The integration of several monitoring techniques has resulted
in a system that can produce an almost instant indication of the health
of an industrial gas turbine and of faults that are developing within
The computer based analysis system allows a turbine operator to detect
the onset of faults that could lead to expensive down-time and repairs.
The failure of a blade in a gas turbine, for example, could destroy
much of the engine within seconds unless the turbine is switched off
The three collaborators in the project, known as GASTEM, have for
the first time analysed, almost in real time, acoustic and thermal
signals from a turbine along with readings of the aerothermodynamic
conditions in the engine, using expert system computer techniques.
Conditioning monitoring techniques based
on the measurement of characteristics such as bearing vibrations
are well established throughout industry, and alert the user to
changes in a machine that could indicate developing faults. The
techniques are, however, often difficult to apply in the harsh environment
within industrial gas turbines, which are used widely for power
generation, pumping, compression or other applications. Yet faults
can develop very quickly in turbines and the failure of one of the
many blades in a machine can have a devastating effect. Even small
faults can require costly repair work, and expensive downtime. Other
faults can reduce global thermodynamic efficiency and degrade the
economical justification of operating such an industrial powerplant.
GASTEM has tackled the problem of how to monitor the health and
effectiveness of gas turbines by developing computer models that
can analyse in an integrated manner both the noises produced by
a turbine and the thermodynamic measurements from the operating
engine. The expert system software combines the different forms
of signals to diagnose faults, such as a blade being twisted out
of its normal position, within about 10 seconds. This allows rapid
action by the operator to prevent further damage to the turbine.
The partners are the acoustics and noise specialists Metravib RDS
of France, the gas turbine manufacturer European Gas Turbines (EGT)
of the UK, and the National Technical University of Athens (NTUA)
which is expert in turbomachinery.
The partners believe there is a huge potential market for the portable
computer system they have developed to monitor the health of turbines.
They estimate that in the European Union there are 1,800 industrial
gas turbines, 850 of them smaller than 10 MW. World-wide there are
believed to be 12,000 industrial turbines, including 4,500 smaller
than 10 MW.
A typical severe fault that requires modification of an engine before
a later scheduled service could easily cost 40,000 ECUs say the
partners. A correct diagnosis of a major fault could avoid a cost
of about 100,000, ECUs or potentially much more if the cost of lost
production is taken into consideration.
Taking engine soundings
One of the important features of GASTEM has been the ability to
use different forms of data collected from a turbine. The innovative
aspect of this development was to combine an acoustic approach with
thermodynamic techniques. Data fusion of the readings from different
measurement techniques is difficult to achieve, but allows faults
to be pinpointed through, in effect, a mutual learning process.
GASTEM cost 1.5 million ECUs and a follow-on VA-GASTEM project,
under the Value programme, to develop a practical version of the
monitoring system, cost 266,000 ECUs. The European Commission contributed
930,000 ECUs of the GASTEM cost, and 130,000 ECUs to the VA-GASTEM
project. GASTEM followed an earlier three-year project by the partners
to explore advanced experimental techniques, particularly a holographic
technique known as acoustic phased imaging. This involves using
an array of microphones to pick up the sounds from an engine, with
the array processed to achieve a significant acoustic directivity,
somewhat comparable to an 'acoustic binocular'. This also allows
the array to be some distance from the hot and dirty atmosphere
around the turbine, therefore reducing the background noise and
producing a high signal-to-noise ratio for easier analysis.
Acoustic signals are collected using a low-frequency array of microphones
for sound in the 100 Hz-3.5 kHz range and a high- frequency array
for signals in the 3.5 kHz-17 kHz range. This improves directivity,
as the acoustic focus is different at different frequencies. Microphones
would not need to be placed more accurately than a few tenths of
a centimetre away.
GASTEM also uses pressure sensors, thermocouples and other sensors
to produce more conventional monitoring signals. The computer software
assesses the signals, with different types of signal being evaluated
together to provide confirmation of the conditions inside the turbine.
Data reduction methods
The system must carry out signal processing to reduce the huge
flow of acoustic and accelerometer data, which is typically 2 M
bits/second, to produce indicators of the mechanical and thermodynamic
conditions within the turbine. This processing involves fast Fourier
transform analysis of the spectra by techniques such as filtering
and detection of signal peaks, and the formation of 'spectra of
spectra'. In the latter case, acoustic tones are selected from the
different spectra of tones to form the so-called 'cepstrum' of the
signal. The software also reduces the aerothermodynamic and general
operating signals to a small number of indices that define the condition
of the gas turbine. The result is a set of 40 parameters that are
integrated and subjected to an expert system analysis to diagnose
faults. This has required the partners to develop a library of faults
against which signals from a turbine under investigation can be
The expert system is known as CLIPS and is derived from computer
language originating from the NASA space agency. Military missile
designers have also integrated different types of monitoring signals
to provide rapid alarms of faults that might require destruction
of the missiles. The CLIPS software identifies the faulty component
on a turbine, such as the compressor, and the faulty part, such
as a burner. The results are displayed on a computer screen, with
fault indication and alarms, depending on the level of fault.
Portable monitoring unit
VA-GASTEM turned the project results into a practical demonstrator
unit. It involved work on a Tornado gas turbine from EGT that generates
power at the beer and lemonade factory of the Bavaria BV company
in Holland. The aims were to reduce the cost of the system, make
it more user friendly and more suitable for industrial conditions.
The resulting system costs about 25,000 ECUs, compared with the
200,000-ECUs cost of the GASTEM version. The system now weighs about
75 kg, rather than the original 300 kg, making it practical for
use in the field. The information displayed on the personal computer
screen of the system has also been reduced to simplify use. Particularly
important is the ability of the system to deliver results within
10 seconds, thus providing an almost real-time health survey. This
speed is needed to detect vibration of a turbine blade, for example.
Also the system can learn the parameters of the engine that it might
be used for in less than 15 minutes, compared with the one hour
needed on the GASTEM version.
The health monitoring system is now ready for the commercial market.
An estimated 40,000 ECUs would be needed to extend its use on turbines
other than the Tornado, and a further 20,000 ECUs for the equipment
needed, such as acoustic arrays. The partners say that the potential
savings from better maintenance could far outweigh the costs incurred
to bring it to market. GASTEM could be a stand-alone system for
use on different engines, or integrated with the turbine's management
control system. The expected development of computer technology
offers the potential for GASTEM's health monitoring software to
be available as a marginal extension of the control system computing
power, says one of the partners.