Professional Documents
Culture Documents
GT2002-30063
GT2002-30063
ASME TURBO
Proceedings of ASME TURBO EXPO EXPO
2002 2002
The 47th ASME International June 3-6, 2002,
Gas Turbine Amsterdam,
& Aeroengine TheCongress
Technical Netherlands
June 3-6, 2002, Amsterdam, The Netherlands
GT-2002-30063
GT-2002-30063
∂u 1 Compared to eqs. (4) and (5) this equation has the advantage that
+ (u • ∇)u = − ∇p , (2) we only have to solve for the acoustic pressure p but not for
∂t ρ
the three components of the sound particle velocity u . However,
eq. (6) disregards loss mechanisms like area jumps. This simpli-
when the pressure p = p̄ + p , density ρ = ρ̄ + ρ and velocity fication and neglecting entropy waves does not mean a general
u = ū + u each are written as the sum of their mean and their restriction of the technique, though. The governing equations
fluctuating part. The fluctuations of density can be attributed to could be extended to include the mentioned effects. In the frame-
the fluctuations of pressure and entropy. If the entropy fluctua- work of this work we focus on the purely acoustical mechanisms.
tions are solely caused by heat release fluctuations, the following The substantial derivative of the heat release rate fluctuations in
equation holds for the substantial derivative of the density with eq. (6) implies that changes in the heat release can be caused both
respect to time (cf. [19]): by local changes of the heat release rate and by a moving heat re-
lease zone. Equation (6) can be solved if a relation describing the
Dρ 1 Dp κ − 1 coupling of the heat release to the flow field is provided. Giving
= 2 − 2 q̇ . (3)
Dt a Dt a a very general formulation of that relation, the fluctuating heat
release may depend on both the pressure and the velocity, addi-
tionally allowing for a time delay:
Here, a = κp/ρ is the velocity of sound, κ is the ratio of spe-
cific heats and q̇ = q̇¯ + q̇ is the rate of heat addition per unit
volume. Combining eq. (3) with the continuity equation eq. (1), q̇ (x,t) = q̇ ((p, u) (x,t, τ(x))) . (7)
linearizing the resulting expressions, and using the relation for
isentropic changes of state ρ /p = a2 , yields the linearized equa-
tions for the conservation of mass, Here, the delay time τ does not necessarily have to be a single
value but can rather be a delay time distribution. This may be im-
portant for a realistic model of a gas turbine burner flame as has
∂p
+ u • ∇ p̄ + ū • ∇p + κp ∇ • ū + κ p̄∇ • u = (κ − 1)q̇ , (4) been shown by Sattelmayer [2] and Flohr et al. [21]. Equation (6)
∂t involves p only linearly. In order to make unstable perturbations
not grow into unlimited oscillations, it is therefore important to
and momentum, include non-linearities occuring in the heat release.
Our technique for the simulation of combustion oscillations
∂u p 1 can be summed up as follows: provide a non-linear model for
+ (ū • ∇)u + (u • ∇)ū + 2 (ū • ∇)ū = − ∇p . (5)
∂t ρ̄ā ρ̄ the fluctuations of the heat release rate (eq. (7)) to solve the wave
equation eq. (6) in the time domain together with appropriate
boundary conditions. The following paragraph will give a short
These equations do not account for entropy waves, which can
description of the numerical tool we use to realize this idea.
lead to some inaccuracy only for very low frequencies. For a
discussion of this topic see Sattelmayer [2] and Polifke et al. [20].
Equations (4) and (5) can further be simplified for the case Numerical tool
of uniform mean flow. Omitting the terms involving the spatial As in the case of combustion oscillations in annular com-
derivatives of the mean quantities, applying ρ̄ā2 ∇• to the mo- bustors instabilities in the low frequency regime are dominating,
mentum equations eq. (5) and subtracting the result from the sub- and their wave lengths are of the same order as the characteris-
stantial derivative of the continuity equation eq. (4) with respect tic lengths of the combustor, even standard numerical methods
can lead to satisfying results. To solve the equations described
before, we employ the commercial tool FEMLAB. This is a fi-
1 Notation: ∇ • u = ux + vy + wz , ∇u = (ux , vy , wz )T nite element package designed to solve an arbitrary set of partial
Boundary conditions
At the walls, the outward normal component of the sound
MODEL ANNULAR COMBUSTOR particle velocity n • u has to disappear. This is ensured by the
Geometry boundary condition
To verify the feasibility of our approach, we have chosen to
carry out calculations in a generic geometry. This model is de- n • ∇p = 0 . (11)
rived from an annular combustor test rig that in future will give
us valuable experimental information for the understanding of For the inlet and outlet boundary conditions we follow Aka-
combustion instabilities and verification of models (see fig. 1). matsu and Dowling [6] and assume that the flow both enters and
The test rig consists of an annular plenum chamber and an annu- exits the computational domain from and to a large plenum so
lar combustion chamber connected by twelve swirl burners. The that pressure perturbations there can be neglected. Then the out-
plenum chamber is supplied with an externally premixed fuel- ward normal component of the velocity fluctuation will be pro-
air mixture, so fuel inhomogeneities do not occur and entropy portional to the pressure fluctuation:
waves do not play a role in this combustor. The hot gas exits the
combustor through twelve nozzles, allowing operation at slightly
elevated pressure to establish a choked end if desired. n • u = K p . (12)
In the numerical model we avoid a detailed reproduction of
the swirl burners, exit nozzles and inlet orifices. This leads to a Using the momentum equation in the normal direction, this trans-
simple geometry consisting of an inlet plane, followed by an an- lates to
nular plenum, twelve straight ducts representing the swirler sec-
tion, the annular combustion chamber, and finally an exit plane K ∂p
n • ∇p = − , (13)
(see fig. 2). K ūn + ρ̄1 ∂t
-σuB
0 σuB u
D
-σuB
Non-linear heat release model In contrast to Dowling, we introduce the additional parameter σ
A variety of flame and heat release models is available that which allows to ‘turn on’ saturation already for fluctuations that
could be used to provide a relation for eq. (7). The assumptions are smaller than the mean value. This seems meaningfull as a real
and requirements underlying the model employed here are as fol- flame usually will be driven into the non-linear regime even for
lows: moderate fluctuations. A very extreme example has been given
by Hantschk and Vortmeyer [9]. Their numerical simulations
1. The zone of heat release is a predifined fixed region at each show that the total heat release rate might undergo non-linear
burner. There is no movement of the flame. modifications even for fluctuations which are only a few percent
2. The instantaneous rate of heat released per unit volume at of the mean value. There, this most likely has to be attributed to
each burner is related to the burner exit velocity uB at the the fact that the flame is very long so that regions with strong pos-
corresponding burner, involving a single delay time τ. itive fluctuations could compensate regions with strong negative
3. The response of the flame to velocity fluctuations is increas- fluctuations. Though we do not account for a spatial variation
ingly damped with increasing frequency. of the heat release rate and time delay, such effects to some ex-
4. The maximum and minimum heat release are limited, i.e. tent can be modelled with the parameter σ. For all calculations
saturation effects introduce a non-linearity. presented here, σ = 0.75.
0.8
0.6
0
0.4
0.2
(0,1,0) (1,0,0) 0
-0.05
-0.2
-0.4
-1.0
1
(1,1,0) (0,2,0)
0.5
Figure 4. FIRST FOUR ACOUSTICAL EIGENMODES OF THE COM-
B
BUSTOR (NORMALIZED PRESSURE DISTRIBUTION). u´ / u 0
B
-0.5
Mode (0,1,0) (1,0,0) (1,1,0) (0,2,0)
f 0.353 0.390 0.656 0.668 -1
T 2.83 2.56 1.53 1.50
0 10 20 30 40 50 170 180 190 200 210
t
Mode (l,m,n) in this table refers to the contribution of modes of
order l, m and n in axial, circumferential and radial direction, re-
spectively. The eigenfrequencies for the cases with and without Figure 6. TIME EVOLUTION OF A SELF-EXCITED INSTABILITY FOR
mean flow are virtually identical, so just one frequency is given. τ = 1.9, ω0 = 5, NO MEAN FLOW.
This has to be attributed to the low Mach numbers which are
0.011, 0.14 and 0.035 in the plenum, burner section and com-
bustion chamber, respectively. FEM mesh and initial condition
The pressure distribution for these first four modes can be The combustor geometry has been discretized with a finite
seen in fig. 4. Note that the modes involving a circumferen- element mesh consisting of approximately 70,000 tetrahedral el-
tial component are doubly degenerate, i.e. there belongs a two- ements. This mesh is shown in fig. 5.
dimensional eigenfunction space to one eigenvalue. For reasons In this figure also the pressure distribution used as the initial
of simplicity we only show one of the two eigenfunctions. The condition can be seen. The pressure fluctuation at the node points
second is identical to the first rotated by π/(2m) around the axis is randomly set to values between -0.05 and 0.05. This generates
of the combustor, m being the order of the circumferential mode. an initial perturbation with a broad spectrum of wavenumbers
When there is a circumferential component, for low frequen- that is expected to trigger the growth of unstable modes.
cies the mode is prominent only either in the plenum chamber or
in the combustion chamber, not in both. This is caused by the dif-
ferent velocities of sound in the two annuli which are not integer RESULTS
multiples of each other. Here, the first two pure circumferen- Limit cycle evolution
tial modes have strong amplitudes only in the plenum chamber. In the time-domain simulation, the random perturbation in
What might appear as the first circumferential mode just of the the pressure field imposed as an initial condition induces a self-
combustion chamber under close examination reveals to be the excited instability if the parameters of the heat release model are
first combined axial-circumferrential mode, its amplitude in the set appropriately. This can be seen in fig. 6 where the fluctuation
plenum chamber only being very low. of the burner exit velocity at one specific burner in terms of the
0.8
0.6
0.4
0.2
0˚ 45˚ 90˚ 135˚
0
-0.2
-0.4
-0.6
-0.8
-1.0
180˚ 225˚ 270˚ 315˚
Figure 7. NORMALIZED PRESSURE DISTRIBUTION AT DIFFERENT PHASE ANGLES FOR ONE PERIOD OF THE LIMIT CYCLE FOR τ = 1.9,
ω0 = 5, NO MEAN FLOW.
mean burner exit velocity is shown. The calculation has been 1.5 0.15
carried out for no mean flow conditions with a delay time τ = 1.9
and a cut-off frequency ω0 = 5. The initially random fluctuations 1 0.1
0.5 0.05
When the velocity fluctuations become so strong that saturation
p´ / p
occurs, the growth rate gets smaller until finally the amplitude 0 0
remains constant. At this point the limit cycle is reached.
-0.5 -0.05
More information about this limit cycle can be obtained
from the evolution of the pressure distribution during one pe-
-1 -0.1
riod of the oscillation, shown in fig. 7. It turns out that the mode
excited in this case is the first combined axial-circumferential -1.5 -0.15
240 240.5 241 241.5 242 242.5 243 243.5 244
mode (1,1,0). Here this mode appears as a spinning mode, i.e. t
the node line is not stationary but rather rotates by 2π during one
cycle. This leads to a phase difference between the fluctuations
Figure 8. FLUCTUATION OF BURNER EXIT VELOCITY (—), PRES-
at the burners and the combustor exit, which best can be recog-
SURE (· · · ), AND HEAT RELEASE RATE (−−) DURING THE LIMIT CY-
nized from the fact that the node line is oblique and not parallel to
CLE FOR τ = 1.9, ω0 = 5, NO MEAN FLOW.
the combustor axis. The pressure at the exit lags the pressure at
the burners because the information takes a certain time to travel
from the heat release zone to the exit.
The mechanism underlying the self-excited instabilty be- the heat release fluctuation is limited to a value slightly below
comes clearer examining the relation of the burner exit velocity 0.75 q̇¯ which is the value given for the onset of saturation. This
fluctuations, pressure fluctuations, and heat release fluctuations value is not fully reached because of the additional damping in
at one burner. Figure 8 shows these quantities in terms of their the heat release model. The velocity fluctuations, however, ex-
mean values for a few periods of the limit cycle. Here, the pres- ceed the saturation limit and gain an amplitude just above the
sure fluctuation at the center of the heat release zone of the cor- mean velocity, meaning that there is a short part of the cycle
responding burner is given. It can be seen that the amplitude of where reflow can be observed. The pressure fluctuations reach
B
u´ / u
Normalized Magnitude
B
-1
10
-10
0 20 40 60 80 100
-2
10
t
1
0.5
q´ / q
-3
10 0
the inlet and exit. Including other loss effects would very likely
yield lower pressure fluctuations.
0.01
From fig. 8, it is also obvious that the pressure fluctuation
lags the burner exit velocity fluctuation almost exactly by T /4,
u´B / uB
0.3 0.7
Cycle Increment
0.2 0.6
0.1
0.5
f
0
-0.1 0.4
-0.2
0.3
0.8 1
0.6
0.8
u´B,max / uB
0.4
0.6
0.2
0˚ 90˚ 0 0.4
-0.2 0.2
-0.4
0
1 2 3 4 5 6 7 8 9 10 11 12
-0.6 Burner Number
-0.8
Figure 15. AMPLITUDES OF THE BURNER EXIT VELOCITY FLUCTU-
-1.0
ATIONS FOR THE CASES WITHOUT (•) AND WITH () MEAN FLOW
180˚ 270˚
(τ = 1.9, ω0 = 5). LIMIT FOR ONSET OF SATURATION (−−).