You are on page 1of 13

Body Force Modeling of the Fan Stage of a

Windmilling Turbofan

Massyl Lagha & Guillaume Dufour∗


ISAE-SUPAERO, Université de Toulouse, France

ABSTRACT
The determination of the rotational speed and massflow of the fan of a turbofan at windmill is critical in the
design of the engine-supporting structure and the sizing of the vertical stabilizer. Given the very high bypass ratio
obtained at windmill, the flow in the fan stage and bypass duct is of prime interest. Classical CFD simulations
have been shown to predict such flows accurately, but extensive parametric studies can be needed, stressing the
need for reduced-cost modeling of the flow in the engine. A Body Force Modeling (BFM) approach for windmilling
simulations is examined in the present contribution. The BFM approach replaces turbomachinery rows by source
terms, reducing the computational cost (here by a factor 5). A shaft model is coupled to the BFM source terms,
to drive the simulation to a power balance of the low-pressure shaft. The overall approach is thus self-contained
and can predict both the massflow and the rotational speed in the windmilling regime. Comparisons with en-
gine experimental results show the proposed model can predict the rotational speed within 7 %, and the massflow
within 5 %. Local analysis and comparisons with experimental data and reference blade calculations show that the
work exchange, in term of total temperature variation, is predicted within 0.5 K, and the overall total pressure ratio
within 1 %. However, the losses in the stator are largely underestimated, which explains the discrepancy for the
massflow predictions.

1 Introduction
When a turbofan engine flames out during flight, the ram pressure at the fan inlet creates an internal flow that causes
spool rotation. This regime is called windmilling operation [1], and corresponds to a free-wheeling mode of the rotor. Early
in the engine integration design process, two essential engine-out performance characteristics are needed: (i) the engine drag
at windmill, as it sets the sizing of the vertical stabilizer [2] and (ii) the windmilling rotational speed of the fan, as it impacts
the design of the supporting structure of the engine [3] or the shaft bearing lubrication systems. The engine internal massflow
is also essential, as it sets the internal drag (when combined with the velocity change in a classical momentum analysis [4]),
but also the external drag (through spillage effects over the nacelle [5]). Furthermore, energy extraction potential from the
engine can be required for integration issues. Finally, regarding the combustor design, the assessment of the ability to relight
the engine in flight requires the evaluation of the massflow, pressure and temperature in the combustion chamber. In the
present contribution, the focus is on the prediction of the windmilling rotational speed and massflow.
In a turbofan at windmill, the establishment of these two characteristics can be conceptually divided into two “features”.
First, for a given flight point, the overall loss from the fan inlet to the bypass nozzle exit sets the massflow that passes through
the engine. Indeed, by expressing conservation of mass, Gunn & Hall [6] show that the massflow at windmill depends only
on the flight Mach number, area ratio and overall total pressure ratio. Second, for a given massflow, the fan rotor then sets the
rotational speed such that (almost) zero work is exchanged, hence the name free-windmilling used in ref. [7]. However, these
two aspects are in fact coupled, as the losses in the stator (which are the dominant ones [5, 4]) depend on the flow delivered
by the rotor. Finally, as the bypass ratio of a windmilling turbofan is very high (typically ranging from 50 to 80 [8, 4]), the
contribution of the core is usually neglected, and the flow in the fan stage and bypass duct is of prime interest.
Several methods based on first principles are proposed in the literature to tackle the prediction of windmilling charac-
teristics. Zachos [4] devised a model based on frictional flow theory (Fanno flow) to predict engine massflow and drag at
windmill, but the approach is not developed to predict the rotational speed, and relies on available engine data to calibrate the
overall loss. Gunn & Hall [6] applied the Euler turbomachinery equation at an average radius (based on zero work exchange)

∗ Address all correspondence to this author: guillaume.dufour@isae-supaero.fr.

TURBO-21-1086 DUFOUR 1
to express the rotational speed of the rotor as a function of massflow, but the method requires an estimation of the pressure
loss across the engine and of the deviation across the fan rotor. A similar approach is proposed by Binder et al. [7], which
fully accounts for the 3D rotor geometry and can also be applied to load-controlled regimes (as opposed to free windmilling
only). However, an empirical deviation model is required to obtain the rotational speed, and the method needs to be coupled
to a separate pressure loss model to predict the massflow. Finally, Prasad [5, 9] proposed a global framework using physical
modeling coupled to CFD calculations to predict both the massflow and the rotational speed. Altogether, those methods are
not self-contained, in the sense that they all require inputs from either empirical or experimental data, or additional CFD
computations. Moreover, none of these approaches can be extended to non-axisymmetric inflow in a straightforward way,
whereas taking into account inflow distortion at windmill may prove necessary for future aeropropulsive architectures, such
as boundary layer ingestion or distributed propulsion. In the present paper, the proposed approach aims at being: (i) efficient
in term of accuracy-to-cost ratio, (ii) more informative than 1D or 2D models, to possibly allow for fan design iterations,
(iii) able to account for inflow-distortion effect and (iv) self-contained.
Given these constraints, the chosen modeling approach is the Body Force Modeling (BFM) method: (i) the basic prin-
ciple of the BFM approach is to replace turbomachinery rows by source terms in the problem equations, thus alleviating
the need to actually mesh blades and thereby reducing the computational cost; (ii) the model input is the full blade camber
surface, which makes it sensitive to blade design parameters and allows 3D computations; (iii) by nature, the BFM approach
filters out blade-to-blade effects (the effect of the blades are smeared out across the pitch), and can thus tackle whole-annulus
inflow distortion problems with a steady approach. Finally, (iv) we propose to couple a shaft model to the BFM source
terms, such that the proposed approach is self-contained and can predict both the massflow and the rotational speed in the
windmilling regime.
A previous study by the authors’ team [10] used a similar approach, but with a simpler BFM model, and only focused on
the rotor. As such, it only predicted the rotational speed for a specified massflow. In contrast, the present contribution uses a
more general BFM model and tackles the modeling of the whole stage (ie, the rotor and the stator), such that the massflow is
now also a result.
In this context, the goals of the present contribution are to: (i) propose a BFM approach coupled to a shaft model to
predict both the rotational speed and the massflow of a fan at windmill, and (ii) validate and analyze the BFM predictions
with respect to classical CFD computations and experimental results.
After introducing the test case, the numerical methods and the BFM model are briefly described, together with the shaft
model. Next, global windmilling characteristics are presented and validated against experimental data. Finally, the results are
analyzed and discussed in terms of work exchange and loss predictions, using reference CFD calculations and experimental
data.

2 Test case: the DGEN 380 Turbofan


The DGEN 380 is a high bypass ratio, unmixed-flow, geared turbofan developed by Price Induction, now Akira. The
engine is designed for a cruise altitude of 4 500 m and a flight Mach number of 0.338. The fan, whose diameter is 352 mm,
is driven by a single-stage uncooled low-pressure turbine stage. The core engine consists of a centrifugal compressor and a
single-stage high-pressure turbine. Figure 1 presents a meridional cut of the engine, showing the overall architecture, and
the stations of interest for the analysis: station 2A is located just upstream of the spinner, station 2R is downstream of the
rotor, and station 21A is downstream of the stator.
A fully-instrumented version of the DGEN 380 is on a test rig at ISAE-SUPAERO, specially suited to reproduce wind-
milling operation in an ambient ground set up. The engine is equipped with conventional measurement systems to character-
ize the cycle performance, and with directional five-hole probes to characterize the flow across the fan stage. To reproduce
windmilling conditions, a 75 kW centrifugal fan is used to blow air through the engine up to 6 kg/s, which is 55 % of the
test-bed nominal massflow rate, and corresponds to a local Mach number M2A = 0.16 at the engine inlet [8].
To give an equivalent flight Mach number, the ratio between the inlet total pressure (imposed by the blower) and the
outlet static pressure (imposed by the atmosphere) can be used. In the present study, the range explored gives equivalent
values of flight Mach number M∞ between 0.14 and 0.28, corresponding to 41 % and 84 % of the nominal flight Mach
number of the engine. These relative values match well with what is reported in the literature for windmilling studies of
larger engines (see for example refs. [11, 12, 5]), where the relative values examined go down to 46 %. Table 1 summarizes
the range of the windmilling points investigated, where the reduced massflow and rotational speed are defined as:


ṁ rTt−2A ND
ṁred−2A = and Nred = √ (1)
pt−2A A2A γrTt−2A

TURBO-21-1086 DUFOUR 2
Parameter Min. Max.
Equivalent flight Mach number M∞ 0.14 0.28
Inlet Mach number M2A 0.08 0.16
Inlet reduced massflow ṁred−2A 0.08 0.165
Fan reduced rotational speed (% design) Nred 8 16

Table 1. Characteristics of the windmilling points investigated

Fig. 1. Meridional view of the engine, with stations of interest: engine inlet (2A), rotor outlet (2R), stator outlet (21A).

3 Numerical and modeling approaches


3.1 Meshing strategies
For the conventional CFD computations with the blade meshed, hereafter referred to as “Baseline CFD ”, the numerical
domain is partially shown in Fig. 2(a): it starts three fan chords upstream of the spinner and ends (i) in the bypass duct, about
one fan chord before the actual nozzle exit plane, and (ii) in the core duct, at the inlet of the high-pressure compressor. A
mixing plane is located between the rotor and the stator. The geometry of the fan includes the hub fillet and the tip clearance
gap.
The numerical domain is discretized with a multi-block structured mesh generated with Numeca’s Autogrid 5TM , con-
sisting of O -type blocks around the blades, and H -type blocks to fill the passage, using a single-blade periodic domain. The
first cell at all the walls has a size of 10 µm, ensuring a ∆y+
w value up to 2 over the blades, and below 1 for the hub and shroud
for all the simulations; the expansion ratio across the boundary layer mesh is about 10 %. In the spanwise direction, 189
points are used in the rotor (of which 25 points are in the tip gap), and 185 points in the stator. In the azimuthal direction, 81
points are used in the rotor, and 73 points in the stator. In the streamwise direction, 297 points are put around the rotor blade,
and 245 points around the stator blade. Overall, the mesh is comprised of about 12 million points.
For the BFM computations, the numerical domain is shown in Fig. 2(b). It is similar to that of the baseline CFD case,
but without the blades, the corresponding zone being shown in green and blue. The BFM domain starts just after the spinner
leading edge to avoid zero radius, which eases the mesh generation as only 3 cells are used in the azimuthal direction (note
that this is not a limitation of the approach). The meshing strategy of the BFM domain consists in a multi-block structured
mesh. The boundary layer discretization at the end-walls is similar to the baseline CFD mesh (10 µm at first cell and 10 %
expansion ratio). In the spanwise direction 177 points are used, and in the streamwise direction 73 points are used to
discretize the blocks corresponding to the blades. The total number of points is about 335 000.

3.2 Numerical set up


The baseline CFD simulations are performed with the Euranus solver of the Fine/TurboTM package of Numeca. This
multi-block structured solver is thoroughly presented by Hirsch et al. [13]. The BFM computations are performed with the
Fine/OpenTM package of Numeca, which is an unstructured solver. This solver includes a user-interface (called OpenlabsTM )
that allows to modify the flow governing equations. The BFM model is implemented thanks to this module.
For both the baseline CFD and BFM simulations, the Reynolds-averaged Navier–Stokes equations (RANS) are used,
with the one-equation model of Spalart & Allmaras for turbulence closure. The RANS equations are solved in the rotating
frame for the baseline CFD, while they are solved in the absolute frame with added source terms in the BFM approach.
A pseudo-time-marching method is applied to obtain the steady state solution, using a four-stage Runge–Kutta scheme
with implicit residual smoothing. Local time stepping and a three-level multigrid technique are used to accelerate con-
vergence to the steady state. The discretization in space is based on a cell-centered finite-volume approach. The convective
fluxes are determined by a second-order centered scheme with added artificial dissipation and the viscous fluxes are centered.

TURBO-21-1086 DUFOUR 3
(a) Baseline CFD (mixing plane in yellow) (b) BFM approach

Fig. 2. Views of the numerical domains and grids.

In the experiments, the blower used to reproduce windmilling conditions is connected to the engine with a long duct,
along which boundary layers develop. The duct is not included in the simulations, but at the inlet of the baseline CFD numer-
ical domain, radial profiles of total pressure and velocity direction are imposed according to the experimental measurements,
to match the experimental inflow conditions of the fan stage.
For the baseline simulations, the rotational speed is imposed from the experimental data, so that a reference can be used.
Obviously, this is not a predictive approach, contrary to what will be discussed for the BFM simulations.
To avoid representing the actual outlet nozzle ejecting the bypass flow in the atmosphere, static pressure is imposed
at the outlets of the numerical domains. The static pressure values for the core and bypass outlet boundary conditions are
adjusted so that the baseline CFD results match the experimental total pressure ratio profiles downstream of the stator (station
21A), for each windmilling operating point.
For the BFM approach, since the domain starts at the spinner leading edge, radial profiles are extracted from the calibra-
ted baseline CFD simulations, and imposed as inlet boundary conditions. At the bypass outlet, the static pressure condition
imposed is identical to that of the baseline CFD computations. Therefore, both the baseline and the BFM computations are
run at the same back pressure, so that they represent equivalent flight points.
In a predictive approach for engine design, nozzle characteristics would need to be available to impose the outlet bound-
ary condition, as described by Prasad [5] for instance. Here, using the value from the baseline CFD can be seen as a substitute
to such characteristics, and this is what allows the massflow to be considered a result of the present simulations.
Finally, in the core duct of the BFM calculations, a zero massflow condition is used (as in ref. [6] for instance), because
this is not a piece of information that would be available in a predictive approach.

3.3 Body Force Methodology


In the body force approach, source terms are added to the right-hand side of the governing equations. Here, the steady
RANS equations are used:

1
∇ · (N~V + T) = − (~V · ∇b)N + F (2)
b

where:
 
0
   
ρ 0
N =  ρ~V  , T = −~~τ + pI , F =  ρ~f  (3)
 
ρht −~~τ · ~V ρ fθ ωr

The BFM method derives from a formulation proposed by Marble [14], and a seminal expression for the force for-
mulation was derived at MIT by Gong [15]. From the methodology standpoint, different extensions were proposed, such
as: inclusion of blade metal blockage effects [16, 17], advanced calibration [18, 19], wake effects [20], immersed-boundary
formulation [21, 22], multiple tone noise generation [23], or coupling with engine-level prediction [24], to name but a few.
The accuracy of the approach has accordingly been demonstrated for several applications: compressor surge [25, 26, 27],
fan–intake interactions [28, 18, 17, 29], noise predictions [23, 30, 31], generic inflow-distortion/fan interaction [32, 33] or
boundary layer ingestion [34, 35, 36], for instance.
The body force source-term formulation used here is an adaptation of the Hall model [36], proposed by Thollet [37].
Several studies [38, 39, 40] have demonstrated the ability of this model to represent full annulus distortion transfer with a

TURBO-21-1086 DUFOUR 4
steady approach. In particular, ref. [41] validates this model against experimental results for 360◦ inflow distortion simu-
lations of a low speed fan at windmill.
The first right-hand side term in Eq. (2) models the blades metal blockage, while F represents the blade force field. This
force field is formulated as normal and parallel forces in the local frame of the relative flow vector. The normal force fn
induces flow deviation (i.e., work exchange), and the parallel force f p represents the losses:

W2
fn = KMach πδ (4)
sb|nθ |
W2
fp = Cf (5)
sb|nθ |

where W is the local relative velocity norm, δ is the local flow deviation (i.e., the angle between the local relative velocity
and the blade direction) , s is the blade pitch, b is the blade metal blockage, nθ is the azimuthal projection of the unit vector
normal to the local blade camber ~n, and KMach is a compressibility correction factor based on the Prandtl-Glauert correction
for subsonic Mach numbers and the Ackeret correction for supersonic Mach numbers:

  
1
min √ ,3 , if Mr ≤ 1


2
KMach =  1−Mr  (6)
min

 √4 , 3 , if Mr > 1
2
2π Mr −1

The local friction coefficient C f is derived from an empirical expression based on the turbulent boundary layer developement
over a flat plate [42]:

ρWx
C f = 0.0592Re−0.2
x , with Rex = (7)
µ

It should be emphasized that the only input to the body force model is the geometry of the blade through the blade
camber surface and the thickness distribution (i.e., no a priori knowledge of deviation or loss is required). The blade camber
surface is defined by its normal, which is a function of the axial and radial directions ~n(x, r). Therefore, at each node of the
domain corresponding to the blade, a force can be defined locally, which takes the local geometry as input, and reacts to the
local flow conditions.
To define the relative velocity in the rotor, the rotational speed of the blade is used. It can be either an input from the
user, or obtained from a separate shaft model, as described in the next section.
In terms of cost, the baseline CFD simulations take about 96 CPU hours, while the BFM computations are performed
in about 16 CPU hours. The cost reduction is thus a factor 6. It should be emphasized that, in the present study, only
axisymmetric inflow conditions are treated, which makes the benefit limited as steady computations can be performed for
both the baseline and BFM calculations. In the perspective of inflow distortion cases, a whole annulus domain will be
required and, more importantly, blade computations will necessitate unsteady sliding mesh methods. On the other hand,
BFM computations can still be performed with a steady approach. In that situation, the benefit would be larger (for instance,
it is one order of magnitude larger in references [39] and [41] for such applications).

3.4 Zero-work and shaft model


In the literature, 2D analyses are derived to express the rotational speed of a windmilling rotor as a function of the
massflow (see refs. [7, 9] for instance). However, as discussed in the introduction, this requires coupling with another model
to predict the massflow, which contradicts the present objective of having a self-contained approach. Furthermore, such
approaches can not be extended to non-axisymmetric inflow conditions. Instead, the approach proposed here is to drive the
BFM computations to the windmilling state by changing the rotational speed during the calculation until the shaft is power
balanced.
We start from the fundamental mechanics equation for the transient of the low-pressure spool [1], which can be written
as:

dω ẆLP−turbine + Ẇfan + ẆSF


=− (8)
dt ωXJ

TURBO-21-1086 DUFOUR 5
where ω is the rotational speed. The term Ẇfan = ṁ2A c p (Tt2R − Tt0 ) is the power of the work done by the fan on the flow
(which implicitly depends on ω), and ẆLP−turbine = ṁcore c p ∆TtLP−turbine is the power of the work done by the low-pressure
turbine, with ṁ2A and ṁcore the overall and core massflow, respectively. Finally, ẆSF is the power dissipated by the shaft
friction (which also depends on ω), and XJ is the polar moment of inertia of the entire low speed spool.
The power contribution of the LP turbine is usually neglected because of the very high bypass ratio [5, 4, 8, 9]. Under
this assumption, the preceding relation can be expressed as follows:

dω Ẇfan + ẆSF
=− (9)
dt ωXJ

In the present case of the fan at windmill, this equation simply leads to a steady state solution ( dω 2
dt = 0 rad/s ) where the
fan power compensates the shaft friction losses. A specific model is used here to take these losses into account, since they
scale according to a third-degree polynomial of the rotational speed for the tested engine [43]. The equation can be further
simplified by assuming no friction losses on the shaft (ẆSF = 0 W), in which case the steady state solution gives a fan
providing zero work.
This transient equation can be implemented in the simulation to compute ω at each pseudo-time step, and input it to the
BFM model. The power of the fan is computed using the massflow, and mass-averaged values of total temperature at the
inlet and outlet of the domain. Therefore, in a single computation, the BFM model and the zero-work model will drive the
calculation to a power-balanced or zero-work situation, with both the rotational speed and the massflow being results.
In practice, omitting the 1/ω term in the right-hand side of Eq. (9) was found to improve the stability of the approach.
Although this sacrifices the dimensional consistency of the implemented equation, this term is meaningless when the steady
state is attained. In the same line, the XJ term was selected to optimize convergence and stability, regardless of its actual
value (which is unknown here). The initial value for the modeled rotational speed can be set up to about ±50 % than the
expected value. Finally, the initial flow solution is a constant field.
As can be expected, adding the rotational speed model slows down the convergence rate of the BFM calculation by a
factor of 1.33.

4 Validation and analysis of windmilling predictions


4.1 Prediction of the rotational speed and massflow
With the shaft model and the boundary conditions described above, windmilling points obtained experimentally are sim-
ulated with the BFM approach. Figure 3 shows the fan reduced rotational speed as a function of the inlet reduced massflow
for the BFM model and the engine experimental data, where each point corresponds to an equivalent flight Mach number
condition in the range described in Tab. 1. The BFM computations are performed with either the zero-work assumption or
the shaft friction model.
With the shaft-friction model, the discrepancy between the BFM predictions and the experimental data increases with
the flight Mach number, from 0.5 % to 6.5 % for the rotational speed, and from 2.6 % to 4.6 % for the massflow.

2.6
BFM (Shaft model) M∞ = 0.28
2.4 BFM (Zero-work model)
Reduced rotational speed Nred [-]

EXP
2.2

2.0

1.8

1.6

1.4

1.2

1.0
M∞ = 0.14
0.8
0.06 0.08 0.10 0.12 0.14 0.16 0.18
Inlet reduced massflow ṁred−2A [-]

Fig. 3. Validation of the prediction of the windmilling characteristics: massflow and rotational speed predicted by the BFM approach com-
pared with the experimental values over a range of equivalent flight Mach number from 0.14 to 0.28.

TURBO-21-1086 DUFOUR 6
The two curves present only a slight offset, which indicates that for a specified massflow, the value of the rotational
speed is within 1 % of the experimental data. In other words, it can be hypothesized that the BFM approach reproduces
accurately the work exchange, and that the principal error is an overestimation of the massflow, which can be attributed to
an underestimation of the losses. The following analysis aims to assess this hypothesis.

4.2 Analysis of the work exchange across the rotor


To assess the accuracy of the overall work exchange predictions, the performance of the fan rotor in terms of total
temperature rise against reduced massflow is shown in Fig. 4. Two iso-speed lines are considered, corresponding to the
experimental rotational speeds at windmill for the first and last points of figure 3: (i) the 10 %Nn line obtained for M∞ = 0.14
and (ii) the 20 %Nn line obtained for M∞ = 0.28. For the baseline and BFM computations, the rotational speed is thus
imposed and the back pressure varied. The experimental total temperature rise is not represented here since its measurement
uncertainty (±2 K) exceeds the plot range.

0.6
Total temperature variation ∆Tt [K]

0.5
iso Nred = 16 %Nred−nom
0.4

0.3

0.2

0.1

0.0

−0.1

−0.2 Baseline CFD


BFM iso Nred = 8 %Nred−nom
−0.3
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200 0.225
Inlet reduced massflow ṁred−2A [-]

Fig. 4. Work exchange across the rotor.

The maximum discrepancy between the baseline simulations and the BFM predictions is 0.02 K, which confirms the
accuracy of the BFM approach to predict the overall work exchange in such sub-idle operating conditions.
To analyze the local accuracy of the approach, radial profiles downstream of the rotor are presented in Fig. 5, for the
first (M∞ = 0.14) and last (M∞ = 0.28) windmilling points presented in Fig. 3.
c p ∆Tt
The radial profiles of work coefficient ψ = (ωR) 2 of Fig. 5(a) show two important features of the flow structure in
a windmilling rotor: (i) the inboard sections operate in compressor mode while the outboard sections operate in turbine
mode, the overall work being close to zero, as extensively discussed in the literature [5, 8, 7, 44, 6]; and (ii) the flow is
self-similar [7, 9] since, for a given model, the curves for the two windmilling are nearly superimposed. This features are
reproduced by the BFM approach. Close to the tip, larger discrepancies between the BFM and baseline results can observed,
though they correspond to total temperature variation differences of about 0.5 K. As discussed in previous work [10], this is
related to a separated flow in the rotor in this area.
Figure 5(b) shows the radial profiles of the absolute flow angle, including the experimental data. The experimental
uncertainties reported in ref. [8] are shown as shaded areas around the experimental points. Both the BFM and baseline
CFD results are within the experimental uncertainty (2.8◦ ) over most of the span. Close to the blade tip (above 80 % span),
discrepancies between the BFM and experimental results can reach 8◦ , and 10◦ with respect to the baseline CFD.
This analysis confirms that the overall work exchange is well captured (within 0.02 K), even though local discrepancies
are more significant (up to 0.5 K).

4.3 Analysis of the loss predictions


The focus is now on analyzing how the losses across the stage are predicted by the BFM approach.
The performance predictions for the rotor, the stator and the stage are presented in Fig. 6, for the same iso-speed lines as
in the previous section. The experimental data obtained on the engine is included as a single point for each rotational speed,
corresponding to the windmilling points of figure 3.

TURBO-21-1086 DUFOUR 7
1.0 1.0

0.8 0.8

0.6 0.6

h/hmax [-]
h/hmax [-]

Turbine mode

Zero work
0.4 0.4
Compressor mode

EXP M∞ = 0.28
0.2 0.2 Baseline CFD M∞ = 0.28
Baseline CFD M∞ = 0.28 BFM M∞ = 0.28
BFM M∞ = 0.28 EXP M∞ = 0.14
Baseline CFD M∞ = 0.14 Baseline CFD M∞ = 0.14
BFM M∞ = 0.14 BFM M∞ = 0.14
0.0 0.0
−0.4 −0.2 0.0 0.2 0.4 −30 −20 −10 0 10 20
Work coefficient ψ [-] Absolute flow angle [deg]

(a) Work coefficient. (b) Absolute flow angle.

Fig. 5. Radial profiles downstream of the rotor for the windmilling points. For the sake of clarity, only one out of five points are plotted for the
BFM results.

1.0075 0.016
iso Nred = 16 %Nred−nom Baseline CFD
0.014 BFM iso Nred = 16 %Nred−nom
Rotor total–total pressure ratio [-]

1.0050
EXP M∞ = 0.28
Stator loss coefficient Kp [-]

EXP M∞ = 0.14
1.0025 0.012

1.0000 0.010

0.9975 0.008
iso Nred = 8 %Nred−nom iso Nred = 8 %Nred−nom
0.9950 0.006

0.9925 Baseline CFD 0.004


BFM
0.9900 EXP M∞ = 0.28 0.002
EXP M∞ = 0.14
0.9875 0.000
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200 0.225 0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200 0.225
Inlet reduced massflow ṁred−2A [-] Inlet reduced massflow ṁred−2A [-]

(a) Rotor pressure ratio (b) Stator total-pressure loss coefficient

1.010
Stage total–total pressure ratio [-]

1.005

1.000

0.995

0.990

0.985
Baseline CFD
BFM
0.980 EXP M∞ = 0.28
EXP M∞ = 0.14
0.975
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200 0.225
Inlet reduced massflow ṁred−2A [-]

(c) Stage pressure ratio

Fig. 6. Global loss predictions across the stage.

TURBO-21-1086 DUFOUR 8
Figure 6(a) shows the rotor total pressure ratio against the non-dimensional massflow. For low massflows, the BFM
and baseline predictions match well. Around the windmilling points, the discrepancy between the experiments and the BFM
P −P
predictions is about 0.1 %. Figure 6(b) shows the stator loss coefficient K p = t−2RPt−2Rt−21A , indicating very large discrepancies
of about a factor 10. Again, this can be attributed to a largely separated flow, associated to a negative angle of about −20◦ ,
as discussed in ref. [44] for the same configuration.
However, for the stage, Fig. 6(c) shows that the errors for the components combine to an error of about 0.1 % for the total
pressure ratio at the M∞ = 0.14 windmilling point, and about 1 % at the M∞ = 0.28 point. In other words, while the loss in the
stator is largely understimated by the BFM approach, the overall stage pressure ratio is underestimated by 1 % at maximum,
which is associated to the 4.6 % error on the massflow discussed before. For massflows higher than the windmilling points,
the errors of the BFM further increase, which can be related to a sharp loss increase, as discussed by Ortolan et al. [45] for
load-controlled windmilling.
To further understand the discrepancies in the stator, radial profiles downstream of the stator are presented in Fig. 7, for
the operating points corresponding to the windmilling points presented in Fig 3.

1.0 1.0

0.8 0.8

0.6 0.6
h/hmax [-]

h/hmax [-]

0.4 0.4

EXP M∞ = 0.28
0.2 Baseline CFD M∞ = 0.28 0.2
BFM M∞ = 0.28
EXP M∞ = 0.14
Baseline CFD M∞ = 0.14
BFM M∞ = 0.14
0.0 0.0
0.95 0.96 0.97 0.98 0.99 1.00 1.01 0.00 0.05 0.10 0.15 0.20 0.25 0.30
Absolute pressure ratio [-] Absolute Mach number [-]

(a) Absolute pressure ratio (b) Mach number

Fig. 7. Radial profiles downstream of the stator for the windmilling points. For the sake of clarity, only one out of five points are plotted for
the BFM results.

Concerning the stage total pressure ratio shown in Fig. 7(a), the results of the baseline CFD are within the experimental
uncertainties (0.2 %) over the whole span for both operating points. The different levels of losses up to 40 % span and at the
tip for the two windmilling points are well reproduced by the BFM approach. The pressure ratio deficit observed between
40 % and 95 % span is due to the significant flow separation over the stator, which is typical of the windmilling operation
of a fan stage as reported in the literature [5, 7, 44, 9]. Such a separated flow from the blade cannot be reproduced by a BFM
model, but its impact could be modeled by resorting to calibration. This was not done in the present study to remain in line
with the goal to assess a self-contained approach.
Similar observations can be done concerning the Mach number profiles of Fig. 7(b), which clearly confirm the zones
where the massflow is overestimated correspond to the zones where the losses are underestimated.
As a conclusion, it appears that the BFM approach largely underestimates the global and local losses in the stator at the
free windmilling points, but this only amounts to an overall 1 % error in the total pressure ratio across the stage.

Conclusion and perspectives


In this paper, a body force model (BFM) coupled to a shaft model is used to predict two essential global windmilling
characteristics (rotational speed and massflow) of a turbofan. The proposed approach is self-contained as it does not require
previous knowledge of engine data or additional CFD calculations. The only model input is the blades’ geometry. The shaft
model is coupled to the BFM equation thanks to the rotational speed and the power exchanged by the fan. It is included in

TURBO-21-1086 DUFOUR 9
the calculation by a transient equation for the rotational speed, which is resolved within the pseudo time-marching process
to the steady state. The simulation cost is about 5 times less than a classical blade simulation. Validation against engine
experimental data shows that the results are within 0.5 % to 6.5 % for the rotational speed, and within 2.6 % to 4.6 % for the
massflow. The work exchange is predicted with 0.5 K, and the stage pressure ratio within 1 %. The main discrepancy cause
is a large underestimation of the losses of the stator, which degrades as the upstream Mach number increases.
The main perspective of the present study is to study windmilling characteristics for inflow distortion cases and to im-
prove the loss modeling accuracy of the BFM formulation. For such non-axisymmetric cases, full-annulus computations are
required, and the cost reduction of using a BFM approach will increase, as it allows a steady approach whereas blade com-
putations require unsteady sliding-mesh methods. This will justify the use of calibration with isolated blade computations to
improve the model accuracy.

Acknowledgements
This study is founded by the French Ministry of Defense through the DGA/AID. The authors would also like to thank
Airbus, and in particular Florian Blanc and William Thollet, for supporting activities on the body force method at ISAE-
SUPAERO.

TURBO-21-1086 DUFOUR 10
NOMENCLATURE
BFM Body Force Model/Modeling
EXP Experimental data
SF Shaft Friction
LP Low Pressure
A Local passage section [m2 ]
b Local blade metal blockage [-]
Cf Local friction coefficient [-]
cp Specific heat capacity at constant pressure [J/K/kg]
D Fan Diameter [m]
fn Force normal to the relative velocity [N/kg]
fp Parallel-to-relative-velocity force [N/kg]
fθ Azimuthal projection of the force field [N/kg]
h Spanwise distance from the hub [m]
ht Total specific enthalpy [J/kg]
I Identity matrix [-]
KMach Compressibility correction factor [-]
Kp Stator loss coefficient [-]
M2A Inlet Mach number [-]
Mr Local relative Mach number [-]
M∞ Equivalent flight Mach number [-]
ṁ2A Engine overall massflow [kg/s]
ṁcore Engine core massflow [kg/s]
ṁred Reduced massflow [-]
Nn Fan nominal rotational speed [rpm]
Nred Reduced rotational speed [-]
~n Unit vector normal to the local blade camber surface [-]
nθ Azimuthal projection of the unit vector ~n [-]
p Static pressure [Pa]
pt Total pressure [Pa]
Rex Local blade Reynolds number [-]
r Perfect gas constant [J/K/Kg]
r Radial coordinate [m]
R Fan tip radius [m]
s Blade pitch [m]
Tt Total temperature [K]
t Time [s]
~V Absolute velocity vector [m/s]
W Norm of the relative velocity [m/s]
Ẇ Power received by the flow [W]
XJ Polar moment of inertia [kg.m2 ]
x Axial coordinate [m]
γ Specific heat ratio [-]
δ Local flow deviation angle [rad]
µ Dynamic viscosity [kg/m/s]
ρ Density [kg/m3 ]
~
~τ Viscous stress tensor [Pa]
ψ Work coefficient[-]
ω Rotational speed [rad/s]

TURBO-21-1086 DUFOUR 11
References
[1] Walsh, P. P. and Fletcher, P. Gas Turbine Performance. Blackwell Science, Oxford, U.K. (2004).
[2] Daggett, D. L., Brown, S. T. and Kawai, R. T. “Ultra-Efficient Engine Diameter Study.” Technical Report No.
NASA/CR-2003-212309. NASA. 2003. URL https://ntrs.nasa.gov/citations/20030061085.
[3] von Groll, G. and Ewins, D. J. “On the Dynamics of Windmilling in Aero-Engines.” IMechE Conference Transactions,
Vol. 6: pp. 721–730. 2000. Professional Engineering Publishing; 1998.
[4] Zachos, P. K. “Modelling and Analysis of Turbofan Engines Under Windmilling Conditions.” AIAA Journal of Propul-
sion and Power Vol. 29 No. 4 (2013): pp. 882–890. DOI https://doi.org/10.2514/1.B34729.
[5] Prasad, Dilip and Lord, Wesley K. “Internal Losses and Flow Behavior of a Turbofan Stage at Windmill.” Journal of
Turbomachinery Vol. 132 No. 3 (2010): 031007 (10 pages). URL https://doi.org/10.1115/1.3147106.
[6] Gunn, Ewan J. and Hall, Cesare A. “Loss and Deviation in Windmilling Fans.” Journal of Turbomachinery Vol. 138
No. 10 (2016). URL https://doi.org/10.1115/1.4033163. 101002.
[7] Binder, N., Courty-Audren, S. K., Duplaa, S., Dufour, G. and Carbonneau, X. “Theoretical Analysis of the Aerody-
namics of Low-Speed Fans in Free and Load-Controlled Windmilling Operation.” Journal of Turbomachinery Vol. 137
No. 10 (2015). URL https://doi.org/10.1115/1.4030308.
[8] Garcı́a Rosa, N., Dufour, G., Barènes, R. and Lavergne, G. “Experimental Analysis of the Global Performance and the
Flow Through a High-Bypass Turbofan in Windmilling Conditions.” Journal of Turbomachinery Vol. 137 No. 051001
(2015). URL https://doi.org/10.1115/1.4028647.
[9] Prasad, Dilip. “Aerodynamic Similarity Principles and Scaling Laws for Windmilling Fans.” Journal of Turboma-
chinery Vol. 140 No. 12 (2018). DOI 10.1115/1.4041375. URL https://doi.org/10.1115/1.4041375.
121004.
[10] Dufour, Guillaume and Thollet, William. “Body Force Modeling of the Aerodynamics of the Fan of a Turbofan
at Windmill.” Vol. Volume 2C: Turbomachinery. 2016. URL https://doi.org/10.1115/GT2016-57462.
V02CT39A045.
[11] Braig, W., Schulte, H. and Riegler, R. “Comparative Analysis of the Windmilling Performance of Turbojet and Turbofan
Engines.” Journal of Propulsion and Power Vol. 15 No. 2 (1999): pp. 326–333. URL https://doi.org/10.
2514/2.5430.
[12] Mishra, R. K., Gouda, G. and Vedaprakash, B. S. “Relight Envelope of a Military Gas Turbine Engine: An Experimental
Study.” ASME Conference Proceedings Vol. 2008 No. 43116 (2008): pp. 55–60. URL https://doi.org/10.
1115/GT2008-50256.
[13] Hirsch, Ch., Lacor, C., Dener, C. and Vucinic, D. “An Integrated CFD System for 3D Turbomachinery Applications.”
Technical report no. AGARD-CP-510. 1991.
[14] Marble, F. “Three Dimensional Flow in Turbomachines.” Press, Princeton University (ed.). High Speed Aerodynamics
and Jet Propulsion, Vol. X: pp. 83–166. 1964.
[15] Gong, Y. Y. “A Computational Model for Rotating Stall and Inlet Distortions in Multistage Compressors.” Ph.D.
Thesis, Massachussets Institute of Technology. 1998.
[16] Kottapalli, A. P. “Development of a Body Force Model for Centrifugal Compressors.” Master’s Thesis, Massachussets
Institute of Technology. 2013.
[17] Thollet, W., Blanc, F., Dufour, G. and Carbonneau, X. “Body Force Modeling for Aerodynamic Analysis of Air
Intake–Fan Interactions.” Proc. of the 50th 3AF Interntional Conference on Applied Aerodynamics. 2015.
[18] Peters, A., Spakovszky, Z. S., Lord, W. K. and Rose, B. “Ultra-Short Nacelles for Low Fan Pressure Ratio Propulsors.”
Journal of Turbomachinery Vol. 137 No. 2 (2014). URL https://doi.org/10.1115/1.4028235.
[19] López de Vega, L., Dufour, G., Blanc, F. and Thollet, W. “A Machine Learning Based Body Force Model for Fan–
Airframe Aerodynamic Interactions.” GPPS Forum. GPPS-2018-90. 2018.
[20] Loiodice, S., Tucker, P. G. and Watson, J. “Modelling of Coupled Open Rotor Engine Intakes.” Proc. of the 48th AIAA
Aerospace Sciences Meeting. 840. 2010. URL https://arc.aiaa.org/doi/pdf/10.2514/6.2010-840.
[21] Cao, Teng, Hield, Paul and Tucker, Paul G. “Hierarchical Immersed Boundary Method with Smeared Geometry.”
Journal of Propulsion and Power Vol. 33 No. 5 (2017): pp. 1151–1163. URL https://doi.org/10.2514/1.
B36190.
[22] Cui, Jiahuan, Watson, Rob, Ma, Yunfei and Tucker, Paul. “Low Order Modeling for Fan and Outlet Guide Vanes
in Aero-Engines.” Journal of Turbomachinery Vol. 141 No. 3 (2019). URL https://doi.org/10.1115/1.
4042202.
[23] Defoe, J. J. and Spakovszky, Z. S. “Shock Propagation and MPT Noise From a Transonic Rotor in Nonuniform Flow.”
Journal of Turbomachinery Vol. 135 No. 1 (2012). URL https://doi.org/10.1115/1.4006497.
[24] López de Vega, Luis, Dufour, Guillaume and Garcı́a Rosa, Nicolás. “Fully Coupled Body Force–Engine Performance
Methodology for Boundary Layer Ingestion.” Journal of Propulsion and Power Vol. 37 No. 2 (2021): pp. 192–201.
URL https://doi.org/10.2514/1.B37743.
[25] Gong, Y. Y., Tan, C. S., Gordon, K. A. and Greitzer, E. M. “A Computational Model for Short-Wavelength Stall

TURBO-21-1086 DUFOUR 12
Inception and Development in Multistage Compressors.” Journal of Turbomachinery Vol. 121 No. 4 (1999). URL
https://doi.org/10.1115/1.2836726.
[26] Chima, R. V. “A Three-Dimensional Unsteady CFD Model of Compressor Stability.” ASME Turbo Expo 2006. 2006.
URL https://doi.org/10.1115/GT2006-90040.
[27] Liu, Xiaohua, Zhou, Yanpei, Sun, Xiaofeng and Sun, Dakun. “Calculation of Flow Instability Inception in High
Speed Axial Compressors Based on an Eigenvalue Theory.” Journal of Turbomachinery Vol. 137 No. 6 (2015). URL
https://doi.org/10.1115/1.4028768.
[28] Hsiao, E., Naimi, M., Lewis, J. P., Dalbey, K., Gong, Y. Y. and Tan, C. S. “Actuator Duct Model of Turbomachinery
Components for Powered-Nacelle Navier-Stokes Calculations.” Journal of Propulsion and Power Vol. 17 No. 4 (2001):
pp. 919–927. URL https://doi.org/10.2514/2.5825.
[29] Vadlamani, Nagabhushana Rao, Cao, Teng, Watson, Rob and Tucker, Paul G. “Toward Future Installations: Mutual
Interactions of Short Intakes With Modern High Bypass Fans.” Journal of Turbomachinery Vol. 141 No. 8 (2019).
URL https://doi.org/10.1115/1.4044080.
[30] Defoe, Jeffrey J. and Spakovszky, Zoltán S. “Effects of Boundary-Layer Ingestion on the Aero-Acoustics of Tran-
sonic Fan Rotors.” Journal of Turbomachinery Vol. 135 No. 5 (2013). URL https://doi.org/10.1115/1.
4023461.
[31] Tyacke, James C., Naqavi, Iftekhar Z. and Tucker, Paul G. “Body Force Modelling of Internal Geometry for Jet
Noise Prediction.” Radespiel, Rolf, Niehuis, Reinhard, Kroll, Norbert and Behrends, Kathrin (eds.). Advances in
Simulation of Wing and Nacelle Stall: pp. 97–109. 2016. Springer International Publishing, Cham. URL https:
//doi.org/10.1007/978-3-319-21127-5_6.
[32] Ma, Yunfei, Cui, Jiahuan, Vadlamani, Nagabhushana Rao and Tucker, Paul. “A Mixed-Fidelity Numerical Study for
Fan–Distortion Interaction.” Journal of Turbomachinery Vol. 140 No. 9 (2018). URL https://doi.org/10.
1115/1.4040860.
[33] Hill, D. J. and Defoe, J. J. “Scaling of Incidence Variations With Inlet Distortion for a Transonic Axial Compressor.”
Journal of Turbomachinery Vol. 142 No. 2 (2020). URL https://doi.org/10.1115/1.4045464.
[34] Plas, A. P., Sargeant, M. A., Madani, V., Crichton, D., Greitzer, E. M., Hynes, T. P. and Hall, C. A. “Performance of a
Boundary Layer Ingesting (BLI) Propulsion System.” 45th AIAA Aerospace Sciences Meeting and Exhibit. 2007. URL
https://doi.org/10.2514/6.2007-450.
[35] Chima, R. V., Arend, D. J., Castner, R. S. and Slater, J. W. “CFD Models of a Serpentine Inlet, Fan and Nozzle.” 48th
AIAA Aerospace Sciences Meeting. 2010. URL https://arc.aiaa.org/doi/pdf/10.2514/6.2010-33.
[36] Hall, D. K., Greitzer, E. M. and Tan, C. S. “Analysis of Fan Stage Conceptual Design Attributes for Boundary Layer
Ingestion.” Journal of Turbomachinery Vol. 139 No. 7 (2017). URL https://doi.org/10.1115/1.4035631.
071012.
[37] Thollet, William. “Body force modeling of fan-airframe interactions.” Ph.D. Thesis, ISAE-SUPAERO. 2017.
[38] Godard, Benjamin, De Jaeghere, Edouard and Gourdain, Nicolas. “Efficient Design Investigation of a Turbofan
in Distorted Inlet Conditions.” Vol. Volume 2A: Turbomachinery. 2019. URL https://doi.org/10.1115/
GT2019-90471. V02AT39A011.
[39] Benichou, Emmanuel, Dufour, Guillaume, Bousquet, Yannick, Binder, Nicolas, Ortolan, Aurélie and Carbonneau,
Xavier. “Body Force Modeling of the Aerodynamics of a Low-Speed Fan under Distorted Inflow.” International
Journal of Turbomachinery, Propulsion and Power Vol. 4 No. 3 (2019). URL https://doi.org/10.3390/
ijtpp4030029.
[40] Awes, Amaury, Dufour, Guillaume, Daon, Renaud, Marty, Julien, Barrier, Raphaël and Carbonneau, Xavier. “Unsteady
Body Force Methodology for Fan Operability Assessment under Clean and Distorted Inflow Conditions.” AIAA Scitech
2021 Forum. URL https://doi.org/10.2514/6.2021-0388.
[41] Benichou, Emmanuel, Binder, Nicolas, Bousquet, Yannick and Carbonneau, Xavier. “Improvement of the Parallel
Compressor Model and Application to Inlet Flow Distortion.” International Journal of Turbomachinery, Propulsion
and Power Vol. 6 No. 3 (2021). URL https://doi.org/10.3390/ijtpp6030034.
[42] Schlichting, H and Kestin, J. Boundary-layer Theory. McGraw-Hill classic textbook reissue series, McGraw-Hill
(1979).
[43] Boudet, René. “Transmission de puissance - Introduction.” Techniques de l’Ingénieur Vol. 33 No. 0 (1999): p. 13. URL
https://www.techniques-ingenieur.fr/res/pdf/encyclopedia/42184210-bm5600.pdf.
[44] Dufour, G., Garcı́a Rosa, N. and Duplaa, S. “Validation and Flow Structure Analysis in a Turbofan Stage at Windmill.”
Proc IMechE Part A: Journal of Power and Energy Vol. 229 No. 6 (2015): pp. 571–583. URL https://doi.org/
10.1177/0957650915587144.
[45] Ortolan, Aurélie, Courty-Audren, Suk-Kee, Lagha, Massyl, Binder, Nicolas, Carbonneau, Xavier and Challas, Flo-
rent. “Generic Properties of Flows in Low-Speed Axial Fans Operating at Load-Controlled Windmill.” Journal of
Turbomachinery Vol. 140 No. 8 (2018). URL https://doi.org/10.1115/1.4040678. 081002.

TURBO-21-1086 DUFOUR 13

You might also like