Chinese

Journal of
Aeronautics
Chinese Journal of Aeronautics 22(2009) 18-21 www.elsevier.com/locate/cja

Aircraft Engine Sensor Fault Diagnostics Based on Estimation of

Engine’s Health Degradation
Xue Wei, Guo Yingqing*
School of Power and Energy, Northwestern Polytechnical University, Xi’an 710072, China

Received 16 January 2008; accepted 22 April 2008

Abstract
A duty in development of an on-line fault detection algorithm is to make it associate with estimation of engine’s health degra-
dation. For this purpose, an on-line diagnostic algorithm is put forward. Using a tracking filter to estimate the engine’s health
condition over its lifetime, can be reconstructed an onboard model, which is then made to match a real aircraft gas turbine engine.
Finally, a bank of Kalman filters is applied in fault detection and isolation (FDI) of sensors for the engine. Through the bank, the
real faults that have occurred can be detected and isolated. The on-line fault detection algorithm has the ability of maintaining
the effectiveness over the engine’s lifetime and is verified by simulation using a nonlinear engine model.

Keywords: aerospace propulsion system; Kalman filter; health degradation; sensor fault diagnostics

1. Introduction1 limited applications in practices. A Kalman filter was

applied to aircraft sensor and actuator fault diagnosis
Fault detection and isolation (FDI) logic plays a key by C. Hajiyev and F. Caliskan[8]. This approach was
role in enhancing the working safety and reliability based on the faults affecting the mean of the Kalman
and reducing the operating cost of aircraft propulsion filter innovation sequence. A sensor fault that shifted
systems. However, it is a challenging to ensure high the mean of the innovation sequence could be detected
reliability of the FDI task[1]. For this purpose, a variety and isolated.
of approaches were proposed[2-13]. In general, in-flight diagnostic systems are designed
W. C. Merrill, et al.[5] used a bank of Kalman filters under a nominal health, or a non-degrading condition,
for aircraft engine sensor FDI. That study successfully which constitutes a reference baseline for diagnostics.
improved the control loop tolerance to sensor failures, Any deviation observed in engine outputs from the
which were considered as the engine failures most reference values may indicate the presence of a fault.
likely to happen under the harsh operating conditions. As an engine degrades overtime, in-flight diagnostic
In that study, actuator failure was not taken into ac- systems may lose their effectiveness. Engine health
count. In the work by T. Kobayashi and D. L. Simon[6], degradation is a normal aging process that occurs in all
an FDI system which used a bank of Kalman filters aircraft engines due to usage, and therefore can not be
was developed for aircraft engine sensor and actuator regarded as faults. However, similar to other faults,
FDI in conjunction with the detection of component degradation causes the engine outputs to deviate from
faults. The results indicated that the proposed FDI the normal. When an engine’s output deviation reaches
system would be promising for reliable diagnostics of a certain level, the diagnostic system may misinterpret
aircraft engine sensors and actuators. An analytical it as a fault and generate a false alarm.
redundancy-based approach for detecting and isolating One approach to maintain the effectiveness of
sensor, actuator, and component faults in complex dy- in-flight diagnostic algorithms, when it is applied to
namical systems, such as aircraft and spacecraft, was degrading engines, is to be periodically updated or
developed by E. C. Larson, et al.[7]. This method had redesigned based on the estimates of health degrada-
tion. Health degradation can be judged by trend moni-
*Corresponding author. Tel.: +86-29-88492749. toring systems. Through the updating based on the
E-mail address: [email protected] estimated health degradation, the health baseline of an
Foundation item: APTD Program in-flight diagnostic system can be adapted to the de-
1000-9361/$ - see front matter © 2009 Elsevier Ltd. All rights reserved.
graded engine, and thereby the system is able to diag-
doi: 10.1016/S1000-9361(08)60064-3 nose a fault with effectiveness.
No.1 Xue Wei et al. / Chinese Journal of Aeronautics 22(2009) 18-21 · 19 ·

In this article, the approach to diagnose sensor faults

based on the engine’s health degradation is discussed.
Using the Kalman filter to estimate the deterioration of
the engine, the onboard model can be reconstructed,
and a bank of Kalman filters is applied in FDI of sen-
sors for the aircraft gas turbine engine.

2. Engine Models

The engine model used in this study is for the Fig.1 Model-based control and diagnostics logic.
nonlinear simulation of an advanced military twin-
spool turbofan engine. The deterioration of engine x    Bu
Ax   w °½
(3)
performance is modeled by adjustments of efficiency
y    Du  v ¾°
Cx
or flow coefficient scalars of the following four com- ¿
ponents: fan (FAN), booster (BST), high-pressure tur- where
bine (HPT), and low-pressure turbine (LPT). These
ª xº  ª A Lº  ªBº 
scalars representing the component performance dete- x «h» , A «0 0»,B « 0 » ,C >C M@
rioration are the health parameters. The engine state ¬ ¼ ¬ ¼ ¬ ¼
variables are low-pressure spool speed (XNL) and The estimated state vector x e , the sensor meas-
high-pressure spool speed (XNH); health parameters
urements of ye, and the Kalman filter gain matrix can
are FAN, BST, HPT, and LPT efficiencies; actuators
be found with the Kalman filter of the form:
are fuel flow (WFB) and nozzle area (A8), and sensors
used in the study are XNL, XNH, booster exit pressure x e    Bu
Ax  K y y ½
e e
(P31), LPT exit pressure(P6) and LPT inlet temperature
 °°
(T45). ye Cx e  Du ¾ (4)
The FDI logic uses the Kalman filter approach to  T 1 °
K PC R °¿
estimate the state variables and health parameters, en-
gine outputs from a set of sensors, and controlling where matrix P is the solution of the following steady-
commands. A linear model under consideration is rep- state Riccati equation:
resented by the following state-space equations:
  PA T  PC T R 1CP
AP  Q 0 (5)
x Ax  Bu  Lh  w ½
¾ (1)
y Cx  Du  Mh  v ¿
4. Fault Detection Algorithm for Sensors
where the vectors x, h, and u represent the state vari-
ables, health parameters, and controlling commands, When a fault occurs, the first step is to detect it as
respectively, y the sensor measurement vector, w and v soon as possible. The approach used for this model-
the process and sensor noise, both assumed to repre- based fault detection is composed of two steps as fol-
sent Gaussian white noise. Their covariance matrices lows:
are: (1) To generate residual signals from the sensor
E[ w (k )] 0 ½ measurements and their Kalman filter estimates.
E[v (k )] 0 ° (2) To compare the residuals with thresholds to
° make fault detection[9].
¾ (2)
E[ w (k  W ) w (k )] QG(kW ) °
T
This article uses a model-based approach with a
E[v (k  W )v T (k )] RG(kW ) °¿ bank of Kalman filters for sensor FDI. The sensor and
actuator faults are “soft fault”. A “soft fault” is defined
as inconsistency between true and measured sensor
3. Estimation of Health Degradation values that are relatively small in magnitude and thus
difficult to detect by a simple range-checking approach,
As shown in Fig.1, as an important part in the whereas a “hard fault”—larger in magnitude and thus
model-based control and diagnostics logic, the on- more readily to detect.
board model and tracking filter use two sets of input Each Kalman filter is designed for a specific sensor
signals, namely sensor measurements and actuator fault. In the case that a fault does occur, all filters ex-
position. The engine’s degradation can be tracked by cept the one using the correct hypothesis will produce
one Kalman filter based on the input signals. After the large estimation errors. By monitoring the residual of
estimation of the Kalman filter, the onboard model can each filter, the specific faults that have occurred can be
be shifted to the vicinity of the degraded engine. detected and isolated[8,10]. Fig.2 shows the structure of
In the Kalman filter problem setup, the engine state sensor FDI using a bank of Kalman filters. The bank
vector is augmented with health parameters as follows of Kalman filters contains five Kalman filters where
· 20 · Xue Wei et al. / Chinese Journal of Aeronautics 22(2009) 18-21 No.1

five is the number of sensors to be monitored. The When the residual is obtained, the weighted sum of
control input and a subset of the sensor measurements squares residuals for each of the Kalman filters are
are fed to each of the five Kalman filters. The sensor calculated by
which is not used by a particular filter is the one being Wi Vi (ei )T (6 i ) 1 ei (7)
mentioned by that filter for fault detection. For in-
stance, Filter i uses the sensor subset that excludes where 6 i diag[V i ]2 .The vector Vi is the noise stan-
Sensor i. Hence each Kalman filter estimates the aug- dard deviation, and the additional weight Vi the
mented state vector using four sensors. Filter 1 uses all weighting factor.
sensors except Sensor 1, Filter 2 does the same except
Filter 2, and so on. Therefore, Filter 1 is able to esti-
mate the augmented state vector from fault-free sensor 5. Simulation Results
measurements, whereas the estimates of the remaining
filters are distorted by the fault in Sensor 1. The bank of Kalman filters function on the nonlin-
ear dynamical model of an aircraft engine with faults
in sensors. Fig.3 shows the estimation of a degrading
engine. The nonlinear dynamical model generates five
sets of real signals at a given state. The sensor fault
can be added on those signals directly. There are five
sensors that might be at fault: the sensor of XNH,
XNL, P31, P6, and T45. The other engine states detected
by sensors are fuel flow (WFM), afterburner flow
(WFAs), altitude (ALT), Mach number (MA), variable
stator vanes (A1), fan inlet guide vanes (A2), main
Fig.2 Sensor fault detection isolation using bank of Kal-
man filters.
burner exit temperature (T41), fuel/air ratio (WFAR),
compressor surge margin (SMC), fan surge margin
For each filter, the residual vector is: (SMF), net thrust (FN), and specific fuel consumption
ei yei  yi (6) (sfc).

Fig.3 Simulation of sensor FDI using bank of Kalman filters.

One Kalman filter, a trend tracking filter, is used to but also by engine degradation. Consequently, the es-
estimate the health degradation of the real engine. As timation of the degrading engine is critical to the FDI
shown in Fig.4, the values of W1-W5 will turn zero system.
given absence of degradation and fault. However, if As shown in Fig.6, after the Kalman filter accurately
the HPT efficiency decreases by 2% and the onboard estimated the degradation of HPT, the onboard model
model fails to shift to the vicinity of the degraded en- shifts to the vicinity of the degraded engine, and the
gine, the in-flight diagnostic systems may lose their in-flight diagnostic system comes into effect. Over the
effectiveness, as shown in Fig.5, making the values of engine’s lifetime, the engine health condition is esti-
the W1-W5 grow rapidly and exceed the threshold, mated ceaselessly by the Kalman filter and on the basis
which results in a false alarm. This is because shifts in of the estimates, the onboard model is updated. When
measured engine outputs are induced not only by faults the fault is added at 10 steps and stop at 200 steps in
No.1 Xue Wei et al. / Chinese Journal of Aeronautics 22(2009) 18-21 · 21 ·

the LPT inlet temperature measurement sensor and at 6. Conclusions

the same time HPT efficiency decreases by 2%, the
in-flight diagnostic system will detect and isolate the This article investigates the aircraft engine sensor
fault.
fault diagnostics based on the estimation of engine’s
As shown in Fig.7, W1-W4 grow rapidly but W5 re-
health degradation by a tracking filter. With this
mains nearly unchanged, which indicates that there
arrangement, the on-line fault detection algorithm is
being a fault in T45 sensor.
able to maintain its diagnostic effectiveness over the
lifetime of the aircraft engine.
The proposed approach has been simulated using
a nonlinear engine model. The results show that this
approach is capable of retaining on-line fault detec-
tion ability in the process of an aircraft engine’s
health degradation.

References
Fig.4 Values of W1-W5 without fault and degradation.
[1] Lakshminarasimha A N, Boyce M P, Meher-Homji C B.
Modeling and analysis of gas turbine performance de-
terioration. Journal of Engineering for Gas Turbines
and Power 1994; 116(1): 46-52.
[2] Diakunchak I S. Performance detrioration in industrial
gas turbines. Journal of Engineering for Gas Turbines
and Power 1992; 114(4): 161-168.
[3] Tabakoff W, Lakshminarasimha A N, Pasin M. Simula-
tion of compressor performance deterioration due to
erosion. Journal of Turbomachinery 1990; 112(1): 78-
83.
[4] Batcho P F, Moller J C, Padova C, et al. Interpretation
of gas turbine response due to dust ingestion. Journal
of Engineering for Gas Turbines and Power 1987;
Fig.5 Values of W1-W5 with HPT efficiency decreased by 109(7): 344-352.
2% but without fault. [5] Merrill W C, Delaat J C, Bruton W M. Advanced de-
tection, isolation, and accommodation of sensor fail-
ures—real-time evaluation. Journal of Guidance, Con-
trol, and Dynamics 1988; 11(6): 517-526.
[6] Kobayashi T, Simon D L. Evaluation of an enhanced
bank of Kalman filters for in-flight aircraft engine
sensor fault diagnostics. NASA/TM-2004-213203,
2004.
[7] Larson E C, Parker B E, Jr, Clark B R. Model-based
sensor and actuator fault detection and isolation. Pro-
ceedings of American Control Conference. 2002; 5:
4215-4219.
Fig.6 Estimation of Kalman filter with HPT efficiency
[8] Hajiyev C, Caliskan F. Fault diagnosis and reconfigu-
decreased by 2%.
ration in flight control systems. Boston, MA: Kluwer
Academic Publishers, 2003.
[9] Fan S Q, Xu Y H. Aerospace propulsion system control.
Beijing: Aviation Industry Press, 2001. [in Chinese]
[10] Xue W, Guo Y Q. Troubleshooting soft failure of air-
craft engine sensor with sequential probability ratio.
Journal of Aerospace Power 2007; 22(11): 1925-1929.
[in Chinese]
[11] Richardson J H, Sallee G P, Smakula F K. Causes of
high pressure compressor detrioration in service.
AIAA-79-1234, 1979.
[12] Luppold R H, Roman J R, Gallops G W, et al. Esti-
mating in-flight performance variations using Kalman
filter concepts. AIAA-89-2584, 1989.
[13] Lambert H H. A simulation study of turbofan engine
Fig.7 Values of W1-W5 with a fault in T45 sensor and HPT deterioration estimation using Kalman filter techniques.
efficiency decreased by 2%. NASA-TM-104233, 1991.