Documentos de Académico
Documentos de Profesional
Documentos de Cultura
GT2016
June 13 17, 2016, Seoul, South Korea
GT2016-56166
EXPERIMENTAL SET-UP
Two optical access port-holes allow the flame tip and of
The MethaNull test rig the flames rear to be observed i.e. to check whether there is a
The test rig is a compact atmospheric combustion facility diffusion flame or not.
that includes a plenum, a staged burner, a combustor casing and Although the combustor is atmospheric, a robust
an exhaust system. The burner itself is a premixed, lowly- construction was chosen for safety reasons. Theoretically, the
swirled burner placed in a strongly-swirled secondary flow. In rig is designed to operate at up to 6 bar.
accordance with publication [2], the burner shall hereinafter be Some dimensions are only provided to give a sense of
referred to as the pilot burner and the secondary flow as scale. The liner (glass tube) dimensions are 100 mm diameter
main air. Technical details on the burner and on the and 410 mm length. The combustor casing has a 160 mm
MethaNull technology are outlined in the same reference; here diameter (same for the plenum). The length from the front plate
we consider configuration II with a concentric stage burner, to the exhaust is 1240 mm.
where the pilot burner is embedded in the main stage. This pilot
burner is connected to a siren that modulates the air flow. The
complete test rig can be seen in figure 1. Operating points
A flow control panel regulates the air and gas head The operating points described in this publication are
pressure levels. The air supply system is connected to a summarised in table 1. The first is the standard point used for
pressurised air network (8 bar). A commercial propane bottle steady versus pulsed conditions. The second investigated type
(type UN 1965 mixture C, 46.3 MJ/kg, 2 kg/m3) provides the of operation is the lean blow out limit (LBO) for detection of
gas supply. The ignition system is embedded. Redundant the extinction, where the fuel content is reduced by half. The
analogue and digital displays report on the air and gas flow third and last type of operation, a reproducible flashback is
conditions. achieved under these air flow conditions by injecting the same
The plenum air supply is split into cooling air for the quantity of fuel in the main stage as in the pilot. The ability to
whole system and combustion air for the main air supply. The deliberately produce controlled flashback conditions is in fact a
combustion chamber casing is protected from the flame by a key feature of this test rig and is instrumental in this particular
heat-resistant glass tube that is fixed on the head plate, where study. The burner is designed for experimental purposes only
the flame settles. The cooling air flows between the casing and it is not commercial. To make this burner flashback-free, the
the glass tube until it mixes with the burnt gases in the main air must be augmented by a third. All points involve the
convergent section called the mixer. A further chimney section pilot burner only, with the exception of the last one (flashback)
is added to condition the plume flow and to measure exhaust where the main gas is turned on. The ratio of cooling air to
gas contents near the outlet. burning air is 2:1.
The test rig was equipped for this purpose with ten probe
ports with an inner G1' thread placed at different locations
along the plenum and the combustor casing. Special adapters
were designed so that changing the sensor or permutation of
accelerometer with a pressure sensor was easy. This is FIGURE 3: SENSOR LOCATIONS
illustrated in figure 3.
REPLAY
Piezosensor 3 time, as well as for the detailed post-analysis following the
Signal conditionning
Piezosensor 4 Live tests. The software provides the possibilities of generating two
analysis VibroSight
databases on demand to document the test or saving the entire
and
Piezosensor 5 measurement.
recording
Piezosensor 6
The first database is a custom history database. It records
Post-process options
pre-configured extractions (such as true peak-to-peak or RMS)
as well as reference signals (such as the TTL siren reference) at
at wish
Accelerometer 1 one second intervals. Time-wise, it also logs pieces of signal
Accelerometer 2
and/or processed bits of signal such as power spectra. The
database is designed to travel quickly through time, identify
Custom history database High resolution (26kHz) and situate the parts of signal to be processed further in detail.
Ext. Microphone 1 for analysis using database using *.TDMS
VibroSight format
The rate at which the individual data elements are stored can be
Ext. Microphone 2 adapted. Logging rates and data density can be customised to a
great extent in order to achieve the appropriate logging
footprint throughout a test sequence. The logging rate can be
FIGURE 4: MEGGITTS MEASUREMENT CHAIN changed manually or can be automated e.g. alarm-based on
the RMS level of one sensor. If it reaches a certain level, then
the rate of acquisition is automatically augmented. For instance,
during warm-up, one dataset of 0.2 s time waveform (or time
signal) and the corresponding spectrum are saved relatively
slowly at a rate of 10 s. During the core part of the experiment,
In our case, as shown in figure 3 for the list of sensors and the rate is increased to saving every second, providing more
figure 4 for the data acquisition, a total of 11 dynamic channels details but also associated file sizes.
(two accelerometers, six pressure transducers, two external The second means of storage is the continuous and
microphones, one synchronisation TTL signal coming from the seamless time waveform of each dynamic channel at full
siren) and one synchronisation channel (the TTL is duplicated sampling frequency, in this case 25.6 kHz. It is conditioned
for order tracking purposes) are connected to the card. The under the form of single datasets saved under *.TDMS format
sampling frequency of the XMV16 was set to 25.6 kHz for all (a standard developed by National Instruments). With our
11 channels. settings, one file per channel was recorded every minute. Put
Since the domain of interest for this particular burner is the together, the dataset is reconstituted seamlessly for post-
range 0-1 kHz, it makes sense to reorganise one signal chunk processing purposes.
into n sub-sampled signal chunks. The reason for this is that it Two options for post-processing are then possible. The
is thereby possible to obtain a representative spectrum on the *.TDMS dataset is converted back into analog form and fed
frequency domain of interest derived from averaging n spectra, into into the VM600 again. After that, VibroSight performs the
faster than if we were considering the whole accessible analysis based on its configuration in the same way as the
spectrum. As a result, the average frequency peak gain and the original run. After the digitally stored signal is converted into
most represented phase angle related to this peak are an analog signal and fed into the card in replay mode, the
determined with greater precision. With a 25.6 kHz sampling repeatability of the converted signal matches the results from
frequency, the cut-off frequency is 12.8 kHz. If we want the the original acquisition and processing with a fidelity of 1%,
range 0-1 kHz to be relatively alias-free, then n=5 appears to be provided the same configuration is used during the initial and
a correct sub-sampling factor (the cut-off frequency is then the replayed run. This small remaining difference is mainly due
divided by 5, hence 2.56 kHz). The reorganisation follows: the to the stochastic nature of the signals and asynchronous
first sub-sampled chunk, starting with the first value of the sampling and processing cycle. This is the replay option. The
original signal and then selecting every fifth value from then second option is the direct processing of the *.TDMS data in
on. The second chunk starts with the second value of the digital form using standard processing software such as Matlab.
original signal and then selects every fifth from then on and We carried out both options within the framework of this study.
so on. This process highlights the non-stochastic nature of the The practical aspect is the possibility of processing data
data. This interleaved sub-sampling is possible in post- from the same physical test sequences multiple times, with
processing using both methods detailed below. different focuses and multiple processing configurations. This
Figure 4 shows how the signal is processed and stored. The capacity allows experimental test time to be optimised and is
acquisition card communicates with a PC via an Ethernet cable. applicable to real engines as well as test rigs.
Frequency ramps
S Steady P Pulsed
13:50
Flame
13:45
13:40
13:35
Time
(No flame)
13:30
13:25
STEADY
Flame
13:20
13:15
13:10
(No flame)
13:05
200 400 600 800 1000 1200
Frequency (Hz)
FIGURE 6: POWER SPECTRUM ON THE LINE CP03 RECORDED DURING TWO CONSECUTIVE WARM UP
SEQUENCES, STEADY FIRST AND THEN PULSED.
Time
Time
Figure 6 shows how the power spectrum varies with time Frequency (Hz) Frequency (Hz)
as a result of the machines warm-up process. The two last
sequences of warm-up displayed in figure 5 are observable
between 13:10 and 13:56. A first sequence of twenty minutes
under steady state conditions is followed by an extinction and -5 0 5 -2 0 2
cool-down, followed by a second sequence under pulsed FIGURE 7: SCANNING FREQUENCY RAMPS FOR
conditions (365 Hz, the first harmonic at 730 Hz is clearly ORDER-TRACKING OF THE FLAMES REPONSE,
observable). The higher the acoustic power the darker the plot. AMPLITUDE (LEFT) AND PHASE (RIGHT)
The repeated J pattern at about 500 Hz is noteworthy on the
plot; it drifts for about 10 minutes on a frequency span of
15 Hz. This is partly related to the warm-up of the air and
therefore to the higher speed of sound. The external The siren draws a pattern called a ramp when modulating
microphones situated at ambient temperature (1.3 m away from its rotational speed. The excitation frequency grows steadily at
the facilitys exhaust) repeat this pattern (not shown), which 10 Hz/s from 0 to 800 Hz and back. Three consecutive ramps
excludes an artefact on the piezoelectric dynamic pressure are shown in figure 7.
sensors due to their warm-up. It is possible to record the amplitude and phase of a given
This observation highlights why precision in frequency and sensor along the trace of the ramp. The amplitude and phase are
in amplitude is required for combustion monitoring and control. coloured with an out-of-plane component on the plot. It is
The acoustic and structural behaviour adapts to changes of remarkable that the values repeat themselves at a given
operating points and possibly to climatic changes (e.g. air frequency along each branch of the ramps.
temperature at the intake and humidity).
If a pressure transducer is incorrectly mounted, it can The VM600 piloted by VibroSight is designed to perform
report signal artefacts due to mechanical coupling of housing order tracking. Amplitude and phase (= angle difference or
vibration onto the sensing element. If those were to happen they phase shift between the measured signal and the reference
can be proven to be strongly correlated to the reading of the signal, here the TTL of the siren with one upwards front per
accelerometer (mechanical vibration). The systematic single pulsation) along the siren excitation are extracted, similar
comparison of plots similar to figure 6 allows this problem to to what is usually used when performing order tracking
be detected and rectified quickly. vibration analysis on a rotating shaft. This is shown in figure 8.
The peak at 365 Hz is detected, despite the constant fluctuation
in frequency of the siren. This plot was produced in post-
Flame response to frequency scans using the siren processing (replay mode), but can also be generated on-the-fly.
The following method is used to detect at which The resonant frequency with related phase angle is also seen as
frequencies of excitation thermoacoustic couplings are met. a cluster of points at a given polar angle on the Nyquist plot.
This is part of MethaNull technology. It is also a method Note the constant saw tooth pattern of the phase clearly
recommended for combustor characterisation regarding observable in the low frequency domain, this is related to the
combustion stability. variation of the sirens frequency, resulting in a phase slide.
Time range
modulated
15:23
at fixed
15:23 frequency
15:23
15:22
15:22
15:22
15:22
300 350 400 450
Frequency range [Hz]
10 20 30 40 50 60 70
Log (P) [a.u.]
0
practice it is often not known how great the influence actually
-5
is. Does the neighbouring sensor see the same dynamics or not,
14:21 14:22 14:23 14:24 14:25
and if not, how much do they deviate due to the different
4* LBO EXTINCTION 4*FLASHBACK
5
x 10
4
location and orientation?
One result of this study is to directly compare the readings
between sensors in the same experiments and thereby quantify
CP05
how well they match each other for different signal content, at
-5
14:21 14:22 14:23 14:24 14:25 least in terms of magnitude. This could be used to conduct
4
plausibility checks for readings and comparisons with real
x 10
4 turbines as well.
2 Processing parameters: any analysis of combustion
CP06