Professional Documents
Culture Documents
Published by:
http://www.sagepublications.com
On behalf of:
Additional services and information for Proceedings of the Institution of Mechanical Engineers, Part A: Journal of Power and Energy
can be found at:
Subscriptions: http://pia.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://pia.sagepub.com/content/217/2/169.refs.html
What is This?
Y G Li
Department of Propulsion, Power and Aerospace Engineering, School of Engineering,
Cran eld University, Bedford MK43 0AL, UK
Abstract: Most gas turbine performance analysis based diagnostic methods use the information from
steady state measurements. Unfortunately, steady state measurement may not be obtained easily in some
situations, and some types of gas turbine fault contribute little to performance deviation at steady state
operating conditions but signi cantly during transient processes. Therefore, gas turbine diagnostics with
transient measurement is superior to that with steady state measurement. In this paper, an accumulated
deviation is de ned for gas turbine performance parameters in order to measure the level of performance
deviation during transient processes. The features of the accumulated deviation are analysed and compared
with traditionally de ned performance deviation at a steady state condition. A non-linear model based
diagnostic method, combined with a genetic algorithm (GA), is developed and applied to a model gas turbine
engine to diagnose engine faults by using the accumulated deviation obtained from transient measurement.
Typical transient measurable parameters of gas turbine engines are used for fault diagnostics, and a typical
slam acceleration process from idle to maximum power is chosen in the analysis. The developed diagnostic
approach is applied to the model engine implanted with three typical single-component faults and is shown
to be very successful.
Keywords: gas turbine, engine, diagnostics, transient, performance, fault detection, genetic algorithm
ducts. The effect of heat transfer between air (or gas) and
engine component is also included in the simulation. The
engine performance model behaves highly non-linearly and
can be used to simulate the engine behaviour in a variety of
operating conditions. The original code was used to predict
the transient processes of Rolls-Royce Spey and Tay
engines, and satisfactory results were obtained [22]. Details
of the transient prediction methods have been described by
Maccallum and Qi [20] and Pilidis [23].
Engine models can be tuned to simulate real engine
performance accurately. Owing to the fact that current
research is to develop a diagnostic approach with transient
measurement, it is assumed that the engine model is accurate Fig. 3 Variation in HP shaft bearing loss: —, CH P B ˆ 1.01;
enough to simulate real engine behaviour during transient £, CH P B ˆ 1.02; ‚ , CH P B ˆ 1.03; E , CH P B ˆ 1.04
processes. Tuning techniques are beyond the scope of this
paper and therefore not discussed. HP shaft extra power consumption relative to the HP turbine
The degradation of compressors and turbines is repre- power at engine maximum power condition is shown in
sented by the drop in isentropic ef ciencies and the change in Fig. 3 for different CHPB varying from 1.01 to 1.04. It can be
ow capacities. Combustor degradation is represented by the
seen that, although the bearing loss does not change too
drop in combustion ef ciency. Shaft bearing failure is
much when the engine accelerates from low to high opera-
simulated and implanted by extracting a certain amount of ting conditions, the percentage of this extra power consump-
power done by increased bearing friction from either the HP tion relative to the HP turbine power is up to 20 times higher
or the low-pressure (LP) shaft. The power is calculated by when the engine is operating at low power rather than at
means of equations (2) and (3). To simplify the analysis, it is high power. Therefore, the impact of bearing failure is larger
assumed that the friction is constant during a transient when the engine is operating at low power rather than at
process for a certain level of bearing failure high power.
WHPB ˆ …CHPB ¡ 1†… fHPB rLP †‰2p…NHP ¡ NLP †Š …2†
WLPB ˆ …CLPB ¡ 1†… fLPB rLP †…2pNLP † …3† 3 PERFORMANCE DEVIATION
When HP shaft bearings fail, the power consumed by
Traditionally, the deviation in gas turbine performance
increased friction is extracted from the HP shaft and applied
parameters is de ned as the shift of the parameters from
to the LP shaft. When LP shaft bearings fail, the power
their nominal values at a steady state condition due to
consumed is only extracted from the LP shaft. Owing to the
performance degradation, and is expressed as
lack of experimental data, it is assumed that the increase in
bearing friction results in an extra power consumption of the z ¡ zn
order of up to 0.4 per cent (corresponding to CHPB ˆ 1:04) Dz ˆ …4†
zn
of HP turbine power when the engine is operating at its
maximum power. The variation in the ratio of the HP shaft
where z is the current value of a performance parameter and
extra power consumption and the HP turbine power during
zn is the nominal value of the parameter. Such a de nition is
the acceleration is shown in Fig. 2, while the variation in the
based on the assumption that the measurements are made at
the same steady state operation point. When an engine is
new, there should be no deviations in any of the measurable
parameters. When the engine is degraded, the deviations in
the measurable parameters are used by a diagnostic system
for fault detection, isolation and quanti cation. Unfortu-
nately, such a deviation concept cannot be easily applied to
transient process diagnostics.
In order to diagnose gas turbine faults by analysing
transient measurements, the analysis of performance devia-
tion for the whole transient process is more useful than at a
certain operation point. Therefore, a new concept of accu-
mulated deviation is introduced to measure the shift of a
parameter trajectory from its nominal position (Fig. 4)
Fig. 2 Variation in the ratio of HP shaft bearing loss and owing to engine component degradation, de ned as the
HP turbine power: —, CH P B ˆ 1.01; £, CH P B ˆ 1.02; area between the two trajectories divided by the area
‚ , C H P B ˆ 1.03; E , CH P B ˆ 1.04 under the nominal trajectory during the transient period
A05302 # IMechE 2003 Proc. Instn Mech. Engrs Vol. 217 Part A: J. Power and Energy
Downloaded from pia.sagepub.com at TUFTS UNIV on September 26, 2014
172 Y G LI
Fig. 10 Shaft speed changes due to HP turbine fault: ———, Fig. 13 Fuel owrate change due to HP shaft bearing failure:
nominal engine; – – –, degraded engine ———, nominal engine; – – –, degraded engine
A05302 # IMechE 2003 Proc. Instn Mech. Engrs Vol. 217 Part A: J. Power and Energy
Downloaded from pia.sagepub.com at TUFTS UNIV on September 26, 2014
174 Y G LI
A05302 # IMechE 2003 Proc. Instn Mech. Engrs Vol. 217 Part A: J. Power and Energy
Downloaded from pia.sagepub.com at TUFTS UNIV on September 26, 2014
176 Y G LI
Table 3 Module setting of DENIAL program the component parameters used to described the faults are
Evolution module listed in Table 2.
Function User function (engine model) Direct search for a gas turbine fault within an 11-
Worst tness 1.E20
Problem type Minimization dimensional space corresponding to 11 potentially degraded
Number of variables 1 or 2 component (independent) parameters with seven measure-
Number of evolutions 300 ments (dependent parameters) is inaccurate because the
Population module number of independent parameters is less than the number
Population size 90
Reproduction model Steady state without duplicates of dependent parameters, and it is very time consuming even
Parent selection method Stochastic tournament if a solution can be obtained. Based on the fact that a single
Replacement method Exponential ranking engine component fault is the most common case in fault
Reproduction module diagnosis of gas turbine engines, the seven fault classes
Number of operators 2
Uniform arithmetical crossover 35.0, 1.0, 0.5 shown in Table 2 with corresponding parameters to describe
Uniform creep 5.0, 0.7, 0.001 the faults are considered, where each single-component fault
is regarded as one fault class. Therefore, seven separate
searches are carried out to isolate a component fault, where
6 FAULT DIAGNOSIS AND ANALYSIS it is assumed that in each search only one class of fault may
occur. By comparing the seven objective functions obtained
In order to diagnose the faults of the model engine by using in the seven searches, the lowest objective function among all
transient measurements during the acceleration process, the fault classes indicates the most likely fault. This approach
typical engine measurable parameters are chosen (Table 1). has the advantages that the searched result for any of the fault
In transient measurement, the time constants of sensors, classes in one- or two-dimensional space (corresponding to
especially thermocouples, will cause delay in the measure- one or two component parameters describing one fault class)
ment and therefore have an in uence on the measurement is unique owing to highly redundant (seven) measurements,
accuracy, which must be taken into account in real applica- and the computation time is much shorter than direct search
tions with proper calibration and correction. In this study, in 11-dimensional space. This method was introduced by
the effect of time constants is ignored for simplicity. Zedda and Singh [14, 15].
Sensor noise is found to be insensible to the accumulated In the present paper, three engine fault cases are studied,
deviations because the in uence of measurement noise and the search results are shown in Table 4. It is obvious that,
becomes very small when the accumulated deviation is calcu- for each of the faults, the diagnostic system searches all
lated with integration [equation (5) or (6)]. Therefore, sensor possible fault classes and only obtains the minimum objective
noise is not considered in this study. In addition, all the sensors function when the correct fault class is searched. Furthermore,
are assumed to be fault free for simplicity, although the the faulty component is easy to pick up correctly because the
diagnosticapproach has the capabilityof tackling sensor faults. objective function corresponding to the faulty component is
To simplify the analysis, it is assumed that only single much smaller than the others. The approach proves to be able
engine components may be faulty and seven major engine to identify any of these component faults effectively. This is
single-component faults are taken into account. Simulated due to the fact that the fault signatures expressed by accumu-
transient measurements of the engine implanted with three lated deviations are magni ed during the acceleration
engine component faults are used to test the effectiveness of compared with the fault signatures expressed by steady state
the diagnostic approach. The three implanted fault cases and deviations. Consequently, the fault signatures of the three
Proc. Instn Mech. Engrs Vol. 217 Part A: J. Power and Energy
Downloaded from pia.sagepub.com at TUFTS UNIV on September 26, 2014
A05302 # IMechE 2003
A GAS TURBINE DIAGNOSTIC APPROACH WITH TRANSIENT MEASUREMENTS 177
engine faults are signi cant enough for the diagnostic 7 Luppold, R. H., Roman, J. R., Gallops, G. W. and Kerr, L. J.
approach to identify and quantify the faults accurately. Estimating in- ight engine performance variations using
Search for some fault classes (FCs), such as FC1 [inter- Kalman lter concepts. AIAA-89-2584, 1989.
mediate pressure (IP) compressor fault], FC2 (HP compres- 8 Kerr, L. J., Nemec, T. S. and Gallops, G. W. Real-time
estimation of gas turbine engine damage using a control-based
sor fault), FC4 (HP turbine fault) and FC5 (LP turbine
Kalman lter algorithm. J. Engng for Gas Turbines and Power,
fault), is within two-dimensional space, and search for other
April 1992, 114, 187–195.
fault classes, such as FC3 (combustor fault), FC6 (HP shaft 9 Bird, J. W. and Schwartz, H. M. Diagnosis of turbine engine
bearing fault) and FC7 (LP shaft bearing fault), is within transient performance with model-based parameter estimation
one-dimensional space. The computation for each search techniques. ASME 94-GT-317, 1994.
takes about 1 h on a personal computer (Pentium III). 10 Lunderstaedt, R. A. and Junk, R. H. Application of the gas-
path analysis (GPA) for the non-stationary operation of a jet
7 CONCLUSIONS engine. 13th ISABE, Chattanooga, Tennessee, September 1997,
ISABE-97-7062.
11 Urban, L. A. Gas Turbine Engine Parameter Interrelationship,
The introduction of transient accumulated deviation has
Ct. 1st edition, 1967, 2nd edition, 1969 (HSD UTC, Windsor
proved to be a useful new concept for measuring performance Locks, Connecticut).
deviation of gas turbine engines during transient processes. 12 Li, G. Y. Performance analysis based gas turbine fault diag-
The comparison of fault signatures expressed by the accumu- nostics: A review. Proc. Instn Mech. Engrs, Part A: J. Power and
lated deviation and the steady state deviation shows that gas Energy, 2002, 216(3).
turbine component fault diagnosis using transient data can be 13 Saravanamuttoo, H. I. H. and Maclsaac, B. D. Thermo-
more effective than using steady state data owing to magni ed dynamic models for pipeline gas turbine diagnostics. J. Engng
fault signatures. Diagnostics with transient measurements for Power, October 1983, 105, 875–884.
becomes more useful than with steady state measurements 14 Zedda, M. and Singh, R. Gas turbine engine and sensor
when steady state measurements are dif cult to obtain and fault diagnosis using optimisation techniques. AIAA-99-2530,
when engine faults contribute little to performance deviation 1999.
15 Zedda, M. and Singh, R. Gas turbine engine and sensor
at steady state conditions but signi cantly at transient
diagnostics. 13th ISABE, 1999, ISABE 99-7238.
processes. A non-linear model based diagnostic approach 16 Gulati, A., Taylor, D. and Singh, R. Multiple operating point
using typical gas turbine transient measurements, combined analysis using genetic algorithm optimisation for gas turbine
with a genetic algorithm, has been shown to be very successful diagnostics. 15th ISABE, Bangalore, India, September 2001,
by applying the approach to a model gas turbine turbofan ISABE-2001-1139.
engine implanted with some typical single-component faults. 17 Gulati, A., Zedda, M. and Singh, R. Gas turbine engine and
sensor multiple operating point analysis using optimization
ACKNOWLEDGEMENT techniques. AIAA-2000-3716, 2000.
18 Grönstedt, T. U. J. A multi-point gas path analysis tool for gas
turbine engines with a moderate level of instrumentation.
The present research is supported by a Platform Grant from
15th ISABE, Bangalore, India, September 2001, ISABE-
the Engineering and Physical Sciences Research Council
2001-1051.
(EPSRC), United Kingdom. 19 Sampath, S., Gulati, A. and Singh, R. Fault diagnostics using
genetic algorithm for advanced cycle gas turbine. ASME GT-
REFERENCES 2002-30021, June 2002.
20 Maccallum, N. R. L. and Qi, O. F. The transient behavior of
1 Merrington, G. L. Fault diagnosis of gas turbine engines from aircraft gas turbines. IMechE Seminar (Seminar S777) on Gas
transient data. J. Engng for Gas Turbines and Power, April Turbines: Technology and Developments, Institution of
1989, 111, 237–243. Mechanical Engineering, London, November 1989.
2 Meher-Homji, C. B. and Bhargava, R. Condition monitoring 21 Fawke, A. J. and Saravanamuttoo, H. I. H. Digital computer
and diagnostic aspects of gas turbine transient response. ASME methods for prediction of gas turbine dynamic response. SAE
paper 92-GT-100, 1992. paper 710550, 1971.
3 Merrington, G. L. Identi cation of dynamic characteristics 22 Maccallum, N. R. L. Computational models for the transient
for fault isolation purpose in a gas turbine using closed- performance of RB183-02 (Spey) and RB183-03 (Tay) engines.
loop measurements. AGARD-CP-448, Engine Condition TR RR/1, Department of Mechanical Engineering, University
Monitoring—Technology and Experience, October 1988. of Glasgow, Glasgow, Scotland, August 1984.
4 Merrington, G., Kwon, O.-K., Goodwin, G. and Carlsson, B. 23 Pilidis, P. Digital simulation of gas turbine performance. PhD
Fault detection and diagnosis in gas turbines. J. Engng for gas dissertation,Department of MechanicalEngineering, University
turbines and power, April 1991, 113, 276–282. of Glasgow, Scotland, November 1983.
5 Merrington, G. L. Fault diagnosis in gas turbines using a 24 Dyson, R. J. E. and Doel, D. L. CF6-80 condition
model-based technique. J. Engng for Gas Turbines and Power, monitoring—the engine manufacturer’s involvement in data
April 1994, 116, 374–380. acquisition and analysis. AIAA-84-1412, 1984.
6 Henry, J. R. CF-18/F404 transient performance trending. 25 Widell, H. GENIAL 1.1: a friendly function optimizer based on
AGARD-CP-448, Engine condition monitoring—Technology evolutional algorithms—user manual. http://hjem.get2net.dk/
and Experience, October 1988. widell/genial.htm, July 1998.
A05302 # IMechE 2003 Proc. Instn Mech. Engrs Vol. 217 Part A: J. Power and Energy
Downloaded from pia.sagepub.com at TUFTS UNIV on September 26, 2014