratio of the impeller theoretical
0.3 head over an ideal
data7
ν=1cSt, theoretical head
Ra=50µm
7 ν=1cSt, Ra=50µm data8
data8
0.2 head rise H i . This ideal head is specified
corresponding to the consumed shaft-power for developing
data9 data9
6
by the following
0 2 expression
4 6 8 10 12 0 2 4 6 8 10 12
Q(L/s) Q (L/s)
ηV M nω
H th = (5)
11 ρgQ
where ηV is the pump volumetric efficiency, ηV =0.865 is given by [24-26]; M n the total torque
0.8
10
upper margin
upper margin
applied to liquid pumped by the impeller blades and the0.7inside shroud and hub, which are calculated
9
based on CFD results; ω the impeller
mean
rotational angular
0.6
mean ω =πn 30
speed, ; Q the pump flow
rate; g8 the acceleration of gravity. The pump head H is defined as the total energy rise of fluid
lower margin
H(m)
ηh
0.5
from the entry ofθ=90
theo suction
data1 pipe to the volute nozzle exit. Likewise, the pumpθ=90
lower margin hydraulic
o
data1 efficiency
7 data2o data2o
θ=75 θ=75
ηh is the ratio of the pump head over the ideal theoretical
data3
θ=60
data4
o
o
0.4 head H
th .
data3
θ=60
o
θ=45 data4
θ=45
o
6 data5o data5o
θ=30 θ=30
data6o 0.3 data6
θ=15
o
θ=15
data7 data7
5 ν=1cSt, Ra=50µm ν=1cSt, Ra=50µm
data8 data8
data9 0.2 data9
4
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Q(L/s) Q (L/s)
10
following sections.
Figure 3 demonstrates the impeller theoretical head H i , hydraulic efficiency ηhi , hydraulic
loss coefficient ξi and slip factor σ in terms of flow rate Q at the kinematical viscosities of 1cSt,
24cSt, 48ct and 60cSt and with wetted surface roughness of Ra =50 μm . As reference, the
theoretical heads of one-dimensional inviscid fluid flow derived from both Eulerian equation of
turbomachinery and the slip factors of Wiesner and Stodola are also plotted in that figure. Wiesner and
Stodola slip factors are depicted with the following formulae
π
Z sin β 2b Stodola
σ = (6)
sin β 2b Wiesner
Z 0.7
Accordingly the impeller theoretical head of one-dimensional inviscid fluid flow is written by
u2 Q
H i1D = (1 − σ ) u 2 − (7)
g ηV A2ψ 2 tan β 2b
where u 2 is the impeller tip speed, u 2 = ωD2 2 , D2 the impeller exit diameter; ηV the
volumetric efficiency, ηV =0.869 based on the experiments in [24, 25]; A2 the impeller exit area,
A2 = πD2 b2 , b2 the blade exit width; ψ2 the blade blockage coefficient at exit, ψ2 =1-
ZS u 2 πD2 , S u 2 the tangential thickness of blade at exit, S u 2 =10mm. The slip factor σ is
defined as the ratio of the slip velocity of fluid at the impeller exit ∆Vu 2 over the impeller tip speed
∆Vu 2
σ= (8)
u2
Even the slip factors can be estimated by comparing the velocity-triangles at the impeller outlet
constructed with CFD outcomes with those of assumed one-dimensional flow of inviscid fluids, these
factors don’t seem to be linked directly to the impeller theoretical head of viscous fluids because at
the impeller outlet the variable u 2Vu 2 g in Euler equation no longer represents the total energy of
viscous fluids. Instead, in the paper, the slip factors will be evaluated by using the impeller theoretical
head of viscous fluids obtained from CFD computations.
Substituting the theoretical head of viscous fluids H i given by CFD computations into (7), the
slip factors corresponding to these heads of viscous fluids are available via the following equation
gH Q
σ = 1 − 2 i +
(9)
u2 ηV u 2 A2ψ 2 tan β 2b
11
The slip factors calculated just in this way are illustrated in Fig. 3. Note that the impeller
hydraulic loss coefficient ξi showed in that figure is defined as the ratio of the impeller hydraulic
loss ∆hi over the impeller velocity head based on the impeller tip speed u 2 , i.e. ξi = g∆hi u 22 .
It is clearly that tremendous difference of the impeller theoretical head between 3D viscous fluid
and 1D inviscid fluid emerges. The impeller theoretical head of viscous fluid no longer increases
continuously with decreasing flow rate, but a peak occurs at a flow rate of 4L/s (around 57% flow rate
of the best efficiency point) instead, causing stall operating conditions at less flow rate than 4L/s
which are frequently present in centrifugal compressors at part-load points. The 3D viscous flow
theoretical head no longer increased monotonously with decreasing flow rate agrees well with the
experiments in [31, 32].
The impeller theoretical head of 1D inviscid fluid based on Stodola slip factor is in good
agreement with those of viscous fluid by CFD, specially nearly at Q =6L/s. The head based on
Wiesner factor is substantially over-predicted. It is implied that Stodola slip factor is more suitable to
the impellers with less number of blades, small blade discharge angle, long and strongly back-curved
blade than Wiesner slip factor.
It should be noticed that since the impeller theoretical head of 3D viscous fluid flow significantly
deviates from the linear theoretical head of 1D inviscid fluid flow except at a flow rate around Q
=6L/s, the slip factors based on the theoretical heads of 3D viscous fluid no longer are a flow rate-
independent constant, instead are varied with flow rate, subsequently, at both high and low flow rates,
larger slip factors can be found.
Both the impeller theoretical head and slip factor of 3D viscous fluid flow are dependent on
liquid viscosity. Except at a viscosity of ν =24cSt, the theoretical head of viscous fluid is degraded
and the corresponding slip factors are arisen with increasing liquid viscosity. At ν =24cSt, however,
the situation is reverse. In this case, as a flow rate is more than 4L/s, the theoretical head of viscous
fluid is higher than that of water (ν =1cSt), accordingly the slip factor is less than that of water.
The impeller hydraulic efficiency ηhi has a peak between 75% and 90% at a flow rate of Q
=6L/s-7L/s, in turn the impeller hydraulic loss coefficient ξi has the lowest values at this flow rate
range. The hydraulic loss coefficient is increased rapidly towards low flow rate due to severely
reverse flow developed in the impeller. For example, the loss coefficient at Q =1L/s is as high as 64
times at Q =7L/s. Further, as the flow rate more than 7L/s, the hydraulic loss coefficient is increased
but grows slowly with flow rate because of the significant contribution of skin friction loss.
The impeller hydraulic efficiency and the hydraulic loss coefficient rely on liquid viscosity as
well. In general, the hydraulic efficiency is decreased but the loss coefficient is increased with
increasing liquid viscosity. However, the impeller performance is improved at ν =24cSt as the flow
rate more than 5L/s since the theoretical head and hydraulic efficiency of oil1 are higher than those of
12
water (ν =1cSt) and the loss coefficient is lower than that of water ( ν =1cSt).
0.5
data1
0.45 data2
data3
1cSt
24cSt
data4
48cSt
data5
0.4 60cSt
data6
16 0.35
σ
0.3
Stodola
0.25
Wiesner
15
0.2
0 2 4 6 8 10 12
Q(L/s)
1
10
0.9
0.8 data1
1cSt
24cSt
data2
48Cst
data3
0.7 0
60cSt
10 data4
0.6
14
ηhi
ξi
0.5 data1
1cSt
24cSt
data2 -1
0.4 48cSt
data3 10
60cSt
data4
0.3
0.2
-2
10
13
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Q (L/s) Q (L/s)
Fig. 3 Impeller theoretical head, slip factor, hydraulic efficiency and loss coefficient in terms of
flow rate at θ =45o and Ra =50 μm
1D, Stodola
12
3.3 Effect of Viscosity on Pump Performance
volute hydraulic loss coefficient ξV and pump total hydraulic loss coefficient ξV + ξi in terms of
(m)
11 13
i
flow rate at various viscosities of 1cSt, 24cSt, 48cSt and 60cSt and with a surface roughness of 50
µm . The volute hydraulic loss coefficient ξV is defined as the ratio of the hydraulic loss ∆hV
inside the volute over the velocity head based on the impeller tip speed, i.e. ξV = g∆hV u 22 . The
computational
13 pump head isn’t completely consistent with the experimental performance, causing a
0.7
maximum
12 relative error of 11%, since the slope of the latter is sharper than the former. The
Exp
mechanism
11 behind this phenomenon perhaps is a disagreement
0.6
between practical model
CFD
and design
drawings
10 in the dimensions and/or shape of volute, especially
0.5 the area of the volute throat.
9 In addition, the pump volumetric and mechanical efficiencies were roughly estimated based on
1cSt data1 0.4
the 8experimental
data2data in [24, 25], then the pump efficiency was calculated by them and the pump
H(m)
24cSt
η
48cSt
data3 CFD
hydraulic
7
efficiency
60cSt
data4
given by CFD. Although the predicted
0.3 pump efficiency is very comparable to
data1
1cSt
data2 EXP
24cSt
at ν =1cSt, 48cSt and 60cSt, a significant deviation from the experiment
1cSt data5
those
6 measured
24cSt
data6
48cSt
data3
60cSt
is found at
0.2 data4
48cSt
ν =24cSt
5 where
data7
ν
1cSt
data8the experimental pump efficiency is less than that of water ( 24cSt
60cSt
data5
data6=1cSt). The reason
48cSt
data7
for 4this is perhaps that the less correct disk loss is involved
0.1
in the CFD estimate of pump efficiency.
60cSt
data8
The3 flow computations tending to include this loss should0 be worth being conducted in forthcoming
0 2 4 6 8 10 12 0 2 4 6 8 10 12
investigations. Q(L/s) Q (L/s)
The pump hydraulic efficiency is below 75% in all cases. It is the volute that makes the pump
hydraulic
0.8 efficiency be reduced by at least 15%. In addition, the higher the flow rate, the larger the
reduction. It is more likely that the volute affects the hydraulic
-1 losses more considerable than the
0.7
10
impeller does, specially at the high flow rate for this pump.
0.6 Unlike the impeller hydraulic loss coefficient, the volute hydraulic loss coefficient is raised with
increasing flow rate. For instance, as the flow rate is increased from 1L/s to 11L/s, the hydraulic loss
ξV
ηh
0.5
coefficient is increased by about 3.5 times. -2
10 data1
1cSt
0.4
The pump total hydraulic loss coefficient is minimum at around Q =6L/s. As
data1
1cSt
24cSt
data2
the flow rate
data2
24cSt
48cSt
data3 48cSt
data3
becomes
0.3
low, the total loss coefficient gets significantly large, whereas it rises moderately
60cSt
data4 60cSt
data4 with
increasing flow rate. Comparing the impeller hydraulic loss coefficient with that of the volute, it is
0.2
discovered that the impeller hydraulic loss coefficient is dominated
-3 at low flow rate, at high flow rate,
10
0 2 4 6 8 10 12 0 2 4 6 8 10 12
however, the volute loss coefficient is remarked.
Q (L/s) Q(L/s)
1
Fig. 4 Pump head and hydraulic efficiency, volute
10
hydraulic loss coefficient and pump total hydraulic
loss coefficient in terms of flow rate, the symbols
indicate the experimental head and efficiency of the
data1
1cSt
24cSt
data2
pump
48cSt
data3
0
10 60cSt
data4
ξ i +ξ V
-1
10
10
-2
14
0 2 4 6 8 10 12
Q (L/s)
As shown in Fig. 4, even both the pump computed head and hydraulic efficiency are deteriorated
gradually as increasing liquid viscosity, at ν =24cSt (oil 1), they are improved compared with those
at ν =1cSt (water). The phenomenon, regarding that the pump performance is improved at a
particular high viscosity of liquid compared to pumping water, is named as “sudden-rising head
15
effect”. It has been observed that the pump head is improved at ν =24cSt-60cSt (oil1, oil 2 and oil3)
in the experiments of [24] and indicated the sudden-rising-effect-head is onset at this viscosity.
Obviously, this phenomenon is well predicted by the numerical computations. Therefore, this effect
has been confirmed to exist theoretically and experimentally. Since the effects of flow turbulence and
surface roughness on the boundary layer flows in the pump are estimated approximately, the sudden-
rising-head effect has to merge just at ν =24cSt in the numerical computations. Nevertheless, it is
hopeful that the analytical prediction may be improved with the development in both turbulence
model and boundary layer flow investigations.
For an absolutely smooth surface ( Ra =0 μm ), the computed pump head is decreased
continuously with increasing liquid viscosity, subsequently, the sudden-rising-head effect is excluded.
For a roughness of Ra =100 μm , however, the sudden-rising-head effect occurs once more (Fig.5).
It is believed that the surface roughness of flow channels plays a critical role in the occurrence of
sudden-rising-head effect.
11 11
10 10
9 9
8 8
H(m)
H(m)
7 data1
1cSt 7
24cSt
data2 data1
1cSt
48cSt
data3 24cSt
data2
6 60cSt
data4
6
48cSt
data3
60cSt
data4
5 5
4 4
3 3
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Q(L/s) Q(L/s)
Fig. 5 Pump head in terms of flow rate at surface roughness of 0 μm and 100 μm
Figure 6 shows the averaged skin friction factors applied on the liquid pumped by the wetted
surfaces of the impeller and volute in terms of flow rate. The impeller mean skin friction factor C fi
is defined as the ratio of the averaged shear stress τ wi applied by the impeller wetted surfaces on the
liquid over the velocity head ρu 22 2 , that is C fi = 2τ wi ρu 22 . Likewise, the volute mean skin
friction factor C fV is the ratio of the averaged shear stress τ wV applied on the liquid by the volute
over the velocity head ρu 22 2 , C fV = 2τ wV ρu 22 . Both mane skin friction factors include the
16
effects of liquid viscosity and surface roughness on the skin friction loss in the flow passages of
impeller and volute. For the same passage, the larger the mean skin friction factor, the more the skin
friction loss.
In Fig. 6, even though the magnitudes of C fi and C fV are comparable, their relations with the
flow rate are quite different. For C fi , it has a minimum value at Q =4L/s, and then is increased
rapidly with increasing flow rate, C fV , however, grows slowly but continuously from low to high
flow rate. The flow velocity in the impeller channels may be increased by increasing blockage due to
the reserve flow onset at much low flow rate. It should be unsurprised to notice that an increasing
C fi can be seen at a flow rate less than 4L/s. Besides the vortex and shock losses, the increased
impeller skin friction loss is likely responsible for the huge hydraulic loss coefficient of impeller at
low flow rate shown in Fig. 3.
-3 -3
x 10 x 10
7 7
6 data1 6
1cSt data1
1cSt
data2
24cSt 24cSt
data2
48cSt
data3 48cSt
data3
5 60cSt
data4 5 60cSt
data4
C fV
4 4
fi
C
3 3
2 2
1 1
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Q(L/s) Q(L/s)
Fig. 6 Mean skin friction factors of both impeller an volute in terms of flow rate at various liquid
viscosities and Ra =50 μm
The mean skin friction factors are raised with increasing liquid viscosity, at ν =24cSt, however,
both C fi and C fV start to be lower than those at ν =1cSt (water) respectively from Q =4.5L/s
and Q =6.5L/s. The lowered mean skin friction factors correspond to the decreased skin friction loss
in the pump, accordingly the pump performance shows a higher head developed and increased
hydraulic efficiency. It should be concluded that the deceased skin friction loss in both impeller and
volute is responsible for the sudden-rising-head effect.
Based on the experimental outcomes of turbulent flow of the boundary layer over flat rough
plates [30], in the hydraulic rough regime, the skin friction factor is related to the surface roughness
17
only. As the liquid viscosity increasing, the Reynolds number is decreased, causing the flow regime is
the hydraulic transition zone, where the skin friction factor not only depends on both Reynolds
number and surface roughness but also is decreased with decreasing Reynolds number. As Reynolds
number is decreased further, the flow regime is in the hydraulic smooth zone, in which the factor is
just dependent on Reynolds number but is raised as decreasing Reynolds number. It can be imagined
that the sudden-rising-head effect should take place in the transition zone rather than in the others.
Two critical parameters to decide the sudden-rising-head effect onset are the high enough surface
roughness and properly decreased Reynolds number. If the surface roughness is too small or the
Reynolds number is decreased too much, causing the flow regime is in the hydraulic smooth zone, the
sudden-rising-head effect will unlikely happen. Otherwise, if the surface roughness is too high and
Reynolds number is too large, the effect still won’t occur for the flow regime is in the hydraulic rough
zone.
In fact, the sudden-rising head effect was also found with experiments in [33]on the centrifugal
oil pump with a specific speed of 807 (USA) or 57 (China) and confirmed with the estimation of the
friction loss of water and viscous oils flow in the impeller and volute by using the routine boundary
layer theory. At the moment, this effect also exists in the experimental performance of the pump with
the specific speed of 93 and is confirmed with CFD results once again. Therefore, the sudden-rising
head effect may be an essential characteristic of a low specific speed centrifugal pump handling
highly viscous oils.
It should be noticed that even though the sudden-rising head effect occurs, the pump gross
efficiency still is gradually decreased due to the possible increased impeller disk friction loss by the
increasing liquid viscosity [24, 25]. The clear mechanism behind it obviously needs to be debuted
further experimentally and numerically.
flow rate. Contrarily, the velocity head H iV has a peak at around Q =4L/s, then a dramatic decrease
in H iV can be found at a flow rate less than 4L/s. Therefore, it is believed this too low velocity head
at a flow rate less than 4L/s should contribute the peak onset on the theoretical head curves.
18
10 4
steady
9 3.5
8 3
H (m)
H is (m)
7 2.5
data1
1cSt data1
iV
1cSt
24cSt
data2 data2
24cSt
48cSt steady
data3 48cSt steady
data3
6 60cSt
data4 2 60cSt
data4
data5
1cSt data5
1cSt
24cSt unsteady
data6 data6
24cSt
48cSt
5 data7 1.5 48cSt unsteady
data7 unsteady
60cSt
data8 60cSt
data8
4 1
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Q (L /s ) Q (L /s)
Fig. 7 Impeller static pressure head and velocity
0 .7 head rises as well as dimensionless tangential
component of absolute velocity of liquid at
impeller exit in terms of flow rate at various liquid
viscosities and Ra =50 μm
0 .6
steady
V u2 /u 2
0 .5
data1
1cSt
24cSt
data2
steady
48cSt
data3
60cSt
data4
0 .4 data5
1cSt
24cSt unsteady
3
data6
3
48cSt
data7
60cSt
data8
0 .3
0 2 4 6 8 10 12
Q (L /s )
unsteady
Q =6L/s Q =3L/s
Fig. 8 Relative velocity vector and static pressure contour over the impeller blades and hub at a low
flow rate ( Q =3L/s) and the best efficiency point ( Q =6L/s) for water (ν =1cSt) in the steady and
unsteady cases
19
To expose the reason for extreme low velocity head occurrence at a low flow rate, the computed
tangential components of absolute velocity of liquid at the impeller discharge have been illustrated in
Fig. 7 too. The tangential component of velocity has been made dimensionless with the impeller tip
20
speed u 2 . It is clearly that a dramatically decreased tangential velocity is caused. It should be noticed
that the LDV measurements also revealed that the decreased velocity happens at the impeller exit with
decreasing flow rate in [26]. The reduction of tangential velocity implies that a volute can apply much
1D
3D
W2
V2
v2 w2
Vr2 vr2
β2b
Vu2 u2
21
Fig. 9 Velocity triangle at impeller discharge for estimate slip factor
For a uniform 1D ideal flow, the tangential component of the relative velocity wu 2 is given by
the following formula
vr 2 Q
wu 2 = = (10)
tan β 2b ηV A2ψ 2 tan β 2b
Based on CFD computational results, the velocity components Vu 2 and Vr 2 are available, the
tangential component of the relative velocity of 3D viscous fluid flow is
Wu 2 = u 2 −Vu 2 (11)
With the slip factor definition (8), the slip factor extracted from the velocity triangles is
expressed by
Vu 2 Q
σ =1 − − (12)
u 2 η V A2ψ 2 tan β 2b u 2
The computed slip factors based on both unsteady and steady flows have been shown in Fig. 10
for the impeller with volute at various viscosities and a surface roughness of Ra =50 μm . The slip
factors are reduced with increasing flow rate. The similar characteristics of slip factor versus flow rate
were revealed by experiments in [36, 37]. Moreover, the slip factors extracted from velocity triangles
considerable differs from those based on the impeller theoretical head by CFD computation (see Fig.
3).
As the flow rate is less than 5L/s, the slip factors are rapidly increased as high as 0.6. Otherwise,
the slip factors asymptotically approach the Wiesner’s slip factor. The slip factors due to the unsteady
flow model much more depend on the liquid viscosity than those due to the steady model. Except at
ν =24cSt, the slip factor are reduced with increasing liquid viscosity, causing an good agreement
with that was found experimentally in [38]. At ν =24cSt, although the slip factor is the largest, the
highest static pressure rise is developed across the impeller (Fig. 7). Since the static pressure rise is
0.65
dominated compared to the velocity head rise in the impeller, an increased slip factor doesn’t mean a
0.6
lowered total head developed by the impeller. data1
1cSt
24cSt
data2
steady
0.55 48cSt
data3
60cSt
data4
0.5 1cSt
data5
24cSt
data6
unsteady
0.45 48cSt
data7
60cSt
data8
0.4 data9
data10
0.35
0.3
Stodola
0.25
Wiesner
0.2
0 2 4 6 8 10 12
ν =60c
St
Fig. 11 Absolute velocity (left) and static pressure (right) contours in the plane through the impeller
23 at ν =1cSt , 24cSt, 48cSt and 60cSt, the
blade mid-span at the best efficiency point ( Q =6L/s)
velocity is colored by velocity magnitude (m/s), and the static pressure is colored by liquid column
height (m)
3.7 Effect of Surface Roughness
Figure 12 illustrates the pump head and hydraulic efficiency in terms of flow rate at various
viscosities and Ra=0 μm , 50 μm and 100 μm . For Ra=0 μm , the strongest surface roughness
effect can be found. Otherwise, as the liquid viscosity increasing, the roughness effect is suppressed
24
substantially. Moreover, once the surface roughness Ra > 50 μm , the effect on the performance
becomes less dominated. The experimental data in [39, 40] seem to show the similar effect.
11 0.9
10 0.8
9 0.7
8
0.6
H (m)
data1
1cSt
data2 1cSt
data1
ηh
24cSt 0.5
7 data3 Ra=0 µm 24cSt
data2
48cSt Ra=0 µm
60cSt
data4 48cSt
data3
data5 60cSt
1cSt data4
24cSt
data6 0.4
6 data5
1cSt
48cSt
data7 Ra=50 µm
data6
24cSt
60cSt
data8 Ra=50 µm
48cSt
data7
data9
1cSt 0.3 60cSt
data8
5 24cSt
data10
Ra=100 µm 1cSt
data9
48cSt
data11
60cSt
24cSt
data10
data12 Ra=100 µm
0.2 48cSt
data11
4 60cSt
data12
0 2 4 6 8 10 12 0.1
0 2 4 6 8 10 12
Q(L/s)
Q (L/s)
Fig. 12 Pump head and hydraulic efficiency in terms of flow rate for surface roughness of 0 μm , 50 μm
and 100 μm and at various liquid viscosities
data1
1cSt data1
1cSt
24cSt
data2 24cSt
data2
48cSt
data3 48cSt
data3
60cSt
data4 0.8 60cSt
10 data4
H (m)
ηh
0.7
9
0.6
0.5
0 20 40 60 80 100 0 20 40 60 80 100
R a (µm ) Ra (µm)
25
Fig. 13 Pump head and hydraulic efficiency in terms of roughness Ra at Q =6L for four viscosity
values
Table 2 shows the admissible roughness Ra* of both impeller and volute in terms of Ra at the
best efficiency point Q =6L/s for various viscosities based on Schlichting formula [39]. Since the
mean relative velocity of flow in the impeller is close to that in the volute, the admissible roughness
Ra* of the impeller is comparable to the volute. The admissible roughness is increased with
increasing liquid viscosity. Fig. 13 illustrates the pump head and hydraulic efficiency in terms of
surface roughness Ra at Q =6L for four viscosities. At ν =1cSt, the performance is degraded until
Ra less than 40 μm . Beyond this roughness the performance is slightly affected. At ν =24cSt, the
performance is improved as Ra >25 μm . Based on Table2, the admissible roughness Ra* of both
impeller and volute is around 100 μm at ν =24cSt. Consequently, when Ra is between 25 μm and
100 μm , the flow regime of boundary layer in the impeller and volute should be in the transition
zone where less skin friction loss is expected. As the viscosity more than 48cSt, the performance is
basically unaffected by the surface roughness because the rough is much less than Ra* and the
corresponding flow regime of boundary layer in the impeller and volute should be in hydraulic
smooth zone.
If a pump is used to transport highly viscous liquids, it may be acceptable that the pump flow
channels maintain a relative large roughness. Otherwise, the surface roughness should be kept as
small as possible.
4 Discussions
4.1 Flow Regime in Section Pipe
In the computations, the Reynolds number at the suction pipe is varied in a range of 3.9 × 102-3.3
× 10 5
from water to oil 3 and from a high flow rate to a low one. The Reynolds number at around 14
operating points is less than 2300, and the flow in the suction piper should be laminar, subsequently,
the flow in the impeller would be either laminar or turbulent at these conditions. Due to lack of
information of flow transition in impellers, even the Reynolds number is so low, the flow in the pump
26
still is assumed to be turbulent in the paper.
28
equations with an aid of the standard turbulence k − ε model combined with the non-equilibrium
wall function is applicable to handle highly viscous liquid effect on the performance, more accurate
results are still expected to be present in future.
Reference
[1] Lee S C and Chen D, 1993, Off-design Performance prediction for radial-flow impellers,
International Journal of Turbo & Jet Engines, 10, 45-60
[2] Zhou W D, Zhao Z M, Lee T S and Winoto S H, 2003, Investigation of flow through centrifugal
pump impellers using computational fluid dynamics, International Journal of Rotating
Machinery, 9(1), 49-61
[3] Denus K and Osborne C, 2002, Hydraulic development of a centrifugal pump impeller using the
agile turbomachinery design system, Task Quarterly, 6(1), 79-100
[4] Byskov R K and Pederson N, 2003, Flow in a centrifugal pump impeller at design and off-design
conditions-Part II: large eddy simulations, ASME Journal of Fluids Engineering, 125, 73-83
[5] Anagnostopoulos J S, 2006, Numerical calculation of the flow in a centrifugal pump impeller
using Cartesian grid, Proceedings of the 2nd International Conference on Applied and
Theoretical Mechanics, 20-22 November 2006, Venice, Italy, 124-129
[6] Guleren K M and Pinarbasi A, 2004, Numerical simulation of the stalled flow within a vaned
centrifugal pump, Proc Instn Mech Engrs-Part C: Journal of Mechanical Engineering Science,
218, 425-435
[7] Huang S, Islam M F and Liu P F, 2006, Numerical simulation of 3d turbulent flow through an
entire stage in a multistage centrifugal pump, International Journal of Computational Fluid
Dynamics, 20(5), 309-314
[8] Majidi K and Siekmann H E, 2000, Numerical of secondary flow in pump volute and circular
casings using 3D viscous flow techniques, International Journal of Rotating Machinery, 6(4),
245-252
[9] Eduardo B M, Joaquim F F, Jorge L P G and Carlos S M, 2000, Numerical simulation of
centrifugal pumps, Proceedings of ASME 2000 Fluids Engineering Division Summer Meeting,
June 11-15, 2000, Boston, USA, 1-5
[10] Tamm A, Ludwig G and Stoffel B, 2001, Numerical experimental and theoretical analysis of
efficiencies of the industrial a centrifugal pump, Proceedings of ASME 2001 Fluids Engineering
Division Summer Meeting, May 29-June 1, 2001, Louisiana, USA, 1-7
[11] Dick E, Vierendeels J, Serbrugyns A and Voorde J V, 2001, Performance prediction of centrifugal
pumps with CFD-tools, Task Quarterly, 5(4), 579-594
[12] Oh H W, Yoon E S, Park M R, Sun K and Hwang C M, 2005, Hydrodynamic design and
performance analysis of a centrifugal blood pump for cardiopulmonary circulation, Proc Instn
Mech Engrs-Part A: Journal of Power and Energy, 219, 525-532
29
[13] Asuaje M, Bakir F, Kergourlay G, Noguera R and Rey R, 2006, Three-dimensional quasi-
unsteady flow simulation in a centrifugal pump: comparison with experimental results, Proc Instn
Mech Engrs-Part A: Journal of Power and Energy, 220, 239-255
[14] Kergourlay G, Younsi M, Bakir F and Rey R, 2007, Influence of splitter blades on the flow field
of in a centrifugal pump: test-analysis comparison, International Journal of Rotating machinery,
Volume 2007, 1-13
[15] Cheah K W, Lee T S, Winoto S H and Zhao Z M, 2007, Numerical flow simulation in a
centrifugal pump at design and off-design conditions, International Journal of Rotating
machinery, Volume 2007, 1-8
[16] Ippen A T, 1946, The influence of viscosity on centrifugal-pump performance, Transaction of
ASME, 68, 823-848
[17] Kamimoto G, Matsuoka Y and Shirai H, 1959, On the flow in the impeller of centrifugal type
hydraulic machinery, Transaction of JSME, 25(153), 383-388
[18] Itaya S and Nishikawa T, 1960, Studies on the volute pump handling viscous fluids, Transaction
of JSME, 26(162), 202-209
[19] Acosta A J and Hollander A, 1965, Observations on the performance of centrifugal pumps at low
Reynolds numbers, Proceedings of the Symposium on Pump Design, Testing and Operation,
12th-14th 1965, Glasgow, UK, 379-393
[20] Tanaka K and Ohashi H, 1984, Performance of centrifugal pumps at low Reynolds number,
Transaction of JSME, Series B, 50(449), 279-285
[21] Aoki K, Yamamoto T, Ohta H and Nakayama Y, 1985, Study on centrifugal pump for high
viscosity liquids, Transaction of JSME, Series B, 51(468), 2753-2758
[22] Ohta H, Aoki K and Yamamoto T, 1985, Study on centrifugal pump for high viscosity liquids
(effect of clearance ratio), Transaction of JSME, Series B, 51(472), 4295-4300
[23] Ohta H and Aoki K, 1990, Study on centrifugal pump for high viscosity liquids (effect of
impeller blade number on the pump performance), Transaction of JSME, Series B, 56(526),
1702-1707
[24]Li W G and Hu Z M, 1996, Experiments on the performance of a centrifugal oil pump, Chinese
Fluid Machinery, 25(2), 3-8
[25] Li W G, 2000, Effects of viscosity on the performance of a centrifugal oil pump and the flow
pattern in the impeller, International Journal of Heat and Fluid Flow, 21(2)), 271-275
[26] Li W G, 2002, Flow of viscous oil in the volute of centrifugal pump, Journal of Thermal Science,
11(1), 10-15
[27] Steck E and Flelsch K O, 1989, Calculating 3d laminar flow through centrifugal pumps,
Proceedings of the 6th International Conference on Numerical Methods in Laminar and
Turbulent Flow, 11th-15th July, 1989, Swansea, UK, 1321-1331
[28] Li W G, 1997, Numerical computations of 3d turbulent flow in a centrifugal pump impeller,
30
Chinese Pump Technology, (5), 20-29
[29] Day W, Lemire P P, Flack R D and McDaniel J C, 2003, Effect of Reynolds number on
performance of a small centrifugal pump, Proceedings of the 4th ASME-JSME Joint Fluids
Engineering Conference, Honolulu, Hawaii, USA, July 6-11, 2003
[30] Cebeci T and Bradshaw P, 1977, Momentum transfer in boundary layers, Washington:
Hemisphere Publishing Corporation, 179-191
[31] Peck J F, 1951, Investigation concerning flow conditions in a centrifugal pump and effect of
blade loading on head slip, Proceedings of Institution of Mechanical Engineers, 164, 1-30
[32] Bowerman R D and Acosta A J, 1957, Effect of the volute on performance of a centrifugal-pump
impeller, Transaction of ASME, 79, 1057-1069
[33] Li W G, 2000, The sudden-rising head effect in centrifugal oil pumps, World Pumps, (409), 34-
36
[34] Sakai T and Watanabe I, 1967, The slip factor of centrifugal and mixed-flow compressors,
Transaction of JSME, 33(249), 735-744
[35] Caridad J A and Kenyery F, 2005, Slip factor for centrifugal impellers under single and two-
phase flow conditions, ASME Journal of Fluids Engineering, 127, 317-321
[36] Kasai T, 1935, On the exit velocity and slip coefficient at the outlet of centrifugal impeller with
some references to the characteristics of pump, Transaction of JSME, 1(3), 234-246
[37] Kasai T, 1934, On the flow conditions at the exit of centrifugal impeller, Journal of JSME,
37(202), 60-69
[38] Ohta H, 1999, Effect of Reynolds number on slip factor of centrifugal pump for high viscosity
liquids, Transaction of JSME, Series B, 65(639), 3697-3704
[39] Shirakura M, 1959, Effects of surface roughness of impeller channel on volute-pump
performance, Journal of JSME, 62(485), 890-899
[40] Ida T, 1965, The effects of impeller vane roughness and thickness on the characteristics of the
mixed-flow propeller pump, Bulletin of JSME, 8(32), 634-643
31
Appendix Effect of Number of Cells on Performance
To check the effect of number of cells in flow domains on the numerical results, the meshes with
respective 540,000 cells, 910,000 cells and 1,290,000 cells have bee generated by using GAMBIT at a
relative position between impeller blade and volute tongue ( θ =90o), then the numerical
computations of water steady turbulent flow in the model pump were launched with the standard
k −ε turbulence model and the non-equilibrium wall function for each mesh. The wetted surfaces of
the pump is subject to a roughness Ra =50 µ m.
The pump performance curves as a function of flow rate have been illustrated in Fig. 14. The
pump head seems to be independent on all the mesh sizes, but the hydraulic efficiency starts to be
independent on the mesh size when the number of cells is increased up to 910,000. This mesh size is
used in all numerical computations in this paper.
0.8
0.7
0.6
11 0.5
η
0.4 data1
540000
910000
data2
1290000
data3
0.3
0.2
0 2 4 6 8 10 12
Q (L/s)
Fig. 14 Pump head and hydraulic efficiency in terms of flow rate under three different numbers of cells
in the flow domains
10
9 32