2007-01-2542

Towards a stochastic in-flight thrust determination process

João Carlos Hoff, Ph.D.
Independent Researcher

Copyright © 2007 Society of Automotive Engineers, Inc

ABSTRACT Cv Nozzle thrust coefficient

Far Fuel air ratio
Traditionally jet engines in-flight thrust FAR Federal Airworthiness Regulation
determination is a deterministic process that starting from FG Engine gross thrust (N)
the fan and core measured pressures and temperatures FHV Fuel heating value (J/Kg)
calculates, by many different ways, the intermediate FN Engine net thrust (N)
turbomachine parameters up to the exhaust nozzle pressure FR Ram drag (N)
and temperature while iteratively search for the engine FTB Flight Test Bed
mass flow, attending simultaneously energy, mass and GAMA Ratio of specific heats
momentum conservation. With this data and the previously GLTF Ground Level Test Facility
determined nozzle coefficients, the engine installed thrust HAR Water air ratio
is calculated. These methods are industry standards that are HPX Power extraction (mechanical)
well documented on SAE reports AIR 1703A and AIR ht Enthalpy (ht13, ht15, ht28, ht6,….) (J/Kg)
5450. A new approach is proposed in this work, a more J Cost Functional
stochastic approach, which in fact estimates the engine fan °K Kelvin (temperature unit)
and core pressures and temperatures from initial air mass m Meter (distance unit)
flow and gross thrust values. Using the Output-Error m8 Flow Function at nozzle
Method the values of gross thrust and air mass flow are N Number of samples
iteratively updated by a modified Newton-Raphson N Newton (force unit)
algorithm minimizing the error between the fan and core n Starter generator efficiency
calculated and measured air pressures and temperatures. nc Combustor efficiency
The new technique solves in fact, by optimization, the Pa Pascal (pressure unit)
inverse formulation of the in-flight thrust determination P Pressure (Pa)
problem. The advantages of the new technique over the PS0 Ambient pressure (Pa)
traditional one is that it has stochastic properties allowing PS16 Static pressure at engine station 16 (Pa)
to process the noisy flight test data samples without PS6 Static pressure at engine station 6 (Pa)
previous averaging. This paper presents the new technique PS6M Static pressure at mixer station (Pa)
and demonstrates its application to a set of engine Deck PS8 Static pressure at the engine nozzle (Pa)
data transformed to sampled data by application of noise. Pt0 Total pressure (freestream) (Pa)
Pt15 Total pressure at engine station 15 (Pa)
NOMENCLATURE Pt6 Total pressure at engine station 6 (Pa)
Pt6M Total pressure at mixer (Pa)
A15 By-pass area at engine station 15 (m2) Pt8 Nozzle total pressure (Pa)
A16 By-pass area at the mixer station 16 (m2) R Residual Covariance Matrix
A6 Core area at station 6 (m2) Ratio Algorithm internal parameter
A6M Mixer area (A16+A6) (m2) REM Residual Error method
AFM Airplane Flight Manual s Seconds
ATF Altitude Test Facility SAE Society of Automotive Engineers
A8 Nozzle area – nominal (m2) SQRT Square root
A8eff Nozzle area – effective (m2) T Temperature (°K)
A8EQ Nozzle area – equivalent (m2) Tt0 Total temperature (°K)
cL Vector of unknowns Tt15 Total temperature at engine station 15 (°K)
Cd Nozzle flow coefficient Tt6 Total temperature at engine station 6 (°K)
Tt6M Total temperature at engine mixer station (°K) Error Method (REM), which are of the denominated ‘gas
Tt8 Nozzle total temperature (°K) flow path’ method group.
V16 Air speed at station 16 (m/s)
V6 Air speed at station 6 (m/s) All methods used as industry standards need the
V6M Air speed at mixer station (m/s) nozzle characterization, which is done via nozzle model
V8 Air speed at nozzle (m/s) tests in a test stand. The nozzle calibration process is done
Vt True airspeed (m/s) by determining the nozzle coefficients Cv and Cd, thrust
x̂ Vector of estimated states and mass flow coefficient, respectively. These coefficients
z Measurement vector are dimensionless groups that relates actual nozzle thrust to
W Air mass flow (Kg/s) ideal nozzle thrust and actual mass flow to ideal mass flow.
W1 Air mass flow at engine intake (Kg/s) In-flight the ideal thrust and mass flow are iteratively
W15 Fan air mass flow (Kg/s) calculated and then, via the coefficients, the actual thrust is
W6 Core as mass flow (Kg/s) determined. Before this step it is required to calculate the
W6M Air mass flow at mixer (Kg/s) nozzle total pressure and temperature, which may be
W8 Air mass flow at nozzle (Kg/s) calculated from turbo machine maps, energy, mass and
Wf Fuel flow (Kg/s) moment conservation.
δ Pressure ratio
∂x This work deals with the inverse formulation of the
States partial derivatives relative to cL problem, that is, starting from initial values of gross thrust,
∂c
or nozzle total pressure and temperature, and fan and core
air mass flows, fan and core pressure and temperatures are
Subscripts
calculated. Via an optimization algorithm the error
15, 16, 6, 6M Engine stations numbering as defined on
between the calculated and the measured fan and core
SAE ARP 775C [1]
temperature and pressure are minimized by inserting
appropriately the values of gross thrust (or nozzle total
INTRODUCTION
pressure and temperature) and air mass flow (fan and
core).
The development of an airplane requires the
estimation of the installed propulsive thrust for its flight-
The advantages of the new technique over the
operating envelope. For transport category airplanes, the
traditional techniques presented in reports SAE AIR
FAR 25 [2], through AC 25-7A [3], requires the
1703A [5] and SAE AIR 5450 [6] is that it has stochastic
measurement of the installed thrust for a range of altitude,
properties allowing to process the noisy flight test data
airspeed, fan speed, and power extraction to be used in the
samples without previous averaging.
calibration of the engine cycle model (Deck). The FAR 23
[4] is less severe and may accept a calibrated Deck based
on the inlet and nozzle characterization, GLTF and FTB DEVELOPMENT
tests. The Deck data is later use in the development of the
airplane certified performance data presented in the It is proposed the use of an optimization algorithm to
Airplane Flight Manual (AFM). Jet engine installed thrust the estimation of in-flight installed engine gross thrust by
is not directly measured. In fact it is indirectly determined minimizing the error between the calculated and the
by a process that requires the nozzle characterization from measured fan and core pressure and temperatures. As the
nozzle model tests and the measurement of actual engine engine thrust and mass flow doesn’t have specific
pressures and temperatures at the same stations measured equations or state space models the Output-Error is the
in the model during nozzle model characterization tests. It ideal algorithm to be used to identify the unknowns, thrust
is not an easy task and in general requires a refined plan, and air mass flow. The problem is stated as an optimization
an accurate and expensive instrumentation, the nozzle problem where given the measured quantities as fan and
model calibration, engine calibration on GLTF, FTB, ATF, core pressures, temperatures, fuel flow, etc., the algorithm
several hours of expensive flights, and many hours of inserts values of gross thrust, fan and core air mass flow,
engineering analysis. The literature presents several by minimization of a cost functional based on the
methods of determining, deterministically, the installed calculated and measured fan and core pressures and
thrust of jet engines. Today, the SAE AIR 1703A [5] and temperatures. The cost functional is derived from the
SAE AIR 5450 [6] are the guides and reflect the state of negative logarithm of the Likelihood Function which after
the art and industry standard on the subject. Among the simplifications takes the form of Eq. 1.
methods there described the most accurate in use by the
1 N
industry are the denominated W T (Weight temperature J= ∑ ( z − xˆ ) T R −1 ( z − xˆ ) (1)
method), Ap (Area pressure method) and the Residual N − 1 K =1
 Equation 1 is minimized by a modified Newton- 8. The static pressure PS6M at the mixer station is
Raphson [7] type algorithm and the unknown gross thrust calculated from the mixer pressure model;
and air mass flow are calculated iteratively by Eq. 2 from a 9. The enthalpy ht6M at the mixer station is
given initial value: calculated from the mixer enthalpy model;
10. The mixer total temperature is calculated from the
∂ J 
2
−1 mixer enthalpy Tt6M= f(ht6M, Far, HAR) ,
∂J Far=f(Wf, W6M)
c L = c L −1 −  2  (2)
 ∂c  ∂c 11. The total pressure Pt6M at the mixer station is
calculated from PS6M Tt6M, A6M and W6M: Pt6M =
The first and second derivatives of J relative to the f(Tt6A, PS6M,A6M,V6M)
unknowns are calculated by: 12. Pt8 is calculated from Pt6M considering duct loss.
13. At the nozzle station, Tt8 = Tt6A
∂J 2 ∂ ( z − xˆ ) 14. The flow function is calculated as: m8 = f(Pt8/Pt0,
= ∑ N ( z − xˆ ) T R −1 (3) GAMA), GAMA=f(Tt8, Far. HAR)
∂c N − 1 k =1 ∂c
15. A8eff = W8*SQRT((Tt8)/(Pt8*m8)) with W8=W6M
16. CALL CDCV(Pt8,PS0,Cd,Cv) subroutine to
∂ 2J 2 ˆ ) T −1 ∂ ( z − xˆ )
N ∂ (z − x determine Cv and Cd coefficients;
= ∑ k =1 R (4)
∂c 2 N −1 ∂c ∂c 17. A8EQ = A8*Cd;
18. Iterate W1 repeating steps 2 to 17 until A8eff=A8EQ
In Eq. 1, 3 and 4 above R is the matrix of the 19. Ps8 , V8 are calculated from Pt8, Tt8, A8 and W8.
covariance of the residuals which is calculated by: 20. FG = Cv*( W8*V8 + (PS8-PS0)*A8eff)
Gross
Thrust
1 N 21. FR = W1*Vt
Ran Drag
R= ∑ ( z − xˆ )( z − xˆ ) T (5) 22. FN = FG-FR
Net Thrust
N − 1 K =1
Figure 1 below summarizes the direct method, or
The estimates accuracy may be evaluated by the deterministic algorithm.
Cramér-Rao lower bound (variance theoretical minimum
values), which can be calculated from terms of Eq. 2, or its
improved version for colored noise, as presented in
Morelli, Klein [7].

The algorithm allows processing a set of samples

recorded during an engine/airplane stabilization taking into
account the noise of the recorded data. To demonstrate the
use of the estimation process a simple but effective (and
intentionally inaccurate) in-flight thrust algorithm
applicable to a long cowl turbofan with mixer has been
defined and the deterministic and the more statistical
algorithms are compared.

THE DETERMINISTIC ALGORITHM

The algorithm that follows is a simplified

deterministic, and intentionally inaccurate process of in-
flight thrust and mass flow determination from measured
values of fan and core pressure and temperatures:

1. Given the measured values Tt0, Tt15, Pt15, Tt6, Pt6,

Vt, bleeds and Wf;
2. Assume an initial value for the air mass flow W1;
3. From energy and flow conservation W1 is split in
W15 and W6, the fan and core mass flow,
respectively;
4. Pt16 is calculated from Pt15 using the duct loss.
5. Tt16 = Tt15
6. W16 = W15 - Bleed
7. At the mixer station W6M = W6+W16. Figure 1 – Direct algorithm
 Figure 2 summarizes the inverse problem or
stochastic algorithm;
THE INVERSE PROBLEM

From initial values of Pt8, Tt8, Ratio, W15 and W6 the

temperatures and pressure Pt15, Tt15, Pt6 and Tt6 are
estimated.

1. Given the measured values of Wf and bleeds;

2. Given the initial values of Pt8, Tt8, W15 and W6;
3. Tt6M = Tt8, W6M=W15+W6-Bleed
4. Far = f(Wf, W6M);
5. The mixer enthalpy is calculated from the known
total temperature, ht6M = f(Tt6M, Far, HAR);
6. Enthalpy model at mixer station: ht6M = f(Tt6M,,
ht16, h6, Far, HAR)
7. From the energy conservation:
W1*ht0 + Wf*FHV*ηc = W15* ht15 + W6* (1+Far) *
ht6 + W13 * ht13 + W28 * ht28+ W3 * h3+ HPX/n
8. Solving 6 and 7, the enthalpy at stations 15 and 6,
ht6 and ht15 respectively, can be determined from
which Tt6 and Tt15 are estimated. In addition;
9. Tt15=Tt16;
10. Pt6M is calculated from Pt8 considering duct loss;
11. With Pt6M, Tt6M, A6M calculate PS6M, V6M;
12. From PS6M , Tt6, Tt16 and the pressure model, the
total pressures Pt6 and Pt16 are estimated;
13. Pt15 is calculated from Pt16, taking into account
the pressure loss.
14. Mass flow function m8=f(Pt8/PS0, Gama);
15. A8eff is calculated from m8;
16. The algorithm shall attend A8eff=A8EQ where A8EQ Figure 2. The Inverse problem.
= A8*Cd where A8EQ is used as a pseudo
measurement. Cd and Cv from subroutine;
17. PS8 and V8 are calculated from Pt8, Tt8, A8 and
W8. RESULTS
18. Calculate the partial derivatives of Pt15, Tt15, Pt6
and Pt16 to the unknowns (Pt8, Tt8, W15, and W6). The above described estimation process has been
19. Calculate the derivatives given by Eq. 3 and 4 applied to a set of engine data. The data set has been
using the partial derivatives of step 18, the derived from an engine Deck from which one set of fan
measurement and the estimates, processing all and core pressure and temperature and the associated gross
available samples; thrust and air mass flow has been obtained. White noise
20. Update the estimates by Eq. 2 has been added to the data set generating 50 samples.
21. Update R using Eq. 5 Noise characteristics have been obtained from real engine
22. Check for a convergence criterion – perform new data taken from in-flight thrust measurement tests. Initially
iteration if required starting at step 1 using the the data has been used as it was a GLTF data calculating
updated Pt8, Tt8, W15, W6; the nozzle Cv and Cd coefficients to be used latter in both
23. FG = Cv*( W8*V8 + (PS8-PS0)*A8eff))
Gross estimation processes. The new algorithm has been
Thrust implemented together with the direct algorithm, that is, the
24. FR = W1*Vt
Ram Drag algorithm that calculates thrust and mass flow from the
25. FN = FG-FR
Net Thrust measured fan and core pressure and temperature. The
calculated Tt8, Pt8, W15 and W6 of the direct algorithm
A minimum criterion shall be defined for the cost were used as initial values to the inverse problem, that is,
functional J or for two consecutive estimates and used to the stochastic algorithm. The new estimation process
stop the iteration process. Upon reaching final values of produced gross thrust and air mass flow very close to the
Pt8, Tt8, W15 and W6 the gross thrust is calculated from the results of the direct method and both reproduced very well
flow function, Cv, Cd, etc., ending the whole process.
the source data. Table 1, below presents the resultant gross
thrust and air mass flow for both algorithms.

Table 1 – Estimated Gross Thrust and Mass Flow

Direct Stochastic
Algorithm Algorithm
Gross Thrust (N) 7792.0 7799.0
Fan Mass Flow (Kg/s) 21.94 21.94
Core Mass Flow (Kg/s) 4.60 4.60

Figure 3, below shows the Cost Functional values

plotted against the algorithm iteration number. Figure 4 Figure 4. Total temperature Tt8 (°K) versus
shows the Tt8 temperature along the iterations and its iteration
convergence to final values. Figure 5 shows the nozzle
total pressure estimated values along the algorithm
iterations. Figure 6 shows an algorithm internal parameter.
All estimated parameters converged to final values
indicating the good comportment of the algorithm and
converged always to the same final values when perturbing
one or more initial values. Included in the figures is the
Cramer-Rao bound of the estimated parameter at the last
presented iteration, as a vertical bar.

Two different algorithm configurations have been

implemented, with and without the constraint A8EQ=A8eff.
Both presented similar results, however, the inclusion of
the constraint may improve algorithm identificability since
it is equivalent to add one state and one measurement to
the problem. The constraint has been implemented as a
pseudo measurement, that is, A8EQ=A8*Cd was used as an
Figure 5. Total pressure Pt8 (Pa) versus
accurate measurement. Partial derivatives were numerically
iteration
calculated by central differences.

Figure 6. Ratio – algorithm internal parameter

FINAL COMMENTS
Figure 3. Cost Functional values versus
A probabilistic formulation to in-flight thrust
algorithm iterations
estimation has been presented. The new process finds the
unknowns, the thrust and air mass flow, through an
optimization process minimizing the error between the REFERENCES
calculated and measured values of fan and core pressures
and temperatures. The main advantage of the new [1] SAE ARP 775C. Gas Turbine Engine Performance
estimation process it that it allows the use of a large set of Station Identification and Nomenclature. SAE, USA,
data samples without previous averaging of the samples. December 1997.
As a statistical process the calculated thrust and air mass
flow reflects the whole data set and is updated iteratively [2] PART 25 – Airworthiness Standard: Transport
as more samples from the measurements are processed Category Airplanes. FAA, Department of Transport, USA.
during the calculation process.
[3] AC 25-7A – Flight Test Guide for Certification of
The process has been demonstrated for a simple and Transport Category Airplanes. FAA Department of
intentionally inaccurate process (to demonstrate the new Transport, USA, 6/3/99.
formulation and not the thrust calculation process).
However, may be applicable to more complex in-flight [4] PART 23 – Airworthiness Standard: Normal, Utility,
thrust formulation. The main difficulty a user will have is Aerobatic, and Commuter Category Airplanes. FAA,
in the formulation of the inverse problem, that is, Department of Transport, USA.
calculating the pressures and temperatures from given
values of thrust (or nozzle total pressure and temperature) [5] SAE AIR 1703A. In-Flight Thrust Determination. SAE,
and air mass flow because more information is required USA Feb. 20, 2006.
from the turbomachine - the inverse process may require
more information than the direct process because it has [6] SAE AIR 5450. Advanced Ducted Propulsor In-Flight
more unknowns. However, the optimization process is Thrust Determination. SAE, USA, Draft 2006-06-15.
capable of identifying the additional unknowns added by
the new formulation. The nozzle characterization, that is, [7] Morelli, E., A. Klein, V. Determining the accuracy of
the determination of the nozzle coefficients from nozzle Maximum Likelihood parameter estimates with colored
model tests, shall take into account a direct process residuals. NASA Report 194893, 1994.
equivalent to the inverse problem. It has been observed
some convergence difficulties due to high sensitivity of the RESPONSIBILITY
calculated states, fan and core pressures and temperatures,
to the nozzle temperature in the inverse problem what has The author is the only responsible for the printed
been solved by introducing a reduction factor to the material included in this paper.
correction vector generated at every update step, as
presented in Eq. 6 below (FAC<<1.0); ACKNOWLODGEMENT

 ∂ 2 J  ∂J The author would like to express its grateful thanks

C L = C L −1  2  FAC (6) to Eng. Rodrigo De La Fuente and Dr. Hugo Borelli
 ∂C  ∂C Resende which provided the means and encouraged the
study of in-flight thrust determination in Brazil.
The algorithm is also dependent of good initial
estimates, due to the sensitivity explained above. To ABOUT THE AUTOR
overcome this problem the direct algorithm has been used
as startup. However, disturbing the initial values the João C. Hoff has a B.Sc. and M.Sc. in Mechanical
algorithm is able to converge always to the same solution. Engineering, and Ph.D. in Flight Dynamics. Is in the
The algorithm is fast and may process one hundred aerospace industry since the 70’s developing mainly flight
samples in few seconds. The problem of convergence has test related activities (aircraft performance, armament and
been associated to difficulties in the ht15 and ht6 enthalpy system tests). [email protected]
calculations where depending of the amount of correction
introduced by the algorithm update at the end of an
iteration cycle ht15 and ht6 may assume very large or
negative values, invalidating results and partial derivative
calculation. Additional software development may solve
the problem.