Professional Documents
Culture Documents
1992-170
AlAA 92-0170
A Flight Test Study to Determine
Aircraft Propulsive Efficiency and
Aerodynamic Drag Using Two
Parameter Identification Methods
and Lock's Propeller Model
T. Taylor, G. Bennett, D. Coble,
K. Hall, and P. Bridges
Raspet Flight Research Laboratory
Mississippi State University
Mississippi State, MS
6 - $0
Nomenclature
AR
C
c,
C,
Climb Angle
Ambient Air Density
Blade Angle
8 ab
Graduate Student
AlAA Student Member
t Director, Raspet Flight Research Laboratory
AIAA Member
Associate Professor, Mississippi State University
AIM Member
Introduction
It has always been difficult to separate thrust and
drag forces acting on a propeller driven aircraft. The
usual flight test method used to determine aircraft drag
LJ
'
W
The slender nacelles of this aircraft should minimize
nacelle effects on the propeller.
High quality
instrumentation and data acquisition systems were
installed for data collection. In addition, extensive
calibration of the sensors and airdata boom were
conducted to minimize measurement errors. This aspect
of the project is being reported by Chaney, et. al. Is.
All post flight data processing was performed on a
25 megahertz 386 class personal computer equipped with
a 387 math coprocessor and Microsoft@Professional
Development Fortran Version 5.1. The typical fifty
iteration rnn time for both parameter identification
algorithms was approximately twenty minutes for fifty
five data points.
The discussionwhich followssummarizes briefly the
development of an aircraft performance analysis method
using Lock's propeller r d e l and hvo parameter
identification algorithms. More details are given in
Sabzehparvar 12, Taylor 16, B o b I', and Coble I'. The
fundamental assumption made is that all propeller-nacelle
installations will have blade lift and drag polars of
similar shapes; consequently, the measured aircraft
performance data can be processed using a gradient
search or Kalman filter parameter identification
techniqueto determine simultaneously the coefficientsof
v
the blade lift and drag polars and the aircraft drag
coefficients. Lock's single element propeller model is
summarized and verification of the concept is shown.
The aircraft performance model is presented along with
the cost function used to drive the parameter
identificationmethods. Finally, the results of the study
are presented.
Theorv
T=0.5p NbcW2Rx(CL,cos$
- C,.,sin$)
(3)
P, =0.5pN,cRxV3cos2V~(Cl,cosp - C,,,Sin$)
vcsinp
(1 +-
COSd )%b
(4)
c, + C p ,
+ C,a,
I*
12.d
(2)
0.560703
C, = 0.103223
C, = 0.009942
c, = 0.002204
(Table 1)
C, = 0.001327
BuIlE ANcm
[.I " u m m
*I 702 RU)M
"
j 8 = 3s
M V A N C E RATIO [I]
~
_
_
_
Figure 5. Comparison of Predicted and Calculated
Propeller Efficiencies for 690 Propellers.
L-w,.
(6)
h = vsin7
v7 .
(7)
["
-W
8 dt
(Ut
Cl
SI
for j = 1, 2,
variable:
dx,
{dx: + dx: + ... + dx:
... , 7.
(19)
A C l = s l A x l = s,m,Al
(20)
(22)
The quantitiy af, I aCi is the first partial derivative
evaluated at C, using a central difference formula. The
new values of each parameter are computed
Cl = Cp + AC,
(a)
\c/
u-)
x:
K,
P,(-)HZ&(-N
128)
H k ( ~ ( - ) ) P ~ ( - ) H ~ ( ~ +R,
( - ) )
Using the gain matrix, the bRst estimate is computed
next. The state vector is put through the measurement
matrix
to form a forecast of the measurement.
Subtracted from the original measurement is the forecast
that yields a correcting factor. The correcting factor
passes through a forward gain of & to produce an
v
&(-)
PJ+) = [I
- KP&(-))IP~(-)
(30)
2.
s, = 0.700
s, = 0.300
s, = 0.050
s,
0.020
(Table 2)
s, = 0.010
s, = 0.030
s, = 1.600
Discussion of Results
Coawuter Generated Data
In order to determine the performance
characteristics of both parameter identification
techniques, a set of data was generated for both level
flight conditions and constant blade angle conditions
using a known set of parameters. A computer program
was written to generate data similar to the target flight
conditions using the initial guesses for the propeller
curve fit coefficients and assumed values for the zero
SI
0.600
s, = 0.103
s, = 0.043
s, = 0.019
s, = 0.019
s, = 0.030
s, = 0.700.
L.,
S
(Table 3)
,.m,
4
PERTURBATION IN COEFFICIENTS
2OX PERTURBATION IN COEFFICIENTS
u 10% PEENRBATION IN CODnClENTS
W 6): PERTURBATION IN COEfWClENTS
- EXACT VALUE OF COEFFICIENT
-SO%
v
Figure 9. Gradient Search Cast Function Iteration
History Using Computer Generated Level Flight Test
Data.
ITClUTIONS
rapidly and did not deviate ooce they reached steadystate. The other parameters, which describe the
propeller lift and drag, behaved in a similar manner.
Figures 17 and 18 show the convergence of the propeller
efficiencies for the level flight and constant blade angle
cases. By the fiftieth iteration both cases had converged
to the correct values.
It was found that the Q matrix affected the speed
that the data converged; therefore, the Q matrix was
modified by trial and error to provide the best
convergence for all the parameters. This technique is
often referred to 8s "tuning the filter". The modification
of the Q matrix was necessary for the computer
generated data and the actual flight test data.
5% PEF3URBAnON IN COEFFICIENTS
10% PERTURBATION IN COEFFICIENT3
20% PERTURBATION IN COEFFICIKNTS
u 50% PERNRBATION IN COEFiIClXNTS
EXACT VALUE OF COEFFICIENT
ITERATION NUYBER
PDRIURBITION IN COOFICIENTS
10% PERNRBATION IN COEFFlClENn
20% PERTURBATION IN COEFrlClENtS
509 PRRTURBAnON IN COFPFIO?MT
FUCT vaue OF COEFFICIENT
.
I5%
I
**1*1
- E
Results. The initial
conditions chosen for the extended Kalman filter were
the values that were used to generate the computer data.
These values were then perturbed by 5 % . 1096, 20%,
and SO% of their exact values to test the convergence
ability of the filter. Each of the data points was passed
through the filter fifty times lo produce Figures 15
through 18.
Figures 15 and 16 show that the Oswald efficiency
factor and the zero lift drag coefficient converged very
_
.
I
6.OW
10 000
FP DATA
IS DATA
16:OOO FT DATA
b4U4 20.000 IS DATA
*u. ALTIWDES COYDINED
12
jow
Lorn
6000 Fp DATA
1O:OOO IT DATA
1S.000 IT DATA
M+W 20 ooo
DATA
Uf. ALTITUDES COMBINED
_D
rr
Ejm,
80,
@*
O
L 2 0
om0
WEBATIONS
0.beo 5.000
W 10.000
PI
DATA
IT DATA
15.000 FT DATA
UIIO 20.000 IT DATA
C
A I L UTITUDES COMBINED
0.110
Data.
5,000 IT DATA
10 WO IT DATA
t 5 ' W IT DATA
M+MZO'OW IT DATA
ALTITUDL3 COMBINED
_D
3 a.s*
4
P
0.w
_D
d.
13
5 . W FI DATA
1 O . W FI DATA
1 5 . W T I DATA
t U W 20 W R DATA
UnlVDCs COMBWED
04eW
Ad
0.-
munow
Figure 26.
Extenended Kalman Filter Iteration
History of Zem Lift Drag CWrrcient for SteadyState Level Fliiht Test Data.
It is not clear why the propeller drag coefficients
did not converge after fifty iterations since the computer
generated test case produced good results. One
explanation is that there was a bias in the data that was
not detected. Also the manufacturer's values of jetengine thrust were used without being experimentally
confirmed. An inadvertent error during programming
proved that the weight plays a big role in the
convergence of the Oswald efficiency factor and the lift
drag coefficient. Since the weight was taken to be a
constant value, this could be another source of error.
In addition, the engines were assumed identical and not
modeled separately. Lastly, there may be additional
terms that were not taken into account in this model.
Conclusions and Recommendations
Two parameter identification techniques have been
developed to extract and predict aircraft performance
characteristics from steady state flight test data. Both
methods use Lock's propeller model to determine
propeller thrust. which has been proven as an accurate
model with manufacturer's propeller performance data.
The following conclusions and recommendations are
made after careful evaluation of this research.
1. A single element propeller model has the capability
of accurately describing propeller performance over a
wide range of operating conditions with two low order
curve fits.
2. A robust flight test method that fully exercises the
propeller flight envelope was attempted as an alternative
to usual level speed-power maneuvers.
Some
difficulties encountered with this method were that a
skilled pilot and good atmospheric conditions are
necessary to obtain steady-state flight test data.
References
'd
L/