Professional Documents
Culture Documents
Two-Dimensional Diffuser
K. Pedersen
Table of contents
1. Introduction
1.1 Selected paper
2. Experimental setup
2.1 Geometry & apparatus
2.2 Flow conditions
2.2.1 Primary inlet
2.2.2 Jet inlet
2.2.3 Outlet
2.3 Experiment parameters & configurations
3. CFD modelling setup
3.1 Geometry
3.2 Mesh
3.3 Fluent: setup
3.3.1 General settings
3.3.2 Model
3.3.3 Solver
3.3.4 Boundary conditions
4. Results
4.1 CFD results (with & without flow control)
4.1.1 14 degree diffuser
4.1.2 30 degree diffuser
4.2 Comparison with experimental data
4.2.1 Pressure recovery factor
4.2.2 Velocity profiles
5. Conclusions
Appendix I: Nomenclature
Appendix II: Original graphs & figures (Fiedler et al.)
Appendix III: Primary inlet velocity profile UDF
1. Introduction
In this project, our goal is to use CFD software to model fluid flow through a diffuser, and to compare
our results with those of experiments in the literature.
First, we will select a paper presenting the results of such an experiment. We will then attempt to
replicate the experiment's geometry and flow conditions in ANSYS Fluent. Finally, we will compare the
results computed in Fluent with those measured in the original experiment, and attempt to understand
and explain any discrepancies between the two sets of results.
In order to keep the problem from becoming too complex, we will choose an experiment that allows us
to make the following simplifications:
2D, symmetric diffuser geometry
Incompressible flow.
Examining the literature, we find that nearly all work on expansion through a diffuser is conducted
using air as the fluid medium. Air exhibits incompressible behavior if M 0.3 , where M is the fluid's
Mach number.
1 +
[]
Speed of sound in dry air: cair = 331.3 273.15 [m/s]
J
Rair = R/M air 287.1 [ kgK ]
air = 1.4
[] - Temperature in degrees Celsius.
Taking a maximum Mach number of M max = 0.3 , and assuming experiments are conducted around
room temperature (25), the maximum experimental fluid velocity will be:
U max = M max cair 0.3 346.2 [m/s] = 103.86 [m/s] = 340.7 [f ps]
We seek a research paper which presents results of flow control experiments using air flow velocities
below 103.86 [m/s] (340.7 [f ps]) . The majority of experiments on diffuser flow are conducted with a
focus on applications in the field of aeronautics, at sonic or supersonic fluid velocities. Only a small
number of studies have been done on active flow control on diffuser flow at M 0.3 .
1.1 Selected paper:1
Fiedler RA, Gessner FB., "Influence of Tangential Fluid Injection on the Performance of
Two-Dimensional Diffusers," ASME. J. Basic Eng. 1972; 94(3):666-674.
Notes:
Dimensions in the original paper are provided in US standard units. All calculations for this
work will be conducted using SI units.
Where relevant, dimensions in this work will be presented both in US and in SI units. This is to
avoid confusion, enable easy comparison with the original data, and prevent accumulating
errors from units conversion.
2. Experimental setup
In this section, we will describe the experimental apparatus and its geometry, and determine boundary
conditions at the inlets and outlet.
Figures 1 and 2 present the experiment setup and geometry. It is important to note that Fiedler et al.
conducted their experiment on a rectangular diffuser. This is a departure from previous experiments
by Ackeret (1926) and Nicoll (1970), both of whom used conical diffusers.
1
Note: Two other studies on active flow control with 2D incompressible flow through diffusers are:
Ackeret, J., "Removing Boundary Layer by Suction," NACA Tech. Mem. 395, 1926.
Nicoll WB, Ramaprian BR., "Performance of Conical Diffusers With Annular Injection at Inlet,"
ASME. J. Basic Eng. 1970;92(4):827-835.
In order to approximate an infinite, 2-dimensional diffuser, a large inlet aspect ratio of 8:1 (width to
height) was chosen. According to previous research done by Reneau et al.2 (1967), this ratio should
ensure that flow through the diffuser at its center will not be significantly influenced by the side walls.
The diffuser is symmetric relative to the horizontal midplane. In Fig. 2, the lower half of the diffuser is
shown. The upper half is an identical reflection of the lower half.
Jets of air at an average velocity of U j are injected tangentially to the walls of the diffuser (see figures
2 and 15), through injection slots of width B. There are two identical injection slots that run the width of
the diffuser: one on each symmetrical half of the diffuser.
Distance of injection slot from the diffuser inlet: xj = 2.64 [in] = 0.0671 [m]
Diffuser width: 24 [in] = 0.6096 [m] .
A complete list of symbols used in this work and the original paper is reproduced in Appendix I.
2
Reneau LR, Johnston JP, Kline SJ., "Performance and Design of Straight, Two-Dimensional Diffusers," ASME.
J. Basic Eng. 1967;89(1):141-150. doi:10.1115/1.3609544.
Incompressible flow ( M 0.3 ).
Constant air properties (including kinematic viscosity , ambient temperature T env ) for all
experiments.
Isothermal expansion through the diffuser: T 1 = T 2 = T env .
Negligible influence of gravitational forces.
Average velocity U 1
In all the result sets that we will refer to, the Reynolds number at the diffuser inlet is given to have the
value of Re1 = 9 104 .
UD
Reynolds number definition: Re
Re1
U1 = W1
For our range of air values, we get: 15.3 [m/s] U 1 22.5 [m/s]
Since Fiedler et al. present their findings almost exclusively in the form of non-dimensionalized
performance coefficients, it is difficult to extract information about original experimental conditions.
There is only one set of conditions for which the inlet velocities are given:
In Figure 6, we see that U j 100 [f t/s] = 30.48 [m/s] . In this graph, it is also stated that U j /U 1 = 1.74 .
Therefore, for the experimental conditions of Figure 6, the velocity U 1 is:
Neither U 1 nor U j are stated for other experimental configurations. Results are presented using the
nondimensional velocity ratio U j /U 1 . In order to evaluate the system for more than one configuration,
we will make the assumption that U 1 is constant for all experiments, while U j varies according to the
velocity ratio.
We know this assumption is at least roughly accurate, because the inlet Reynolds number for all
configurations is stated to have the value of Re1 = 9 104 . Since W 1 and air are assumed constant
for all experiments, U 1 must also stay at least approximately the same between experiments in order
to preserve a constant Reynolds number.
The primary inlet Reynolds number, Re1 = 9 104 , is well into the turbulent range. Indeed, Fiedler et
al. state that flow conditions were selected to ensure 'sustained turbulent boundary flow at the diffuser
inlet'. Furthermore, a strip of sandpaper is placed around the periphery of the duct, 4 inches upstream
from the diffuser inlet, to ensure the flow is 'tripped' into turbulence before entering the diffuser.
We can estimate the boundary layer thickness from the displacement thickness, which is given to be
constant for all experimental configurations: 2 1 * /W 1 = 0.020 .
We will assume a power-law profile for the velocity boundary layer at the inlet:
y 1/7
u
U = ()
According to the power-law profile, the relationship between the displacement thickness 1 * and the
velocity boundary layer thickness will be:
1 8 1 * = 0.080 W 1
1 0.0061 [m] = 6.1 [mm]
To define the inlet velocity profile, we must find U . Mass flux through boundary layer region:
0.0061 0.0061
y 1/7
u= U ( ) = [1.813 U y 8/7 ]00.0061 = 0.00533775 U
0 0
m1 = 2 0.00533775 U + U (W 1 2 1 ) = 0.074675 U
U 1 W 1 = 0.074675 U
17.520.0762
U = 0.074675 = 17.8778 [m/s]
Our approximation for the velocity profile at the velocity inlet will be:
y 1/7
f or y 0.0061 [m] : ux = 17.8778 ( 0.0061 ) [m/s]
For the jet inlet, we will assume a uniform injection flow of velocity U j , disregarding the effect of
boundary layers that may have formed prior to injection into the diffuser. This simplification is
supported in several places in the text:
Uj = [ ]
Uj
U1
U1 = [ ]
Uj
U1
17.52 [m/s]
2.2.3 Outlet
The static pressure at the diffuser outlet, p2 , is assumed to be constant across the outlet
cross-section, and equal to atmospheric pressure. Fiedler et al. make this assumption in their
calculations (pg. 668).
Here, we use these parameters to define 6 different configurations of the experimental system at
steady state. Each of the 6 configurations will be modelled individually and compared to selected
results from the original experiments. (Atypical values are highlighted for ease of reading.)
Parameter Definition 1 2 3 4 5 6
2 Diffuser angle. 14 14 14 14 30 30
2B/W 1 Normalized slot width. 0.0 0.063 0.063 0.063 0.0 0.063
B - slot width
W 1 - diffuser inlet width
Re1 Reynolds number at entry. 9 104 9 104 9 104 9 104 9 104 9 104
3.1 Geometry
We will sketch the profile of the diffuser using ANSYS DesignModeler. Four different geometric
configurations are needed:
We could model only one half of the diffuser and assume identical behavior in the other, symmetrical
half. However, according to research referenced by Fiedler et al., stall tends to occur asymmetrically
in diffusers of the proportions specified. For this reason, we will model the full diffuser instead of only
half of it.
3.2 Mesh
ANSYS Meshing is used to define the mesh over each geometry. We want the mesh to be fine next to
the diffuser walls, and coarser toward the center of the diffuser.
At the edge of the viscous sublayer, y + 5 . If we aim for a y + value of approx. 2 for the width of the
wall-adjacent cells, the centroid of these cells will fall within the viscous sublayer region, as
recommended when using enhanced wall treatment (see next section).
4 1/8
u = U e
0.039/Re1 0.25 = 17.8778 * 0.038 * (9 10 ) = 0.8374
Fluid: air
Gauge pressure is measured relative to an atmospheric pressure of
p = 1 [atm] = 101, 325 [P a]
Density: = 1.177 [kg/m3 ]
Dynamic viscosity: 1.846 105 [kg/ms]
3.3.2 Model
Our problem involves viscous flow phenomena; therefore we will select a viscous model to solve it.
Below is a comparison of some of the turbulence solution models provided in Fluent:
Model Comments
Standard k Though widely used in CFD, the standard k model is not suitable for flows
involving pressure gradients or flow separation.
RNG k More accurate than standard k model. Suitable for modelling pressure
gradients, separation flows.
Realizable k Provides similar benefits to RNG k model, but is typically easier to converge.
3
Source: Modeling Turbulent Flows, Introductory FLUENT Training, ANSYS
Reynolds stress Suitable for complex 3D flows with strong streamline curvature and/or swirl.
Provides the highest accuracy, but uses a lot of processing power/time, and is
often difficult to converge.
We will select the Realizable k model for our calculations. We will turn on the "Enhanced wall
treatment" setting, since we wish to model phenomena such as boundary layer separation that occur
within the region close to the wall.
3.3.3 Solver
Our problem is 2D, steady state, and incompressible. We will choose a pressure-based solver with an
implicit solution formulation. Since our problem is 2-dimensional, we will select the SIMPLE solution
scheme. The spatial discretization methods selected are:
Pressure Standard
Primary inlet
Again, our approximation for the velocity profile at the primary inlet will be:
y 1/7
f or y 0.0061 [m] : ux = 17.8778 ( 0.0061 )
We do not have information about the profiles of the turbulent kinetic energy ( k ) or dissipation rate (
). However, the paper specifies an approximate value for the inlet core flow turbulent intensity ( I < 1%
), and we have already estimated an inlet boundary layer thickness of 99 0.0061 [m] . Using these,
we can specify the following inlet turbulence values:
Turbulent intensity: I 1%
Turbulence length scale: l 0.4* 99 = 0.00244 [m]
Jet inlet
We will define a constant velocity profile of U j , tangent to the diffuser wall, at each jet inlet.
Therefore, if the inlet jet flow were fully developed, the turbulent intensity would be I 5.5% . The jet
flow at the inlet is not fully developed, however, so the intensity will be smaller than this estimate. We
will estimate a turbulent intensity value of I = 2% .
Using the guidelines in the Fluent User Guide for "flows downstream of turning vanes, perforated
plates, etc.", we will specify a turbulent length scale of l B = 0.0024 [m] .
Outlet
We will define a pressure outlet boundary condition at the diffuser outlet. As explained previously, we
assume a constant and uniform pressure at the outlet of:
We will choose to specify the turbulent intensity and hydraulic diameter of the outlet. The turbulent
intensity will be high at the outlet, so we will guess an approximate intensity value of I 10% .
2AB
Formula for hydraulic diameter of a rectangle: DH = A+B
4
ANSYS Fluent User Guide
Outlet hydraulic diameter:
4. Results
Velocity magnitude
[m/s]
Pressure coefficient
( C p )5
Main inlet
Path lines (colored
by source) Jet inlets
14 degrees is a sufficiently moderate diffuser angle that no boundary layer separation occurs from the
diffuser walls even without flow control. In both cases, the path lines of the flow are roughly similar.
However, significant improvement to the performance coefficient is observed: from C p = 0.625 without
flow control, to C p = 0.805 with flow control (a 28.8% improvement).
5
Cp calculated using average inlet pressure & velocity.
4.1.2 30 degree diffuser
Velocity
vector plot
(shaded by
velocity
magnitude,
[m/s])
Static
pressure
[Pa]
Pressure
coefficient6
6
Cp calculated using average inlet pressure & velocity.
Path lines Main inlet
(colored by
path line Jet inlets
source)
Outlet
At the larger angle of 30 degrees, we observe boundary layer separation from both diffuser walls
when no flow control is used. However, with a relative jet injection velocity of U j /U 1 = 1.92 , the flow
remains attached throughout the entire length of the diffuser. Two intermediate jet velocities are also
presented in order to observe the transition between the two states.
The four values of jet intensity span the range of diffuser flow regimes - from jet flow, to
fully-developed stall, to unstalled flow7:
No flow Jet flow Flow separates from diffuser walls near inlet and
control continues as a jet through the diffuser. Large stationary
eddies form between the jet and the diffuser walls.
1.74 Fully-developed stall Flow remains attached to one wall, but separates from
the other wall near the inlet. A large, stable eddy forms on
the side of separation.
In the next section, we will observe how these regime transitions impact the pressure recovery
coefficient.
The following graphs show a comparison between the experimental data and the results obtained in
ANSYS Fluent.
7
Source: Robert D. Blevins, Applied fluid dynamics handbook
4.2.1 Pressure recovery factor
pp1
The pressure recovery factor at a point in the diffuser is defined as: Cp = q1
p - Static pressure at the point in question.
p1 - Area-averaged static pressure at a reference section - in our case, the diffuser inlet.
q 1 - The dynamic pressure at a reference point, calculated from the average inlet flow
velocity. In our problem, q 1 is
constant for all configurations:
2
q 1 = 21 U 1 = 180.578 [P a]
When we compare the changes in flow behavior with the pressure recovery coefficients, we observe a
clear improvement in diffuser performance with increased jet injection. This trend confirms our
expectation that preventing boundary layer separation reduces the diffuser's head loss and flow
resistance.
Preventing head loss in diffusers has the potential to lead to savings both in operational and in capital
costs, for many different kinds of machines and facilities. Less head loss means less power is
required to circulate fluid through a system. Reducing pressure losses would also allow a system to
be designed for lower operational pressures, reducing the stress components are required to
withstand. The economic practicality of such a system would depend on the efficiency, reliability, and
cost of the jet blower.
Inlet pressure values - average vs. centerline: The C p values calculated from the average inlet
pressure (in green) represent the more correct data set, according to the definition of the pressure
recovery factor. Interestingly, the centerline pressure (red) fits the experimental results significantly
better than the area-averaged values for most of the range examined. The exception is the lower
range of U j /U 1 values, in which the diffuser is stalled, and for which we would expect inaccurate
results due to boundary-layer separation.
This fact is likely due to bias in the results, caused by possible discrepancies between the CFD model
and the original experiment. However, I will note that the precise definition of pressure recovery
coefficient is the subject of some debate. A number of researchers favor centerline measurements
rather than averaged measurements for the dynamic pressure q 1 8, and most studies do not make an
effort to measure and differentiate between average or centerline inlet static pressure.
8
Deniz, S., et al. "Effects of Inlet Flow Field Conditions on the Performance of Centrifugal Compressor
Diffusers: Part 1Discrete-Passage Diffuser." Journal of Turbomachinery 122 (2000): 1.
4.2.2 Velocity profiles
The following graphs present flow velocity profiles at selected locations in the diffuser, for two different
relative jet velocities: U j /U 1 = 1.74 and U j /U 1 = 0.95 . The original data from Fiedler et al.'s graphs are
shown in blue, and the data extracted from the Fluent simulations are shown in red.
We can see that the velocity profiles calculated in Fluent resemble the experimental results, but
consistently underestimate the actual fluid velocity. The CFD simulation improves in accuracy as
U j /U 1 increases.
In the original experiment, discrepancies were also observed between Fiedler et al.'s own theoretical
velocity profiles and the experimental profiles measured. Their theoretical profiles resemble the ones
generated in the current Fluent simulations (see Appendix II), with similar differences between the
calculated and measured data.
Fiedler et al. speculated that the differences in the profiles were caused by side wall effects, and that
a truly infinite, 2D diffuser (as simulated in this project), would conform to their predicted velocity
profiles.
5. Conclusions
CFD simulation in ANSYS Fluent successfully illustrated the benefits of flow control with jets on
diffuser performance.
In the 30 degree diffuser, jet injection caused dramatic changes in the diffuser flow regime.
Previously-stalled flow attached fully to the diffuser walls as the jet velocity was increased, with large
gains in pressure recovery.
Significant improvement was also observed in the 14 degree diffuser. Though the flow regime in the
more moderate diffuser did not change, its performance also saw a marked improvement with
increasing jet velocity.
The computer simulation produced results that resembled those obtained in the original experiment by
Fiedler et al. in 1972. Selected velocity profiles obtained in Fluent are similar to the measured
experimental velocity profiles, with discrepancies possibly attributable to side-wall effects in the
experimental diffuser. Simulated pressure recovery ( C p ) as a function of jet blowing intensity
produced curves almost identical in shape to those measured in the experiment, albeit with a
tendency to overestimate the diffuser performance over most of the relevant range.
Appendix I: Nomenclature
Presented here are the symbols used by Fiedler et al. in their paper.
Appendix II: Original graphs & figures (Fiedler et al.)
Fig. 4: Experimental and predicted velocity profiles Fig. 5: Experimental and predicted velocity profiles
downstream of injection slot for U j /U 1 = 1.74 , 2 = 14 , downstream of injection slot for U j /U 1 = 0.95 , 2 = 14 ,
Re1 = 9 104 , 2 * 1 /W 1 = 0.020 , 2B/W 1 = 0.063 Re1 = 9 104 , 2 * 1 /W 1 = 0.020 , 2B/W 1 = 0.063
Fig. 8: Influence of jet blowing on diffuser pressure recovery, 2 = 14 , Re1 = 9 104 , 2 * 1 /W 1 = 0.020
Fig. 10: Influence of jet blowing on diffuser performance, 2 = 20 , Re1 = 9 104 , 2 * 1 /W 1 = 0.020 , b = 0.6
4
Fig. 11: Influence of jet blowing on diffuser performance, 2 = 30 , Re1 = 9 10 , 2 * 1 /W 1 = 0.020 , b = 0.6
Appendix III: Primary inlet velocity profile UDF
The velocity profile at the inlet was specified in all cases using the same user-defined function (UDF).
The UDF defines the velocity profile at the inlet as was described previously in this work (see section
2.2.1). The code is written in C, and reads as follows:
#include "udf.h"
DEFINE_PROFILE(x_velocity,thread,position)
{
real x[ND_ND], y;
face_t f;
begin_f_loop(f, thread)
{
F_CENTROID(x, f, thread);
y=x[1];
if (y >= ymax-del)
F_PROFILE(f, thread, position) = ufree*pow((ymax-y)/del,P);
else if (y <= del+ymin)
F_PROFILE(f, thread, position) = ufree*pow((y-ymin)/del,P);
else
F_PROFILE(f, thread, position) = ufree;
}
end_f_loop(f, thread)
}
This UDF was modified by code provided by user 'upeksa' on the CFD Online forum.
Steps to import UDF into ANSYS Fluent program:
9
or 'Compiled'.