National Conference on Condition Monitoring (NCCM),

October 4 - 5, 2013, Bangalore

NCCM-2013-27

Detection of Incipient Stall in the Axial Compressor of

a Gas Turbine Engine

C. Jagadish Babu#1, Sanjay G. Barad*2, Mathews P. Samuel#3, R.K. Giridhar*4

#
RCMA (Eng), CEMILAC
C.V. Raman Nagar, Bangalore, 560093, India
1
jagadish.gtt @gmail.com
3
[email protected]
*
GTRE, DRDO,
C.V. Raman Nagar, Bangalore, 560093, India
2
[email protected]
4
[email protected]

ABSTRACT
Aerodynamic instabilities such as compressor stall and surge are the major functional
failure modes in a gas turbine engine. Ensuring adequate margins between the operating
point and the stability limit, so as to prevent such failures is an important airworthiness
requirement. Along with meeting this safety criterion, detection and control of stall improves
the compressor performance. In order to detect the onset of stall, a few parameters related to
the aerodynamic stability of axial compressor are sensed and processed further to derive
information on certain statistical moments that are indicative of the incipiency of compressor
stall. Details of approach and procedures employed for condition monitoring in a typical gas
turbine engine by using the statistical moments of parameters influencing onset of stall has
been presented in this paper.

Compressor section of the gas turbine engine studied here consists of three stages of
low-pressure compressor followed by six stages of high-pressure compressor. The given
compressor handles mass flow of 78 kg/s with an overall pressure ratio of 21.5 at the rated
speed at sea level static conditions. In this study, engine behaviour during stall has been
analysed both in time and frequency domains. The pressure fluctuations were captured
through Kulites placed at the casings of the low-pressure compressor, intermediate
compressor entry and inlet guide vanes. A feed forward back-propagation neural network
model was developed to predict the statistical moments of the parameters of interest. Using
the engine operating conditions as input, the model has been trained to predict the statistical
moments of the pressure fluctuations that are encountered during normal operation to the
error level of 10-5. The methodology has been successfully demonstrated during actual engine
test in which the onset of stall has been detected by comparing the output of neural network
model with the statistical moments directly evaluated under the given operating conditions.
The methodology described in this paper opens up an area of research involving
computationally viable and operationally reliable condition monitoring systems to mitigate
aerodynamic instability of axial compressors used in gas turbine engines.
Key Words: Compressor stability, Stall detection, Statistical moments, Condition
monitoring.
1. INTRODUCTION
The term stability denotes the capability of a compressor to recover from the
disturbances that alter its operational equilibrium. Loss of stability of a compressor is
generally identified as stall or surge. In ideal cases, a stable system when exposed to small
transient disturbances, returns to its original operating point and if subjected to moderate
disturbances, the compressor stabilises at a new operational equilibrium point. The instability
issues of the compressor in a gas turbine engine have different connotations from
performance as well as a safety point of view. The primary issue is the aerodynamic stability,
which deals with deterioration in safety margins due to flow-separation, stall or surge. The
second issue is the performance stability that deals with matching of compressor performance
with downstream components. Research on improving the aerodynamic stability aim at
extending the usable range of compressor performance map by alleviating rotating stall and
surge [1-4]. Apart from the aerodynamic effect on flow, these instabilities impart heavy stress
on the components of the order of 2 to 5 times that of its normal operating range, even
leading to catastrophic destruction of the hardware. Such safety hazards motivated several
studies [5-8] to detect incipient stall and monitor the reduction in the margins before
developing into a full blown surge. This paper is another independent attempt to develop an
alert on the incipient stall based on pressure measurements in the flow path.
Non realising stable compression system that can handle high pressure ratios is one of
the major stumbling blocks in the advancement of gas turbine engines. The overall pressure
ratio has increased from the range of 3-5 in the initial days of gas turbine engine to the order
of 25-30 in the modern engines. When all other conventional methods of improvements reach
a level of saturation, engine performance can be enhanced by operating the compressor close
to the instability region to take the advantage of higher pressure ratio. However in such cases,
systems to detect and control incipient instability are very essential and the research reported
herein and similar works reported elsewhere addresses such critical issues of operating an
engine in the vicinity of a potentially dangerous regime. When a compressor stalls, the flow
over the blades separates with reduction in both pressure rise capability and flow capability.
With further increase in instability, the compressor cannot hold the high pressure in the
downstream and surge occurs with violent flow reversal. In order to avoid stall/surge,
airworthiness engineers insist on demonstration of adequate surge margins. But from the
design perspective, enforcement of large surge margins is a strong constraint that limits the
efficiency and necessitates over designing the compression system. Hence the research on
reducing the surge margin without compromising operational safety is a topic of interest to
the designers. Cousins et al.[6] experimentally showed that surge and rotating stall begin
through flow separation within a blade row and this local separation grows into a full rotating
stall or to an extreme event of surge affecting the entire machine. Day [9] further
substantiated this and claimed that surge is always preceded by rotating stall. Results of these
studies imply that if one can reliably detect incipient stall, mechanisms can be instituted such
that the distance between surge line and the operating line can be brought down from the
current levels considerably. This paper focuses on the first task ie., detection of incipient stall.
Once the stall is detected, active stall control such as operation of blow-off valves or guide
vane adjustment [9] can be resorted to as a means of preventing its approach towards the
stability limit. There are several attempts reported in literature to predict the stall and surge in
axial compressors using a variety of measurements such as velocity [3, 10], vibration [11]
and pressure[8, 12]. In this paper we further peruse on the detection of aerodynamic
instability based on the measurement of pressure signals in the airflow path.
 The remaining portion of this paper is organised as follows. Section 2 provides the
rationale of the investigation and the methodology adopted in this work. It further discusses
the experimental set up and main indications of compressor stall, leading to identification of
the parameters to be taken up in this investigation. Section 3 discusses the various methods
for stall identification considered in this study. Data acquisition and training of the neural
network is taken up in section 4 and the online detection of compressor instability is
presented in sections. The article is concluded in section 6 highlighting the important findings
of this study.

2. EXPERIMENTAL METHODOLOGY
2.1 Investigation Rationale
It is proposed to detect instability of the compression system based on the extracted
measurable parameters y(t) at time t through a four step process as shown in Figure 1.
Initially the signal x(t) obtained from the measurement probes is pre-processed to remove the
noise and prepare the data for feature extraction. Then the information relevant to the
detection process is extracted from the pre-processed data x’(t) to a feature vector y(t).
Feature extraction facilitates easy and reliable detection. The final step is to identify the
instantaneous state of the system qi(t) in terms of stability by comparing the feature vector
y(t) extracted from the real time data with the feature vector y’(t), the output of the artificial
neural network model. The detection criterion for incipient stall is the departure from known
stable operation. Hence the feature vector y’(t) corresponding to stable engine operation for a
given set of stage variables at inlet and exit of the compressor need to be evaluated using
artificial neural network (ANN) model.

Stage ANN Model

Variables Simulation
P1, T1, y’(t)
P2, T2
Pre Feature Instability
processor Extraction Detection
Sensor x(t) x’(t) y(t) Model
qi(t)
(Signal) Comparison
Signal Statistical Detect stall if
Kulites, conditioning moments y(t) vs. y’(t) any

Figure 1. Flow diagram for the instability detection process.

2.2 Experimental Setup

The compressor discussed here consists of three stages of low-pressure compressor
followed by six stages of high-pressure compressor with transonic blades. Variability is
provided for inlet guide vanes as well as the first two stages of compressor stators. This
compressor can handle mass flow of 78kg/s with an overall pressure ratio of 21.5. The
measurements are taken during the test runs carried out in a normally aspirated ground test
facility. Since this paper is focused on detection of stall rather than the stall initiation
mechanism and the compressor characteristics leading to instability, further discussion on the
design features of the compressor is excluded here. However the salient features of the
experimental setup relevant to this study is given in detail. Data sensing and acquisition setup
was created as an augmentation in the regular engine test facility. Pressure sensors (Kulite
probes) are located at Low Pressure (LP) casing, Intermediate Casing Entry (ICEN), Inlet
Guide Vanes (IGV) as indicated in Figure 2. The pressure measurements were done using the
Kulite XCQ-062 which allows for a very rugged package suited for probes, pressure rakes
and other similar test set ups. This transducer is well suited for both dynamic and static
pressure measurements in benign or harsh environments. The main parameters of interest in
pressure measurement are provided in Table 1. The data acquisition hardware consists of NI-
USB 9233 with 50kS/s sampling rate. The NI USB-9233 data acquisition device[13] provides
a USB interface for four channels of 24-bit analog inputs with integrated signal conditioning.
It consists of two components viz., an NI 9233 module and an NI USB-9162 carrier. The
block diagram of the data acquisition system used in the current work is shown in Figure 3.
LP1 ICEN IGV

Figure 2. Location of pressure sensors on LP Compressor

Table 1. Specifications of Pressure Probe used

Parameter Range/Value
Pressure Full Scale Range(s) 5 – 1000 psi
Sampling rate 50 kS/s
Output Excitation 100 mV
10 V DC
10 V AC
15 V DC
15 V AC

Temp Range -55 to +120°C

Figure 3. Data Acquisition System Schematic Diagram.

2.3 Indicators of Stall
Stall is characterised by the presence of large wakes on portions of the blade that
arises due to separation of suction surface boundary layer. In the entire range of compressor
speeds there is a minimum flow point for each speed, where the compressor will stall/surge
depending upon the characteristics of the compressor. Generally compressor stall may
manifest as either an individual blade stall or rotating stall. Individual blade stall is
characterised by the development of large separated-flow regions or zones of low flow in the
wake of each blade. Rotating stall consists of large stall zones covering several blade
passages and propagating at some fraction of rotor speed in the absolute direction of rotor
rotation. The locus of minimum-flow points just before the performance of the compressor
deteriorates abruptly is known as stall-limit line or surge line. There are several causes that
can lead to a stall. Occurrence of instability in the compressor is manifested as shown in
Figure 4.

From the above, it is clear that the gas dynamic instability behind the blades is the
origin of stall and if allowed to grow it can lead to a surge. Therefore, the most important
point is to identify the origin of stall. One such attempt reported in literature[10] is the
velocity measurement using hot wire anemometer which could identify the precursor for stall
about 10ms prior to it developing to a serious instability. Though 10ms duration was too short
to initiate any action to prevent the surge, this study provided further impetus to carryout
research in identifying precursors in terms of pressure signals which appear 1 or 2 seconds
before the compressor malfunction. Dhingra et al.[8] and Dremin et al. [12] showed that
statistical parameters of pressure measurements can predict incipient stall. In this study we
further explore the possibility of using statistical moments of pressure measurements as a
precursor for detecting the incipient stall.

Compressor LP & HP Incompatible VG/ Over Pressure High JPT

Damage Mismatch Nozzle schedules fuelling Fluctuation Vibration increase

Stall

Off- Design Down stream Inlet Transient Mass flow High Strain Speed
Conditions disturbance Distortion Operation Fluctuation Levels Fluctuation

Figure 4. Causes and Effects of stall in Gas turbine engines

2.4 Parameters Considered in this Study

On obtaining the data as per the scheme shown in Figure 3, statistical features of
Standard Deviation (SD), Skewness (SK), Kurtosis (KT) are extracted from the signal
acquired. The nomenclatures used for these features in this paper are as given below:

LP1(SD), LP1(SK), : refers to standard deviation, skewness and kurtosis of the

LP1(KT) pressure sensor located at LP1 respectively.

ICEN(SD), ICEN(SK), : refers to standard deviation, skewness and kurtosis of the

ICEN(KT) pressure sensor located at ICEN respectively.
IGV(SD), IGV(SK), : refers to standard deviation, skewness and kurtosis of the
IGV(KT) pressure sensor located at IGV respectively.

NL, P1, T1, P2, T2 : refers to low pressure compressor rotor speed, pressure and
temperature at the inlet of compressor and exit of the
compressor respectively.

The data extracted needs to be conditioned to remove spikes and noise. This has been
done through manual processing. Since our focus is on instability phenomena, the transient
data has been carefully captured and the steady state RPM data has been ignored, as it is not
relevant in this study. The selected data has been condensed to 87 points between speed range
of 960 to 9560 RPM. This conditioned data has been further used for training the neural
network.

3. IDENTIFICATION OF STALL
Engine behaviour during stall has been analysed both in the time and frequency
domains from the signals recorded in various engine runs. In the time domain method various
statistical parameters are extracted from the signal. From time domain signals, stalls are
clearly observed at ICEN, IGV locations and ICEN location as high as the intensity of about
0.4 Psi. These signals looked like pressure pulsations in time domain. A sample of the time
domain signal captured during engine run for illustrating the contrast for with and without
stall is presented in Figure 5(a).

Stall Signal
Normal Signal

(a) Signals with and without stall

Normal Signal

Stall Signal

(b) Probability density function

Figure 5. Contrast between Stall and Normal Signals at ICEN

 Probability density function as shown in Figure 5(b) has also been utilised to
represent the stall in time domain. From the probability density shown at the right hand side,
one can clearly see the difference in skewness of the stall signal, a characteristic property that
has been utilised in this paper to detect stall.

In the frequency domain analysis, fast Fourier transform (FFT) of the time pressure
signal is used to detect stall frequencies. For a typical data sample captured during engine
test, Spectrogram and Campbell diagram are plotted to depict the stall frequencies and
resonance points respectively. The effect of stall in the spectrogram of the ICEN pressure is
seen in the FFT plot of the signal against time as shown in Figure 6(a). Stall frequency is seen
around 0.61 order and the stall is getting intensified as the engine speed is increasing.

(a) Spectrogram (b) Cambell diagram

Figure 6. Frequency domain analysis of the Stall Signal at ICE

Campbell diagram obtained by plotting the variations in the natural frequencies of the
component against rotor shaft speed and the engine order line is utilised to detect the
resonance point. Intersection point of order line, component frequency and speed represent a
resonance point. In the Campbell diagram (shown in Figure 6(b)) obtained by plotting FFT of
the IC Entry pressure signal against the rotor speed (NL), the order lines are stall frequency
lines (0.61 order line) and their harmonics.

During the experiments a distinct difference was observed between the measurements
from the static frame of reference and the rotating frame of reference. Let us denote the stall
frequency in static frame of reference as Fs, rotating frame of reference as Fr and engine
frequency as Fn. Then the relationship between these frequencies are given by Fs = Fn-Fr. In
the current experimental setup, stall in the low pressure compressor was sensed by pickups
(Kulite) mounted on casing (stationary frame of reference) and the strain gauges mounted on
the rotor blades (rotating frame of reference). Consequently, Figure 7(b) shows strain gauge
signal with a stall frequency of 0.39 order, where as Figure 7(a) shows a stall frequency of
0.61 order arising from the pressure signal mounted on casing.
 (a)Static frame of reference (b) Rotating frame of reference
Figure 7. Static and rotating frame of reference

4. DATA ACQUISITION

The data from three Kulite pressure sensors (located at LP1, ICEN, IGV) as well as
the data on low speed spool (NL), air properties at the inlet (P1, T1) and at the exit (P2, T2) of
the LP compressor have been acquired with NI-USB 9233 / NI 4472 as per the setup
configuration shown in Figure 8(a) and the acquired wave form is displayed in the Figure
8(b). As per the investigation plan presented in section 2.1, conditioned data is further
processed to extract the features and then compared with the output of the artificial neural
network (ANN) model. Hence the next point to be discussed is stall detection using ANN
model.

(a) Data acquisition settings (b) Signals acquired

Figure 8 Data acquisition setup and signals acquired

The stall detection programe is conceived to have time-domain analysis with simple
extracted statistical parameters and neural networks. The software application has been
developed using LabVIEW 7.1 and Matlab 2007a has been used for Neural Network (NN)
tool bar. However, the new versions of LabVIEW (version 8.6 and above) has NN tool bar
incorporated in it. The input vector consists of five parameters such as low pressure
compressor’s speed, inlet conditions P1, T1 and exit conditions P2, T2. The target vectors are
nine statistical features viz., LP1(SD), LP1(SK), LP(KT), ICEN(SD), ICEN(SK), ICEN(KT),
IGV(SD), IGV (SK), IGV(KT). Further details of neural network configuration for training
are given Table 2. This feed forward back propagation neural network model has been trained
using the normal engine run data till the error level has been brought below 10-5.

Table 2: Details of neural network configuration

Sl.No Network Configuration Details
1 Network type Feed forward back-propagation
2 Number of layer Two hidden layers
3 Number of neurons Input neuron =5; Target Neurons = 9
Neurons in hidden layer 1 = 87
Neurons in hidden layer 2 = 50
4 Transfer function Tansig in all layers
5 Network inputs NL, P1, T1 P2, T2
6 Targets Target vectors are LP1(SD), LP1(SK),
LP(KT),ICEN(SD), ICEN(SK), ICEN(KT),
IGV(SD), IGV (SK), IGV(KT)
7 Training function TRAINLM
8 Adaption learning function LEARNGD
9 Epochs & goal Epochs = 1000; Goal = 0.00001

5. ON-LINE DEMONSTRATION
One of the developmental test runs of the engine was choose for demonstration for the
methodology discussed in this paper. For the specific demonstration run, test schedule
consists of four dwell segments. The engine started and then dwelled at 4500 RPM for some
time and followed by dwell at 6500 RPM and at 9000 RPM. While coming down again
dwelled at 6500 RPM and at 4500 RPM before engine shut down. During this run, stalls were
noticed between fist dwell and second dwell and hence throttle excursions made around 4500
RPM to 6500 RPM engine to get rid of stall as shown in Figure 9. Now let us look into the
details how the proposed methodology was useful in capturing these stalls. As presented in
the instrumentation scheme (Figure 2), there are three measurement locations. The data form
all these sensors were analysed as a part of this work. However for the brevity of this paper,
we discuss only one typical case in the remaining portions of this paper. As a representative
measure, signals form the ICEN location (ie middle location) is selected for detailed
discussion on incipient stall detection.

The stall detection system monitors the engine compressor continuously through
acquired pressures signals from the Kulite transducers. For every instantaneous speed, the
statistical features from these signals are calculated and also predicted using neural network.
In this paper, the statistical parameters of actual observations calculated and the predicted
(simulated) values from the neural network model are indicated by _A and _S suffixed to
parameter symbols respectively. During the demonstration test run, the system detected onset
of stall by comparing the deviation in magnitude of calculated statistical parameters of actual
observations with those values predicted by the neural network. When the stall vanished i.e
above 5457 RPM, both simulated and actual results are coinciding. The snapshots taken
during demonstration trial w. r. to the standard deviation ICEN(SD) and skewness ICEN(SK)
of the signal at ICEN location are presented in Figure 10 with the marking on the stall
inception points for easy identification.
 Throttle manipulation
to get rid of Stall cells

Figure 9. Cyclogram for the demonstration test.

Now let us have detailed look into the stall inception captured by the system. Plots are
prepared for demonstrating the stall inception based on the departure from the normal trend in
standard deviation against speed and time. Following observations can be drawn from these
graphs presented in Figure 11. Figure 11(a) shows variation of two parameters ICEN(SD)_A
(plotted in red colour) and ICEN(SD)_S (plotted in green colour) vs. Speed(RPM). Similarly
Figure 11(b) shows variation of three parameters ICEN(SD)_A (plotted in red colour),
ICEN(SD)_S (plotted in green colour) and Speed NL (plotted in blue colour) vs. time/block
number.

(a) ICEN Std deviation (b) ICEN Skewness

Figure 10. Std deviation and skewness at ICEN location

First let us look at the indications based on the standard deviation of the pressure
signal. A marked deviation of ICEN(SD)_A from the ICEN(SD)_S was observed during stall
and the deviation persisted as long as stalls were present even with varying RPM. Intensity of
the stall cells (magnitude) increased with increase in speed. Maximum intensity of the stall at
ICEN location (NL=6387; SD=0.354565) was found to be more than the maximum intensity
noticed on data collected from IGV location. Further it was observed that stall cells
disappeared at a point where RPM=5457, ICEN (SD)_S=0.083786 Psi, and ICEN (SD)_A=
0.098827 Psi. The sudden disappearance of stall cells was evidenced by the sudden
convergence of ICEN(SD)_A and ICEN(SD)_S. This was an indication of clean/smooth flow
in the compressor ie., no marked deviation of ICEN(SD)_A from the ICEN(SD)_S was
observed when stall cells were not present even with varying RPM.
 In a nutshell, departure of two curves ICEN(SD)_A (plotted in red colour) and
ICEN(SD)_S (plotted in green colour) in these graphs tentatively indicates the presence of
stall cells and convergence of these two curves w.r.to changes in speed as well as time
indicates the absence of stall cells. However let us further explore the pattern of skewness of
the signals before confirming the onset of the stall.

(a) ICEN Std deviation Vs Speed (b) ICEN Std deviation Vs Time
Figure 11. Std deviation against RPM and time at ICEN location

The study on the skewness of the variable of interest at ICEN location is presented in
Figure 12(a) and (b). In these plots, variation of skewness against speed and time are
considered for detailed study. In Figure 12(a), variation of two parameters
ICEN(SK)_A(plotted in red colour) and ICEN(SK)_S (plotted in green colour) vs. speed is
considered. Similarly in Figure 12(b), variation of three parameters ICEN(SK)_A(plotted in
red colour), ICEN(SK)_S (plotted in green colour) and Speed NL(plotted in blue colour) vs.
speed is accommodated. A marked deviation of ICEN(SK)_A from the ICEN(SK)_S was
observed during stall and the deviation persisted as long as stalls were present even with
varying RPM. A point of onset and disappearance of stall was clearly detectable through the
study of skewness as shown in Figure 12.

(a) ICEN Skewness Vs Speed (b) ICEN Skewness Vs Time

Figure 12. Skewness against Speed and time at ICEN location

During the stall, skewness was observed to be negative (around -1), which can be
corroborated with the periodic pulses of the stall signals with negative amplitude. Further
sudden disappearance of stall cells is indicated in the plots by the sudden convergence of
ICEN(SK)_A and ICEN(SK)_S. There was no significant deviation of ICEN(SK)_A from
ICEN(SK)_S observed when stall cells were not present even with varying RPM. During the
demonstration run, the stall cells were detected just in the vicinity of 4500 RPM during stair
up cycle and these stall cells disappeared at a point where the speed reached 5457 RPM.
These events are explicitly captured by the study of skewness of the pressure signals and the
program could identify both onset and vanishing point of stall.

From the study reported herein, it is inferred that the standard deviation is one of
sensible statistical parameter for detection of stall. However, increase in the value of standard
deviation alone cannot be taken as a conclusive confirmation on stall, unless it is
corroborated by evidence in skewness. Before concluding this paper, let us highlight a few
off shots of this study. In addition to the online detection of the incipient stall, this study
provides a platform for detailed off-line analysis using two by-products ie., an offline signal
player and an offline simulator. The offline signal player is helpful in analyzing the
previously recorded data. Similarly the off-line simulator generates simulated parameters
from neural network for further analysis and study of the system behaviour.

6. CONCLUSION
The work presented herein has provided a basic framework required for incipient stall
detection in gas turbine engine and demonstrated its capability in the real time environment.
The capability to detect engine stall with simple statistical parameters such as standard
deviation and skewness has been established through the on line and off line studies carried
out. Of course, sensitivity and reliability in detection of stall lies in the reliability of neural
network model which represent the engine behaviour. Among the variables considered,
standard deviation of the pressure signal could detect the presence of stall. However, a
combination of standard deviation and skewness could detect the incipient stall as well as the
disappearance of stall conclusively. The third statistical quantity studied here namely,
kurtosis was not found sensitive enough to the incipience of stall to consider it as a parameter
for detecting the onset of stall. Out all the three statistical moments considered in this study,
skewness was found to be the best indicator of stall in the given compressor settings.

ACKNOWLEDGEMENTS
The authors wish to express their sincere thanks to Dr. K. Tamilmani, Chief Executive,
CEMILAC and Chief Controller R&D (Aero), who has shown keen interest in this study and
provided whole hearted encouragement during the course of this investigation. Authors also
express their gratitude towards the scientists at vibration engineering group at GTRE for their
support during this study.

REFERENCES
[1] Bright, M. M., Qammar, H.K., Weigl, H.T. and Panduano, J. (1997). Stall Pre-cursor
Identification in High speed Compressor Stages Using Chaotic Time Series Analysis
Methods. Journal of Turbomachinery, Vol 19, Issue 3, pp. 491- 499.
[2] Owen A.K., and Davis Jr. M.W. (1994). Modeling the Dynamic Behaviour of an Axial-
centrifugal System. AIAA paper # 94-2802, AIAA/ASME/SAE/ASEE joint Propulsion
conference, Indianapolis, Indonesia.
[3] Nicolas Gourdian, Stephane Burguburu, Francis Lanoenf and Guy Jean Michon (2010),
Simulation of Rotating Stall in a Whole Stage of an Axial Compressor. Computers and
Fluids 39, pp. 1644-1655.
[4] Methling F.O., Prute R., and Stoff, H. (2000). The Onset of Aerodynamic Instability in a
3-Stage Transonic Compressor. Published in RTO MP-051, RTO AVT Symposium on
Active control Technology for Enhanced Performance Operational Capabilities of
 Military aircraft, Land Vehicles, and Sea vehicles, held in Brunschweig, Germany 8-11
May 2000, pp.8-1 – 8-12.
[5] Cousins W.T., Brien W.F. (1985). Axial Flow Compressor Post-stall Analysis. AIAA
paper 85-1349.
[6] Cousins W.T., Jones M.G. and Belling T.L.(1995). Stall and Surge Characteristics of
Axial Compressors: The enhancement of Engine Stability, AGARD Symposium on
Loss Mechanisms in Turbomachinery, (AGARD CP-571) Derby, UK.
[7] Breuer, T., Servaty, S. (1995). Stall Inception and Surge in High Speed Axial Flow
Compressors. AGARD Symposium on Loss Mechanisms in Turbomachinery,
(AGARD CP-571) Derby, UK. 26/ 1-17.
[8] Dhingra, M., Neumeier, Y. and Prasad J.V.R. (2003). Stall and Surge Precursors in
Axial Compressors. Paper No. AIAA 2003-4425, 39th AIAA/ASME/SAE/ASEE joint
Propulsion conference and Exhibit, 20-23 July, Huntsville, Alabma, pp1-8.
[9] Day (1993). The Unstable Behaviour of Low and High Speed Compressors. Journal of
Turbomachinery, Vol 16, Issue 2, pp. 194- 201.
[10] Georgantas, A. (1994). A review of Compressor Aerodynamic Instabilities. Canada:
National Aeronautical Establishment, NRC.
[11] Marshall, D.F. and Sorokes, J.M. (2000). A Review of Aerodynamic Forces Acting on
Centrifugal Compressors and Resulting Vibration Characteristics of Rotors. Proc. 29th
Turbomachinary Symposium, September 2000, pp. 263- 280.
[12] Dremin, I.M., Furletov, V.I., Ivanov, O.V., Nechitailo, V.A., Terziev, V.G. (2002).
Precursors of Stall and Surge Processes in Gas Turbines Revealed by Wavelet Analysis.
Journal of Control Engineering Practice, Vol 10, pp. 599-604.
[13] National Instrument. web site [online]. [accessed June 15, 2013]. Available in www-
format: <URL: http://www.ni.com/pdf/manuals/371569d.pdf

C. Jagadish Babu is currently working as a Technical Officer at the

Regional Centre for Military Airworthiness (Engines), CEMILAC,
Bangalore. His research interests include condition monitoring and
airworthiness assurance.

Sanjay G. Barad is currently working as a scientist at Gas Turbine

Research Establishment (GTRE), Bangalore. His research interests include
vibration analysis, rotor dynamics, signal processing and condition
monitoring of aero engines.

Dr. Mathews P. Samuel is currently working as a scientist at the Regional

Centre for Military Airworthiness (Engines), CEMILAC, Bangalore. His
research interests include condition monitoring, aero engine life cycle
management, system modeling and airworthiness assurance.

R.K. Giridhar is currently working as a scientist at Gas Turbine Research

Establishment (GTRE), Bangalore. His research interests include vibration
analysis, rotor dynamics and condition monitoring of aero engines.