Professional Documents
Culture Documents
d
London SW7 2AZ, UK mark.wilson@rolls-royce.com London SW7 2AZ, UK
klee2@ic.ac.uk m.vahdati@imperial.ac.uk
ite
ed
ABSTRACT 600k points per passage using steady Reynolds-averaged Navier-
The original Spalart-Allmaras (SA) model is known to predict Stokes (RANS) simulations.
py
premature stall when applied to fan or compressor, which is in The Spalart-Allmaras (SA) model [1] is a widely used
line with the observation of other researchers who use the SA turbulence model for industrial flow computations and it has
model. Therefore, to improve the prediction of the stall been proven reliable in turbomachinery applications due to its
Co
boundary, the original SA model was modified by scaling the simplicity, robustness and efficiency [2-7]. However, it is well
source term based on the local pressure gradient and the velocity known that the original SA model predicts a larger separation
helicity of the flow. Furthermore, a generalized wall function zone, which leads to the blockage of passages, limiting the static
ot
valid for non-zero wall pressure gradient was implemented to pressure rise, a considerable total pressure loss and premature
improve the accuracy of boundary conditions at the solid wall. stall [5, 6]. Some studies have been made on this subject. As an
This work aims to produce a turbulence model which can be used example, Liu et al. [5] proposed a modified SA model which uses
tN
to model flows near the stall boundary for the transonic fan rotors helicity to model the energy backscatter, which can be substantial
on relatively coarse grids of around 600k points per passage. in the region of corner separation in the compressors. They
Initially, two fan rotors with different design and operating showed that by using the modified SA turbulence model, it is
rip
speeds were used to optimize the new parameters in the modified possible to suppress the corner separation in a compressor
turbulence model. The optimization was based on improving the passage which appeared with the original SA model but was not
correlation between measured and numerical radial profiles of present in the experiment. The numerical results showed a good
sc
the pressure ratio. Thereafter, steady computations were agreement with the measured data in terms of the size of
performed for two other fans (by using the same parameters) and separation and the distribution of the surface static pressure and
nu
the predictions were compared with the experimental data for all turbulent viscosity. Li et al. [6] improved the accuracy of the SA
the four fan rotors. Numerical results showed a significant model in shock-wave/boundary layer interaction flow by
improvement over those obtained with the original SA model, including the effect of the adverse pressure gradient in the
Ma
when compared against the measured data. Finally, for turbulence model. Their results indicated that by modelling the
completeness it was decided to test the performance of the effects of the pressure gradient, the production of turbulent
modified model by comparing the result with measured data for viscosity becomes larger in the shock and separation regions, for
a simple canonical case. both impinging shock-wave/turbulent boundary layer interaction
ed
dynamic (CFD) model which can be used for predictions of pressure and skin friction. Oriji and Tucker [7] devised new
flutter and non-synchronous vibration (due to prestall modules for the SA model to account for acceleration, roughness
ce
disturbances or rotating stall) for real industrial fan rotors in a and streamline curvature. The modified model improved the
reasonable time frame, with the aim of mapping out a stability prediction of separation and reattachment point in an intake
Ac
boundary within a week. It is obvious, but the development of under crosswind and showed a good agreement with the
such a model is of great importance to all engine manufacturers experiment. They reported that the larger separation bubble for
as it will allow the identification of potential vibration and noise the original model is due to the fact that the model cannot
problems at early stages of development and consequently generate sufficient turbulent viscosity to reattach the flow.
reduce the number of rig/engine tests. The approach taken here Although, the previous modifications were successful for
is to modify the SA turbulence model based on correlation with avoiding premature stall in the specific flows for which they
the measured data, so that it can predict the stability boundary as were designed, they were not general enough to apply to
well as radial profiles of flow on relatively coarse grids of around transonic fan rotors. Therefore, a more reliable modified SA
model for fan rotors is developed in this paper. The aim of the
1 TURBO-17-1114 LEE
model is to suppress the unphysical separation zones, and hence The CFD code used for this work is AU3D [11], which is a
predict an accurate mean flow near the stall boundary. three-dimensional, time accurate, viscous, compressible
Furthermore, a more generalized wall function [8] in the log unsteady RANS solver, based on a cell-vertex finite volume
region which accounts for the effect of the wall shear stress as methodology and mixed-element unstructured grid. The flow
well as the wall pressure gradient on the surface of the fans is variables are stored on the nodes of a generic semi-structured
implemented. grid and numerical fluxes are computed along the edges of the
This paper is organized as follows. First of all, the numerical grid. The numerical fluxes are computed using Roe’s flux vector
model and the test cases are introduced. In the second part, the difference splitting coupled to Jameson–Schmidt–Turkel (JST)
modifications made to the turbulence model are introduced and scheme. The solution method is implicit with second-order
d
the proposed model is validated. The numerical steady state accuracy in space and time. For steady-state flow computations,
results, for several fan rotors, are compared with the measured the solution is advanced in pseudo-time using local time
ite
data in terms of pressure ratio, mass flow and radial profiles. In stepping, while dual time stepping is used for unsteady
the final part of this paper, the effects of the current modifications computations to ensure time accuracy. Solution acceleration
ed
for a simple canonical case, transonic flow over an axisymmetric techniques, such as residual smoothing and local time stepping
bump, [9-10] for which extensive measured data is available are are used for steady-state flow calculations and in the inner
considered. Although such an analysis of turbulent structures is iteration for the unsteady computations. It is assumed the
py
not the prime objective of this paper, it is informative to see how frequencies of interest are sufficiently far away from the
the current modification of the SA model change the viscous frequencies of turbulent flow structures. The resulting CFD code
quantities such as, separation/reattachment point and Reynolds has been used over the past 20 years for flows at off design
Co
stresses. conditions with a degree of success [2-4, 12-14].
ot
Flow solver hexahedral elements around the aerofoil, in the boundary layer
region, and prismatic elements in the passage. The rotor-to-rotor
tN
grid is unstructured, but in the boundary layer region a body-
fitted two-dimensional O grid (which is refined toward leading-
edge and trailing-edge) is used. The O grid contains 335 points
rip
tip gap.
Four different fan rotors are considered for the steady
validation of the proposed numerical model. These include a
Ma
2 TURBO-17-1114 LEE
d
The transport equation for the original SA (OSA) model [1]
is given in Eq. 1.
ite
𝜕𝜈̂ 𝜕𝜈̂
+ 𝑢𝑗 = 𝑐𝑏1 (1 − 𝑓𝑡2 )𝑆̂𝜈̂ − 𝐷𝑒𝑠𝑡𝑟𝑢𝑐𝑡𝑖𝑜𝑛
ed
𝜕𝑡 𝜕𝑥𝑗 FIG. 2 β FUNCTION (SEE EQ. 5)
+𝐷𝑖𝑓𝑓𝑢𝑠𝑖𝑜𝑛 (1)
turbulent viscosity in the region of shock-wave/boundary layer
py
Here, 𝜈̂ is the corrected turbulent viscosity, 𝑢𝑗 is the velocity interaction, and consequently predicted accurate separation
component in the jth direction, t is the time, 𝑥𝑗 is the Cartesian points, showing a good correlation with the experiment. With
reference to the previous modifications, the original production
Co
coordinate vector, 𝑐𝑏1 is the empirical constant, 𝑓𝑡2 is the trip
term (Eq. 3) is modified by using both the adverse pressure
function and 𝑆̂ is the modified vorticity. The first term on the gradient and the velocity helicity to increase the turbulent
left hand side in Eq. 1 is the time derivative and the second term viscosity in selected areas, such as the shock, separated region
is the advection term. On the right hand side, the first term is the
and tip leakage flow. The original 𝑆̂ term is modified according
ot
production term followed by the Destruction and the Diffusion
to:
terms. In the work presented here, the Destruction and the
tN
Diffusion terms remain unchanged from the original model and ̂
𝜈
only the production term is modified. The turbulent viscosity is 𝑆̂𝑚 = 𝛽𝑆 + 𝑓 (4)
𝜅 2 𝑑 2 𝑣2
computed from the equations:
rip
𝛽 = 𝐶𝑠 × 𝐶𝑣ℎ (5)
𝜈𝑡 = 𝜈̂ 𝑓𝑣1 (2)
𝑐ℎ1 ×tanh[𝑎1 𝑃∗2 ] 𝑐ℎ2 ×tanh[𝑎2 𝐻 2 ]
𝐶𝑠 = + 1.0, 𝐶𝑣ℎ = + 1.0 (6)
sc
3 3 3
where 𝑓𝑣1 = 𝜒 /(𝜒 + 𝑐𝜈1 ) and 𝜒 = 𝜈̂ /𝜈. tanh(1.0) tanh(1.0)
𝑑𝑝 𝜌2 𝑢 3 1
Here, 𝑓𝑣1 is the Mellor and Herring’s damping function, 𝑐𝜈1 is where 𝑃 ∗ = / , 𝑛= (𝑢𝑥 , 𝑢𝑦 , 𝑢𝑧 ) and
nu
𝑑𝑛 𝜇 |𝑢|
an empirical constant and 𝜈 is the molecular kinetic viscosity.
𝐻∗ = 𝑢 ⃗ /|𝑢||𝛺|
⃗ ∙𝛺
The production term 𝑆̂ is given by:
Ma
̂
𝜈 Here, ch1 and ch2 are (the new) constants of the turbulence model
𝑆̂ = 𝑆 + 𝑓 (3) to be determined, a1 and a2 are the scaling factors, u is the
𝜅 2 𝑑 2 𝑣2
1 𝜕𝑢𝑖 𝜕𝑢𝑗 velocity vector, n is the unit vector in the flow direction, P* is the
where 𝑆 = √2Ω𝑖𝑗 Ω𝑖𝑗 , Ω𝑖𝑗 = ( − ) and
non-dimensional pressure gradient in the direction of flow, H is
ed
2 𝜕𝑥𝑗 𝜕𝑥𝑖
𝑓𝑣2 = 1 −
𝜒
. the helicity and 𝜴 is the vorticity. After a detailed numerical
1+𝜒𝑓𝑣1 study of Fan A and Fan B, the constants were chosen as ch1 = 0.5
pt
and ch2 = 0.7 and were kept fixed throughout the rest of this
Here, 𝑓𝑣2 is the damping function. The rest of the damping work. Beta is a hyperbolic tangent function of the adverse
functions and coefficients in Eq. 2 and Eq. 3 are unchanged in pressure gradient and helicity. The scaling factor a2 was chosen
ce
this work, and can be found in Ref. [1]. as 3 and produces a 'continuous' Heaviside function (see Fig. 2).
Liu et al. [5] modified the SA model using the velocity It comes from correlation between CFD and measured data;
Ac
helicity to take into account the energy backscatters. Li et al. [6] larger values of a2 may result in a lack of convergence for the
introduced the adverse pressure gradient into the production term CFD solver, whereas lower values do not create viscosity fast
in the original SA model so that the over prediction of separation enough. The constant a1 is chosen to give a similar profile as Fig.
for the original SA model can be suppressed. Both modifications 2 for the pressure gradient contribution, but it is mesh
improved the accuracy of the SA model. The modification with dependent. As can be seen in Fig. 2, β increases gradually in the
helicity suppressed the unrealistic corner separation in selected areas according to the magnitude of the adverse pressure
compressor passages and showed good agreement with the gradient and the helicity. For a smooth flow field, β approaches
measurement in terms of the size of separation bubble. The one, and hence the 𝑆̂𝑚 term converges to the original 𝑆̂ term
modified model with adverse pressure gradient increased the
3 TURBO-17-1114 LEE
in the SA turbulence model. Fig. 3 shows the Mach contour on Shih et al. [8] suggested a more generalized wall function
the suction surface of the benchmark low-speed fan rotor, Fan A, based on the unified wall velocity by including both the wall
near the stall boundary, together with the two components of the stress and the wall pressure gradient. They showed that both the
β function given in Eq. 6. It is seen from this plot that, the adverse and favorable pressure gradients have a significant effect
modified SA model can identify the regions of strong adverse on the mean velocity and the turbulent shear stress. Therefore,
pressure gradient (the shaded region in Fig. 3(a) and the green the unified wall function will be a more adequate boundary
region in Fig. 3(b)) and helicity (three-dimensional flow shown condition for more complex flows. With reference to their work,
by the arrows in Fig. 3(a) and the green region in Fig. 3(c)). The the standard wall function is modified as:
two components of the β function are superimposed on the fan
d
as in Eq. 5, and consequently increase the intensity of the 𝑢𝑐 = 𝑢𝜏 + 𝑢𝑝 = √|𝜏𝜔 |/𝜌 + [(𝜈/𝜌)|𝑑𝑃𝜔 /𝑑𝑛|]1/3 (9)
turbulent viscosity in the selected regions, as shown in Fig. 3(d).
ite
Fig. 4 shows the general behavior of the β function as the
mass flow decreases and the fan moves towards stall. In this plot,
ed
the Mach number profiles on the suction side of the fan together
with the streamlines superimposed on them are plotted at three
mass flows. Also shown in this figure are the corresponding β
py
functions. It is seen from the plot that on the design working line,
the streamlines on the suction side of the fan are almost uniform,
and hence the β function has a small influence on the flow. On
Co
the other hand, as the mass flow decreases and the incidence to
the fan increases, the shock strength increases and the flow on
the suction side of the fan becomes three-dimensional, and
consequently the 𝛽 function increases and becomes more
ot
dominant over a larger part of the fan, as shown in Fig. 4. It is
important to mention that, for all the mass flows shown in Fig 4,
tN
the shock is expelled from the passage and hence the β function
does not have any significant effect on the flow on the pressure
surface of the fan.
rip
the wall shear stress and the wall pressure gradient. The standard
wall function given in Eq. 7 [16], however, is only valid for a
nu
𝑈 1 𝑢𝜏 𝑦
= 𝜅 𝑙𝑛( )+ 𝐶 (7)
Ma
𝑢𝜏 𝜈
where 𝑢𝜏 = √|𝜏𝜔 |/𝜌
velocity and y is the normal distance from the wall. Tennekes and
Lumley [17] derived the wall function for a boundary layer with
a strong adverse pressure gradient and zero wall shear stress. In
pt
this case, the adverse pressure gradient controls the surface flow:
ce
𝑈 𝑢𝑝 𝑦
= 𝛼 𝑙𝑛( )+𝛽 (8)
𝑢𝑝 𝜈
where 𝑢𝑝 = [(𝜈/𝜌)|𝑑𝑃𝜔 /𝑑𝑥|]1/3
Ac
4 TURBO-17-1114 LEE
d
performed with the original as well as generalized wall
functions. Single passage steady computations from the intake
ite
through to the throttle (Fig. 1) were performed by using the MSA
model. The grid for the intake, OGV, ESS and nozzle were the
ed
same for both fan grids. Fig. 5 shows the comparison of the radial
profiles of pressure rise between the two grids. It is seen from
Fig. 5(a, b) that the wall-resolved solutions show close
py
agreement with the results of the MSA+GWF for both flows,
showing the accuracy of prediction using the GWF and
confirming the applicability of the generalized wall function.
Co
APPLICATION TO TRANSONIC FAN ROTORS
In this section, the proposed modified SA model is validated
ot
by comparing the steady-state solution with the measured data in
terms of overall performance and radial profiles. In this section,
a summary of the results obtained is presented. It should be
tN
FIG. 4 GENERAL BEHAVIOR OF β FOR FAN A: (A) MACH mentioned that, for Fan A and B, the correct geometry is used at
CONTOUR AND STREAMLINES AND (B) β CONTOUR each speed (i.e. allowing for change of stagger with speed).
rip
Fan A
CASE SETUP Fan A has 18 rotors, the tip Mach number is below 1 at
The steady-state solutions from different versions of the SA design speed and the hub-tip ratio at inlet to the rotor is about
sc
model introduced in the previous section are presented. 0.3. Fig. 6(a) shows the comparison of the constant speed
According to the version of SA model and the wall function used, characteristic between the measured data and the computed
nu
they are defined as: results from different SA models at 80% speed. It is seen from
the plot that all the turbulence models fail to predict the positive
1) Original SA turbulence model + original wall function slope part of the characteristic. Furthermore, it is noticeable from
Ma
(OSA) this plot that the OSA and the OSA+GWF predict the stall
2) Original SA turbulence model + generalized wall boundary at a higher flow coefficient than the MSA and the
function (OSA+GWF) MSA+GWF (see the arrows in Fig. 6(a)). The under prediction
3) Modified SA turbulence model + original wall function of the stall boundary is attributed to the fact that the OSA predicts
ed
(MSA) a larger separation zone than the MSA, which leads to the
4) Modified SA turbulence model + generalized wall blockage of the passage near the tip region of the fan. It is also
function (MSA+GWF) observed from this plot that, both the OSA and the MSA over-
pt
for the steady validation of the numerical model proposed in the such flows. Finally, it is seen from this figure that the results with
previous section. These include a low-speed fan rotor (referred the MSA+GWF show a good correlation with measured data in
to as Fan A [13]), two high-speed fan rotors (referred to as Fan
Ac
5 TURBO-17-1114 LEE
d
For the computations at the higher flow coefficient of 1.07 (Fig.
6b), the use of GWF improves the prediction of the total pressure
ite
ratio as compared to the standard wall function, the latter over-
predicting the total pressure ratio by around 1%. Fig. 6(c) shows
ed
the comparison of the radial profiles at mass flow of 0.95. For
this operating point, the comparison can be made only for the
two MSA models as the OSA cannot predict the steady flow at
py
this point due to premature stall (as shown in Fig. 6(a)). There is
a close agreement between the measured data and the
MSA+GWF, confirming that both modifications, the turbulence
Co
model and the wall function, are necessary to obtain results in
accord with the measured data for such flows.
Fig. 7 compares Mach contours on the suction surface of Fan
A at 80% speed at mass flow of 1.02. Superimposed on these are
ot
the surface streamlines. It is clearly seen from these plots that
both MSA models (Fig. 7(c, d)) reduce the flow separation and
tN
the radial migration near the tip and trailing edge. This is obvious
as the turbulent viscosity in the MSA models is increased by the
effect of the β function in the selected areas. In addition, the
rip
For clarity of plot, the measured data are compared with the OSA
and the MSA+GWF. Again, it is seen that the computed results
by the modifications show a significant improvement in
Ma
accuracy over the OSA. Fig. 8(b, c) shows the steady radial
profiles at mass flows of 1.31 and 1.17 for 100% speed. It is clear
from these figures that the modified model shows much better
agreement with the measurements for both points.
ed
pt
ce
Ac
6 TURBO-17-1114 LEE
d
ite
ed
py
Co
ot
FIG. 7 MACH CONTOURS ON THE SUCTION SURFACE OF
FIG. 8 FAN A CHARACTERISTIC MAP AT 100% SPEED (A)
tN
FAN A AT MASS FLOW OF 1.02: (A) OSA, (B) OSA+GWF, (C)
MSA AND (D) MSA+GWF AND DISTRIBUTION OF STAGNATION PRESSURE AT
NORMALIZED MASS FLOWS OF (B) 1.31 AND (C) 1.17
rip
Fan B and C
The calculated and measured results for the OSA and the
sc
accord with the measured data for all the points on the
characteristic. In addition, it shows a better agreement with test
data than that of the OSA in the high mass flow region (1.07 to
Ma
and stalls the flow in this region, as shown in Fig. 10(a). Again,
at the lower mass flow, the OSA cannot predict the steady flow
at this point but the modified model agrees well with the
Ac
7 TURBO-17-1114 LEE
d
ite
FIG. 10 MACH CONTOURS ON THE SUCTION SURFACE OF
FAN B AT MASS FLOW OF 1.00: (A) OSA AND (B) MSA+GWF
ed
estimate the whole characteristic line. It is believed that the
py
under-estimation of the characteristic (at this speed) is attributed
to the unsteadiness (asymmetry) in the low mass flow region (see
the data points at mass flow of 0.79 in Fig. 11(a) which has been
Co
obtained using full annulus unsteady computations).
Computed radial profiles for two mass flows of 0.9 and 0.81
are compared with the measured data in Fig. 11(b, c). It is seen
from Fig. 11(b) that the flow features in the measurement is well
ot
predicted at mass flow of 0.9 by the modified model. On the
FIG. 11 FAN C CHARACTERISTIC MAP AT 70% SPEED (A)
other hand, at mass flow of 0.81 (Fig. 11(c)) the computed results
tN
AND DISTRIBUTION OF STAGNATION PRESSURE AT
show a lower total pressure than the measured data for the top NORMALIZED MASS FLOWS OF (B) 0.9 AND (C) 0.81
20% of the fan. The under-prediction of the total pressure near
the tip region is attributed to the fact that the (symmetric) steady
rip
NASA Rotor 67
Ma
8 TURBO-17-1114 LEE
d
shows a reasonably good agreement with the measurement. It is
also seen from Fig. 15 that the modified SA model tends to over-
ite
predict the pressure downstream of the axial position of x/c=0.8,
which was also observed by other researchers who used two-
ed
equation models [9-10]. The original SA model, however, shows
less deviation due to the over-prediction of flow separation. This
FIG. 13 DISTRIBUTION OF STAGNATION PRESSURE AND is attributed to the fact that the original SA model cannot produce
py
TEMPERATURE AT 100% SPEED FOR MASS FLOW OF (A) sufficient turbulent viscosity, and hence over predicts the size of
0.99 AND (B) 0.93
the separation bubble, as was already mentioned in the
modified model shows a noticeable improvement in the total introduction of this paper. This is confirmed by the profiles of
Co
pressure ratio and the total temperature ratio between 50% to the streamwise velocity (non-dimensionalized by the freestream
90% height in comparison with the OSA model. velocity) at different axial positions, as shown in Fig. 16(a). It is
The results shown in this paper represent a selected clearly seen from Fig. 16(a) that the original SA model
overestimates the reverse-flow velocity whereas the modified
ot
summary of the steady state computations performed and
compared with the measured data for different fans and at model shows a good agreement with the measured data at all
different operating speeds. Other results, not shown here, give
tN
similar improvements in prediction of flows at design and at off-
design flow conditions. This study has demonstrated that these
modifications of the SA model and wall function can be
rip
CANONICAL FLOW
In this section, the implications of the current modifications
of the SA model for a simple canonical case for which extensive
Ma
measured data is available are considered. Although an analysis FIG. 14 GEOMETRY AND FLOW FEATURES
of turbulent structures is not the objective of this paper, it is be
informative to see how the current modifications of the SA model
will change the viscous quantities such separation/reattachment
ed
features for this test case. It is seen from Fig. 14 that a normal
shock interacts with the boundary layer which results in the
detachment of the boundary layer over the bump with a
subsequent reattachment on the downstream side. Based on the
experience of other researchers presented on the TMR web site
[9], the second-finest grid level 2 × 721 × 321, was used for the
computation. The inlet Mach number is 0.875, the Reynolds
number based on the chord length of the bump (c) is 2.763 × 106 FIG. 15 PRESSURE DISTRIBUTION
and the inlet temperature is 300K.
9 TURBO-17-1114 LEE
The result of this paper show that, the proposed model which
was based on correlation with experimental data, is capable of
predicting the steady mean flow around fan rotors with a good
degree of accuracy.
ACKNOWLEDGMENTS
The authors thank Rolls-Royce plc. and CleanSky for both
sponsoring this work and allowing its publication. They would
also like to thank their colleagues Professor Nick Cumpsty, Dr.
d
Sina Stapelfeldt and Mr. Wei It Kuan at Imperial College for very
useful discussion.
ite
NOMENCLATURE
ed
𝜈̂ = corrected turbulent viscosity
𝜈 = molecular kinetic viscosity
𝑓𝑡2 = trip function
py
𝑓𝑣1 = Mellor and Herring’s damping function
𝑓𝑣2 = damping function
𝑐𝑏1 , 𝑐𝜈1 = empirical constants
Co
𝑆̂ = modified vorticity
𝑐ℎ1 , 𝑐ℎ2 = amplitude of beta function
P* = non-dimensional adverse pressure gradient
ot
FIG. 16 PROFILES OF (A) STREAMWISE VELOCITY AND (B) 𝒏 = direction of flow
REYNOLDS STRESS 𝐻 = velocity helicity
tN
𝜴 = vorticity
positions. Fig. 16(b) shows the corresponding profiles of the 𝑢𝜏 = skin friction velocity
non-dimensional Reynolds stress. Both SA models under predict 𝑢𝑝 = wall pressure gradient velocity
rip
the measured data, which was also observed by other researchers 𝑦 = normal distance from the wall
[9-10]. On the other hand, the modified model shows a better u = velocity
agreement with the measurement data in terms of the alignment 𝜌 = density
sc
of the peak Reynolds stress for all positions. In addition, it is 𝜏𝜔 = wall shear stress
confirmed from Figs. 15 and 16 that the modified SA model 𝑃𝜔 = wall pressure
without wall function shows a close agreement with the results 𝐶𝑝 = wall pressure coefficient
nu
helicity in the production term, to improve accuracy of flow Observations in a High Speed Compressor—Part II:
solutions near the stall boundary. In addition, the standard wall Numerical Study," ASME J. Turbomach., 137(5), p. 051003.
function in the log region was modified to include the effects of [3] Choi, M., Vahdati, M., and Imregun, M., 2011, "Effects of
pt
pressure gradient. Fan Speed on Rotating Stall Inception and Recovery," ASME
Steady simulations were conducted for four fan rotors to J. Turbomach., 133(4), p. 041013.
ce
validate the numerical approach used in this study. The steady [4] Choi, M., Smith, N. H., and Vahdati, M., 2013, "Validation
results were compared with the experimental data and were well- of Numerical Simulation for Rotating Stall in a Transonic
matched in terms of the pressure ratio, mass flow and radial Fan," ASME J. Turbomach., 135(2), p. 021004.
Ac
profiles. The test cases show that the modified model not only [5] Liu, Y., Lu, L., Fang, L., and Gao, F., 2011, "Modification of
increases the stall margin compared to the original model, but Spalart–Allmaras Model with Consideration of Turbulence
can also improve the prediction of mean flow around the rotor Energy Backscatter Using Velocity Helicity," Phys. Lett.,
significantly. For the low-speed fan, however, the modified 375(24), pp. 2377-2381.
model fails to predict the positive slope region of the [6] Li, M., Lipeng, L., Jian, F., and Qiuhui W., 2014, "A Study
characteristic. This is because the experimental flow is on Turbulence Transportation and Modification of Spalart–
asymmetric in the region. Allmaras Model for Shock-Wave/Turbulent Boundary Layer
Interaction Flow," Chinese J. Aeronaut., 27(2), pp. 200-209.
10 TURBO-17-1114 LEE
d
and Validation of a Second-Moment-Closure Model,” AIAA
J., 54(5), pp. 1524-1540.
ite
[11] Sayma, A. I., Vahdati, M., Sbardella, L., and Imregun, M.,
2000, “Modeling of Three-Dimensional Viscous
ed
Compressible Turbomachinery Flows Using Unstructured
Hybrid Grids,” AIAA J., 38(6), pp. 945–954.
[12] Vahdati, M., Sayma, A., Freeman, C., and Imregun, M.,
py
2005, "On the Use of Atmospheric Boundary Conditions for
Axial-Flow Compressor Stall Simulations," ASME J.
Turbomach., 127(2), pp. 349-351.
Co
[13] Lee, K.-B., Wilson, M., and Vahdati, M., 2016, "Numerical
Study on Aeroelastic Instability for a Low Speed Fan,"
ASME Paper No. GT2016-56462.
[14] Vahdati, M. and Cumpsty, N., 2015, “Aeroelastic Instability
ot
in Transonic Fans,” J. Eng. Gas Turb. Power., 138, pp.
022604.
tN
[15] Strazisar, A. J., Wood, J. R., Hathaway, M. D., and Suder,
K. L., 1989, "Laser Anemometer Measurements in a
Transonic Axial-Flow Fan Rotor." NASA Technical Paper
rip
2878.
[16] Millikan, C. B., 1938, "A Critical Discussion of Turbulent
Flows in Channels and Circular Tubes," Proc. 5th Int. Cong.
sc
135(1), p. 011033.
[20] Pardowitz, B., Tapken, U., Sorge, R., Thamsen, P. U., and
Enghardt, L., 2014, "Rotating Instability in an Annular
pt
[21] Mailach, R., Lehmann, I., and Vogeler, K., 2001, "Rotating
Instabilities in an Axial Compressor Originating from the
Fluctuating Blade Tip Vortex," ASME J. Turbomach.,
Ac
11 TURBO-17-1114 LEE