View PDF

. .
; .111E .AMERICAN soeinv
346 E 4701 Si, New York, N.Y. 10017 ' •
- , ., ,
• , . ,
The Society Wail not be responsible for statements or opinions advanced in papers,or discussion at meetings of the Society, or ot i tis DI;IsiOne or ‘,
SectionS, or, printed In its publications. Discussion IS printed only if the paper—Is published
to photabbni:..1
• Authorization'
. in....•an . 'ASME. Journal:
material: for internal or personal' use under circumstance not falling within the fair use provisions of the Copyright Act IS granted by :ASMS to; .' ...-:.
Hbrades and Other users registered with the Copyright Clearance Centei(CCC)Trallsactiohal Rationing Service provided that the base fee of 60.30 ‘.-..'
per Page is paid directly to the CCC 27 Congress Street, Salem' MA 01970 Requests special permission or bulk tepaidiiiilonehoUldbe ,addtessecl , .
to the ASMETrichileal RiblisNng dipartmeht 1 .
Thight ASME
. Copy
... •
1. 0
All. Rights Resolved -'...
. '•
1 •
,•Frintecint1J.S.A ' .,' I-
C. Hornsby & E. R. Norster
III 1111111011,11111 11 111111
European Gas Turbines Ltd
Industrial Turbine Group
P.O. Box 1
Thorngate House
Lincoln, LN2 5DJ
United Kingdom
This paper describes the methodology and
application of Computational Fluid Dynamics (CFD)
to Dry Low NOx (DLN) combustion systems
throughout the range of small industrial gas
turbines produced at European Gas Turbines
(EGT) Lincoln UK.
The use of CFD in the development of such
systems has been encouraged not only by the
availability of a variety of general purpose CFD
codes, but also by the inherent difficulties
associated with direct measurement in such a
harsh environment. Combusting flow analyses
provide detailed predictions of local temperature
and velocity fields together with exhaust emissions,
enabling numerous conceptual studies to be
undertaken without the usual associated
mechanical difficulties.
In particular, the work EGT has concentrated on
concerns the prediction of fuel / air mixing quality
upstream of the flame front, in order to assess the
effect of fuel injector design variables on NOx
production. This methodology has accelerated
injector development resulting in less than 10
ppmV NOx combustors.
Validation of the detailed features of the flow field is
currently underway, though parametric
comparisons have already proved consistently
accurate in displaying the trends necessary for the
development of an ultra low NOx combustion
system. Correlations of rig emissions data with
overall predictions have shown to be in good
( kgs- ]
area of cell face i
mass fraction of fuel
through cell face i
mass flow weighted mean
general function
mass flow of fluid through cell face i
standard deviation
combustor primary zone
1. Introduction
Despite the complexities of the internal flow
associated with the gas turbine engine the pressing
need for aerothermodynamic improvements has made
the aggressive use of CFD inevitable. The specific
problems associated with combustor flows differ from
their external aerodynamic counterparts in three
principal ways: (1) they usually involve complicated
geometries and boundary condition constraints, (2)
there is close coupling of flow elements involving
three-dimensional unsteady secondary flows and (3)
there is significant energy exchange from multi-phase
chemical reactions. While this seems formidable and
could cause restraint one finds quite the opposite.
Since the combustor environment is difficult to
produce experimentally CFD is being employed as a
technology to give insight and control, and developing
new solution methodologies employing economic
computational techniques. The ultimate use of CFD is
as a design/analysis tool which can provide substantiel
Presented at the International Gas Turbine & Aeroengine Congress & Exhibition
Orlando, Florida — June 2-5, 1997
Downloaded From: on 12/29/2014 Terms of Use:
reductions in time and cost in the combustor
development cycle. This paper describes the use
of CFD in advancing the development of an
injection system to optimise the low NOx
performance of a generic combustor design. The
combustor and injection system are analysed for
fuel/air mixing, operating temperature and emission
species. The measured NOx emissions from rig
and engine tests are correlated with the predictions
of a NOx model and also the improved mixing
performance of the injector as indicated through
CFD analyses.
through this method allowed optimisation of designs to
proceed in minimal time.
Combustion System Description
The particular DLN combustion system under
consideration is a can-annular arrangement in
common with the conventional combustion systems
across the range of EGT small gas turbines
comprising the Typhoon, Tornado and Tempest.
Figure 1 illustrates the Typhoon installation of the G30
DLN combustors with details of the burner
2. Background
The combustor is of the lean-premixed variety with
55% of the total flow admitted through a radial swirler
to control the reaction zone stoichiometry. Limitations
to the length of the can means that a separate
premixing zone in common with other systems of this
type is not possible, instead a "fast mixing" fuel
injection scheme is incorporated prior to burning in
order to achieve low emissions capability. In addition
to the fast mixer, an additional diffusion flame pilot
system is utilised for flexible machine operation.
In recent years most manufacturers of gas turbine
engines have developed lean-burn combustors
with the basic low NOx production benefits of low
flame temperatures associated with this approach.
The level of NOx in these systems is known to be
sensitive to the degree of mixing of fuel and air.
However until recently the degree of mixing
achieved in a given injection system has been
poorly defined. In European Gas Turbines
combustor development programmes extensive
use has been made of CFD to improve the
prediction of fuel-air mixing in addition to the
internal flow pattern of the combustor. The
introduction of lean-burn technology on EGT small
engines has not only been a requirement to meet
legislative commitments but also to address
competitive forces. The accelerated introduction of
the lean-burn approach is being applied across a
range of engines; Typhoon, Tornado & Tempest,
and the time to optimise such systems is at a
Rather than develop a CFD code in house, as has
previously been popular for many turbomachinery
applications, it was felt that a general purpose
commercial CFD code including combustion modelling
capabilities would be appropriate, allowing EGT to
work on the engineering without the software
development lead time, particularly where fully elliptic
flow is concerned.
The combustion factors influencing NOx production
are normally expressed through time, temperature
& turbulence. Since for lean-burn flames the most
significant production of NOx occurs at the flame
front (2 to 4 milli seconds ) post flame changes
have little influence. Bulk temperature, however,
which is dependent on overall stoichiometry has a
major influence. In addition, turbulence or
effectively mixing of the fuel and air provides
localised regions of burning which are above and
below the overall mixture ratio. The NOx
production rate increases exponentially with
temperature and it is not surprising that under such
conditions a higher NOx production results than for
the case of fully pre-mixed conditions. The overall
sensitivity of NOx production to the degree of
mixing can be established through a statistical
mixture distribution parameter such as the
standard deviation Ref. 1-4. The influence of both
mixing and stoichiometry on NOx production can
be quantified as illustrated typically in Ref. 5.
Correlations of the influence of various injector
factors determined through CFD and compared
The particular code chosen for this purpose was
STAR CD. This code provides not only the necessary
physical modelling, but also incorporates an
unstructured meshing format, allowing models to be
built without the constraints of a rectangular coordinate system. Furthermore, embedded meshing
techniques allow refinement to be carried out locally in
areas of particular interest (e.g. around the injection
holes). The recent addition to these meshing tools of
"arbitrary interfacing", whereby blocks of mesh can be
connected however they are configured, has further
enhanced the speed with which the models can be
constructed and analysed.
Computational Hardware
The software was run on workstations and cluster
machines. A Silicon Graphics Indigo R4000 machine
(64 Mbytes RAM, 1 Gbyte disk space) was used for
mesh building and post processing. Hewlett Packard
cluster machines were used for the numerical
Downloaded From: on 12/29/2014 Terms of Use:
the sensitivity of the flow to these conditions. A
further problem is that of geometry, in particular how
accurately the model represents the geometry and if
the grid is sufficiently refined to capture the nature of
the flow. Again sensitivity checks can be made to
either highlight or circumvent such problems. The
physical modelling in the CFD code itself can also
cause problems. If the parameter in question
depends strongly on a phenomenon which is not
included in the mathematical models then the
analysis could prove futile.
analysis, a HP735 (144 Mbytes RAM, 1 Gbyte disk
space) for small jobs and a HPK200 (1 Gbyte
RAM, 7 Gbytes disk space) for larger analyses.
Test Facilities
Testing was conducted on the High Pressure Air
Facility (HPAF) in Lincoln. Air is delivered at full
engine pressure (up to 19 bara and 5 kg/s flow)
from a compressor driven by an EGT T65000
mechanical drive Gas Turbine. Indirect heating of
the air (up to 550 °C) allows engine conditions to
be simulated at each of the three high pressure
test rigs installed in the facility.
6.2 Time Constraints
A single combustor Typhoon test rig comprising an
uncooled combustor casing with the combustor
exhausting into a water cooled drum and exhaust
section was used for testing. Provision was made
for instrumentation to measure the relevant
combustor parameters.
The purpose of the CFD analysis is to provide a
faster route for development. Hence if the analysis
required is so complex that the CPU time it takes is
greater than the time to manufacture and test the
same thing then it becomes redundant. Sufficient
computing power must be available to keep the CPU
time low enough, or a simpler analysis must be
6.3 Experimental Data Quality
Iteration Loop
To calibrate the analyses effectively the test data
must be consistent and reliable. Data quality
problems also carry the more global issue of how to
assess if the product is improved, and indeed what
the original performance was.
Figure 2 illustrates the procedure used for CFD
parametric studies. The initial step is to perform a
baseline study. This consists of analysing a current
configuration and comparing the results with
available test data. Commonly the test data and
analysis will not agree, this can be for a variety of
reasons ranging from the coarseness of the
computational grid to the applicability .of the
physical models used. In such circumstances it is
necessary to perform a further analysis of a
different configuration also with corresponding test
data, and for which the parameter in question is
measurably different from the original case.
Comparison of the variation in the parameter
allows a calibration of the modelling to be made.
Further modified models are subsequently
analysed and changes in the parameter noted.
Where a notable improvement in the parameter
occurs the modified geometry is tested, and the
data compared to the analysis. If this still proves
consistent the product is improved and taken as a
new baseline for further enhancements.
6. Potential Problems
7.1 Approach
CFD modelling was used as a tool to aid engineering
development, rather than a detailed analysis to
provide quantitative performance data. The approach
taken was to use a number of relatively simple models
to provide parametric comparisons with a base model.
This approach is by far the most cost effective and
widespread use of CFD, it enables a number of
comparative analyses to be performed in a relatively
short space of time, providing the engineer with a fast
"what if" tool, The altemative approach of detailed
quantitative analysis not only carries with it a large
time penalty, but also a requirement for
correspondingly detailed data measurement both
upstream of the combustor for boundary condition
definition, and inside the combustor for validation
A number of problems can occur in performing
such a procedure, which can be sectioned into
three distinct areas:-
With this aim the following physical models, together
with their associated asSumptions, were applied:-
6.1 CFD Models
The first, and possibly most common, problem
with CFD is the availability of adequate boundary
conditions. This does not necessarily mean that
all the details of the flow entering the calculation
domain must be accurately known, but where
they are not checks should be performed as to
Analysis is steady state
k-s turbulence model
Density, p calculated as an ideal gas f(T,p)
Wall functions
Downloaded From: on 12/29/2014 Terms of Use:
Properties at model inlets assumed to be flat
based upon bulk flow properties from test
Combined Eddy Break-Up / chemical kinetics
reaction rate model.
Fuel is CH 4 using the 2-step reaction:-
These models were run non-combusting under full
engine inlet temperature and pressure conditions in
order to assess the fuel/air mixing characteristics
upstream of the burning zone (test results and fluid
velocity calculations both suggesting that no burning
could take place in the swirler slots).
CH4 +
3/2 02 -> CO + 2H 20
CO +
1/202 -> CO2
The analyses were parameterised using the mass flow
weighted standard deviation of fuel distribution and
fraction of fuel coverage at the exit of the slot defined
by :-
using the reaction rate data of Dupont et al (Ref.
std deviation, S = E(C 7) - E7(C) )
4IQmci441 =
A1 (CH41° 7 102] 8 eXP('E81
A2 [CO]' 5
E(c) = mean = ECM/
& E(C7) = EC2M; / EMi
eXP(-Ea2 / RD
fractional coverage = S f(C i , A ) / SA;
= 5.0114 x 10"
= 202.32x 106
= 1.2604 x 10' 7
= 179.75 x 106
f(0,A1 )=A;
m' 5 kmora5 s-1
J kmo1-1
m725 kmo1475
J kmoll
The boundary conditions for this model were applied
as flat profiles derived from bulk flow test data as
detailed measurements of these conditions were not
available. Sensitivity checks of turbulent intensities
applied at model inlet revealed the model to be very
insensitive to such variations as the turbulence
generated downstream of the model inlet swamped
the convected quantities.
Combustion is adiabatic and radiation is not
7.2 Single Swirler Slot Fuel/Air Mixing Models
A particularly fast and useful parametric method
was achieved by modelling only part of the
combustor. In this case only a single slot of the
radial swirler was modelled thereby imposing the
assumptions that :•
the flow of air and fuel into each swirler slot is
identical, and
the downstream interactions of the swirler slots
are of little consequence to the mixing at the
injection point dose to the swirler entry.
7.3 Full Combustor Fuel/Air Mixing Models
non-combusting models of a full combustor and
transition duct were developed. These models remove
the assumptions of symmetry imposed by the single
slot models, but retain the simplistic approach to the
physics of the problem thereby keeping time penalties
for the extended model to a minimum.
The base model is a three dimensional geometry of a
complete Typhoon DLN combustor and transition duct.
It includes an entry plenum upstream of the swirler in
common with the single slot models, and comprises
approximately 200,000 fluid cells (fig 4). The
assumption is made that the pressure drop is identical
over each fuel injection orifice consistent with the
single slot models, this ignores any distribution bias in
the fuel feed manifold. Although this model is
constructed using almost double the number of cells
as the single slot models the definition in the swirler
slot is reduced, it is however still possible to make
qualitative comparisons between the two models.
The base model is a three dimensional geometry of
a 30 0 sector of the G30 12 slot swirler with entry
plenum. It is determined to have cyclic symmetry
along an axial plane either side of the slot, and
comprises approximately 118,000 fluid cells (fig 3).
Typically modifications to the base model involved
changes to:•
C1 >0
C1 =0
fuel injection location and orifice size
fuel injection pressure
swirler inlet geometry
Downloaded From: on 12/29/2014 Terms of Use:
fuel coverage by using three injection points in place
of two.
These models were run at similar conditions to the
single slot models. The fuel/air mixing
characteristics upstream of the buming zone were
assessed in two ways:•
The full combustor fueVair mixing models were
similarly successful in the prediction of fuel
distribution, the patterns at slot exit being consistent at
slot exit if not quantitatively similar. The NOx
predictions at prechamber exit, using a correlation via
the standard deviation of fuel distribution, proved
reliable up to a point but as the mixing improved and
the fuel pattern altered the predictions became
inconsistent due to the recirculatory nature of the flow.
Thus the more precise method of NOx prediction
using the NOx correlation at zero standard deviation
became necessary. This proved successful and
carried with it the added bonus of predicting a burning
zone via the subset of the model created as shown in
figure 7.
Mass flow weighted standard deviations were
taken at the exit of the prechamber and NOx
values derived using the overall stoichiometry
and a NOx / Mixedness correlation (illustrated
• A subset of the domain was taken under viable
conditions for combustion (as regards flame
speed and stoichiometry). A NOx value per cell
was then extracted using the NOx correlation,
assuming the fuel in each computational cell to
be perfectly premixed. A volume integration was
then carried out to obtain a bulk NOx value. This
NOx value was then used with the stoichiometry
and correlation to obtain a standard deviation.
The final more rigorous method of full combusting flow
analysis was less accurate in predicting NOx due to
underprediction of the temperature field, though
predictions were more consistent but at a significant
time penalty. This method did however prove useful in
providing detailed predictions of the CO and
temperature patterns (figs 8 and 9).
7.4 Full Combusting Flow Models
A more rigorous approach in order to predict
temperatures and CO production was also used.
This was done using full combusting flow models
as a further extension to the scalar transport
models above.
NOx predictions based on the CFD models above are
illustrated in figure 10 using the NOx correlation chart
(Ref. 5) and compared to those obtained directly from
test. The test results use the design point
stoichiometry, (1), the measured NOx value and the
correlation to predict a standard deviation, S. Where
the CFD predictions use the design point
stoichiometry, and either :
• the standard deviation, S. and the correlation to
predict a NOx value, or
• the correlation at zero standard deviation and
volume integrate over the field to predict values for
both NOx and standard deviation.
Data from both GE and Siemens is also included in
figure 10 as comparison.
Details of temperature, CH4 and CO fields were
obtained directly from these models. Values for
NOx and standard deviation were extracted in a
similar manner to those above by:• Taking a subset of the domain based on the
temperature field (e.g. T> 1800K)
• Obtaining a NOx value per cell via the
correlation assuming the fuel to be perfectly
premixed in each computational cell.
• Performing a volume integral over the subset to
obtain a bulk NOx value.
• Using the calculated bulk NOx value, overall
stoichiometry and the NOx correlation to obtain
a standard deviation of fuel distribution.
9. Discussion
The methodology used to apply CFD to DLN
combustion systems has proved fast and effective in
optimising the emissions characteristics of the
combustor. The flexibility and cost effectiveness of this
approach has more than justified its use, and provided
a good deal of insight into the operation of the system.
There are drawbacks to such analysis work, notably
that CFD remains relatively specialised and the
inability to provide very fine flow details, or quantitative
performance data, without a good deal of time and
effort still leaves many unconvinced of its usefulness
despite the parametric success.
The single slot models proved a good guide as to
the distribution of fuel in the swirler slot, in
particular the penetration of fuel jets into the high
velocity air stream could easily be assessed.
Figure 5 depicts an isosurface of gas fuel mass
fraction as it is injected into the slot providing a
clear picture of the fuel penetration. The trends of
the parameters corresponded to the trends of the
NOx produced though it was not possible to extract
values for NOx directly from these models. Figure
6 shows the mass fraction of fuel for sections taken
at the slot exit and highlights the improvement in
Downloaded From: on 12/29/2014 Terms of Use:
However, there are a variety of ways of increasing
the'accuracy of the predictions including:-
The authors would like to acknowledge the support of
their colleagues of the DLN team, Lincoln.
• 'Further grid refinement
• Enhanced turbulence models i.e. Chen k s,
second moment closure.
• Enhanced combustion models i.e. 3-step
reaction schemes, probability density functions.
• Transient analyses.
1 Mikus T, Heywood J B & Hicks R E Nitric
Oxide Formation in Gas Turbine Engines: A
Theoretical and Experimental Study; NASA C
R 2977 April 1978
Though these, and many other, additional models
would undoubtedly improve the accuracy of the
results, almost all would be at a considerable time
penalty, which must be weighed-up against
available computing power. Clearly as the trend
towards faster computers continues, more complex
and accurate models will become feasible.
2 Lyon V J Fuel-Air Nonuniformity Effect on
Nitric Oxide Emissions. AIAA Journal Vol. 20
pp 660-665 1982; NASA Technical Paper
1978, 1981.
3 Fric T F Effect of Fuel-Air Unmixedness on
NOx Emissions, AIM 92-3345 July 1992.
10. Further Plans And Developments
Current developments of the reported CFD
modelling has resulted in predictions of fuel/air
mixedness of N ° 2 distillate both in liquid and
vapour forms via Lagrangian 2-phase evaporative
models. Hopefully in the near future this will lead to
the G30 DLN combustor operating on liquid fuels
without the need for diluent injection for NOx
4 Kesseli J, Norster E R & Landau M Low
NOx Combustor Design and Test with a
Recuperated Gas Turbine Engine; ASME
Cogen-Turbo Vol. 7 Book No. 100333, 1992.
5 Norster E R and De Pietro S M Dry Low
Emissions Combustion System for EGT
Small Gas Turbines, Institute of Diesel and
Gas Turbine Engineers Publication No 495,
March 1996.
6 Dupont, V. Pourkashanian, M. Williams, A.
Modelling of Process Heaters Fired by Natural
Gas: J. Inst. Energy, pp 20 - 28 March 1993.
Downloaded From: on 12/29/2014 Terms of Use:
Figure 1 - Installation of G30 DLN combustor
Figure 4 - Full combustor CFD mesh
Figure 2 - Flowchart detailing CFD procedure
MCA er
Figure 5 - CFD predicted penetration of fuel jet
Figure 3 - Single slot CFD mesh
Figure 6 - CFD pred'cted fuel distribution at swirler
slot exit
-MI 0 T3
.4.3 10
man not
Ma OLE C0411119TOR - MOLE Pileilfffi SIDT
2 Injection points
3 Injection points
Downloaded From: on 12/29/2014 Terms of Use:
Figure 9 - CFD predicted temperature distribution
in combustor
Figure 7 - NOx calculation subset
1131103a1CU MAYO=
100111111001 SET
Figure 8 - CFD predicted CO distribution in
Figure 10 - NOx / Mixedness con-elation
111111/1111 I
gi Ma
1,12511111 LB
SC t-00
wanes nrof-in
MOS ROCIE01-100D0 OF 0111043011
TR 1
t it
A - EGT, 2 njection points, high pressure rig data
B - EGT, 2 njection points, prechamber exit mbcing CEO
C - EGT, 2 injection points, zonal mixing CFD
D - EGT, 2 injection points, zonal combusting CEO
E - EGT, 3 injection points, high pressure rig data
F - GE high pressure rig data, ASME 92-GT-121
G - Siemens theory (S=0.15), IMechE Sept 95
Downloaded From: on 12/29/2014 Terms of Use: