21st Applied Aerodynamics Conference AIAA 2003-4079

23-26 June 2003, Orlando, Florida

CFD Analysis for Ducted Fans with Validation

I-Chung Chang ∗
NASA Ames Research Center, Moffett Field, CA
R. G. Rajagopalan †
Iowa State University, Ames, IA

An axisymmetric, incompressible Navier-Stokes solver was developed to calculate the

flow field of a ducted fan. The fan effect is modeled via the momentum source concept.
The effects of the spinning fan blades were introduced into the flow field as time-averaged
sources in the momentum equations. These source terms were not known a priori but are
the result of the flow solution at each iteration. This approach simplified the modeling
of the fan blades and provided a very rapid solution procedure for the flow problem.
A new grid generator capable of recognizing the duct and nacelle geometry was de-
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

veloped and tested for several industrial ducted fans of current interest. Prediction of
hover performance for a ducted fan model was calculated and compared with available
wind tunnel test data. The comparison was good. Preliminary results showed that the
Computational Fluid Dynamics (CFD) program could be used as an axial flow analysis
tool for ducted fan design.

Introduction parallel to the duct axis. In addition, the fan is

treated as an actuator disk with constant inflow. The
The concept of ducted fans as a suitable propulsive results obtained with these ideal assumptions are es-
device for certain vertical/short take-off and landing sentially useful only for approximate comparison with
(V/STOL) applications has been explored for more experiments and for predicting the upper limit on the
than a quarter century; the BELL X-22A is a good performance. In such simplified analysis the geomet-
example of a successful application. In the 1970s, the ric characteristics of the duct, blades, hub and guiding
ducted propeller was seen as a “quiet” propulsive de- vanes are not properly represented.
vice with the potential to meet the noise requirements
of the 1980s and beyond. The 1990s saw the ducted fan Some of the problems associated with the representa-
as a viable concept for several unmanned air vehicles tion of the geometry and surfaces of the ducted fan
(UAV) such as Sikorsky’s Cypher and Micro-Craft’s were eliminated by potential flow methods.4–6 How-
Lift Augmented Ducted Fan (LADF). The UAVs have ever, the inherent assumptions of potential flow (in-
a small footprint and are light, making them efficient viscid, irrotational and incompressible) preclude esti-
for vertical take-off and landing. The noise character- mation of important characteristics of the duct such
istics of the ducted fans also make them suitable for as friction drag, location of the center of pressure, and
undetected surveillance. stall characteristics of the duct. Also, due to the lin-
The viscous flow through a ducted fan is very complex. ear nature of the analysis, mutual interference effects
The clearance between the tip of the fan blades and between duct and fan are ignored. As a result, all ba-
the duct is several orders of magnitude smaller than sic design studies were made using costly wind tunnel
the fan radius. It strongly affects the characteristics experiments.
of the flow through the ducted fan and consequently In the early 1960s, extensive wind tunnel experi-
the performance of the ducted fan system. The in- ments7–10 were conducted and a wealth of data on the
flow lip region is another geometric characteristic that thrust, drag and efficiency of scale models and proto-
influences the performance of the system. The shape types was collected. But, specific application was the
of the duct, geometry of the nacelle and position of guiding principle of the work, and parametric study
the fan within the duct all play important roles in the proved costly. In the 1970s, Davis11 reported on the
determination of the performance of the ducted fan. work performed in the United Kingdom to validate
For analysis tools to be successful in ducted fan de- the claim that the ducted propeller is a quiet and
sign, they must be capable of including as much of the cost effective propulsive device. NASA also conducted
detail of the geometry of the system as possible. several experimental studies12–16 to investigate inte-
Simple momentum balance methods1–3 assume uni- grated performance of ducted propellers mounted on
form inflow and outflow with the direction of flow wings and full aircraft. These research efforts were
∗ Senior AIAA member
very informative regarding the control characteristics
† AIAA member of aircraft with ducted propellers. Nevertheless, most
of the experimental research was designed for specific

This material is declared a work of the U.S. Government and is not subject to copyright protection in the United States.
 applications and not for the basic research of the flow z momentum equation:
through the ducted fan.
∂
In the next section, the viscous flow analysis code for (ρVz ) +
a ducted fan will be described. ∂t
    
1 ∂ ∂Vz ∂ 2 ∂Vz
r ρVr Vz − µr + r ρVz − µr
Computational Procedure r ∂r ∂r ∂z ∂z
The choice of the flow solver and the technique used ∂p
=− + Sz0 (6)
for modeling the fan are central to the present CFD ∂z
code. Their essential details are described here. where Sr0 , Sθ0 and Sz0 are source terms through which
Flow Governing Equations the influence of the fan is introduced into the surround-
ing flow field.
The flow around a ducted fan is mostly low-speed,
except near the fan blade tips. The flow field is basi- Discretization of the Flow Governing Equations
cally governed by the unsteady, laminar, incompress-
Consider the following generic governing equation for
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

ible Navier-Stokes equations. For incompressible flow,

an axisymmetric system:
conservation of mass and momentum are sufficient con-
ditions for defining the flow field. The conservation of ∂
mass applied to a fluid passing through an infinitesimal (ρT ) +
∂t
fixed control volume yields the equation of continuity:     
  1 ∂ ∂T ∂ ∂T
∂ρ ~ =0 r ρVr T − µr + r ρVz T − µr
+ ∇ • ρV (1) r ∂r ∂r ∂z ∂z
∂t
=s (7)
Newton’s second law applied to a fluid passing through
the control volume yields the momentum equation: where s represents the terms on the right side of the
momentum equations, excluding the fan sources.
∂  ~ ~V~ = ρf~ + ∇ • Πij + S 0 (2)
ρV + ∇ • ρV Defining the following total fluxes,
∂t
 
∂T
For an axisymmetric system, the flow equations in Jr = r ρVr T − µr
∂r
scalar form are as follows:  
∂T
Jz = r ρVz T − µr
Continuity equation: ∂z
 
1 ∂ ∂
(r ρVr ) + (r ρVz ) = 0 (3) Equation 7 can be rewritten as
r ∂r ∂z
 
∂ 1 ∂ ∂
(ρT ) + (Jr ) + (Jz ) = s (8)
r momentum equation: ∂t r ∂r ∂z

∂
(ρVr ) + Integrating the above equation over the control volume
∂t shown in Figure 1 yields:
    
1 ∂ 2 ∂Vr ∂ ∂Vr
r ρVr − µr + r ρVz Vr − µr Ze Zt Zn Zt1
r ∂r ∂r ∂z ∂z ∂
2 (ρT ) r dt dr dθ dz+
∂p ρVθ µVr ∂t
=− + − 2 + Sr0 (4) w b s to
∂r r r
Zt1 Ze Zt Zn
1 ∂
θ momentum equation: (Jr ) r dr dθ dz dt+
r ∂r
to w b s
∂
(ρVθ ) + Zt1 Ze Zt Zn
∂t 1 ∂
     (Jθ ) r dr dθ dz dt+
1 ∂ ∂Vθ ∂ ∂Vθ r ∂θ
r ρVr Vθ − µr + r ρVz Vθ − µr
r ∂r ∂r ∂z ∂z to w b s

ρVr Vθ µVθ Zt1 Ze Zt Zn

=− − 2 + Sθ0 (5) 1 ∂
r r (Jz ) r dr dθ dz dt
r ∂z
to w b s
 Jn , Js , Je , Jw are given by:

Jn = (Jr )n (r ∆θ ∆z)n
Js = (Jr )s (r ∆θ ∆z)s
Je = (Jz )e (r ∆r ∆θ)e
Jw = (Jz )w (r ∆r ∆θ)w

In a similar fashion the continuity equation (Equa-

tion 3) can be integrated over the control volume:

Zt1 Ze Zt Zn
1 ∂
(rρVr ) r dr dθ dz dt+
r ∂r
to w b s
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

Zt1 Zn Zt Ze
1 ∂
(rρVz ) r dz dθ dr dt = 0
r ∂z
to s b w

The integration yields:

1 o
f (Fn − Fs ) + (1 − f ) (Fn − Fs ) +
Fig. 1 A typical control volume in an axisymmetric 1 o
grid f (Fe − Fw ) + (1 − f ) (Fe − Fw ) = 0 (12)

Zt1 Ze Zt Zn
= s r dr dθ dz dt (9) where the total mass fluxes Fn , Fs , Ft , Fb , Fe and Fw
to w b s are given by:

The integration over the time interval can be general- Fn = (ρVr )n (r ∆θ ∆z)n
ized by the following assumptions: Fs = (ρVr )s (r ∆θ ∆z)s
Fe = (ρVz )e (r ∆θ ∆r)e
Zt1
  Fw = (ρVz )w (r ∆θ ∆r)w
TP dt = f TP 1 + (1 − f )TP o ∆t (10)
to
Multiplying Equation 12 by TP and subtracting it from
where f is a weighting factor between 0 and 1; super- Equation 11 results in the following equation:
script 1 indicates the new values of the corresponding h i
1 o
variables at t=t1 seconds; superscript o refers to old (ρT ) − (ρT )
values of the variables at to . rP ∆r ∆θ ∆z+
∆t
1 o
Using the assumption in Equation 10, the integration f [Jn − Fn TP ] + (1 − f ) [Jn − Fn TP ] +
of Equation 9 divided by ∆t yields the following: 1
f [Js − Fs TP ] + (1 − f ) [Js − Fs TP ] +
o

h i
(ρT )1 − (ρT )o f (Je − Fe TP )1 + (1 − f ) (Je − Fe TP )o +
1 1 o
rj ∆r ∆θ ∆z + f [Jn − Js ] + f (Jw − Fw TP ) + (1 − f ) (Jw − Fw TP ) +
∆t
1 o
(1 − f ) [Jn − Fn TP ] + (1 − f ) [Jn − Fn TP ] +
o 1 o
(1 − f ) [Jn − Js ] + f [Je − Jw ] + (1 − f ) [Je − Jw ] 1 o
(1 − f ) [Js − Fs TP ] + (1 − f ) [Js − Fs TP ] +
= f [(sconst + scoef TP ) rj ∆r ∆θ ∆z]1 + 1
(1 − f ) (Je − Fe TP ) + (1 − f ) (Je − Fe TP ) +
o

1 o
(1 − f ) [(sconst + scoef TP ) rj ∆r ∆θ ∆z]o (11) (1 − f ) (Jw − Fw TP ) + (1 − f ) (Jw − Fw TP ) +
1
= f [(sconst + scoef TP ) rP ∆r ∆θ ∆z] +
where (sconst + scoef TP ) is the linearized source term. o
The term, sconst , is the constant part of the linearized (1 − f )[(sconst + scoef TP ) rP ∆r ∆θ ∆z] (13)
source term and scoef , is the coefficient of the inde-
pendent variable T evaluated at the point P . The Following the procedure explained in Chapter 5 of
integrated total fluxes over the control volume faces Reference,17 the terms between the brackets in Equa-
 tion 13 can be replaced by:

(Jn − Fn TP ) = AN (TP − TN )
(Js − Fs TP ) = AS (TS − TP )
(Je − Fe TP ) = AE (TP − TE )
(Jw − Fw TP ) = AW (TW − TP )

For convenience, the superscript for new values is

dropped. Consequently, the final discretized generic
equation can be written as:

aP TP = aE TE + aW TW + aN TN + aS TS +
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

aT TT + aB TB + b
X
= anb Tnb + b (14)

where E, W , N and S refer to east, west, north, and

south grid points respectively; nb refers to the points
neighboring the point P . The coefficients aE , aW , aN
and aS contain the convection and diffusion terms and
are given by the following relations:

aE = De A (|Pe |) + [[−Fe , 0]]

aW = Dw A (|Pw |) + [[Fw , 0]]
aN = Dn A (|Pn |) + [[−Fn , 0]]
aS = Ds A (|Ps |) + [[Fs , 0]] (15)
Fig. 2 The staggered grid
aP = f (aE + aW + aN + aS ) +
ρ rP ∆r ∆θ ∆z (D) given by:
− f scoef ∆V
∆t
Fe = (ρVz )e (r ∆θ ∆r)e
b = f sconst ∆V + (1 − f ) sconst o ∆V +
X µe
(1 − f ) anb o Tnb o + De = (r ∆θ ∆r)e
δze
ρ rP ∆r ∆θ ∆z o Fw = (ρVz )w (r ∆θ ∆r)w
TP µw
∆t Dw = (r ∆θ ∆r)w
δzw
Fn = (ρVr )n (r ∆θ ∆z)n
where Pe , Pw , Pn and Ps are the Peclet numbers, µn
Dn = (r ∆θ ∆z)n (17)
which are the ratio of F (the flow rate across the sur- δrn
face of the control volume) and D (the diffusion across Fs = (ρVr )s (r ∆θ ∆z)s
the surface of the control volume). The symbol [[ ]] µs
Ds = (r ∆θ ∆z)s
denotes the greater of the quantities within. δrs

The power-law scheme was selected for the function where the suffixes e, w, n and s refer to control volume
A (|P |), which is given by faces; δr, δz are the distances between grid points as
illustrated in Figure 2. The value of diffusion pertain-
5
A (|P |) = [[0, (1 − 0.1 |P |) ]] (16) ing to control volume faces is determined through the
use of harmonic mean.
Until now, all the integrations were applied to the
The above function is a curve fit of the exact solution of general control volume in Figure 1. It is important
the steady one-dimensional convection/diffusion prob- to note that the grid used for solving the momentum
lem of the general dependent variable T . The variation equations is staggered. In other words, the control vol-
of T is dependent on the convection (F ) and diffusion ume is different for each of the momentum equations.
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

Fig. 4 Control volume numbering

The source terms of Equations 4-6 are given by:

∂p ρVθ 2
r − momentum source terms = − +
∂r r
µVr
− 2
r
ρVr Vθ µVθ
θ − momentum source terms = − − 2
r r
∂p
z − momentum source terms = −
∂z
Fig. 3 a. Vr grid cell b. Vz grid cell
Integration of the r-momentum Equation Source
Terms
Therefore, a shift in indexing needs to be applied to The source terms of the r-momentum equation must
the discretized equations. The staggered grid for the be discretized on the Vr -staggered grid. Therefore, the
momentum equations is shown in Figure 3. source terms are first integrated on the control volume
shown in Figure 4, and then shifted by one-half control
In addition, by taking advantage of an axisymmetric volume in the negative r-direction.
system, only one control volume in the θ-direction is  
used. For convenience, the term ∆θ is taken to be 1.0 1. − ∂p∂r term.
 
radian.
The integration of − ∂p ∂r on a control volume
Discretization of the Momentum Equation Source yields:
Terms
i+ 12 1 j+ 12
Z Z Z
∂p
− r dr dθ dz
The aim in this section is to integrate, discretize and ∂r
i− 12 0 j− 12
linearize the right-hand side of the momentum equa- 1 1
Z 2 Z1 j+
i+ Z 2
tions (4-6). The integration process first takes place
∂ (p r)
on a control volume similar to the one in Figure 1; =− dr dθ dz
however, here we abandon the E, W , N , S, grid point ∂r
i− 12 0 j− 12
referencing and instead use the classical i, j referenc- i+ 1 1

ing (see Figure 4). Since each of the three momentum Z 2 Z1 j+

Z 2
equations are solved on a different staggered grid, a + p dr dθ dz
shift in indexing is applied on the discretized source i− 12 0 j− 1
h 2 i
terms to properly match the corresponding staggered
= − (p r)j+ 1 − (p r)j− 1 ∆zi + p̃ ∆rj ∆zi
grid. The integration in the θ-direction will be taken 2 2

from 0. to 1. radian. Approximating:

  
pj+ 1 +pj− 1
p̃ by 2 2 This source term is dependent on the Vr velocity.
2
Therefore, the proper formulation is to include it
  into the scoef part of the linearized source term.
∆rj
rj+ 12 by rj + 2
  Integration of the θ-momentum Equation Source
∆rj Terms
rj− 12 by rj − 2
Here, the source terms of the θ-momentum is inte-
gives : grated on the control volume shown in Figure 4, and
     then shifted by one-half control volume in the negative
∆rj ∆rj
= − pj+ 12 rj + − pj− 12 rj − ∆zi θ-direction to match the θ-staggered grid.
2   2
  ∆rj
+ pj+ 12 + pj− 12 ∆zi 1. − ρVrrVθ term.
  2 Integrate over the control volume to obtain:
= pj− 12 − pj+ 12 rj ∆zi 1
Z 2 Z1 j+
i+ Z 2
1

ρVr Vθ
For the Vr -staggered grid we shift the indexing − r dr dθ dz
r
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

by one-half control volume in the negative i− 12 0 j− 12

r-direction, to get: = − (ρVr Vθ )i,j ∆rj ∆zi
  shift by one-half control volume in the negative
∆rj
= (pj−1 − pj ) rj − ∆zi θ-direction to get:
2
= − (ρVr )i,j− 1 (Vθ )i,j ∆rj ∆zi
2
This term will be referred to as (pP − pN )An in
later sections. To satisfy the positive coefficient rule, that is scoef
 2 must be positive, the linearization of this source
2. ρVrθ term. term is as follows:
 2
Integrate ρVrθ over the control volume to get: sconst = [[− (ρVr )i,j− 1 ∆rj ∆zi , 0]] (Vθ )i,j
2
1 1
Z 2 Z1 j+
i+ Z 2 scoef = −[[(ρVr )i,j− 1 ∆rj ∆zi , 0]]
ρVθ 2 2
r dr dθ dz
r
i− 12 0 j− 12 If Vr happens to be negative then the term sconst

will be included in the constant source term. On
= ρVθ 2 i,j ∆rj ∆zi
the other hand, if Vr is positive, then the term
A shift of a one-half control volume in the negative scoef will be included in the dependent source
r-direction yields

term.
= ρVθ 2 i,j− 1 (rj − rj−1 ) ∆zi
2
2. − µV
r 2 term.
θ

The term obtained above is independent of the Integrate over the control volume to obtain:
velocity Vr . Therefore, it is included into the con- 1
Z 2 Z1 j+
i+ Z 2
1

stant part of the linearized source term, sconst . µVθ

− r dr dθ dz
r2
3. − µV r
r 2 term.
i− 12 0 j− 12
" #
Again we integrate over the control volume to ob- rj+ 12
tain: = − (µVθ )i,j ln ∆zi
i+ 1 1 rj− 12
Z 2 Z1 j+
Z 2
µVr
− r dr dθ dz Integration of the z-momentum Equation Source
r2
1 0
i− 2 1
j− 2 Terms
" # ∂p
rj+ 12 1. − ∂z term.
= − (µVr )i,j ln ∆zi Integrate
rj− 12
i+ 12 1 j+ 12
Z Z Z
and then we shift by one-half control volume in ∂p
− r dr dθ dz
the negative r-direction to get  ∂z
i− 12 0 j− 12
rj
= − (µ)i,j− 1 VRi,j ln ∆zi 1 j+ 12
2 rj−1 Z1 Z  
=− r pi+ 12 ,j − pi− 12 ,j dr dθ
1 The reason the index j − 12 is given to µ while the index
j is given to Vr is: Vr is defined at the face of the control 0 j− 12
!
volume,(staggered grid), on the other hand µ is defined at the   (rj+ 12 )2 − (rj− 12 )2
grid points. Therefore, the value of µ must be interpolated at = pi− 12 ,j − pi+ 12 ,j
j − 12 . 2
     rj+ 1 + rj− 1 
= pi− 12 ,j − pi+ 12 ,j rj+ 12 − rj− 12 2 2 Using this method it is necessary to divide the blade
  2 into many spanwise blade elements, where the blade
= pi− 12 ,j − pi+ 12 ,j rj ∆rj properties (aerodynamic and geometric) are assumed
to be constant for each element. Using velocities from
shift by one-half control volume in the negative the previous time-step, the relative wind for each blade
z-direction, to get: element can be computed and then used to determine
the lift and drag (using a look up table for the CL
= (pi−1,j − pi,j ) rj ∆rj and CD ) at each of the blade elements. The forces are
resolved into the coordinate directions and included as
sources in the corresponding momentum equations for
Rotor Modeling
the next solver iteration.
The fan modeling is based on the momentum source
Since the blade does not occupy a given cell at all
concept developed by Rajagopalan.18–20 In order to
times as it sweeps the disk, only a fraction of the fan
obtain the wake of the fan, not known a priori, the
source needs to be allocated to that particular cell.
action of the rotating blades has to be implicitly in-
This fraction is determined as follows.
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

troduced into the governing equations. In other words,

the effect of the spinning blades is in the form of the The time taken by the center of the blade element to
force F(x,y,z,t) imparted by the blade to the fluid par- traverse one revolution is:
ticles in the path of the fan. Realizing that the momen-
tum equation governs the balance of the rate of change 2π
t1.rev = , (19)
of momentum and the external forces experienced by ω
the fluid element, the effects of the spinning blade on
its path at a given time can be modeled by including where ω is the angular speed of the blade in radi-
the force F (imparted by that fan at that particular ans/sec. The time the center of a blade element spends
time) to the momentum source at the cells occupied in a given cell is:
by the fan at that particular time. The force F(x,y,z,t) ∆θ
t∆θ = (20)
can be described by its components in each of the co- ω
ordinate directions. These components of F(x,y,z,t) where ∆θ is the angular measure of the path traced by
are introduced in the scalar momentum equations as a blade element as it passed through that cell. There-
implicit sources. fore, the fractional time that the blade element spends
The force -F exerted by the fluid on the blade varies in a cell is:
along the span of the blade and in general, may vary ∆θ/ω
as a function of time as well. These variations are tf rac =
2π/ω
largely due to the local flow conditions, variations in
the blade’s chord, aerodynamic twist, and geometric ∆θ
= , (21)
twist along the span. As a result, the different seg- 2π
ments of the blade will also see different strengths For a fan with N blades, this fraction becomes:
and directions of the relative wind, and the flow field
around the fan itself may be inherently unsteady due ∆θ
to the presence of bodies. For a time-accurate calcu- tf rac = N . (22)
2π
lation, the source terms in functional notation can be
written as:
This time-averaging technique represents a convenient
Si = Si (Cl , Cd , α, α̇, vabs , ω, and reasonable approximation in the fan modeling pro-
(18) cedure. While it is true that the flow through a fan is
R, t, c, ρ, µef f , Re, M )
unsteady in nature, the flow in the immediate vicin-
where Cl and Cd are airfoil characteristics of the fan ity of the fan disk is mostly in the direction normal
blade, α is the angle of attack made by the fan blade to the disk (except at the fan tip region), coupled
to the relative velocity vector, α̇ is the time rate of with a strong swirling motion due to the shear im-
change of α as the blade moves through a revolution, parted by the rotating blade. Since the mean of the
vabs is the absolute velocity of the fluid at the instan- flow quantities is large, the unsteady components can
taneous blade location (R, t), ω is the angular velocity be averaged without compromising the physics of the
of the fan, and c is the chord of the blade. Even though solution. This time-averaged technique allows suffi-
the complete Navier-Stokes equations are solved every- ciently large time steps to be taken to advance the
where in the flow field, the dependence of Si on µ and solution, without constraint from the rotational speed
Re are considered only implicitly through the airfoil of the blade, since no attempt is made to consider the
sectional characteristics Cl and Cd in this analysis. specific blade locations at different times.
 (1)
Starting
Point
(0)

(3) (2)

Fig. 5 The curve adaptive procedure used in grid

generation.

Cartesian Grid Generation

The shape of the duct plays a very important role in

the performance of a ducted propeller system. There-
fore, proper representation of the duct geometry is
critical. For maximum accuracy in the Cartesian co-
ordinate system, the grids around the duct need to be
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

arranged such that the boundaries of the duct pass

through opposite corners of the surface cells. For this Fig. 6 A sample grid around a duct generated
purpose, an option was added to the grid generation using curve adaptive option.
program to enforce the above criteria, known as “the
curve adaptive option”.
The following methodology is used to implement the
curve adaptive option: First, an initial grid based on
a user specified number of points on the body and
cluster ratio is constructed. Then, from each speci-
fied point, the grid generator finds the next grid point
on the body surface by moving in x or y directions.
This procedure stops when: (1) two neighbor points
are within a given tolerance, (2) the slope angle of the
surface at the point is 0 or 90 degrees or (3) exceed-
ing a preset number of movements. This procedure is
depicted in Figure 5. Figure 6 shows the grids near a
duct generated using the above procedure. As shown
in the figure, the points on the boundary pass the cells
diagonally.
Using the curve adaptive methodology, the resolution Fig. 7 A sample grid around the duct after
of the grid in the convex areas of the body surfaces smoothing process.
is higher than other regions. In other words, this
Gauss-Seidel method. The details of the SIMPLER
method may generate cells with high clustering ratio,
algorithm can be found in Reference;17 only highlights
which may result in numerical instabilities and grid
and important principles are presented here. A brief
shocks. To avoid these problems, a smoothing process
description of the momentum and pressure equations
is implemented into the grid generation program which
used in the SIMPLER algorithm follows.
ensures that the ratio between two adjacent grid cell
sizes is no greater than 2.0. In this process, the grids Momentum Equations
around the duct may be changed slightly. The duct By employing the formulation for discretizing
boundary does not cross the opposite corners of a few the general differential equation, the discretized r–
cells. An example of the final grid generated around momentum equation on a staggered grid shown in
the duct surface is shown in Figure 7. Figure 3.a can be written as:
Solution Procedure of the Discretized Equations X
an Vr n = anb Vr nb + br + (pP − pN )An (23)
In this research, the discretized equations are solved
using a finite-volume approach called SIMPLER, de- where the neighboring coefficients anb account for the
veloped by Patankar.17 This approach uses an iter- combined convection–diffusion influence at the control
ative procedure to solve for the primitive variables volume faces, br represents the discretized source term,
(velocity and pressure fields). The discretized equa- the term pP −pN is the pressure gradient acting on the
tions are solved using a line-by-line method combining control volume, and An is the area on which the pres-
the Tri-Diagonal Matrix Algorithm (TDMA) and the sure acts. The coefficients of the momentum equation
 (ae , aw , an , as ), being functions of the velocity compo- 5. Calculate the source terms of the pressure correc-
nents, make the momentum equation non-linear. This tion equation and solve for the pressure correc-
non-linearity is handled by adopting an iterative strat- tions.
egy and by lagging the coefficients.
6. Correct the velocities using the velocity correction
Similarly, the θ–momentum equation and the z– equations.
momentum equation can also be discretized to yield
the following: 7. Return to step 2 and repeat until convergence.
X
at Vθ t = anb Vθ nb + bθ (24) 8. Start with the next time.

X Boundary Conditions
ae Vz e = anb Vz nb + bz + (pP − pE )Ae (25)
The velocities are set to freestream values at the in-
flow boundaries. The downstream boundary values are
The discretized momentum equations can be solved
extrapolated from the interior grid points and adjusted
iteratively if the pressure field is known. If the cor-
to conserve mass flow through the computational do-
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

rect pressure field is employed the resulting velocity

main. All the control volumes that lie in the solid
field will satisfy the continuity equation. However, the
region are blocked off with zero velocities everywhere.
pressure field is unknown and an equation for pressure
In other words, no-slip viscous boundary conditions
must be derived to solve for it.
are applied to all solid bodies.
Pressure Equation and Pressure Correction Equation
Since the main driving force for the velocity from one Results
cell to the next cell is the pressure difference between
the two cells, an accurate solution of the pressure field The Navier-Stokes flow code described earlier was ap-
is important. By manipulating the continuity equation plied to simulate the following two configurations:
and the momentum equation, an equation for pressure
• Trek Aeropspace’s ducted fan
is derived and discretized to yield a form similar to
Equation 14:17 • Micro-Craft’s Lift Augmented Ducted Fan Un-
manned Air Vehicle (LADFUAV)
aP pP = aE pE + aW pW + aN pN + aS pS + bp (26)

For each case, the configuration and computational

If the correct velocity field is used in the above equa-
grids are described, and CFD results in the form of
tion the correct pressure field will result. For a guessed
pressure contour plots, velocity vectors and graphs
velocity field, a pressure correction equation is required
are presented. CFD results are compared with cor-
to correct the velocity field obtained from solving the
responding experimental data.
momentum equation. Once more, the pressure correc-
tion equation is derived by manipulating the continuity A typical run on a good resolution grid (183 x 176) will
and the momentum equations, and it is cast in a form approximately take 3 CPU hours on the Cray SV1ex
similar to Equation 14. The purpose of the pressure machine (500 MHz clock speed which is equivalent to
correction equation is to improve the pressure field 2.0 GHz for a personal computer (PC) machine) at
such that the velocity field will be corrected every it- Ames Research Center. Specifically, it takes 2.17 CPU
eration and progressively get closer to satisfying the seconds for each iteration step and 5000 iterations for
continuity equation. a complete calculation. The present code is based on
a sequential algorithm and is not easily parallelized.
SIMPLER Algorithm
Therefore, taking advantage of multiple processors is
The sequence of steps for the unsteady SIMPLER al- not possible. However, the code can be run on a PC
gorithm can be summarized as follows: machine. It takes 4.7 CPU hours to run the same cal-
culation on a PC machine with a 750 MHz processor.
1. Start with a given (guessed) velocity field. The turn around time will improve for machines based
on the latest Intel processors (3.06 GHz, or even faster
2. Calculate the coefficients of the momentum and clock speed).
pressure equations.
Trek Aerospace’s Ducted Fan
3. Solve pressure equations to obtain the pressure
field. The first configuration chosen to validate the solver
was Trek Aerospace’s ducted fan, shown in Figure 8.
4. Using the calculated pressure field, solve the mo- This configuration consists of a duct, nacelle and fan.
mentum equations to get the velocity field. The duct has a diameter of 38.073 inches with a fan
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

Fig. 8 Trek Aerospace’s ducted fan configuration

to duct tip clearance of 0.0367 inches. The fan has 5

blades with a tip diameter of 38 inches.
For this case, the flow solver was applied to the Trek
Aerospace’s ducted fan for seven different operating
conditions in hover to construct the overall fan perfor-
mance map. Both duct and nacelle were considered
in the simulation while the fan was modeled by source
terms in the momentum equations, as explained ear-
lier.
The velocity profiles at three different fan speeds
(2000, 3000 and 4000 RPM) are shown in Figure 9.
In all cases, flow upstream and inside the duct are
streamlined and reversed flow in the wake is observed
downstream and outward from the duct. This reversed
flow becomes stronger as the fan speed increases. In-
side the duct, on the suction side near the fan tip, flow
accelerates, and the velocity magnitude increases at
higher RPMs.
Figure 10 depicts the pressure distribution around the
duct and nacelle for the same operating conditions.
The pressure increases as the flow passes through the
fan as a consequence of the work on the fluid by the
fan. Calculations reveal that the change in the pres-
sure through the fan is higher near the fan tip region
(see Figure 10), indicating that most of the pressure
is created by the fan tip. In the wake region, high
gradient pressure regions are observed at the same
location where the reversed flow occurs. These high
pressure regions are related to the vortex shedding and
are stronger as the fan speed increases.
Fig. 9 Velocity vector plot around Trek
The torque and thrust were calculated from the load Aerospace’s fan at RPM=2000, 3000, 4000
distribution on the blades using blade element theory.
A comparison between the CFD-predicted ducted fan
torque and wind tunnel measurements21, 22 at differ-
ent fan speeds is shown in Figure 11. The computed
 60

50

40

Torque (ft-lb)
30

20

Computation
10
Wind Tunnel

0
1500 2000 2500 3000 3500 4000 4500 5000
RPM

Fig. 11 Comparison of fan torque from CFD re-

Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

sults and wind tunnel testing for Trek Aerospace’s

fan at different RPMs
300

Fan

250 Wind Tunnel

Duct
Ducted Fan
200

Thrust (lb)
150

100

50

0
1500 2000 2500 3000 3500 4000 4500 5000
RPM

Fig. 12 Comparison of fan thrust from CFD results

and wind tunnel testing for Trek Aerospace’s fan
at different RPMs

ducted fan torques are within 2 percent of the experi-

mental data. The corresponding comparison of thrust
is plotted in Figure 12. In computation, the thrust of
a ducted fan consists of two parts, one for fan and the
other for duct. The wind tunnel data is in closer agree-
ment with the fan thrust rather than the total thrust
distribution. This phenomenon could possibly be due
to the disrupted flow in the lower quadrant of the test
article, and the presence of support structures within
the duct. After using ad hoc correction formula,23 the
wind tunnel hover values become:21
Thrust Coefficient = 0.213, Power Coefficient = 0.082,
Figure of Merit = 0.7068.
The computed thrust coefficient, power coefficient, and
figure of merit distributions are shown in Figure 13,
respectively. Overall, the thrust coefficient is within 3
Fig. 10 Pressure distribution around Trek
percent of the corrected test data; the power coefficient
Aerospace’s fan at RPM=2000, 3000, 4000
and the figure of merit, within 2 percent of those of
corrected experimental data.
Increasing the blade pitch can increase thrust levels at
the cost of higher power levels to turn the fan. A per-
 0.5 0.35

Ducted Fan Baseline

Fan Pitch +3 deg
0.4 0.31
Duct
Wind Tunnel Corrected
Thrust Coefficient

Thrust Coefficient
0.3 Wind Tunnel
0.27

0.2 0.23

0.1 0.19

0 0.15
1500 2000 2500 3000 3500 4000 4500 5000 1500 2000 2500 3000 3500 4000 4500 5000
RPM RPM
0.12 0.2

Computation 0.175 Baseline

Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

Pitch + 3 Deg
Wind Tunnel Corrected
0.1 0.15
Wind Tunnel

Power Coefficient
Power Coefficient

0.125

0.08 0.1

0.075

0.06 0.05

0.025

0.04 0
1500 2000 2500 3000 3500 4000 4500 5000 1500 2000 2500 3000 3500 4000 4500 5000
RPM RPM
90 80

Baseline
80 Pitch +3 deg
75

70
Figure of Merit (%)

Figure of Merit (%)

60 70
Computation
Wind Tunnel Corrected
50
Wind Tunnel
65
40

30
60
1500 2000 2500 3000 3500 4000 4500 5000
1500 2000 2500 3000 3500 4000 4500 5000
RPM RPM

Fig. 13 Comparison of thrust, power, and figure Fig. 14 Comparison of thrust, power, and figure
of merit distributions from CFD results and wind of merit distributions of both the baseline fan and
tunnel testing for Trek Aerospace’s fan at different the fan with new pitch blades at different RPMs
RPMs
Micro-Craft’s LADFUAV

formance comparison of the Trek Aerospace’s ducted The Micro-Craft LADFUAV configuration has a duct
fan and a new ducted fan with blade pitch increased with a diameter of 9.0 inches and a two-blade fan with
by 3 degrees is shown in Figure 14. At the 3500 RPM, a diameter of 8.985 inches. A simulation was carried
the new ducted fan delivered 25 percent more thrust out in hover at 16,000 RPM.
but required 37.5 percent more power. The figure of
merit was only improved 1.43 percent at this RPM. The velocity-vector plot of the flow going through the
Comparing with the baseline, the new ducted fan per- duct is presented in Figure 15. Due to viscous effects, a
formed better for RPMs greater than 2700 but worse reversed flow occurs near the inner surface of the duct.
for RPMs less than 2700. The fan blade design en- Similar to the previous case, a recirculating region is
hancement is not simple. Even the final pitch setting also observed downstream of the duct. Figure 16 shows
warrants an optimization code. a close region near the fan tip. It can be seen that
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

Fig. 15 Velocity distribution of Micro-Craft’s

Fig. 17 Micro-Craft’s LADFUAV duct pressure
LADFUAV in hover
distribution

0.14

0.12
Torque (ft-lb)

0.1

0.08

0.06
14000 15000 16000 17000 18000 19000
RPM

Fig. 16 Flow near the fan tip region of Micro- Fig. 18 Predicted fan torque of Micro-Craft’s
Craft’s LADFUAV in hover LADFUAV at different RPMs
recirculating flow occurs at the clearance between the
duct and the fan.
6
Figure 17 depicts the pressure distribution on the duct
with a section of the computational grid in the r − z Fan
cutplane. The effect of the fan on the pressure distri- 5 Duct
Ducted Fan
bution on the duct and nacelle is shown in this figure.
The torque and thrust were calculated from the load 4
Thrust (lb)

distribution on the blades using blade element theory.

The predicted fan torque and thrust at different fan 3

speeds are shown in Figure 18 and Figure 19, respec-

tively. Further aerodynamic performance analysis was
2
reported in Figure 20.

Conclusions 1
14000 15000 16000 17000 18000 19000
RPM
An axisymmetric Navier-Stokes flow code was devel-
oped to calculate the flow field of a ducted fan. Pre- Fig. 19 Predicted fan thrust of Micro-Craft’s
dictions of hover performance for two industrial ducted LADFUAV at different RPMs
fan models of current interest were calculated. One of
 0.08
fan geometry including control panels and support
structure within the duct.
0.07

0.06
Fan Acknowledgements
Thrust Coefficient

Duct
Ducted Fan The wind tunnel model of ducted fan was pro-
0.05 vided by Michael M. Moshier and Robert W. Bulaga
of Trek Aerospace. The geometry of the Lift Aug-
0.04
mented Ducted Fan Unmanned Air Vehicle was given
by Larry Lipera of Micro Craft Inc.. Helpful dis-
0.03
cussions and suggestions were received from Anita
0.02
Abrego, Frank Caradonna, Wayne Johnson, Roger
14000 15000 16000 17000 18000 19000 Strawn, Chee Tung and William Warmbrodt. Their
RPM
0.02 close cooperation to advance the ducted fan technol-
ogy is gratefully acknowledged.
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079

0.02
References
1 B.W.
Power Coefficient

McCormick. Aerodynamics of V/STOL Flight. Aca-

demic Press, New York, 1967.
0.02 2 S. Thurston and R. C. Amsler. Review of marine propellers

and ducted propeller propulsive devices. Journal of Aircraft, 3,

May-June 1966.
3 D. L. Kohlman. Introduction to V/STOL Airplanes. Iowa
0.01
State University Press, Iowa, 1981.
4 M. R. Mendenhall and S. B. Spangler. Theoretical study

of ducted fan performance. NASA CR-1494, January 1970.

0.01 5 M. R. Mendenhall and S. B. Spangler. Theoretical study
14000 15000 16000 17000 18000 19000
RPM of ducted fan performance. NASA CR-1495, January 1970.
90
6 R. B. Gray and T. Wright. A vortex wake model for

optimum heavily loaded ducted fans. Journal of Aircraft, 7,

80
March-April 1970.
7 P. F. Yaggy and K. W. Mort. A wind tunnel investigation

of a 4–foot diameter ducted fan mounted on the tip of a semispan

Figure of Merit (%)

70 wing. NASA TN D-776, March 1961.

8 P. F. Yaggy and K. W. Goodson. Aerodynamics of a tilting

ducted fan configuration. NASA TN D-785, March 1961.

60 9 K. J. Grunwald and K. W. Goodson. Aerodynamic loads on

an isolated shrouded–propeller configuration for angles of attack

from −10 deg to 110 deg. NASA TN D-995, January 1962.
50 10 K. W. Mort. Performance characteristics of a 4–foot di-

ameter ducted fan at zero angle of attack for several fan blade
angles. NASA TN D-3122, December 1965.
40
11 D. G. M. Davis. Ducted propulsors – Progress in the United
14000 15000 16000 17000 18000 19000
RPM Kingdom. Business Aircraft Meeting, Wichita, Kansas, Society
of Automative Engineers, Paper No. 750534, 1975.
Fig. 20 Predicted thrust, power, and figure of 12 K. W. Goodson and K. J. Grunwald. Aerodynamic
merit distributions of Micro-Craft’s LADFUAV at characteristics of a powered semispan tilting-shrouded-propeller
different RPMs VTOL model in hovering and transition flight. NASA TN D-
981, January 1962.
them was compared with available wind tunnel test 13 K. W. Mort and P. F. Yaggy. Aerodynamic characteristics
data. The comparison was good. The thrust coeffi- of a 4–foot diameter ducted fan mounted on the tip of a semispan
cient was within 3 percent of the corrected test data; wing. NASA TN D-1335, April 1962.
14 Jr. Newsom, W. A. Aerodynamic characteristics of four–
the power coefficient and the figure of merit within
duct tandem VTOL–aircraft configurations. NASA TN D-1481,
2 percent of those of the corrected wind tunnel data. January 1963.
Preliminary results showed that the present code is ac- 15 E. E. Davenport and K. P. Spreemann. Transition
curate and can be used as an axial flow analysis tool characteristics of a VTOL aircraft powered by four ducted tan-
for ducted fan design. dem propellers. NASA TN D-2254, April 1964.
16 K. P. Spreemann. Wind tunnel investigation of longitudinal
The next step in code development for the ducted fans aerodynamic characteristics of a powered four–duct propeller
will consider three-dimensional hover flight conditions VTOL model in transition. NASA TN D-3192, April 1966.
17 Patankar S.V. Numerical Heat Transfer and Fluid Flow.
where the axisymmetric assumption will be removed to
Hemisphere Publishing Corporation, New York, 1980.
account for the fan blade azimuthal position. Further 18 Rajagopalan R.G. and Lim C.K. Laminar flow analysis of
development of the research project will encompass un- a rotor in hover. Journal of the American Helicopter Society,
steady axial and forward flight conditions with duct 36:12–23, 1991.
 19 Rajagopalan R.G. and Mathur S. Three dimensional anal-

ysis of a rotor in forward flight. Journal of the American

Helicopter Society, 38:14–25, 1993.
20 Rajagopalan R.G. and Zori L. Navier-Stokes calculations

of rotor-airframe interaction in forward flight. Journal of the

American Helicopter Society, 40:57–67, 1995.
21 Moshier M. and Bulaga R. W. Wind tunnel perfor-
mance investigation of the Solotrek XFV ducted fan system.
DARPA/DSO, Arlington, VA, 2001.
22 Abrego A. I. and Bulaga R. W. Performance study of a

ducted fan system. AHS Aerodynamics, Acoustics, and Test

and Evaluation Technical Specialists Meeting, San Francisco,
CA, January 23-25, 2002.
23 Mort K. W. Exit pressure data for a 38 inch diameter

ducted fan in 7 x 10 ft wind tunnel. DARPA/DSO, Arlington,

VA, 2001.
Downloaded by NASA AMES RESEARCH CENTER on December 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2003-4079