Advanced Diagnostics and Prognostics for Engine Health

Monitoring

Ashish Babbar Estefan M. Ortiz

University of Hawaii University of Hawaii
2800 Woodlawn Drive 2800 Woodlawn Drive
Suite 163 Suite 146
Honolulu, HI 96822 Honolulu, HI 96822
808-988-8954 808-988-2909
babbar@hawaii.edu estefan@hawaii.edu

Vassilis L. Syrmos Michael M. Arita

University of Hawaii Aloha Airlines
2500 Campus Road P.O. Box 30028
Honolulu HI 96822 Honolulu, HI 96820
808-956-0813 808-836-4282
syrmos@hawaii.edu marita@alohaairlines.com

Abstract—Modern military and commercial aircraft require approach can integrate the diagnostic and prognostic results
advanced diagnostic and prognostic schemes to determine and provide enhanced visibility for the maintainer into
engine performance in order to reduce operational and current and future system health thus improving reliability.1
2
maintenance costs and improve aircraft down time.
Condition Based Maintenance (CBM) can be performed on
an aircraft when a fault/failure is detected. However, to TABLE OF CONTENTS
streamline maintenance procedures and avoid unnecessary
1. INTRODUCTION ................................................. 1
delays the engineers and maintainers need current and best
2. DATA DESCRIPTION .......................................... 2
estimates of engine performance and health before
3. ENGINE HEALTH DETERIORATION .................. 2
upcoming flights.
4. DIFFERENTIAL ANALYSIS ................................. 4
Using flight condition data and situational parameters, 5. ENGINE HEALTH PROGNOSTICS ...................... 6
diagnostic and prognostic schemes can be developed which 6. CONCLUSIONS ................................................... 8
help in determining engine behavior during a specific flight 7. REFERENCES ..................................................... 8
and predict engine performance by estimating the flight 8. BIOGRAPHY....................................................... 9
parameter conditions of future flights. This approach can
provide several key benefits including: 1. INTRODUCTION
ƒ Reducing unnecessary maintenance actions and
prioritize maintenance procedures based on system
Condition based health management is an important aspect
conditions.
of evaluating engine performance for modern and legacy
ƒ Increasing mission readiness and airworthiness.
aircraft. Engine health performance can be better evaluated
ƒ Improving aircraft safety.
by detecting abnormal behavior in engine related parameters
Various flight parameters such as Exhaust Gas Temperature and using prognostic approaches to predict if such behavior
(EGT), Fuel Flow (FF), Engine Fan Speeds (N1 and N2), will continue for the same aircraft in the upcoming flights.
Oil Pressure (OIP) and Oil Temperature (OIT) can be used The final aspect is to relate the engine abnormality to off-
to estimate the engine health and performance. Diagnostic board maintenance actions that were performed in response
schemes which focus on engine health deterioration are the to the engine behavior [1].
basis of discussion for this paper. Once a fault event is
identified the cause of a fault can be traced by linking the Condition Based Maintenance (CBM) can be divided into
in-flight information with the historical maintenance actions two broad categories – Model Based and Data driven
that were performed for such an event. A prognostic approaches. Model based methods are preferred when an
approach based on double exponential smoothing is also accurate mathematical model for the system, sub-system or
discussed to help identify the trends for flight parameters component is available [2,3]. Data driven approaches are
for upcoming flights. An engine parameter differential preferred when operational aircraft data is available and it
analysis is applied to identify significant deviation in the 1
performance of the engines on a single aircraft. This 1
978-1-4244-2622-5/09/$25.00 ©2009 IEEE.
2
IEEEAC paper #1024, Version 4, Updated December 17, 2008

1
represents the systems being monitored. Such data includes a subset of the recorded parameters. The thresholds are used
both on-board flight operational data and off-board aircraft to identify system anomalies which occur when selected
maintenance data. Various methods have been used for data QAR parameters operate beyond these thresholds. This
driven fault diagnosis and prognosis and they include; information can be used for diagnosis of engine health by
statistical methods, machine learning, competitive learning observing in-flight parameters corresponding to 737
and quality control approaches. A literature review of the engines.
data driven fault diagnosis can be found in [3-12].
The maintenance information for 737s can be divided into
Engine health management has been among the top three kinds of reports: Pilot Reports (PiRep), PreMat
priorities for maintenance departments. Engine health Reports (PreMat) and Delay Cancellation (DC) Reports.
management procedures deal with detection, identification These reports in turn provide details like problem
and isolation of faults, prediction of trends in future flights descriptions and corresponding corrective actions. The
and estimating engine health degradation and overhauls. problem description is a summary of both scheduled and
Improved health management procedures can increase unscheduled maintenance actions. It includes aircraft tail
airworthiness, mission readiness and aircraft safety, while number, report date, Aircraft Transport Association (ATA)
reducing aircraft down times and cutting down unnecessary code which identifies the system or sub-system reported to
maintenance actions. Several approaches have been be malfunctioning. The corrective action is the procedure
researched for engine health management with varying the maintenance department followed in order to correct the
degrees of results. indicated problem. It also reports any flight delays or
cancellations caused in response to problem reported by the
The objective of this paper is to provide a methodology to maintainer.
identify engine health deterioration, provide prognostic
trending on engine related flight parameters and link this
Maintenance and Reliability Data is pulled from the
information to off-board maintenance procedures. This maintenance action reports.
approach can provide an improved insight into engine •Includes both PiRep and PreMat reports.
performance and health. Such an insight helps maintainers Maintenance •Aircraft tail number, ATA codes, report dates
Logs
to make accurate, cost saving maintenance decisions. •PiRep cause and corrective action description.
•PreMat removed part description and sub-systems that were
The remainder of the paper is outlined as followed: Section removed.

2 gives a brief description of the types of data available

In Flight data is recorded by the Aircraft Condition
from 737 aircraft. In Section 3 engine health deterioration is Monitoring System (ACMS).
studied using EGT parameter values for 737 engines. •Includes Smart ACMS Recorder (SAR) and Quick Access
Section 4 discusses a difference analysis based on flight Recorder (QAR) data.

parameters during different flight phases. Engine health •SAR records parameters for 3 flight modes.
Flight Data
prognostics based on trending of flight parameters and •QAR records the complete set of in-flight parameters.
double exponential smoothing is discussed in Section 5. •Engine reports giving information about thresholds set on a
subset of the recorded parameters.
Conclusions and Future work are discussed in Section 6.

Figure 1 – Data sources for 737 engine health management

2. DATA DESCRIPTION
Flight data recorders record various flight parameters that
have certain nominal operation values associated with them.
The 737 data used for this analysis includes both on-board- A subset of the engine related flight parameters have
flight operational data and off-board data consisting of pilot specific thresholds associated with them. Operation beyond
reports, delay cancellation reports, problem descriptions and these thresholds leads to generation of standard aircraft
maintenance actions performed. Flight data is captured by reports. Figure 1 illustrates the data sources available for
on-board Aircraft Condition Monitoring System (ACMS) 737 engine health management analysis discussed in this
data recorders which include: paper.
ƒ Quick Access Recorder (QAR).

ƒ Smart ACMS Recorder (SAR). 3. ENGINE HEALTH DETERIORATION

The SAR data can be divided into three flight modes: Take Engine health can be evaluated using different approaches
off, Cruise and Landing. The QAR data includes various [13-18]. Inspection of engine health over a period of time
flight parameters that are recorded at different sampling which includes several flight cycles gives an insight into
intervals. In addition QAR data includes engine reports that engine health deterioration. Examining a set of engine
provide detailed information about thresholds that are set on parameters such as; Exhaust Gas Temperature (EGT), Fan
Speeds and Fuel flows allows for analysis which can detect

2
gradual engine performance deterioration (long term) and
rapid (short-term) deterioration. Gradual deterioration can
sometimes be attributed to overall engine aging. However,
rapid deterioration or a shift in engine parameter trends
from normal behavior can be due to a single system, sub-
system or component behaving abnormally. If historical
flight data is available for a specific aircraft, trending on
selected flight parameters can be done to observe if there
were signs of gradual or rapid deterioration.

To demonstrate the application of engine health

deterioration analysis, the EGT parameter was analyzed for
two different aircraft over a period of one year.
Approximately 300 flights for each aircraft were used for
this analysis. In the first phase EGT averages were
calculated for each flight. As shown in Figure 2 for Aircraft
‘A’ the average EGT values show a gradual deterioration
over a period of one year. The flights are enumerated in Figure 3 – Deterioration of average EGT values for
chronological order. Quadratic curve fitting was used in this Aircraft ‘B’
analysis to provide trending on the average EGT values for
the available time period. As observed from Figure 2 the
EGT average values during each flight show an upward To isolate this rapid shift in EGT values a gradient based
trend as the number of flights increase. This trend can in approach was used. The objective of the gradient approach
general be attributed to aging of the aircraft and thus is to identify the time window corresponding to the rapid
performance degradation of the engine. shift in a given trend. Gradient approach calculates the
difference between successive samples in a given data set.
A similar analysis was performed on data from Aircraft ‘B’. The difference values are then used to isolate samples
As shown in Figure 3 the gradual trend in the average EGT which show a substantial deviation from the mean values.
values changes and there is a rapid shift to a higher level of For this analysis the mean of the average EGT values for all
EGT values followed by the continuation of the gradual available flights was used as the reference. The gradient
trend. This window of time where this rapid trend shift between the average EGT values for each available flight
occurs can be considered as a possible fault event. was calculated. To make sure that the outlier data points can
be isolated from true trend shifts the gradient for three
successive flights should be greater than two standard
deviations away from the mean for the available flight data.
This way the flights for which there is a single spike in the
average EGT value can be isolated and ignored as an
outlier.

The final part of this analysis is to provide links to the

maintenance information which can assist in diagnosing the
cause of the fault event. In addition the maintenance
information can also help the maintainer to judge if the fault
condition was properly addressed. In this analysis, for the
isolated fault event a time window is identified. The time
window, aircraft number and ATA codes pertaining to
engines are then investigated to pull out maintenance
information. The maintenance data includes the problem
description and the corresponding repair action that was
taken to correct the problem as discussed below:

Figure 2 –Deterioration of average EGT values for ƒ ATA code: 72 – Engine General.
Aircraft ‘A’
ƒ Problem Description: During routine inspection found
a nick on Engine Blade #14.
ƒ Corrective Action: Repaired and blended nick as per
given directives.

3
 4. DIFFERENTIAL ANALYSIS

Evaluation of the engine health can also be accomplished by

examining the difference between symmetric engine
parameters for various phases of flight. Identification of
discrepancies between the performances of the two engines
on 737 aircraft provides a means to identify the health of
each engine. This analysis is used to determine those
instances at which the difference between symmetric flight
parameters (EGT, Fuel Flow, etc…) is substantial. Once
identified, those instances are linked to maintenance
information in order to understand the possible cause of
such a disagreement in symmetric engine parameters.

This analysis is based on comparing equivalent or

symmetric flight parameters for each engine on the aircraft.
Flight parameters such as Exhaust Gas Temperature (EGT),
Fan Speeds, Fuel Flows, Engine Vibration, Oil pressure can
be used in this analysis to check for discrepancies in
operating conditions of the two engines. Exhaust Gas Figure 4 – Engine health deterioration process overview.
Temperature will be used as the monitored parameter for the
initial focus of determining engine degradation. This
1 n
analysis is based on dividing operational flight data into the
three major flight modes:
RMS _ Flightmode ( x) = ∑EGT _ diff ( i )
n i=1

1. Take off Mode Where i=1,2,…,n are the number of data samples for each
2. Cruise Mode flight phase and x is the flight date. The R.M.S values are
then observed as a function of the flight dates. Detection of
3. Landing Mode. engine degradation and engine faults are identified as those
RMS values which show a large difference between the
For each identified flight mode of an observed flight the symmetric parameters for a single flight.
difference (or a similarity measure) between the
corresponding flight parameters is calculated. In this When a large discrepancy is observed between the EGT
analysis EGT.1 (engine 1) and EGT.2 (engine 2) are used as parameters for the two engines a possible link between
the tracked parameters. The similarity measure used in flight data and maintenance data is established and used to
determining the difference between the tracked parameter investigate maintenance actions made in response to this
for each engine is given as: discrepancy. However, if no such link can be established
due to a lack of maintenance or a limited time window used
to establish the link the RMS value provides evidence of
EGT _diff ( i) =( EGT.1( i) −EGT.2( i) )
2
engine wear. Large disagreements in engine parameters
provides leverage to the maintenance departments in which
Using the EGT difference values at each sampling interval to define improved maintenance actions based on
for a given flight phase the root mean square value for the operational conditions of the engine. This will enhance the
EGT difference is calculated for each flight using: current state of aircraft maintenance procedures by basing
maintenance actions on a measured value of engine health
rather than following Fault Isolation Manuals (FIM) which
are based on set time durations not specific to the observed
engine.

The differential analysis results based on EGT values from

two distinct 737 aircraft are shown here. The analyzed flight
data is divided into the three different flight modes using
pre-defined aircraft triggers for each flight mode. The
results reported are based on the cruise mode of flight only.
Figure 5 shows the difference analysis for Aircraft ‘A’ for
approximately 450 flights of which the average flight time
is 5 hours. The sample mean and sample standard deviations

4
for the RMS values for the available flights are also plotted
for reference.

Figure 6 – Difference Analysis results for Aircraft ‘B’

To investigate further into the differences between

Figure 5 - Difference analysis results for Aircraft 'A'
symmetric engine parameter values the observed engine
parameter values at cruise are modeled as a Gaussian
Results from analysis seen in Figure 5 show that the distribution whose density function is given as,
concentration of the RMS values are found near the sample
mean of all the RMS values. However, there are a few data ⎧⎪ 1 ⎛ x − μ ⎞ 2 ⎫⎪
1
points where the RMS values have a substantial deviation f ( x) = exp ⎨− ⎜ ⎟ ⎬
from the sample mean. Simple application of this procedure σ 2π ⎪⎩ 2 ⎝ σ ⎠ ⎭⎪
finds that there are large enough differences between the
engine’s EGT values to demonstrate an uneven strain on
The density’s model parameters μ and σ , mean and
one engine during cruise flight mode. Considering the flight
mode (cruise) such a large difference in EGT values standard deviation respectively, are found using maximum
provides the maintainer and engineer a diagnostic tool in likelihood estimators with a 95% confidence interval.
which can be used to identify early engine problems before
they become full engine failure. For consistency, the flights used to model the respective
engine parameter distributions of Aircraft B are the same
Figure 6 reports the difference analysis for Aircraft ‘B’. In flights used in the differential analysis. Modeling Aircraft
this graph the RMS values have two very different B’s engines EGT values as a Gaussian distribution it is
distributions. The early flights show the RMS values to be found that Engine 2 is operating at a much higher
concentrated near a sample mean of about 10 Degrees C. temperature than Engine 1 while at cruise. Figure 7 displays
Then around flight 108 there is a significant shift in the estimated probability densities of the EGT values for
observed RMS values. The elevated difference between the each of the engines. It is apparent that Engine 2 appears to
EGT values for the two engines lasts for about 200 flights. be operating at a different performance level when
Such a divergence between EGT values of the engines on compared to Engine 1.
the same aircraft provides evidence that one of the aircraft’s
engine is operating at a degraded state or the load put on
each of the engines is drastically uneven. In either case the
difference analysis shows significant performance
degradation in one of the engines.

5
 The next step in analysis is to link flight information to the
maintenance actions and check for any evidence describing
the asymmetry in the EGT values. In addition, the
maintenance data will help to determine if any maintenance
actions were taken to address the possible engine
degradation.

As shown in Figure 5 there are instances where the RMS

values for EGT differences are substantial. Evaluating the
maintenance data for these instances the following problem
descriptions and corrective actions were observed:
ƒ ATA code: 77 – Engine Indicating.
ƒ Problem Description: During Cruise mode EGT for
engine #1 was observed to be 50 Degrees hotter than
engine #2.
ƒ Corrective Action: Inspected and cleaned the plugs
per fault isolation manual. Ground run was found OK.

Figure 7 - EGT Estimated Probability Density for Aircraft For Aircraft ‘B’ in Figure 6 there is no substantial evidence
‘B’ from maintenance data that the rapid trend shift was
addressed. However, it provided a starting point to examine
other parameters which demonstrate similar differences that
Modeling the oil pressure for Aircraft B, similar differences can assist the engineers in narrowing down the cause of the
are found. Figure 8 shows these differences for the oil problem. This analysis can in these scenarios help the
pressure parameter. maintainers and engineers identify trend shifts and make
sure they are properly addressed.

5. ENGINE HEALTH PROGNOSTICS

In addition to the diagnostic abilities provided for condition

based maintenance, it is important to develop prognostic
methods to improve the maintainers and engineers
knowledge of the aircraft’s system behavior of upcoming
flights. Prognostic methods provide indicators of engine
health and performance for future engine usage. The
prognostic methods examined are data driven and thus will
be based on current and past engine health as seen from the
flight operational data and maintenance information. A
combination of several engine related parameters can help
define the engine health at any given time. If the
maintainers are provided with information which details the
behavior of these parameters for future flights it can help
Figure 8 - Oil Pressure Estimated Probability Density for streamline the maintenance actions. The maintainers can
Aircraft ‘B’ also include engine specific maintenance actions in case
they observe that the predicted parameter values are
It should be noted that each of the engines EGT values are approaching their defined threshold values.
operating below the manufactures predefined EGT
threshold. However, if hard thresholds are the only In this analysis the results reported are based on using the
diagnostic tool, then asymmetric behavior in EGT values EGT values for a given flight which uses operational flight
will never be detected. Because of this the engines can data from each of the engines. This analysis can easily be
possibly degrade at different rates causing an uneven load extended to other flight parameters such as fuel flow, fan
distribution during flight. This analysis provides a means to speeds, oil temperature, pressure and other related engine
identify early warning signs of engine degradation by parameters. The flight data is divided into three flight
examining asymmetric behavior in parallel engine modes as discussed earlier. In the next step a flight mode is
parameters. selected and the average EGT values for each engine are

6
calculated. If the maintainer or engineer knows the duration θU = y + zα / 2 (σ / n )
of the upcoming flight, this analysis uses double
exponential smoothing to predict the average EGT value for θ L = y − zα / 2 (σ / n )
the selected flight mode of the next flight. In addition, this
analysis can provide a confidence interval around the y = predicted average EGT value.
predicted EGT value. This confidence interval provides the
upper and lower bounds around the predicted average EGT (1 − α ) = 0.90 is a confidence coefficient.
value to give the maintainer an estimate if any necessary θ L & θU are upper and lower confidence limits.
maintenance checks should be done before the flight is
cleared to operate. σ is the standard deviation where n are the number of
samples.
Double Exponential Smoothing
Figure 9, 10 and 11 illustrate the Average EGT values in
Double Exponential Smoothing is a popular scheme to take off, cruise and landing modes for Aircraft ‘A’. These
produce a smoothed time series data. In single moving are used as reference for predicting the average EGT values
averages the past observations are weighted equally, for upcoming flights in the respective flight modes. The
Exponential Smoothing assigns exponentially decreasing EGT values shown here are for engine #1, this analysis can
weights as the observation get older. In other words, recent be repeated for engine #2 also. Table 1 represents the
observations are given relatively more weight in forecasting predicted average EGT values per flight phase and the
than the older observations. In the case of moving averages, confidence intervals associated with these values. The
the weights assigned to the observations are the same and assumption here is that the upcoming flight will be about
are equal to 1/N. In exponential smoothing, however, there 240 minute (4 hour duration).
are one or more smoothing parameters to be determined (or The double exponential smoothing based prediction scheme
estimated) and these choices determine the weights assigned can track the flight parameter values for upcoming flights
to the observations. and assign confidence intervals to the assigned values. This
analysis will therefore help the maintainer in keeping track
In the smoothing scheme y stands for the original of flight parameter behavior in different flight modes. The
observation and Si stands for smoothed observation. The maintainer can then assign possible maintenance actions
sub-scripts refer to time periods, 1,2,…n. The two equations based on projected values and plan for preventative
associated with Double Exponential Smoothing are: maintenance.

St = αyt + (1 − α )( St −1 + bt −1 );0 ≤ α ≤ 1
Table 1 – Predicted EGT values with confidence intervals
bt = γ ( St − St −1 ) + (1 − γ )bt −1;0 ≤ γ ≤ 1
Flight Predicted Upper Lower
Mode Average Confidence Confidence
S1 is in general set to y1 and b1 = [(y2-y1)+(y3-y2)+(y4-y3)]/3; EGT(Deg C) Limit(Deg C) Limit(Deg C)
α and γ are set to 0.95. The first smoothing equation adjusts Take off 811.21 833.40 789.00
St directly for the trend of the previous period, bt-1, by
adding it to the last smoothed value, St-1. This helps to Cruise 639.10 642.50 635.70
eliminate the lag and brings St to the appropriate base of the Landing 438.58 460.60 416.60
current value. The second smoothing equation then updates
the trend, which is expressed as the difference between the
last two values. The equation is similar to the basic form of
single smoothing, but applied here to the updating of the
trend. Using the smoothing equations the upcoming values
for the selected flight parameters can be forecasted. The
one-period-ahead forecast is given by:
Ft +1 = St + bt
The m-periods-ahead forecast is given by:
Ft + m = S t + m × bt
If the duration of the upcoming flight is known this analysis
can help predict how the parameter under observation will
behave in the next flight. After estimating the predicted
average EGT value for the upcoming flight a 90%
confidence interval is defined around this estimate using:

7
 6. CONCLUSIONS

Data driven diagnostics and prognostics algorithms have

been applied for engine health management based on both
on-board flight data and off-board maintenance data. The
engine health degradation analysis is used to distinguish
between rapid and gradual deterioration and identify fault
events in case of rapid trend shifts. The fault events can
then be linked to maintenance actions that were performed
during that event and help narrow the possible cause of the
event or identify if the fault was addressed or not. The
engine difference analysis focused on the disparity in
behavior of the engines on a single aircraft. This analysis
tried to identify significant behavior disparities and identify
such instances for the maintainer to evaluate and make
Figure 9 – Average EGT values during Take off future maintenance decisions. The engine health prognostics
mode for Aircraft ‘A’ provide a layout for trending of critical flight parameters for
upcoming flights. This additional information can assist the
maintainer to make appropriate maintenance and reliability
decisions and prioritize maintenance procedures in order to
reduce aircraft down time. Knowledge of flight situational
parameters will therefore be decisive in avoiding critical
faults/failures during flights and improve estimates of
engine deterioration.

7. REFERENCES

[1] Estefan Ortiz, Ashish Babbar, Vassilis Syrmos, Gregory

Clark, John Vian, Michael Arita, “Multi-Source Data
Integration for Aircraft Health Management”, 2007 IEEE
Aerospace Conference Proceedings, March 2007.
Figure 10 – Average EGT values during Cruise mode
for Aircraft ‘A’
[2] J. Luo. and M. Namburu, “Model-based prognostic
techniques”, Proc.IEEE Systems Readiness Technology
Conference, September.2003, pp.330-340

[3] Stanley C. Ofsthun and Timothy J. Wilmering, “Model-

Driven Development of Integrated Health Management
Architectures,” 2004 IEEE Aerospace Conference
Proceedings, March 6-13, 2004.

[4] Ashish Babbar and Vassilis L. Syrmos, “Data Driven

Approach for Fault Detection and Identification using
Competitive Learning Techniques”, European Control
Conference, Kos, Greece 2007.

[5] Estefan M Ortiz and Vassilis L. Syrmos, “Support vector

machines and wavelet packet analysis for fault detection
and identification” International Joint Conference on
Neural Networks, Vancouver, Canada, 2006.
Figure 11 – Average EGT values during Landing
mode for Aircraft ‘A’
[6] Xudong Wang and Vassilis L. Syrmos, “Optimal
Cluster Selection based on Fisher Class Separability
Measure”, American Control Conference 2005.

8
[7] P. Frank, R. Patton and R.Clark, Fault Diagnosis in 8. BIOGRAPHY
Dynamic Systems,Theory and Application, Prentice
Hall, New Jersey, 1989 Ashish Babbar is a Lead Systems Engineer with Research
Corporation of the University of Hawaii
[8] P. Frank, R. Patton and R.Clark, Issues in Fault (RCUH). He obtained his Masters of
Diagnosis in Dynamic Systems, Springer, Verlag, Science in Electrical Engineering from the
2000 University of Hawaii at Manoa in 2006. He
has been working as a Systems Engineer
[9] J. J. Gertler, Fault Detection and Diagnosis in on the Integrated Aircraft Health
Engineering Systems, Marcel Dekker, New York, Management (IAHM), Sentient Adaptive
1998 Systems Technologies (SAST) and Unexploded Ordnance
(UXO) detection projects. He has over two years of
[10] D. C. Swanson, “A general prognostic tracking experience in diagnostic and prognostic algorithm
algorithm for predictive maintenance,” Proc. IEEE development as applied to health management applications.
Aerospace Conference, vol. 6, March. 2001, pp. His research interests include signal processing, sensor
2971-2977. fusion, data mining, feature extraction and pattern
recognition algorithm development. He has made key
[11] X. R. Yin, J. Z. He, Z. H. Zhou “Using Neural contributions in development and implementation of
Network for Fault Diagnosis,” Proc. of the IEEE- condition based maintenance systems and GUI development
INNS-ENNS International Joint Conference on for health management of commercial aircraft.
Neural Networks, Volume 5, July 2000, pp. 217-220.
Estefan Ortiz is a Principal Research Engineer with the
[12] G. Betta, C. Ligouri, and A. Pietrosanto, “An Research Corporation of the University
Advanced Neural Network based instrument fault of Hawaii. His current responsibilities
detection and isolation scheme,” IEEE Trans. include development of advanced signal
Intrumen. Meas., vol 47, April 1998, pp. 507-512. processing techniques, development of
fault diagnostic and prognostics
[13] Lin Y., et al, 2002, Maintenance and Repair: A algorithms, development of pattern
Simulation Model for Field Service with Condition discovery and recognition methods, and research of
Based Maintenance, Winter Simulation Conference, advanced machine learning. He obtained his Master of
2002. Science in Electrical Engineering from the University of
Hawaii at Manoa, in 2006. He received a Bachelor of
[14] Rajamani R., et al, Condition Based Maintenance for Science degree in Electrical Engineering from St. Mary’s
Aircraft Engines, ASME GT 2004-54127. University, San Antonio Texas in 2003. He received a
Bachelor of Arts degree in Mathematics from St. Mary’s
[15] Volponi, A.J., Extending Gas Path Analysis Coverage University, San Antonio Texas in 2003. Mr. Ortiz is a
for Other Fault Conditions, Gas Turbine Condition HACU General Motors Engineering Excellence Award
Monitoring and Fault Diagnosis, Von Karman recipient and a Society of Mary Scholarship award winner
Institute for Fluid Dynamics Lecture Series 2003. from St. Mary’s University.

[16] Ganguli Ranjan, Trend Shift Detection in Jet Engine Vassilis Syrmos Dr. Syrmos obtained his Ph.D. at Georgia
Gas Path Measurements Using Cascaded Recursive Institute of Technology, Atlanta,
Median Filter with Gaussian and Laplacian Edge U.S.A. in 1991, in Electrical
Detector, Journal of Engineering for Gas Turbines and Engineering. Since 1991, he has been with
Power, January 2004, Volume 126, Issue1, pp 55-61. the Department of Electrical Engineering
at the University of Hawaii at Manoa,
[17] Ganguli Ranjan, Adaptive Myriad Filter for Improved where he is a Professor and the Associate
Gas Turbine Condition Monitoring Using Transient Vice Chancellor for Research and
Data, Journal of Engineering for Gas Turbines and Graduate Education. He also served as the Associate Dean
Power, April 2005, Vol. 127, Issue 2, pp. 329-339. of the College of Engineering and the Interim Director of
Science and Technology for the Research Corporation of
[18] Ganguli Ranjan, Data Rectification and Detection of the University of Hawaii. He spent a sabbatical leave at The
Trend Shifts in Jet Engine Gas Path Measurements Boeing Company as a Research Fellow. His interests
using Median Filters and Fuzzy Logic, Journal of include geometric and algebraic approaches in linear
Engineering for Gas Turbines and Power, October system theory, computational algorithms methods for signal
2002, Vol. 124, Issue 4. and image processing, robust/optimal filter design in
systems, medical imaging, and prognostics and diagnostics

9
methods related to condition based maintenance systems.
He is a Boeing A.D. Welliver Fellow, a member of the
board of Hawaii Technology Development Venture
program, a member of the Sigma Xi research organization,
and a senior member of IEEE.

Michael Arita Honolulu born, Mike was employed at

Hughes Aircraft in El Segundo, CA. for 11 years where he
worked on military airborne radar weapons systems - F-14
Tomcat and F/A-18 Hornet Radar System production
programs in Quality Assurance. Currently at Aloha
Airlines, he heads the Reliability Section in support of
Aloha’s Boeing 737-200 inter-island fleet and 737-700 Next
Generation Extended Range fleet.

10