See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/271874437

Numerical Simulation of Flow Field for a Whole Centrifugal

Fan and Analysis of the Effects of Blade Inlet Angle and
Impeller Gap

Article in HVAC&R RESEARCH · April 2005

DOI: 10.1080/10789669.2005.10391137

CITATIONS READS
42 3,967

6 authors, including:

Weixiong Wang
aseeder
16 PUBLICATIONS 198 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Patient-specific computational modeling of aneurysm and plaque rupture View project

All content following this page was uploaded by Weixiong Wang on 04 March 2015.

The user has requested enhancement of the downloaded file.

VOLUME 11, NUMBER 2 HVAC&R RESEARCH APRIL 2005

Numerical Simulation of Flow Field for a

Whole Centrifugal Fan and Analysis of the Effects
of Blade Inlet Angle and Impeller Gap
Zhao Yu Song Li, PhD Wenqi He

Weixiong Wang Dongtao Huang, PhD Zhichi Zhu

Received March 29, 2004; accepted August 18, 2004

Numerical simulation of the flow field was carried out for a whole centrifugal fan to consider
the interaction of three parts—inlet, impeller, and scroll. A modern optimum design method for
a centrifugal fan has been presented using a computational fluid dynamics (CFD) technique to
replace the conventional method. A sample fan based on the optimum design was manufactured
and tested. The numerical prediction agrees well with the test data. The excellent performance
of the sample fan has proved the validity of the method adopted. Moreover, the effects of the
blade inlet angle and the gap between impeller and inlet on the fan performance are quantita-
tively analyzed. Particularly the energy loss in every passage, including the gap loss and the
leakage flow rate in the gap, are discussed in detail in the paper. It is found that the blade inlet
angle and the impeller gap play an important role for fan performance.

INTRODUCTION
The aerodynamic design of a centrifugal fan has long followed the conventional method,
which is based on the assumption of one- or two-dimensional ideal flow and some empirical
coefficients. The three main parts, i.e., impeller, scroll type of housing (called scroll in the fol-
lowing), and inlet ring (called inlet in the following), are designed, respectively, and their inter-
actions are neglected (Eck 1973). Furthermore, the fan performance cannot be predicted at the
design stage. In fact, there is a three-dimensional, complicated, and highly turbulent flow in a
centrifugal fan, and the effect of interaction between the main parts on fan performance is seri-
ous, which cannot be neglected. Thus, the existing simplified design method is insufficient to
describe the real flow situation, and the performance of a fan designed based on this method is
commonly not very good. Sometimes, in order to obtain an excellent performance, it takes a lot
of time and cost to make several cycles of design-test-redesign. Therefore, the existing conven-
tional design method needs to be improved.
Computational fluid dynamics (CFD) has made great progress in calculating the
three-dimensional flow in centrifugal pumps and fans. Zhang et al. (1996) computed the
three-dimensional viscous flow in a blade passage of a backswept centrifugal impeller at the
design point with the standard k-ε model. The comparison of computed and measured values
showed good agreement. The computation also successfully predicts the jet-wake structure,
though slightly different in location and size from the measured ones. In recent years the k-ε
model has been more widely used in numerical simulation of flow fields in turbomachinery

Zhao Yu and Wenqi He are research assistants, Song Li is a lecturer, Weixiong Wang is a graduate student, and Dong-
tao Huang and Zhichi Zhu are professors at the Fluid Acoustics Lab, Department of Engineering Mechanics, Tsinghua
University, Beijing, China.

263
264 HVAC&R RESEARCH

engineering. It is well known that the k-ε model is only applicable to a flow with small
separation and is unable to capture the details of separation characteristics and the vortex in a
large separation region. But for good aerodynamical design and near design operation, in which
the flow is smoother in the channel and only small separation occurs, the overall peformance
and the approximate velocity and pressure distribution can then be predicted using the k-ε
model. For example, the standard k-ε model has been successfully applied in CFD for
multiblade centrifugal fans (Seo et al. 2003), centrifugal pumps (Gonzalez et al. 2002), and
centrifugal compressors (Zangeneh et al. 2004). It is reported to be capable of predicting the
correct trends of velocity and pressure profiles and gives especially good prediction of overall
performance, such as total pressure rise, shaft power, and efficiency. Other studies by Thakur et
al. (2002) for a centrifugal blower, Muggli et al. (2002) for a mixed flow pump, and Miner
(2000) for axial and mixed flow pumps also proved the validity of the k-ε model in the
prediction of global performance or the shape and magnitude of the velocity and pressure
profiles. To consider the unsteadiness caused by the interaction of the impeller and the scroll,
the “frozen rotor” model and the “mixing plane” model have been developed for quasi-steady
simulation. They are more advanced models than single-blade passage simulation, as they can
reflect the effect of the non-uniformity of flow distribution in different blade passages. Both
“frozen rotor” and “mixing plane” models can provide good prediction for fan performance,
including static pressure rise, power, and efficiency (Thakur et al. 2002). Seo et al. (2003)
studied three-dimensional turbulent flow in a multi-blade (squirrel cage) centrifugal fan by the
“frozen rotor” method, and a finite volume method with the k-ε turbulence model was
employed. The result showed an agreement with the experiment, especially at higher flow rate
coefficients.
Fully unsteady simulation and large eddy simulation (LES) has also been applied to the simu-
lation of flow in centrifugal pumps. Pedersen et al. (2003) used LES to simulate the flow in a
centrifugal pump, and the detailed flow field agreed well with the PIV and LDV results. Gonza-
lez et al. (2002) used unsteady simulation to study the interaction of the impeller and the scroll
of a centrifugal pump and successfully predicted the fluctuation of pressure with time. However,
these advanced skills are too expensive and too complicated to be employed in practical fan
design.
The application of CFD is developing in the design method for centrifugal fans and pumps.
Lin et al. (2002) numerically simulated the internal flow in a forward-curved blade centrifugal
cooling fan used in a laptop computer. CFD and experimental validation were used for optimum
studying of the flow under three blade angles. Visser et al. (2000) employed potential flow
approximation and CFD calculation to analyze and optimize a centrifugal pump impeller. The
redesign of the blade angle resulted in a better cavitaion performance. Goto et al. (2002) used an
inverse design method and CFD to re-distribute the load on the vane of a pump diffuser. Both
numerical simulation and experiment proved that the modification could suppress the flow
separation and improve the pump performance.
Usually the impeller and scroll are considered as the main parts of the centrifugal fan, while
its other part—the inlet—is not considered important for fan performance. Thus, no design
method for the inlet has been published, and it is chosen or fitted based entirely on the designer’s
experience and fan performance test. In fact, the flow field at the entrance of the impeller is very
important to the flow in the whole fan passage and, thus, to the fan performance, and the effects
of the inlet shape and the gap between impeller and inlet on the flow field at the impeller
entrance are serious. Therefore, it is necessary to do numerical simulation of flow field for a
whole fan or pump, including all three parts—impeller, scroll, and inlet—and to consider the
gap between the impeller and inlet. But in spite of the advancement of CFD, the authors have
not yet found a paper about this. In addition to underestimating the effect of the inlet on the fan
VOLUME 11, NUMBER 2, APRIL 2005 265

aerodynamic performance, it may be there are still some difficulties in applying the CFD tech-
nique to the centrifugal fan:
• The geometric model setup and mesh generation for a whole centrifugal fan with the impeller
gap is more difficult. To ensure the computation results, good accuracy is required in order to
predict the small difference in fan performances for different aerodynamic designs. For exam-
ple, the dimension of the gap is extremely small compared to the dimension of the fan. For an
impeller with a diameter of 0.8 m, the gap is typically about 2.5 mm in the radial direction and
about 5 mm in the axial direction. It results not only in difficulties in mesh generation but also
difficulties in computation accuracy and stability. This is a hard task and needs some original-
ity and know-how.
• The precondition for a successful numerical simulation is that there be a good fan aerody-
namic design to make sure only a small separation occurs in its passage. It is because the most
useful turbulent model in practice, including in complicated flow, is the k-ε model, which is
only applicable to the flow with small separation as indicated before. Furthermore, CFD
results should be appropriately analyzed and applied into the next aerodynamic design cycle
in order to further improve the fan performance. Therefore, it needs a CFD engineer who is
proficient not only in the computation of complicated flow fields but also in the fan aerody-
namic design. This condition is not easily fulfilled.
• The agreement between the numerical results and test data is also difficult because the numer-
ical result is very sensitive to the numerical method, including mesh generation, boundary
condition handling, turbulent model selecting, and data processing. The quantitative agree-
ment about the details of the flow field in the passage, such as velocity or pressure distribu-
tion, is hardly possible. But according to our experience, the numerical prediction of the
integral parameters such as fan performance, including flow rate, total or static pressure rise,
and efficiency, could agree with test data well enough in practice if the numerical handling
has been modified several times by use of the prototype fan test data. Certainly it also needs
much time and cost to reach the agreement.
Recently a numerical simulation of flow field for a whole centrifugal fan, including all three
parts—impeller, scroll, and inlet—and considering the gap between the impeller and inlet, has
been completed by the authors. The error between the numerical prediction of the fan perfor-
mance and the test data at design operation is within 3%. In our lab it has been combined with
the aerodynamic design method to develop a modern optimum design method for a centrifugal
fan to replace the old one. Several fan products with excellent fan performance are available in
the market using this new design method. The new design method and its related numerical
method will be explained in detail in the next sections.
With these methods, the effect of different parameters could be studied during the design
stage for optimization. It is well known that the impeller is the heart of a fan, and every fan
designer always does his/her best to improve the flow in the blade passage at first in order to
raise the fan efficiency and to decrease its noise level. The flow field at the impeller entry
directly affects the flow in the blade passage and, thus, has crucial influence on the fan perfor-
mance. Besides the shape of inlet, there are two factors mainly affecting the flow field at the
impeller entrance. One is the blade inlet angle, and the other is the gap between the impeller and
the inlet.
An inappropriate blade inlet angle will cause additional energy loss in the fan passage. Too
small a blade inlet angle also increases flow separation and non-uniformity in blade passages.
According to one-dimensional ideal flow theory, Eck deduced an optimum flow angle at blade
inlet β1 as 35.26°, when there is a minimum relative flow velocity at the blade entry under the
condition of the same flow rate. It is based on Eck’s assumption that the less the relative flow
velocity at the blade entry, the less the energy loss in the impeller passage. It should be noted
266 HVAC&R RESEARCH

that the flow angle at blade inlet β1 is not a blade inlet geometry angle β1j (called blade inlet
angle in the following). The latter minus β1 leaves an angle called attack angle α. In the
conventional method of fan aerodynamic design, α is usually chosen as 2-3 in. by experience.
Therefore, the blade inlet angle β1j commonly should be 37°-38°. However, in many centrifugal
fans with high efficiency developed in recent years by the authors, this angle ranges from 27° to
32° for backward-curved blade fans. It is far from Eck’s theory. Perhaps the latter might be out
of date. Therefore, it is necessary to study the effect of β1j on fan performance using modern
CFD technique.
In centrifugal fans, the rotating impeller and the stationary inlet are separated by a small gap
called impeller entry gap or impeller gap. A small fraction of the flow rate discharged from the
impeller into the scroll will flow along the front disc of the impeller and back into the impeller
again through the gap. This is a secondary flow in the centrifugal fan and results in the leakage
of flow rate and the increase of loss and, thus, in the reduction of fan efficiency and total pres-
sure rise. The main size of the gap is the difference between the radius of the impeller at the
entrance and the radius of the inlet at the exit and is called gap width δ. Certainly at best δ
should be as small as possible. But too small δ is very difficult and expensive to manufacture.
No one has done quantitative analysis about its effect on fan performance because it is too diffi-
cult using the old method. Now it is possible to study this work using modern CFD technique.
The effects of these two factors on fan performance will be discussed later in detail.

MODERN OPTIMUM DESIGN METHOD

An optimum design process should be completed by several cycles. Each cycle is composed
of three successive steps as follows.
1. Aerodynamic Design
According to the fan parameters at design operation, such as total pressure or static pressure
rise, flow rate, impeller diameter, speed and air temperature, etc., and the other restriction condi-
tions, such as the restriction about total fan size, etc., a preliminary aerodynamic design is
worked out based on the conventional method and design experience to make sure the flow in all
main parts of the designed fan is good and only a small separation may occur in all aerodynamic
passages.
The centrifugal fan studied here is a backward curved blade fan. The design parameters of the
fan are given in Table 1.
2. CFD Simulation
Numerical simulation of the three-dimensional turbulent flow field is carried out for a whole
centrifugal fan including inlet, impeller, and scroll and considering the impeller gap. The numer-
ical result should be analyzed in detail to find out how to further improve the next aerodynamic
design cycle.
3. Modification of Some Empirical Design Criteria to Optimize the Aerodynamic Design
Some empirical design criteria about how to choose the blade inlet angle, the ratio of entry
and exit axial breadths of the impeller, and the coefficient of acceleration at the impeller entry,

Table 1. Design Parameters of the Fan

Flow rate coefficient Φ = 0.15
Total pressure coefficient Ψ = 1.30
Impeller diameter d2 = 0.8 m
Rotation speed n = 960 rpm
Air temperature 20°C
VOLUME 11, NUMBER 2, APRIL 2005 267

etc., are used in the aerodynamic design. These criteria are originally obtained by the one- or
two-dimensional ideal flow theory and greatly affect the geometrical shape of the passages and
fan performance. Now these criteria could be improved by CFD simulation and are modified
using the information obtained from the CFD result. Moreover, the modified criteria will be
used to redesign the fan. The process then goes back to step 1 and one design cycle is finished.
In the section, “Effects of the Blade Inlet Angle,” the blade inlet angle is also used as an
example of the criteria to be modified by CFD in order to improve the fan performance. Other
criteria about the impeller, such as the ratio of entry and exit axial breadths of the impeller and
the coefficient of acceleration at the impeller entry, should also be modified using a similar
method in the design.
After repeating several cycles above, the fan performance is gradually optimized. Finally the
design is terminated when a satisfactory fan performance is achieved.

NUMERICAL METHOD
Geometrical Model Setup and Mesh Generation
In a centrifugal fan, the impeller rotates at a high speed while the inlet is stationary. There is a
radial gap between them at the entry of the impeller. The radius of the inlet at exit is slightly
smaller than that of the impeller at entrance and the inlet is inserted a short axial distance into
the impeller. The geometric models for the computational domain used in this paper are divided
into two types. One is Model I, which is constructed without the gap (δ = 0), and the other is
Model II, which is built with this gap. The geometrical models and their meshes are generated
using a commercial software GAMBIT.
For the geometric model I shown in Figure 1, the whole computational domain is composed
of five blocks. The fan passage is divided into three blocks: the impeller, the inlet, and the scroll.
The other two blocks—inlet duct and outlet duct—are attached to the inlet and scroll of the fan,
respectively. These ducts were artificially added in order to simulate the real situation in the
experiment and to provide better boundary conditions for numerical computation. For the com-

Figure 1. Computational mesh for geometric model I (without gap).

268 HVAC&R RESEARCH

Table 2. Grid Dependency Test (Final Design, δ = 0 mm)

Grid Type Impeller Grid Total Grid Total Pressure Shaft Total
Number Number (Pa) Power Efficiency (%)
(KW)
Regular grid 124,844 256,916 1283 4.58 85.59
Fine grid 582,792 795,296 1288 4.61 85.37
Further refined 1,107,615 1,407,721 1291 4.63 85.22
grid

plicated fan passage, the skew degree of the unstructured mesh is usually less than that of struc-
tured mesh, so that computational convergence would be improved. Therefore, unstructured
mesh is used for the whole fan passages in these two models. The impeller region is constructed
using the tetrahedral cells with a refinement near the blade surface. The resulting value of nondi-
mensional friction length y* is of a magnitude between 30 and 60 on the blades’ surfaces. The
scroll and inlet regions are also defined by the tetrahedral cells. The wedge cells are used for the
regions of inlet and outlet ducts.
To validate the dependency of the grid resolutions, grid dependency tests have been per-
formed. Here the final design in Model I (without gap) with three different grid sizes is used as
an example of the test. The refinement is mainly in the impeller region, as it is the most impor-
tant part of a fan. Table 2 shows that the differences between the three results are very small.
The further refined grid has over eight times more grid numbers than the regular grid in the
impeller region, and the differences caused by grid size are kept below 0.62%, 1.09%, and
0.04% for total pressure, shaft power, and total efficiency, respectively. The effect of grid size
on the numerical results of overall performance could thus be neglected in engineering applica-
tion, if the regular or fine grids are adopted.
In this paper, the regular grid is used in geometric model I for the optimal design process to
save computation time and resources. Although this mesh is not good enough to solve the flow
details in the boundary layer, it is sufficient for predicting the integral quantities such as flow
rate, pressure rise, and efficiency of the fan. The fine grid is used in geometric model II (with
gap) for sample fan design and for the discussion of the gap effect. Because of the gap grid the
total grid number of model II is much more than the fine grid number shown in Table 2.
The geometric model II is almost the same as the model I, except the additional block of the
gap region. The narrow space inside the gap is carefully meshed using the wedge cells. There
are 3 rows of meshes in the radial direction, 10 in the axial direction, and 800 in the
circumferential direction, respectively. The small wedge cells in the gap region gradually and
smoothly grow and transfer to the normal size cells on the impeller and inlet surface. The mesh
scheme in the other parts of Model II is the same as that in Model I. Therefore, Figure 2 only
shows the mesh in the regions of the impeller and inlet with an enlarged view of the mesh
enveloping the gap. For model II of the sample fan, the grid number is 642,856 in impeller,
158,720 in scroll, and 80,482 in inlet. The inlet duct and outlet duct have 18,150 and 26,200
cells, respectively. So the grid number for Model II is about 974,408 altogether.

Numerical Model
The numerical computation was performed using a commercial CFD software FLUENT. The
governing equations to be solved here are the three-dimensional, incompressible Reynolds aver-
aged Navier-Stokes equations. The standard k-ε model with a standard wall function is adopted
as a turbulent model in the paper. The numerical algorithm is SIMPLE. In the algorithm the
pressure parameter is discretized by use of the second-order central difference and the
parameters of velocity k and ε are discretized using the second-order upwind scheme.
VOLUME 11, NUMBER 2, APRIL 2005 269

Figure 2. Mesh in the gap at impeller entry for geometric model II (with gap).

Boundary Conditions
Quasi-steady solution is obtained by use of the multiple reference frame coordinates, which is
also known as “frozen rotor” model in other CFD packages. The impeller region is defined as a
rotating reference frame with constant angular velocity, and it is placed at a fixed
circumferential position relative to the cutoff. Other blocks are defined in a stationary reference
frame. The flow fields of the rotor and the other blocks are separately calculated in their own
coordinates and their information is exchanged locally on the interfaces through a reference
frame transformation so as to consider their interactions. All transient terms are neglected.
For the present case with 20 blades, there is 18° between the two adjacent blades. In order to
investigate the effect of the impeller position relative to the cutoff on the performance, three
relative angular positions (0°, 6°, 12°) were calculated, and the results are listed in Table 3.
Fortunately, the results of fan performance for three different cases were nearly the same. It
means that the quasi-steady simulation can be used to predict the fan performance and that the
effect of the impeller’s relative position on the fan performance could be neglected in the
engineering. Therefore, only one impeller circumferential position relative to the cutoff needs to
be calculated. But it is noted that this conclusion is correct only for the numerical simulation of
flow field for a whole centrifugal fan channel, but it is not correct for a single passage of the
impeller-volute channel.

Table 3. Results of Different Angular Positions of the Impeller

Angular Position Total Pressure Shaft Power Total
(degree) (Pa) (KW) Efficiency
(%)
0 1288 4.61 85.37
6 1292 4.61 85.69
12 1289 4.60 85.64
270 HVAC&R RESEARCH

The flow rate is given at the entrance of the inlet duct and there is an assumption of uniform
normal velocity distribution there. The entrance turbulence intensity I is determined using

–1 ⁄ 8
I = 0.16 × Re , (1)

where Re is the Reynolds number adopting the inlet diameter and uniform normal velocity
above as the characteristic length and characteristic velocity, respectively. The flow is assumed
to be fully developed flow at the exit of the outlet duct. Non-slip conditions are imposed on the
walls.
In Model II, the two end sections of the gap region are assigned as “interior” ones so that the
leakage flow can go into or out of the gap region freely.

Convergence Criterion
It appears from our computation that the numerical results would change periodically when a
convergent solution is approximately achieved and the computation goes on. This might be
caused by the unsteady nature of the actual flow in centrifugal fans and the iterative
computation. Sometimes the predicted performance varies so much that it casts doubt on the
accuracy of the prediction. This usually happens if the fan is not well designed or meshed. In
order to increase the credibility of the computation, a strict standard to check the convergence is
employed here. It requires that two convergence criteria have to be satisfied simultaneously: the
residuals of all quantities must fall below 10-5 and the periodic variation of the pressure, which
is the most changed parameter in the computation, must be below 1%. Commonly, a typical
computation could be completed about in ten hours on a Pentium IV1700 PC.

Processing of Results
The fan performance is described by parameters such as total pressure rise, shaft power (or
called brake power), and efficiency at different given flow rates. The total pressure rise (or
called total pressure of the fan) is defined as the average total pressure difference between the
inlet section (entrance of the inlet ring) and the outlet section (exit of the scroll) of the fan. The
average total pressure of a section is calculated with mass flow rate weighted average from all of
the cells on that section.
The shaft power is calculated by multiplying the torque imposed on the impeller and its speed.
The total efficiency of the fan is defined as

P0 × Q
η = ----------------------- , (2)
W × 3600

where η is total (pressure) efficiency, P0 is fan total pressure (rise) in Pa, Q is flow rate of fan in
m3/h, and W is shaft power in watts.

EXPERIMENTAL VALIDATION
The performance experiment of the sample fan finally optimally designed was implemented
according to standard ISO 5801:1997. The tested fan connected an inlet duct and an outlet duct
with its inlet and exit, respectively. The static pressure rise, flow rate, and the shaft power of the
fan are obtained by measurement. The total pressure rise of the fan was obtained by the static
pressure rise plus an average dynamic pressure head, the velocity of which is the flow rate of the
fan divided by the exit section area at the exit. The performance curves of the model fan
obtained from the experiment are shown in Figure 3. The best efficiency is 85.4%, where the
flow rate is 11400 m3/h. At the designed flow rate 11000 m3/h, the total pressure efficiency is
VOLUME 11, NUMBER 2, APRIL 2005 271

Figure 3. Performance curves of the model fan.

85.3%, which is about 2% to 3% higher than that of other similar fans with high efficiency
available in the Chinese market. The computational results and the test data for the fan
performance at the design operation is shown in Table 4. The computation agrees well with the
experiment. The errors of total pressure, total efficiency, and shaft power between the numerical
prediction and test data are –0.8%, –1.8%, and +1.8%, respectively.
In recent years the authors have used this numerical method, which covered the standard k-ε
model in simulating various types of fan products, including two centrifugal fans for the GE
company in the US in 2000-2002, three axial fans for another company in the US in 2003, and
6-41 and 7-35 series of centrifugal fans and fans used for fluidized-bed boilers available in
China. These products have all been tested by us and also by the consumers above in the US and
in China and are now applied to the fan industry. The errors of predicted total pressure and total
efficiency at the design operation are all below 3% to 5% and 2% to 3%, respectively. This fact
also proves the validity of the numerical method adopted here.
It is concluded that the numerical simulation above is capable of giving an accurate prediction
of the fan’s aerodynamic performance at design operation within permitted error in practice. By
the way, this paper focuses on the fan aerodynamic performance near the design operation and
does not deal with noise at all. Our CFD results also agree well with the experiment near the
design operation and at the case of larger flow rate. As an example, for the sample fan discussed
in the paper, the error of performance is kept below 5% from 90% to 120% of the design flow
rate and below 5% to 10% between 120% and 160% of the design flow rate. But the agreement
is bad at a small flow rate.
272 HVAC&R RESEARCH

Table 4. Numerical and Experimental Results at Design Operation (δ = 2.5 mm)

Flow Rate Total Pressure Total Efficiency Shaft Power
(m3/h) (Pa) (%) (KW)
Experiment 11000 1245 85.3 4.42
CFD (2.5 mm gap) 11000 1235 83.8 4.50

EFFECTS OF THE BLADE INLET ANGLE

The blade inlet angle β1j is an important parameter in impeller design. In fact, in order to
guarantee the flow rate, the diameter of the impeller at its entry and the diameter of the inlet at
its exit are changed with the change of β1j and, thus, the inlet shape is also changed. All of these
changes will affect the flow and energy loss in the whole fan passage greatly. Using numerical
simulation of flow field for the whole fan as depicted above, the optimum β1j could be selected
to obtain the optimum fan aerodynamic performance. Since the effects of the impeller gap for
different β1j are similar, for the sake of computation time, the width of the gap is assumed zero
in this section. Thus, the geometric model used here is Model I (δ = 0). The numerical
simulation was carried out under the conditions in which only β1j was changed, ranging from
24° to 36°, and the other design parameters remain unchanged. Finally, β1j was adopted as 27°
in the designed sample fan. The effects of β1j will be discussed in detail below.

Effect of β1j on the Flow Pattern in the Impeller and Scroll Passages
Figures 4a and 4b show three-dimensional streamlines in a part of the impeller-scroll
passages near cutoff at β1j = 25 and β1j = 27, respectively. Figure 4a shows the strong vortices
filled in three blade passages at 1/2 blade height near the back plate of the impeller. The vortices
start to form near the entrance of the blade passages. Then they develop along the passage and
gradually move to the blade suction side. Their effect becomes weaker near the outlet of the
blade passages. The large vortices formed near the cutoff of the scroll result in severe
nonuniformity of flow rate distribution in the 20 blade passages and considerable total pressure
loss (see Figure 6 and Table 6). On the contrary, there are no such vortices seen in Figure 4b,
and the flow in all passages is smooth. It is noted that Figure 4a also basically represents the
flow pattern at β1j < 27 and Figure 4b represents the flow pattern at β1j > 27, so here only these
two three-dimensional streamline pictures are given.
Figure 5 shows two-dimensional streamlines in some blade passages at β1j = 24, 26, 27, and
33, respectively. The flow patterns can be used to confirm the discription above.

Effect of β1j on the Non-Uniformity of Flow Rate in the Blade Passage

The non-uniformity of flow rate in the passages is caused by the asymmetry of the scroll, but
it could be improved or deteriorated due to selection of different design parameters. Figure 6
shows the distribution of the flow rate in blade passages at different β1j, and the sequence of the
passages is shown in Figure 7. Here the designed average flow rate for each blade passage is
550 m3/h. It is seen from Figure 6 that the integral deviation from 550 m3/h for all of 20 blade
passages is minimum at β1j = 27°, the flow rate in about 15 blade passages being near 550 m3/h.
Only five passages near the cutoff are affected by the impeller-scroll interaction. The serious
non-uniformity of flow rate occurs at β1j < 27°, where three or four passages near the cutoff are
blocked by strong vortices shown in Figures 5a and 5b, and their flow rates are less than half the
designed value. When β1j is increased from β1j = 27°, the non-uniformity of flow rate continues
to increase and more passages deviate from the designed flow rate of 550 m3/h, but the integral
deviation is remarkably smaller than that at β1j < 27°.
VOLUME 11, NUMBER 2, APRIL 2005 273

(a)

(b)

Figure 4. Three-dimensional streamline in a part of impeller-scroll passages near cutoff:

(a) β1j = 25°, (b) β1j = 27°.

Effects of β1j on Energy Loss and Fan Performance

In order to discuss the effects of β1j on the energy loss and fan performance in detail, it is
necessary to analyze the energy loss at every stage of the whole fan quantitatively. The method
to calculate the energy loss is shown in Figure 8 and Table 5, where 1-4, 1-2, 2, 6-3, 3-4, and 5-6
are inlet-exit sections of the fan, inlet ring, impeller, scroll, and impeller gap, respectively; P is
an average total pressure of a section, which is calculated with mass flow rate weighted average
from all the cells on that section; Q and q are the flow rate of fan and the leakage flow rate in the
gap, respectively; and W is the shaft power obtained from a computational function of FLUENT.
It is pointed out that the flow rate in the impeller is Q plus q, and the impeller energy loss
percent is obtained by 1–ηimp, where ηimp is the impeller efficency, shown in Table 5. The gap
energy loss is used in next section, so it will be explained there.
Table 6 is very interesting and shows the numerical results of energy loss in different pas-
sages and fan performance for the design fan at different β1j and δ = 0. Its original data are
shown in Table 7 in next section. Following are some comments on Table 6:

1. Both the best total pressure and the best efficiency appear at β1j = 27°, which was adopted in
the final design of the sample fan.
274 HVAC&R RESEARCH

(a)

(b)

Figure 5. Two-dimensional streamlines in some blade passages at different β1j: (a) β1j =
24°, (b) β1j = 26°.
VOLUME 11, NUMBER 2, APRIL 2005 275

(c)

(d)

Figure 5. (continued) Two-dimensional streamlines in some blade passages at different

β1j: (c) β1j = 27°, (d) β1j = 33°.
276 HVAC&R RESEARCH

Figure 6. Flow rate distribution in different blade passages at different β1j.

Figure 7. Sequence of blade passages.

VOLUME 11, NUMBER 2, APRIL 2005 277

Table 5. Formulas for Calculation of Energy Loss or Efficiency for Different Passages
Passage Total Pressure Flow Energy Loss or Effective Energy Loss Percent or Impeller
Loss or Rise Rate Power Obtained Efficiency or Fan Efficiency
Inlet P1-P2 (loss) Q (P1-P2) × Q (P1-P2) × Q/W (loss)
Impeller P3-P2 (rise) Q +q (P3-P2) × Q+(P3-P6) × q [(P3-P2) × Q+(P3-P6) × q] /W
Scroll P3-P4 (loss) Q (P3-P4) × Q (P3-P4) × Q/W (loss)
Gap P3-P6 (loss) q (P3-P6) × q (P3-P6) × q /W (loss)
Fan P4-P1 (rise) Q (P4-P1) × Q (P4-P1) × Q/W

Figure 8. Sketch for calculation of the energy loss in each passage of a centrifugal fan.

2. The energy loss in the impeller passage is the largest one in the three main parts of the fan,
which is composed of three portions: (1) the energy loss in the passage from the exit of the
inlet to the entry of the blade, called impeller loss; (2) the energy loss in the passage from the
entry of the blade to the exit of the blade, called impeller loss or blade passage loss; (3) the
energy loss in the passage from the exit of the blade to the exit of the impeller, i.e., passage of
the vaneless diffuser, called impeller loss, shown in Table 5. The minimum loss in the whole
impeller passage is 7.5% at β1j = 27°, and the loss increases monotonously as β1j deviates the
optimum value. The increase of impeller loss (1) means the bad flow near the entry of the
impeller at an inappropriate blade inlet angle. Impeller loss (2) reflects the loss in the blade
region and its value is closely related to the appearance of the separate flow in the blade
passages. For β1j > 27°, the more serious the separation is, the larger the impeller loss (2) is.
For β1j > 27°, where there is no obvious separation flow, the value of impeller loss almost
remains unchanged (see Figures 4 and 5). Impeller loss (3) is actually the loss in the vaneless
diffuser and also reaches its minimum near β1j = 27°.
3. The scroll loss is also considerable. Its minimum value is 6.0% also at β1j = 27°. The loss also
increases monotonously as β1j deviates from the optimum value. The reason is not yet clear
and it is necessary to analyze the flow field in the whole scroll passage in detail. The
non-uniformity of the outflow of the impeller shown in Figure 6 might be a factor.
278 HVAC&R RESEARCH

Table 6. Energy Losses in Different Passages and Fan Performance at Different β1j (δ = 0)
β1j Impeller Impeller Impeller Total Scroll Inlet Total Total
(deg) Loss (1) Loss (2) Loss (3) Impeller Loss Loss Efficiency Pressure
(%) (%) (%) Loss (%) (%) (%) (Pa)
(%)
24 1.9 7.5 1.0 10.4 7.3 0.5 81.8 1213
25 1.9 7.0 1.1 10.0 7.3 0.6 82.1 1222
26 1.9 6.3 0.9 9.1 7.1 0.7 83.1 1239
27 1.9 4.9 0.7 7.5 6.0 0.9 85.6 1283
29 2.1 5.0 0.5 7.6 6.7 1.2 84.5 1267
31 2.4 5.0 0.6 8.0 6.9 1.6 83.5 1252
33 2.8 5.0 0.6 8.4 7.2 2.1 82.3 1232
35 2.9 4.9 0.6 8.4 7.5 3.0 81.1 1218
36 3.2 5.0 0.9 9.1 7.8 3.4 79.7 1193

Table 7. Total Pressure (Absolute Value) in Different Passages, Leakage Flow Rate and
Shaft Power at Different δ (Q =11000 m3/h, β1j = 27)
δ (mm) P1 (Pa) P2 (Pa) P3 (Pa) P4 (Pa) P5 (Pa) P6 (Pa) q (m3/h) q /Q (%) W (KW)
0 -531 -544 842 752 170 -524 0 0 4.58
2 -524 -533 805 719 63 -121 307 2.79 4.50
2.5 -514 -524 812 721 105 -54 396 3.60 4.50

4. The inlet loss at β1j = 27° is 0.9%. It is normally small compared to the impeller and scroll
losses at an appropriate design, but it increases with the increase of β1j monotonously and
rapidly, as shown in Table 5. It can be explained according to an important formula
determining the design flow rate, deduced from one-dimensional ideal flow theory (Eck
1973), which is

3
d1 Φ
----- = ------------- × ε , (3)
d2
3 tanβ 1

where d1and d2 are the entry and exit diameters of the blade, respectively; the latter is also
called impeller diameter. Φ is the flow rate coefficient, and ε and β1 are the acceleration
coefficient and flow angle at the blade entry, respectively. In β1 = β1j – α, α is called attack
angle. In our design, d2 = 0.8 m and Φ = 0.15 are given before the design, and ε = d1 / 4b1 =
0.68 and α = 0 (β1 = β1j) are selected optimally. Therfore, d1 is decreased monotonously with
the increase of β1j. In our design, both the diameters of the blade entry and the impeller entry
are the same. Since d1 minus two times the radial gap between inlet and impeller is the exit
diameter of the inlet, the latter is also decreased monotonously with the increase of β1j. Then
the flow velocity in the inlet and the inlet length are increased. Therefore, the inlet loss is
sharply increased with increase of β1j, as shown in Table 5.
5. The effect of β1j on the fan performance is remarkable. It is seen from Table 5 that the benefit
of the total pressure and its efficiency can be obtained up to 6% and 5%, respectively, at the
optimum β1j. Therefore, β1j is an important design parameter for improvement of fan
performance.
VOLUME 11, NUMBER 2, APRIL 2005 279

6. The total pressure and its efficiency at β1j = 27° is about 5% better than that at β1j = 35°.
(Note β1 = β1j in the present case) so that the present method to optimally select β1j by use of
CFD technique is much better than Eck’s theory about the blade inlet angle.

EFFECTS OF THE GAP AT IMPELLER ENTRY

As described in the introduction above, the gap size should be as small as possible. For the
designed fan, whose impeller diameter is 800 mm, considering its commercial application and
market competition, the possible small size of δ is 2.5 or 2.0 mm. Therefore, in this section the
effect of δ is discussed using numerical simulation of the flow field for a whole fan only for
three cases of δ = 0, 2.0 mm, and 2.5 mm, and the mesh generation is based on the geometric
model II.

Effect of δ on the Flow Pattern in the Passages

Figure 9 shows two-dimensional streamlines in the meridian plane of the fan passage,
especially near the front disc of the impeller and corresponding scroll passages, at δ = 0, 2.0
mm, and 2.5 mm, respectively. Obviously the flow in both the impeller and the scroll is
remarkably different if there is an impeller gap or not. Thus, it will affect the total pressure loss
in both the impeller and the scroll.

Effects of δ on the Energy Loss and Fan Performance

Table 7 shows the original data obtained from the numerical results about the total pressure in
different passages, leakage flow rate, and shaft power at different δ. Table 8 denotes the total
pressure rise (if plus sign) or loss (if minus sign) and energy loss in different passages and fan
performance at different δ. Following are some comments on Table 8:
1. The effect of on the fan performance is remarkable. The fan total efficiency is decreased by
1.2% and 1.8% for δ = 2.0 mm and δ = 2.5 mm, respectively, mainly because of the gap
energy loss. Meanwhile, the total pressure of the fan is decreased by 40 Pa (3.1%) and 48 Pa
(3.7%) for δ = 2.0 mm and δ = 2.5 mm, respectively. This is because the total efficiency is
decreased and the flow rate in the impeller is increased a little with the increase of the gap
width when the flow rate of the fan remains the same as Q = 11000 m3/h.
2. As seen from Figure 9, a small fraction of the total volume discharged from the impeller into
the scroll will pass the channel between the scroll and the front disk and reenter the impeller
through the gap to form a secondary flow. It is called the leakage flow rate, which could be
estimated by Eck (1973) using a simplified formula as

q = 3600πδd 1 4P 0 ⁄ 3ρ . (4)

Substituting the parameters of the designed fan for the formula (Equation 4), q is 425.8 m3/h
and 532.2 m3/h for δ = 2.0 mm and δ = 2.5 mm, respectively. But according to the numerical
simulation results from Table 7, q is 300.9 m3/h and 396.2 m3/h for δ = 2.0 mm and δ =
2.5 mm, respectively. Eck’s formula overestimates the leakage flow rate by about 42% and
35% for δ = 2.0 mm and δ = 2.5 mm, respectively.
3. The gap energy loss can be determined using the formula in Table 5, once the total pressure
rise through the impeller and the leakage flow rate are obtained by CFD. They are 1.75% and
2.12% for δ = 2.0 mm and δ = 2.5 mm, respectively. The distribution of gap loss in its two
passages for different δ is shown in Table 9. It is seen that the main loss occurs in the second
flow in the passage between the scroll and the front disk where a big vortex ring appears, as
shown in the Figure 9b. More than 80% of the impeller total pressure exhausts in that region.
Only less than 20% of the impeller total pressure exhausts in the gap flow itself, which was
not expected before.
280 HVAC&R RESEARCH

(a) (b)

(c) (d)

Figure 9. Two-dimensional streamlines in the meridian plane: (a) δ = 0; (b) δ = 2.5 mm;
(c) near the gap, δ = 2.0 mm; and (d) near the gap, δ = 2.5 mm.
VOLUME 11, NUMBER 2, APRIL 2005 281

Table 8. Total Pressure Rise and Energy Losses in Different Passages and Fan
Performance at Different δ (Q = 11000 m3/h, β1j =27)
δ Total Pressure Rise (Pa) Energy Loss or Fan Total Efficiency (%)
(mm)
Inlet Impeller Scroll Gap Fan Inlet Impeller Scroll Gap Fan
0 -13 1386 -90 0 1283 0.86 7.54 6.01 0 85.59
2 -9 1338 -86 -926 1243 0.61 7.39 5.84 1.75 84.41
2.5 -10 1336 -91 -866 1235 0.67 7.22 6.18 2.12 83.81

Table 9. Distribution of Gap Energy Loss in Two Passages (%)

δ (mm) 5-6 (Gap) 3-5 (Scroll) 3-6 (Total)
2.0 0.35 1.40 1.75
2.5 0.39 1.73 2.12

4. The existence of the gap can be helpful to improve the flow in the impeller and the inlet. As
seen in Table 8, the impeller loss and the inlet loss are decreased with the increase of the gap
width. It can be explained that the reentering leakage flow could improve the main stream
near the impeller entry and the exit of the inlet. In fact, the flow channel transferred from the
axial to the centrifugal direction is designed as highly divergent for the sample fan, since the
coefficient of acceleration at the impeller entry is only 0.68. Meanwhile, the centrifugal force
at the impeller entry is smaller so that the retarded air in the main stream could accumulate
near the front disk at the impeller entry and the exit of the inlet, and, thus, the thickness of the
boundary layer there will be increased even to form separation flow. The reentering leakage
flow through the gap with high speed can supply additional energy of P6 – P2, such as 412 Pa
and 470 Pa for δ = 2.0 mm and δ = 2.5mm, respectively, to the retarded air and carry it away,
as seen in Figures 9c and Figure 9d; thus, the main flow in the impeller and the inlet can be
improved. It is very similar to the blowing technique used for a slotted wing, i.e., boundary
layer control technique, in aeronautical engineering to improve the separate flow at the rear
edge of the wing (Schlichting 1968). Therefore, the total impeller loss and the inlet loss for
δ =2.0 mm are 0.15% and 0.25% less than that without a gap and for δ = 2.5 mm are 0.19%
and 0.32%, respectively.
5. Someone considers the magnitude of gap effect on the fan efficiency just being the leakage
flow rate percent. It means if the latter is 3%, the fan efficiency should decrease by 3%. But
now from the numerical simulation for the sample fan, the leakage flow rate percent are
2.79% (δ = 2.0 mm) and 3.60% (δ = 2.5 mm), but the reductions of the efficiency are only
1.18% and 1.78%, respectively, mainly because the work done by the impeller for the
leakage flow rate is much less than that for the main flow rate. For example, the total pressure
rise for the former is 926 Pa (δ = 2.0 mm) and 966 Pa (δ = 2.5 mm), but the total pressure rise
for the latter is 1338 Pa (δ = 2.0 mm) and 1322 Pa (δ = 2.5 mm), respectively. Another reason
is the energy loss of the impeller and inlet are decreased, owing to the gap, as indicated
above.

CONCLUSIONS
1. The numerical computation of flow field for a whole centrifugal fan, including all three
parts—impeller, scroll, and inlet—and considering the impeller gap between the impeller and
inlet is completed in order to consider the interaction of these three parts and the effect of the
impeller gap. Thus, the recent numerical simulation of flow field for a centrifugal fan is fur-
ther improved. The computational results are validated by the data from the performance
282 HVAC&R RESEARCH

experiment of the sample fan described in this paper and other centrifugal fans. The errors of
the numerical prediction for total pressure and total efficiency of the fan at the design opera-
tion are within 3% and 2%, respectively.
2. A modern optimum design method for a centrifugal fan is presented in this paper, which
combined a conventional aerodynamic design method with the numerical computation of
flow field for a whole centrifugal fan and the optimum selection of some main design param-
eters, such as the blade inlet angle, the ratio of entry and exit axial breadths of the impeller,
and the coefficient of acceleration at the impeller entry, etc. This new method not only con-
siders the three-dimensional complicated flow in a centrifugal fan but also considers the
interactions of its three main parts and the effect of the impeller gap on the flow. Further-
more, it can predict the fan performance near design operation. In this paper, excellent per-
formance of a sample fan designed by this method is achieved. Not only the performance
completely satisfies the design request, but also the fan total pressure efficiency reaches
85.3%, which is about 2% to 3% higher than that of other similar fans with high efficiency
available in the Chinese market.
3. The effect of β1j on the fan performance is remarkable. Under the conditions described in the
paper, the benefit of the total pressure and its efficiency can be obtained up to 6% and 5%,
respectively, at the optimum β1j, being 27°, if only β1j is changed ranging from 24° to 36°.
Therefore, β1j is also an important design parameter for improvement of fan performance.
4. The total pressure and its efficiency at β1j = 27° is about 5% higher than that at β1j = 35°,
which is the blade inlet angle suggested by Eck (Note: β1 = β1j in the present case), so the
presented method to optimally select β1j by use of CFD technique is much better than Eck’s
theory about the blade inlet angle.
5. According to the numerical simulation results, the percentage of the leakage flow rate in the
impeller gap is 2.79% and 3.6% for δ = 2.0 mm and δ = 2.5 mm, respectively, which is about
42% and 35% less than that from Eck’s formula for δ = 2.0 mm and δ = 2.5 mm, respectively.
6. The effect of on the fan performance is also remarkable. The fan total efficiency is decreased
1.2% and 1.8%, and the total pressure of the fan is decreased 40 Pa (3.1%) and 48 Pa (3.7%)
for δ = 2.0 mm and δ = 2.5 mm, respectively. It is noted that the reduction of the fan total
efficiency owing to the gap is much less than the percent of leakage flow rate. The reasons
are explained in detail in the comments above.
7. It is very interesting that the total pressure and the energy loss in every passage, including the
gap passage, for a centrifugal fan with excellent performance are given out and analyzed in
detail, which is very useful for fan design in the future.

NOMENCLATURE
b1 = blade entrance breadth P0 = fan total pressure (rise)
d1 = blade entrance diameter Q = flow rate
d2 = impeller diameter q = leakage flow rate
I = turbulence intensity Re = Reynolds number
n = rotational speed W = shaft power
P = total pressure y* = nondimensional friction length

Greek Symbols
α = attack angle ηimp = impeller efficiency
β1 = flow angle at blade inlet ρ = density of air
β1j = blade inlet geometry angle Φ = flow rate coefficient
Ψ = total pressure coefficient
δ = gap width
ε = acceleration coefficient
η = fan total efficiency
VOLUME 11, NUMBER 2, APRIL 2005 283

REFERENCES
Eck, B. 1973. Fans. First English edition, Pergamon Press Ltd., pp. 80-83, 92-95.
Muggli, F.A., P. Holbein, and P. Dupont. 2002. CFD calculation of a mixed flow pump characteristic from
shutoff to maximum flow. ASME J. Fluids Eng., Vol. 124, pp. 798-802.
Gonzalez, J., J. Fernandez, E. Blanco, and C. Santolaria. 2002. Numerical simulation of the dynamic
effects due to impeller-scroll interaction in a centrifugal pump. ASME J. Fluids Eng., Vol. 124, pp.
348-355.
Goto, A., and M. Zangenh. 2002. Hydrodynamic design of pump diffuser using inverse design method and
CFD. ASME J. Fluids Eng., Vol. 124, pp. 319-328.
Lin, S.-C., and C.-L. Huang. 2002. An integrated experimental and numerical study of forward-curved
centrifugal fan. Experimental Thermal and Fluid Science, Vol. 26, pp. 421-434.
M. Zangeneh, M. Schleer, F. Pløger, et al. 2004. Investigation of an inversely designed centrifugal com-
pressor stage—Part I: Design and numerical verification. ASME J. Turbomachinery, Vol. 126, pp.
73-81.
Pedersen, N., P.S. Larsen, and C.B. Jacobsen. 2003. Flow in a centrifugal pump impeller at design and
off-design conditions—Part I: Particle image velocimetry (PIV) and laser doppler velocimetry (LDV)
measurements. ASME, J. Fluids Eng., Vol. 125, pp. 61-72.
Pedersen, N., P.S. Larsen, and C.B. Jacobsen. 2003. Flow in a centrifugal pump impeller at design and
off-design conditions—Part II: Large eddy simulations. ASME, J. Fluids Eng., Vol. 125, pp. 73-83.
Seo, S.-J., K.-Y. Kim, and S.-H. Kang. 2003. Calculations of three-dimensional viscous flow in a multi-
blade centrifugal fan by modeling blade forces. Proceedings of the Institution of Mechanical Engi-
neers, Part A—Journal of Power And Energy 217(A3):287-297.
Schlichting, H. 1968. Boundary-Layer Theory, Sixth English edition, McGraw-Hill Book Company, pp.
364-365.
Miner, S.M. 2000. Evaluation of blade passage analysis using coarse grids. J. Fluids Eng., Vol. 122. pp.
345-348. ASME.
Thakur, S., W. Lin, and J. Wright. 2002. Prediction of flow in centrifugal blowr using quasi-steady
rotor-stator models. Journal of Engineering Mechanics, Vol. 128, pp. 1039-1049.
Visser, F.C., R.J.H. Dijkers, and J.G.H. de Woerd. 2000. Numerical flow-field analysis and design optimi-
zation of a high-energy first-stage centrifugal pump impeller. Computing and Visualization in Science,
Vol 3, pp. 103-108.
Zhang, M.J., M.J. Pomfret, and C.M. Wong. 1996. Three-dimensional viscous flow simulation in a back-
swept centrifugal impeller at the design point. Computers & Fluids, Vol. 25, pp. 497-507.
View publication stats