Professional Documents
Culture Documents
ABSTRACT
cold and hot duct lengths
Measurements of vibration and combustion chamber dynamic mode number, m = 1,2 ...
pressures have been taken on a number of 150MW industrial gas Mn Mach number
turbines operating on pm-mixed natural gas, both during long periods M(r) coefficient matrix
of base-load operation and during short duration load-swings. The P pressure; mean pressure
data has been analysed in terms of the frequency and bandwidth of the AP mean pressure drop
principle peak in the vibration and pressure spectra as a function of heat release rate per unit cross sectional area
load and other operating parameters. It is observed that bandwidth, variation with time is exp(rt), r e s + jcu
which is a measure of the damping of the resonant mode of the Re(r)
combustion chamber's acoustic resonance, decreases towards zero as time
the machines approach their combustion stability limits. At time delay
temperature
A theoretical model of the thermoacoustic behaviour of the AT Tit T,
combustion system has been developed to see to what extent the u, U velocity; mean velocity
observed behaviour on the operational machines can be explained in flow direction
terms of an acoustic model of the ductwork and a flame characterised vibration or pressure signal, Pa or minis
simply by a time-delay. This time delay is obtained from the ratio of specific heats
frequency response function of the flame in response to unsteady acoustic impedance, p'/pcu'
perturbations in inlet velocity and is calculated using computational
damping ratio,-Re(r)Ilm(r)
fluid dynamics.
pressure drop coefficient, APPApU 2
gas density
The model has also been used to illustrate the importance of fuel
supply system design in controlling combustion stability. It is shown non-dimensional time delay, c cAt/L,
that stability can be a strong function of the acoustic impedance of the 9( 0) frequency response function between heat release and
fuel supply and that this can lead to enhanced or reduced stability acoustic pressure, cp(co) a Q7c,p21
depending on the flame characteristics. 4.1(0) flame frequency response function (FFRF) relating
heat release and acoustic velocity, w(ce).(AT/T c)/(y-1)
NOMENCLATURE QINCcUi •
angular frequency, lm(r)
cold and hot duct areas
area of cooling holes Subscripts
amplitudes of lines in FFT 1,2,3,4 compressor discharge, burner inlet, burner outlet,
specific heats of air combustion chamber exit locations
velocity of sound air, fuel
fuel calorific value c,h, cool cold duct, hot duct, cooling holes
swirl vane exit conditions
Presented at the International Gas Turbine & Aeroengirte Congress & ExtuNion
Indianapolis, Indiana — June 7—June 10, 1999
This paper has been accepted for publication in the Transactions of the ASME
Discussion of d talil be accepted at ASME Headquarters urdi September 30, 1999
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 06/19/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
I. INTRODUCTION that has been done (e.g. Bloxsidge et al (1988), Lieuwen and Zinn
(1998), Bohn and Deuker (1993)) and its use is explored in the present
Many large industrial gas turbines operating on pre-mixed natural work. The total delay time is made up of a number of contributions
gas suffer from high levels of dynamic pressure or "humming" in the from the chemical kinetics and the fluid dynamic convection and
combustion chambers which can sometimes prevent them from mixing, but in highly turbulent gas turbine combustors, it is the fluid
reaching full load and can lead to serious damage to internal dynamic convection time between fuel inlet and flame front which
components. The high pressure levels are due to acoustic resonances tends to dominate.
inside the combustion chamber cavities which are coupled with
perturbations in heat release at the flame. To help understand the 3. THERMOACOUSTIC MODEL
phenomenon and in particular to identify the geometric and operating
parameters on which the dynamic pressures depend, simple themio- In the Siemens V94.2, combustion takes place within two large
acoustic models have been developed for the combustion chamber silos, symmetrically placed either side of the machine centreline (see
components including the fuel supply. In support of this, a CFD Fig. 1 taken from Becker et al, 1986).
model of the swirling flow from a burner has been used to quantify the
flame parameters affecting overall combustion stability, in particular At the compressor discharge, the air divides and the flow in each
the time lags between pre-mixed fuel injection and heat release in the silo passes upwards along an annulus surrounding the flame tube. It
flame. The model has been used to establish broad generic turns and passes through eight burners fitted with swirl vanes and into
characteristics of combustion instabilities. In addition, it has been the combustion chamber which is lined with ceramic tiles, and then
applied to a Siemens V94.2 gas turbine and used to quantify two downwards to the turbine entry. In premix mode, fuel is supplied
characteristics of the humming signals, namely frequency and damping through a series of distribution pipes immediately upstream of the
and their variation with load. The results have then been compared swirl vanes. A pilot diffusion flame immediately downstream of the
with data obtained on three operational gas turbines. vanes is used to provide stability over a wide range of excess air ratios
(Becker et al, 1986). Pressure losses between compressor and turbine
2. MODELLING OF COMBUSTION INSTABILITY are very small and confined mainly to the swirl vanes. Small cooling
air holes allow air to flow from the cold duct to the hot duct; these
A number of approaches have been taken to the modelling of have not been incorporated into the full model but an estimate of their
thermoacoustic instability. Richards et al (1993) used a lumped influence has been made.
parameter approach incorporating a bimolecular reaction rate law to
produce a non-linear model of a pulsed combustor. This model was 3.1 Linearised acoustic equations
later adapted to look at lean pre-mixed combustion, Janus et al (1997).
Smith and Leonard (1997) applied an axisymrnetric, transient CFD A simple thermoacoustic representation (Fig. 2) of the actual
analysis of the full Navier Stokes equations to the same pre-mixed combustion system has been used in this work. It represents the
system. Transient CFD analyses of a suddenly perturbed flame have simplest possible model for studying the V94.2 as well as for
also been used to predict a so-called flame frequency response examining more generic aspects of combustion instabilities. To do
function (FFFtF) for both a diffusion flame system, Bohn et al (1996), this, a number of assumptions have been made as follows:
and a pm-mixed system, Bohn et al (1997). A simpler quasi-one-
dimensional CFD analysis was used by Liu and McGuirk (1995) to small perturbation analysis - only conditions prior to instability are
predict the unsteady behaviour of an after-burner model. considered; dynamic pressures are much less than 100 mbar and
acoustic velocities less than 2m/s. A linear, small perturbation
However, the most frequent approach to the problem has been analysis is therefore appropriate with, for example, pressure varying as
through the use of the one-dimensional acoustic wave equations based
on a linearised form of the conservation equations for mass, P(x,t) = Pc + P i(x) exP(r0 (I)
momentum and energy. A good description of this approach has been
given by Dowling (1995) and a model based on this approach has been The variable r is complex with Im(r) - - - 0), the frequency of oscillation,
adopted for the current work. If the acoustic properties of the and Re(r )=s, the factor determining growth or decay of the oscillations.
upstream and downstream parts of the combustion system are known, Generally, both the modelling and experiments show that s<4 for all
then the acoustic equations may be solved by any standard technique cases of interest.
to give the fluctuating pressures and velocities either side of the flame
front. In the present case, a transfer matrix technique is used, in the boundary conditions - it has been observed that at the main humming
same manner as described by Chadha et al (1980) for a valve-pipeworIc frequency, 75 to 78Hz, pressures in the two silos are in anti-phase,
system. The acoustic pressures are assumed continuous across the implying that acoustic boundary conditions of zero unsteady pressure
flame front and the acoustic velocities can be related by the jump in Pl=p.m) can be set on the machine centre-plane where the inlet and
volumetric flux associated with the heat release. Specification of an outlet flows from the silos come together. As a consequence of this,
FFRF relating the fluctuating heat release to the fluctuating velocity the acoustic impedances presented to the flow by the turbine inlet and
closes the set of equations and allows a complete solution to be compressor discharge are actually unimportant and can be ignored.
obtained. Note though that these impedances will be important in controlling
other modes where p i and I); are not zero, though their very high
The FFRF would normally have to be found from experiment, values will tend to suppress acoustic velocity fluctuations into and out
either a physical experiment or a numerical experiment such as that by of the compressor and turbine.
Bohn et al referred to above. However, many of the experimental
results obtained (e.g. Buchner et al (1993), Keller et al (1989), compressor discharge ductwork - this annulus is represented by a duct
Bloxsidge et al (1988) and Bohn et al (1996)) have shown that the of equivalent length and (constant) area to the actual passages between
form of the function can be characterised (at least approximately) by the compressor discharge and the burner inlets. The flow of air in the
simply assuming a time delay or phase lag between the heat release duct is frictionless and at low Mach number and the acoustic
and velocity fluctuations. This time delay function has been used in behaviour is assumed to be one-dimensional, though in general
much, probably most, of the modelling of thermoacoustic instability
2
In this expression 1„ c and p refer to the length of the inlet duct, the This represents an eigenvalue p oblem in the unknown complex
velocity of sound and density of the inlet air variable r, the solutions for which can be determined by setting the
determinant of matrix M(r) to zero and solving for r. This has been
combustion chamber - similarly, the hot zone is represented by a duct accomplished using the symbolic mathematics program Mathcad for a
of length and area equivalent to the passages up to the turbine inlet wide range of temperatures and mode numbers up to m=7.
guide vanes. A constant temperature is assumed throughout and the
combustion products given the same properties as air. At this stage, Examination of the imaginary part of r, lm(r) = co, then determines
cooling flows are ignored. the frequency of any thermoacoustic oscillations, stable or otherwise,
whereas the ratio of the real part to the imaginary part gives a value for
.4
swirl vanes - all pressure losses (1.2 to 1.5% of absolute pressure) are the damping ratio ; = -Re(r)/Im(r). Positive values of ; infer stability
assumed to be confined to the vanes (there is no pressure loss across and vice versa. It is these parameters which will be compared with
the flame). The steady pressure drop is assumed to be equal to the experimental values.
dynamic head 'Ap eU,2 of the air leaving the vanes and the unsteady
pressure loss to be quasi-steady, i.e. ep . = p2'-p2' Finally, for any particular solution for r and an arbitrary value for
u 1', it is possible to determine values of pi, u2' etc. and intermediate
fuel supply - the fuel is assumed to be fully pre-mixed and with values of pressure and velocity, to plot the mode shape associated with
constant stoichiometry at all times. any solution to the equations.
flame - for the fully pre-mixed fuel supply the unsteady heat release 4. GENERIC RESULTS OF THERMOACOUSTIC MODEL
per unit inlet area Q' is directly proportional to the perturbation in WITH FULLY PRE-MIXED FLAME
unsteady velocity u2' (see e.g. Dowling, 1995):
Prior to application of the model to the actual gas turbine
Q' ° cp(Th - Tc) Ma' (3 ) geometry, the model was run for some simple ductwork configurations
in order to establish generic characteristics which would apply to a
Dowling also shows that the heat release results in an increase in wide range of situations. In particular, attempts were made to
volumetric flow rate given by establish how frequency and damping of the acoustic modes were
affected by operational and geometric parameters. Now, non-
(4) dimensional analysis leads to relationships of the general form
Ahu'i - Acie2 = Cr-D4 Pcct2).Q.
Also, it is assumed that the distance from the swirl vanes to the flame = func l (Tb/Ta t, geometry) (8)
is small compared to typical acoustic wavelengths and that there is no
pressure loss across the flame. ; = func2(Tb/To tMn, tat t, geometry) (9)
flame frequency response function - Equation 3 is a quasi-steady An alternative expression for a given geometry, when the time delay
approximation to the flame characteristics and is valid only at low scales inversely with swirl vane exit velocity and when us is constant
frequencies, At higher frequencies, it must be modified by a frequency or depends on toAt, is simply ; ftmc3(T2/T c, Mn), i.e. damping
response function w(co) so that depends only on Mach number and temperature ratio.
Q' ° w(co) • cp(Th - Te) Pelt' (5) The purpose of the generic work was to establish the form of the
functional relationships func 1.3 for simple cases when the operating
ago)) tends to unity at low frequencies and can often be approximated parameters were broadly similar to those expected in a typical gas
by a simple time-delay function at higher values: turbine. The results, some of which are presented below, give a
valuable insight into the trends and order of magnitude of effects to be
ty(to)= exp(-jcoAt) (6) expected.
The frequency response is the only potential source of instability so it 4.1 Localised or distributed pressure drops
is the time delay At or phase lag coat between the unsteady velocity u 2'
Localised or distributed pressure losses in the ductwork do not
and the heat release Q' which is the most important ingredient in
determining stability or otherwise. This will be considered in more generally affect the frequency but will contribute to acoustic damping.
detail later. For example, for the case of isothermal flow at (low) Mach number
Mn down a duct with a distributed steady flow pressure drop
3.2 Solutions to equations coefficient given by the expression by t = APPApU 2, the damping of
the mth acoustic mode is given by the expression
The boundary conditions, the acoustic transfer matrices and flame
equations can be gathered together into a 9 by 9 matrix M(r) relating ç= (I/mn) . ;Mn (10)
the unsteady flows, pressures and heat release Cf.
Similar expressions can be derived for cases with concentrated
r(0))2 = B(0)0)/{ (1 - (()/( 00h2 (24)/(00)2 (15) A similar analysis was carried out on gas turbines 0101 and
0T02, making use of either fitted dynamic pressure transducers or the
to the experimental data close to the resonant peak. Here B represents compressor bearing casing vibration signals. During these periods of
a locally constant amplitude (Pa or mm/s). The overall energy y 2na, operation, lasting four or six weeks, there were few significant load
associated with the resonance was also calculated by first converting swings, though the load varied in sympathy with inlet temperature.
the y(co) values into a power spectral density and integrating over a The data was characterised by considerable scatter of up to ±1Fiz in
suitable frequency range. frequency and ±0.005 in damping, due probably to changes in other
independent parameters during the course of the tests. Averaged data
7.3 Preliminary measurements of dynamic pressures has been calculated therefore and this is shown in Figures 9 and 10.
Some preliminary tests were carried out on GTO1 over a period On GT01, frequencies derived from dynamic pressure signals,
when the machine was operating mainly at base load (though some were close to those of 0T03. Like 0103, damping showed a clear
load swings took place), often in cold weather, with the objective of trend downwards but the level was much higher and consistant with a
establishing typical levels of dynamic pressure in the hot zone stability limit about 20MW higher than GT03. 0T02's frequencies,
immediately above the flame, and correlations with load and inlet derived from vibration measurements, were quite different however
temperature. Signals were band-passed filtered and the rms levels and trends in damping not so clear. It is thought that the reason for
within the band 60Hz to 90Hz obtained. The results, presented as long this is the distortion of the acoustic resonance by the presence of a
term averages versus load, are shown in Fig. 5 at 1 MW intervals. It local mechanical resonance with a similar, fixed frequency and
can be seen that there is a gradual rise in dynamic pressure from low difficulties with the curve fitting algorithms.
load up to about 150MW, followed by a rapid rise towards 10mbaz at
the maximum load obtainable during this period, 169MW. Attempts 8. COMPARISON BETWEEN MEASUREMENTS AND THE
to find a second independent parameter (e.g. inlet temperature) against THERMO-ACOUSTIC MODEL
which to correlate were unsuccessful due to the scatter in the data.
However, it was noted that the behaviour of the data was similar to During a load up of a V94.2, at loads above about 110MW the
that expected from a resonant system excited by a roughly constant compressor inlet guide vanes are continuously opened and fuel flow
level of broad band noise but with a damping which declined in an adjusted to maintain a constant turbine exhaust temperature. The non-
approximately linear fashion towards zero at a load of about 173MW dimensional parameters governing stability i.e. temperature ratio,
(see fitted curve). This was an indication that variable damping was excess air ratio and compressor discharge Mach number, remain
playing a role in the problem. almost constant at values of about 2.4, 2.2 and 0.06 respectively. The
increase in load arises from an increase in compressor mass flow and
7.4 Measurements of frequency and damping during a short load discharge pressure. To first order, it would be expected from Eq. 8
swing on GTO3 that frequency would rise in proportion to ; but that damping would
remain fixed. This is not the case in practice and second order effects
In the absence of dynamic pressure, measurements of compressor take place, necessitating a more careful examination of how the
casing vibration were taken on this machine during various phases of a operating parameters vary during a load up. This was achieved by a
load swing (see Fig. 6): on load; load drop; low load; load up; combination of overall gas turbine performance data measurement and
hununing (during load up); load trim. Humming was thought to be a thermodynamic calculations, primarily of flame temperature.
state where the gas turbine operated beyond the combustion stability
limit when amplitudes were governed by non-linear effects. The The thermo-acoustic model was run with these parameters as input
machine became stable when the load was trimmed by slightly and with the time delay estimates from the CFD modelling and the
reducing the gas turbine exhaust temperature. The data points plotted results compared with the experimental measurements of frequency
are times at which the vibration data was analysed. and damping. In Figures 9 and 10 the experimental results for all three
turbines are compared with the model predictions for the 4th acoustic
Figure 7 shows how the frequency of vibration varied during the mode, this mode having frequencies closest to the experimental ones
load swing. A trend upwards as load increased and humming and with trends in damping closest to those observed. (Note that the
commences can be observed. It is clear also that the on-load and load- kink in the theoretical curve correspond to a change in the way in
trim data points are not part of the same family as the other data, with which the machine is controlled.). Allowing for the variability in the
lower values of frequency (about 76.5Hz) compared to 77.5 Hz at a experimental results, it can be seen that the model predictions agree
load of 148MW with humming. Since part of the reason for higher reasonably well with measurements, in particular both the increase in
frequencies is higher hot gas temperatures, it is therefore possible that frequency and decrease in damping with load are reproduced. To
these temperatures were slightly but abnormally high on 0103 and this obtain this degree of agreement some tuning of the model was needed
is likely to be a factor in inducing humming. however. The effective acoustic lengths of the upstream and
very accurately, (the real system not being just two simple ducts), and an acoustically coupled combustion instability. Part 2 Theory", /FM,
the time delay from the CFD modelling is only an approximate value. Vol. 193, pp 445-473.
Since the predicted results are quite sensitive to the ratio of acoustic
lengths and these will always need some refinement, the simple Bohn, D & Deuker, E, 1993, "An acoustic model to predict
acoustic model used here should be seen more as a tool for combustion driven oscillations", 20" Int. Congress on Combustion
understanding the plant behaviour rather than as a truly predictive tool, Engines, London.
at least for this model of gas turbine.
Bohn, D, Deutsch, G & Kruger, U, 1996, "Numerical prediction of the
A sensitivity analysis showed that even small changes in the dynamic behaviour of pre-mixed flames", ASME paper 96-GT-I33.
calculated values of volumetric flow rate, Mach number, temperature
ratio and time delay had considerable influences on the predictions of Bohn, D, Yuhong, L, Matouschelc, G & Kruger, U, 1997, "Numerical
both the frequency and damping. The results therefore are also prediction of the dynamic behaviour of pre-mixed flames using
sensitive to the thermodynamic model and in particular the method of systematically reduced multi-step reaction mechanisms", ASME paper
calculating the effective flame temperature from the gas turbine 97-GT-265.
exhaust temperatures. It is thought that these effects and measurement
uncertainties in the exhaust temperatures were responsible for the Buchner, H, Hirsch, C, Leuckel, W, 1993, "Experimental
premature humming of GT03. investigations on the dynamics of pulsated pre-mixed axial jet flames",
Comb Sci & Tech, Vol. 94, pp 219-228.
9. CONCLUSIONS
Coffee, T P, 1985, "On simplified reaction mechanisms for the
A theoretical model has been developed which has been used to oxidation of hydrocarbon fuels in flames", Comb. Sci. & Tech. Vol.
derive some generic characteristics of combustion instability in terms 43, pp. 333-339.
of damping ratio. It is based on an acoustic model for the combustion
chamber ducts and a frequency response function model for the flame. Chadha, I A, Hobson, D E, Marshall, A & Wilkinson, D H, 1980,
"Acoustic source properties of governor valves", in Flow induced
A CFD model of flow and combustion in a Siemens V94.2 burner vibrations of power plant components, ed. M K Au-Yang, ASME
has been used to calculate a convective time delay which can be used publ. PVP-41.
to derive an approximation to the frequency response function.
Dowling, A P,.1995, "The calculation of thermoacoustic oscillations",
Some generic aspects of the effects of fuel supply impedance have Jill. Sound & Vib., Vol. 180, pt. 4, pp 557-581.
been explored. It has been shown that a fuel supply resonating at a
frequency close to the overall system natural frequency can have a Janus, M C, Richards, G A, Yip, M J & Robey, E H, 1997, "Effects of
considerable influence on stability but time delays in the flame must be ambient conditions and fuel composition on combustion stability",
included when calculating these effects. ASME paper 97-GT-266.
Quantitative measures of the stability of combustion on three Keller, JO, Brainlette, TT, Dec, J E & Westbrook, C K, 1989, "Pulse
operating gas turbines have been obtained by measurements of the combustion: The importance of characteristic times", Combustion &
frequency and bandwidth of peaks in the vibration or dynamic pressure Flame, Vol. 75, pp 33-44.
spectra. Values for damping ratio have been derived from these
measurements. As load increased, frequency increased and damping Lieuwen, T & Zinn, B T, 1998, "Theoretical investigation of
decreased steadily, the latter being the main reason for the rise in the combustion instability mechanisms in lean pre-mixed gas turbines",
signal amplitudes. Damping reached zero on one machine and a true AIAA paper AIAA-98-0641.
instability occurred.
Mason, H B & Spalding, D B, 1973, "Prediction of reaction-rates in
The theory has been applied to the operating gas turbines using turbulent pre-mixed boundary layer flows", Combustion Institute
measured or calculated values for the performance characteristics and European Symposium, Ed. F I Weinberg.
the results of the CFD calculations as input. The results show
reasonable agreement but also a very large sensitivity to the calculated Rayleigh, I W S, 1945, "The theory of sound": Vol.'', pp 224-227,
values of volumetric flow rate and flame temperature as well as some Dover Publications.
of the geometric parameters. These sensitivities, combined with
uncertainties in the calculation of some of the performance parameters Richards, G A, Morris, 03, Shaw, D W, Keeley, S A & Welter, M J,
on operating machines, may be one of the reasons why nominally 1993, "Thermal Pulse Combustion", Combus. Sci. & Tech., Vol. 94,
identical machines have different stability limits. They also limit the pp 57-85.
fully predictive capabilities of the model.
Smith, C E & Leonard A D, 1997, "CFD modeling of combustion
REFERENCES instability in pre-mixed axisymmetric combustors", ASME paper 97-
GT-305.
Bakke, 1 R & Hjertager, B H, 1987, "The effect of explosion venting
in empty vessels", Inc la for Num. Methods in Eng., Vol. 24, pp Westbrook, C K & Dryer, FL, 1981, "Simplified reaction mechanisms
129-140. for the oxidation of hydrocarbon fuels in flames", Comb. Sci. & Tech.,
Vol. 27, pp. 31-43.
Becker, B, Berenbrink, B P & Brandner, H, 1986, "Prernixing gas and
t tit 1 1%raj
4‘z■frifilitt Si
Manhole
Figure 1
Siemens V94.2 silo combustion chamber
3 _
2
!
El.
1
1 1
1
Ii
I
-2
3
-5 -4 -3 -2 -I 0 1 2 3
Re(1/Z)
Figure 3
Effective fuel supply compliance with and without flame
10
"
-0 1
tu
• col
—
It
0 0.8
C
•_
0
0.4
s j 0.2
i
-1 0 I 2 3
axial distance/burner diameter
Figure 4
Distribution of heat release in flame
11
0
40 60 80 100 120 140 160 180
Load, MW
Figure 5
Long term average dynamic pressures on GTO1
12
150
140
eon-load
130 _ c load drop
s low load
• load-up
120
• humming
a load trim
110
100
35 4 4.5 5 5.5 6 6.5 7
Time, hours
Figure 6
Gas turbine GTO3 load swing profile
13
77
eon-load a load up
oload drop •humming
74 slow load A load trim
73
110 120 130 140 150 160
Load, MW
Figure 7
Variation of frequency with load during GTO3 load swing
14
0.005
Figure 8
Variation of damping with load during load swing on GTO3
15
• GTO1
• GTO2
78 _ 0 GTO3
— Theory
74
72
80 100 120 140 160 180
Load, MW
Figure 9
All machines, variation of frequency with load
16
Figure 10
All machines, variation of damping with load
17