Professional Documents
Culture Documents
A Survey of Aircraft Engine Health Monitoring Systems: February 2001
A Survey of Aircraft Engine Health Monitoring Systems: February 2001
net/publication/2325569
CITATIONS READS
72 1,536
2 authors:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Anupa Bajwa on 04 February 2019.
j$gg-31288
AIAA-99-2528
A Survey of Aircraft Engine Health
Monitoring Systems
Downloaded by Anupa Bajwa on February 4, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.1999-2528
AIAA-99-2528
kern Y. Turner *
NASA Ames Research Center
Downloaded by Anupa Bajwa on February 4, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.1999-2528
Anupa Bajwa
NASA Ames Research Center
Recom Technologies, MS 269-3
Moffett Field, California, 94035
(650) 604 1851
abajwa Bmail.arc.nasa.gov
danceand ambiguity of the data to interpret, and due to manceparameters,and transmissionto the ground only
the high number of false alarms that cause the users’ when an exceedanceis observed.Even though the data
reluctance to rely on the results. To overcome these are collected at a sufficiently high sampling rate during
practical problems, researchershave been developing flight, this data is not stored for further analysis. In-
many other advancedtechniques. This paper presents stead, a single value (e.g., rms value) is transmitted to
a survey of the current state of practice and researchin the ground personnelfor maintenancepurposes,in or-
EHM. The purposeof this survey is to identify the cru- der to detect generaltrends over a long period of time.
cial needsin the area of engine health monitoring, and Cockpit instrumentreadingsare taken once a day, or on
the promising areasof researchto resolve the perceived every flight during cruise conditions. Recorded data is
problems in current practice. processedand compared to “normal” data established
by the manufacturer or operator. One problem with
Downloaded by Anupa Bajwa on February 4, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.1999-2528
2
(c)1999 American Institute of Aeronautics & Astronautics
time [2, IO]. Automatic troubleshootingproceduresor with developingfaults. The aim of theseefforts is to re-
expert system diagnostics are used when rules can be place the standardthreshold setting and fault detection
defined adequately,to identify the most probablecause process, by enhancingthe feature extraction capabili-
of the exceedanceand estimate the possible damage. ties. Researchin the areaof fault diagnosisaims to find
Commercial software packageshavebeendevelopedin automateddiagnosistools which provide the automatic
conjunction with the engine manufacturersto accom- generationof more meaningful and accurate fault di-
plish the exceedancedetectionand diagnosistask, once agnosticinformation. The efforts tend to move towards
the exceedancedata has been transferredto the ground combining the knowledgefrom standardexpert systems
station [S]. with theoreticalknowledge and test-rig information, to
The automation of the diagnosisstep using the ex- develop more reliable aud thorough fault libraries and
ceedanceand trend data relies on building trend and classification tools. There are other applications, such
Downloaded by Anupa Bajwa on February 4, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.1999-2528
baseline signature databasesthrough engine manufac- as helicoptersand the spaceshuttle engines,where en-
turer’s data and through field experience [5]. Expert gine monitoring systemshave advancedfurther than for
systemsoftwarepackagesattemptto capturethe knowl- commercial aircraft. Experience from these applica-
edge of the experts, and provide possible diagnosesto tions provides valuable insight into the problems and
the operators, in a quick and automatedfashion [8]. potential solutions that work to improve the effective-
Commercially developedsoftware packagesare avail- nessof enginemonitoring systems.
able for use in aircraft monitoring [ 171.
Research in Engine Parameter Monitoring
Perceived Problems in EMS Practice Monitoring systemscollect large amounts of data
The current stateof practice of EMS is flooded with that are usually analyzedoffline. Well-establishedsta-
problems waiting to be resolved. Many of the prob- tistical methodsare still the norm. The implementation
lems result from the strict regulations airlines and en- of thesemethodshas grown in sophisticationand speed
gine manufacturershave to follow, and the difficulty with increasesin computationalpower [29]. For typi-
in justifying the cost of implementing such systemsfor cal engine parameterssuch as temperatures,pressures,
commercial aircraft. Most of the problems concentrate and speeds,exceedancesand trends are monitored us-
on the high number of false alarms. False alarms are ing commercially availablesoftwarepackages[ 19,541.
causedby several factors. The main factors that most If a predefinedlimit is exceeded,a cockpit caution is
researchersfocus on are: (1) unreliable feature extrac- activated, and performance data are recorded for fur-
tion algorithms for detection of the relevant failure in- ther investigation on-ground [19]. Alerts basedon ex-
dicators; (2) insufficient failure knowledge for diagno- ceedancesand trend reporting have provided improved
sis of failures with expert systems. Additional issues diagnostic capabilities in commercial and military air-
contributing to these factors are insufficient sampling craft I:351.For vibrational diagnostics,health indicators
of engine parameters,cost of transmitting the data to are establishedby meansof signal averaging,by gener-
the ground, cost of implementing more elaboratemoni- ating component-specificvibration signatures.The idea
toring systems,and ambiguity causedby the inadequate is to use a variety of indicators from time-domain (syn-
alert-reportingand interactivetroubleshootingmethods. chronously time-averaged)and frequency-domainsig-
nals, specific to engine components,as well as ampli-
tude and phasemodulation signatures[44]. Alerts are
CURRENT STATE OF RESEARCH generatedwhen changesin indicator trends exceedthe
set thresholds.
To addressthe practical problemsfor EMS in com-
mercial aircraft, researchershave been searching for
better feature extraction and fault diagnosis methods, Helicopter Health Monitoring One of the most
with the purposeof providing a reliable meansto moni- active areasof researchin engine monitoring is in the
tor and diagnoseenginefailures. Researchin the area of developmentof Health Usage and Monitoring Systems
monitoring aims to find better anomaly detectionmeth- (HUMS) for helicopters [ 18, 30, 34, 51, 541. HUMS
ods that discriminate betweendata characteristicsfrom have beenimplementedon a number of helicoptersfor
an acceptablecondition and trendswhich are associated severalyears,providing valuablein-flight experienceto
3
(c)1999 American Institute of Aeronautics & Astronautics
determinethe successand benefits of implementing an exampleis modeling and monitoring arcjet thrustersus-
engine monitoring system on aircraft. While the cost ing geometricalparameters,inlet conditions, electrical
of implementing such systemsis still high, the rate of parametersand performance parameters [ 111. Other
failure to detect vital faults using HUMS is decreas- examples,including rocket engine failure detection by
ing steadily. Benefits due to HUMS include improved means of system identification, and an implementation
safety and reducedmaintenancecosts [30,34]. of a diagnostic system using feature extraction algo-
The HUMS experienceprovides a valuabletestbed rithms, can be found in [ 13, 37,4 1,421.
for collecting fault databasesand testing the effective-
ness of monitoring systems on helicopters. For ex- Research in Automated Fault Diagnosis
ample, researchshows that traditional monitoring sys- The traditional means of achieving automateddi-
tems (e.g., pressure, temperature, torque) can be en-
Downloaded by Anupa Bajwa on February 4, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.1999-2528
Rocket Engine Monitoring Another very active Model-based diagnosis Model-based diagnosis
areais health monitoring of the SpaceShuttle Main En- presentsa powerful complement to expert systems by
gine (SSME) or rocket engines in general. An exam- adding to the knowledge databaseobtained from field
ple is using a model-basedsystem for the SSME [27]. experienceand experts. Model-baseddiagnosis mainly
Using information from a thermodynamicmodel of the concentrateson combining theoretical knowledge with
engine and using sensor measurements,predictions of test rig information [47, 491. Examples of models
sensoroutputs are made and comparedwith actual out- are propulsion system modeling [26], finite element
put to detect failures. A thorough model for propul- modeling [23], and autoregressivemodeling [24]. In
sion systemsis developedin [32] and is applied to the model-baseddiagnosis, an estimated system model is
SSME [33]. In the unmannedspacecraftrealm, moni- compared to a nominal system model. The residual
toring and control determinethe successof the mission. between the two models provides a measure of the
The critical issue is the type and number of parameters deviation between the estimated and nominal models,
to be measuredto characterizethe system’s health. An and is used to make a decision as to whether a failure
4
(c)1999 American Institute of Aeronautics & Astronautics
has occurred [40, 47, 493. An essentialrequirement a decision tree describing the data. The central choice
for model-based diagnosis is the development of an in the algorithm is selecting which attribute to test at
accurate system model. An example is a spectral each node. One popular algorithm is ID3 [43], which
model of the plume of the SSME that is used to grows the tree top-down, at each node selecting the
monitor the engine by extracting chemical data from attribute that best classifiesthe local training example.
the runtime electromagneticspectrum and comparing Some other potential knowledge discovery techniques
it with known signatures[ 131. Other examplesinclude propose the use of fuzzy cognitive maps [31], fuzzy
a model-based vehicle health monitoring system for belief nets [25], and other soft computing techniques
the SSMH, an example of military usage monitoring for diagnostics and prognostics [14]. The methods
of fracture-critical parts using modeling, and the make use of whatever data and knowledge is available
use of model-basedreasoning for gas turbine engine to achieve reliable diagnosis in cases where failure
Downloaded by Anupa Bajwa on February 4, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.1999-2528
5
(c)1999 American Institute of Aeronautics & Astronautics
monitoring tools that better discriminate between cor- engine health monitoring. In 30th
rect and false alarms, and better diagnosethe origin of AIAALASMWSAE/ASEE Joint Propulsion Confer-
faults in engines,in an automatedfashion. A more spe- ence, volume AIAA-94-3007, Indianapolis, IN,
cific goal is to develop a means to use engine moni- 1994.
toring systems to help with engine maintenancedeci- [12] M.J. Barwell. Compass-ground based engine
sions, such as schedulingoverhaul times and predicting monitoring program for generalapplication. SAE:
the remaining life of engine components,by monitor- Aircraft Gas Turbine Engine Monitoring Systems,
ing engine parameterssuch as temperatures,pressures, An Update, 1988.
speeds,and vibration. The authorsbelieve that the thor- [ 131 D. Benzing, R. Hopkins, and K. Whitaker. Opad:
ough survey of practical and researchinformation pre- An innovative rocket engine health monitoring
sented in this paper will help other researchersin the system. In 16th AIAALIEEE Digital Avionics Sys-
Downloaded by Anupa Bajwa on February 4, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.1999-2528
field of condition monitoring of aircraft engines,prior tems Conference, volume 2, Irvine, CA, October
initiating such a researchprogram. 1997.
1141P. P. Bonissone and K. Goebel. Soft computing
techniquesfor diagnosticsand prognostics. Work-
REFERENCES ing Notes for the 1999AAAI Symposium for Use
[I] SAE ARP 1587. Aircraft gas turbine enginemon- of AI in Equipment Maintenance,March 1999.
itoring system guide. AerospaceRecommended [15] J. Broede and H. Pfoertner. Olmos in gaf
Practice,April 1981. mrca tornado - 10 years of experience with
[2] SAE AIR 1839. A guide to aircraft turbine engine on-board life usage monitoring. In 33rd
vibration monitoring systems. AerospaceInfor- AIAALASME/SAWASEE Joint Propulsion Confer-
mation Report, October 1986. ence,Seattle,WA, July 1997.
[3] SAE AIR 1871B. Lessonslearnedfrom develop- [16] G. V. Brown, C. Lawrence, D. W. Manchala,
mental and operationalturbine engine monitoring Alan B. Palazzolo, Steve Klusman, Albert F.
systems. AerospaceInformation Report, January Kascak, and Gerald T. Montague. In operation
1988. detection and correction of rotor imbalance in
[4] SAE AIR 1872. Guide to life usagemonitoring jet engines using active vibration control. SAE
and parts managementfor aircraft gas turbine en- Transactions-Journal of Aerospace, 103(1): 130-
gines. Aerospace Information Report, February 137,1994.
1988. [17] K. Collinge and K. Schoff. TEXMAS-an ex-
[5] SAE AIR 1873. Guide to limited engine mon- pert system for gas turbine engine diagnosis and
itoring systems for aircraft gas turbine engines. more. SAE: Aircraft Gas Turbine Engine Moni-
AerospaceInformation Report, May 1988. toring Systems,An Update, 1988.
[6] SAE AIR 1900. Guide to temperaturemonitoring [ 181 J. D. Cronkhite. Practical application of health and
in aircraft gas turbine engines. AerospaceInfor- usagemonitoring (HUMS) to helicopter rotor, en-
mation Report, 1988. gine, and drive systems. In 1993 American He-
[7] SAE AIR 4061. Guidelines for integration of en- licopter Society Annual Forum, volume 2, pages
gine monitoring functions with on-board aircraft 1445-1455, St. Louis, MO, May 1993.
systems. AerospaceInformation Report, 1996. [19] R.W. Cue and D.E. Muir. Engine performance
[8] SAE AIR 4175. A guide to the developmentof monitoring and troubleshootingtechniquesfor the
a ground station for engine condition monitoring. CF-18 aircraft. ASME Journal of Engineering
AerospaceInformation Report, 1994. for Gas Turbines and Power, 113:11-19, January
[9] SAE AIR 4176. Cost vs. benefits of engine mon- 1991.
itoring systems. Aerospace Information Report, [20] Hans R. DePold and E Douglas Gass. The appli-
1994. cation of expert systems and neural networks to
[lo] SAE AS 8054. Airborne engine vibration mon- gas turbine prognostics and diagnostics. ASME
itoring (evm) system guidelines for performance. Journal of Engineering for Gas Turbines and
AerospaceStandard,1996. Power, 116:374-380,April 1994.
[l l] M. Andrenucci. Arcjet modeling and [21] U. Fayyad, Piatetsky-ShapiroG., Smyth P., and
6
(c)1999 American Institute of Aeronautics & Astronautics
Uthurusamy R. Advances in Knowledge Discov- [32] K. Kolcio and A. Helmicki. Propulsion system
ery and Data Mining. AAAI PressI MIT Press, modelling for condition monitoring and control. i
1996. - theoretical foundations (air breathing engines).
[22] L. Flitter, K. McCool, and D. Haas. Rotor system In 30th AIAA/ASMWSAE/ASEE Joint Propulsion
load monitoring using a neural network basedap- Conference, Indianapolis, IN, June 1994.
proach. In AIAA Modeling and Simulation Tech- [33] K. Kolcio and A. Hehnicki. Propulsion sys-
nologies Conference, New Orleans, LA, August tem modelling for condition monitoring and
1997. control. ii - application to ssme. In 30th
[23] R. Ganguli, I. Chopra, and D. J. Haas. Heli- AIAALASMEXSAELASEE Joint Propulsion Confer-
copter rotor system health monitoring using nu- ence,Indianapolis,IN, June 1994.
merical simulations and neural networks. In 1997 [34] J. E. Land. Five years of daily HUMS-what have
Downloaded by Anupa Bajwa on February 4, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.1999-2528
American Helicopter Society Annual Forum, vol- we learned?In 1997 American Helicopter Society
ume 2, pages 1285-1296, Virginia Beach, VA, Annual Forum, volume 3, pages293-300, Virginia
April 1997. Beach,VA, April 1997.
[24] A. Garga, B. T. Elverson, and D. C. Lang. Fault [35] H. Lucas and J.E. Paas. Introduction and appli-
classification in helicopter vibration signals. In cation of the generalelectric turbine engine mon-
1997 American Helicopter Society Annual Forum, itoring software within KLM royal dutch airlines.
volume 2, pages 1316-l 323, Virginia Beach,VA, ASME Journal of Engineering for Gas Turbines
April 1997. and Power, 110:23-27, January 1988.
[25] K. Goebel and A. Agogino. Towards fuzzy be- [36] H. Luo and S. Hanagud. Fuzzy learning rate neu-
lief nets. Working Notes for the 1999AAAI Sym- ral network training and structural health monitor-
posium for Use of AI in Equipment Maintenance, ing. In 38th AIAA/ASMWASCE/AHS/ASC Struc-
March 1999. tures, Structural Dynamics and Materials Confer-
[26] A.J. Hehnicki, S. Jaweed,and K. Kolcio. Propul- ence,Kissimmee,FL, April 1997.
sion system modelling for condition monitoring [37] D. J. Malloy, C. Biegl, and J. F. Zakrajsek. De-
and control: A status report. SAE Transactions- velopment of a near real-time turbine engine test-
Journal of Aerospace, 103(1):390-399,1994. ing diagnostic system using feature extraction al-
[27] N. Ho, P.Lozano, R. Mangoubi, and M. Martinez- gorithms. In Proceedings of the 13th International
Sanchez. A model-basedvehicle health moni- Symposium of Air Breathing Engines, volume IS-
toring system for the space shuttle main engine. ABE 97-7144,Chattonooga,Tennesse,September
In 34th AW-VASMWSAEZASEE Joint Propulsion 1997.
Conference, Cleveland,OH, July 1998. [38] Z. Mazur and J. Kubiak. Gas turbine failure eval-
[28] K. Huenecke.Jet Engines: Fundamentals of The- uation and repair. In Proceedings of the Amer-
ory, Design, and Operation. Airlife Publishing ican Power Conference, volume 60-B Chicago,
Ltd, England, 1997. IL, 1998.
[29] K. Kato, A. Kanmuri, and K. Kisara. Data anal- [39] C. Meher-Homji. Detect, troubleshootgas-turbine
ysis for rocket engine health monitoring system. blade failures. Power, pages 35-38, December
In 31st AIAA/ASME/SAWASEE Joint Propulsion 1995.
Conference, San Diego, CA, July 1995. [40] G.L. Menington. Fault diagnosis in gas turbines
[30] S. D. Kershner,J. B. Johnson,and M. D. Gamauf. using a model-based technique. ASME Jour-
Sikorsky support to commercial health and usage nal of Engineering for Gas Turbines and Power,
monitoring systems: A summary of forty months 116:374-380,April 1994.
of support. In 1997 American Helicopter Society [41] C. M. Meyer, W. A. Maul, and A. P. Dhawan.
Annual Forum, volume 2, pages 1233-1241,Vir- Ssmeparameterestimation and model validity us-
ginia Beach,VA, April 1997. ing radial basis function neural networks. In Pro-
[31] 0. Kipersztok. Fault propagationusing fuzzy cog- ceedings of the 1994 Conference on Advanced
nitive maps. Working Notes for the 1999 AAAI Earth to Orbit Propulsion Technology, pages
Symposium for Use of AI in Equipment Mainte- 1019-1021,1983.
nance,March 1999. [42] C. M. Meyer and J. F. Zakrajsek. Rocket engine
7
(c)1999 American Institute of Aeronautics & Astronautics
failure detection using system identification tech- [54] C. Trammel, G. Vossler, and M. Feldmann. UK
niques. NASA Contractor Report 185259.AIAA- Ministry of Defence generic health and usage
90-1993,1993. monitoring system (GenHUMS). In 1997 Amer-
[43] T. M. Mitchell. Machine Learning. McGraw-Hill, ican Helicopter Society Annual Forum, volume 3,
1997. pages282-292, Virginia Beach,VA, April 1997.
[44] J.E. Nicks and G. Krishnappa. Some investiga- [55] I.E. Treager. Aircraft Gas Turbine Engine Tech-
tions into the condition monitoring of mechanical nology. Glencoe Series, McGraw-Hill, Ohio,
componentsin gas turbine engines. In Tenth Zn- ,1996.
ternational Symposium on Air Breathing Engines, [56] H. Winston, R. Clark, and Buchina G. Agets mbr:
pages 591-598, Nottinghmam, England, Septem- An application of model-basedreasoning to gas
ber 1991. turbine diagnostics.AI Magazine, 1995.
Downloaded by Anupa Bajwa on February 4, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.1999-2528
[45] K. Park and G. Reich. Localized systemidentifica- [57] D. Zimmerman,T. Simmermacher,and M. Kaouk.
tion and structural monitoring from vibration test Model correlation and system health monitoring
data. In 38th AUA/ASME/ASCE/AHS/ASC Struc- using frequency domain measurements. In 36th
tures, Structural Dynamics and Materials Confer- AIAA/ASMWASCE/AHS/ASC Structures, Struc-
ence, Kissimmee, FL, April 1997. tural Dynamics and Materials Conference, New
[46] V.C. Patel, V. Kadirkamanathan, and H.A. Orleans,LA, April 1995.
Thompson. A novel self-learning fault detec-
tion system for gas turbine. In UKACC Intema-
tional Conference on CONTROL’96, pages 867-
872, September1996.
[47] M. M. Polycarpou. An on-line approxima-
tion approachto fault monitoring, diagnosis, and
accommodation. SAE Transactions-Journal of
Aerospace, 103(1):371-380,1994.
[48] C. Pomfret. Engine life measurementand diag-
nostics. Aerospace Engineering, June 1994.
[49] W.T. Powers, A.E. Cooper, T.L. Wallace, W.L.
Buntine, and K. Whitaker. Spaceshuttle main en-
gine plume diagnostics: OPAD approachto vehi-
cle health monitoring. SAE Transactions-Journal
of Aerospace, 102(1):2021-2033,1993.
[50] I. Salam, A. Tauqir, and A. Khan. Causesof fa-
tigue failure in an aero-enginecomponent. Engi-
neering Failure Analysis, 4(1):39-47,1997.
[51] J. Schoess,F. Malver, B. Iyer, and J. Kooyman.
Rotor acoustic monitoring system-a fatigue crack
detection system. In 1997 American Helicopter
Society Annual Forum, volume 3, pages274-28 1,
Virginia Beach, VA, April 1997.
[52] A. Tauqir, I. Salam, F. Hashmi, and A. Khan. An
air crashcasestudy.Engineering Failure Analysis,
3(3):203-210,1996.
[53] R. S. Teal, J. T. Evernham, T. J. Larchuk, and
D. G. Miller. Regime recognition for MH-47E
structural usage monitoring. In 1997 Ameri-
can Helicopter Society Annual Forum, volume 2,
pages 1267-1284, Virginia Beach, VA, April
1997.