Professional Documents
Culture Documents
ReceiVed September 11, 2007. ReVised Manuscript ReceiVed January 10, 2008
Flaring is used extensively in the energy and petrochemical industries to dispose of unwanted combustion
gases by burning them in an open flame. However, these units may represent an important source of gas
emissions due to inefficient operation under certain conditions such as high crosswind velocities. Several
experimental studies have previously focused on flames burning in a fixed volume by using wind tunnels. In
these experiments, the entire plume of combustion products was collected, sampled, and analyzed to calculate
the combustion efficiency. Present work simulates these wind-tunnel experiments by using the commercial
computational fluid dynamics (CFD) software package Fluent 6.2. Several three-dimensional (3D) computational
models are developed, and suitable turbulence and chemistry models are applied to simulate the complex
combustion phenomena and flame downwash. The computational work was greatly reduced by applying the
laminar flamelet model, which assumes that a turbulent flame is an ensemble of small laminar structures called
flamelets. Inefficient combustion is observed at high crosswinds, and simulation results are in very good
agreement with experimental data. These results show that CFD can successfully simulate these wind-tunnel
flare experiments. The resulting simulation models could be used to estimate the hydrocarbon emissions from
chemical and petrochemical flares at crosswind conditions, an environmental issue of great importance in air
pollution models.
of the governing transport equations for mass, energy, species, new experiments are characterized by larger burner diameters,
and momentum. Even though research has been done on the in some cases comparable to industrial flare diameters.
subjects of CFD and industrial combustion, there is very little
discussion of industrial flares, except by Baukal et al.,6 which 2. Governing Equations
discusses CFD applications in industrial combustion.
CFD has been recently applied to simulate the effect of steam CFD relies on solving conservation or transport equations
addition and air addition for several laboratory-scale, turbulent for mass, momentum, energy, and participating species. If the
nonpremixed flames.7 Detailed finite-rate chemistry models were flow is turbulent, model equations for specific turbulent quanti-
applied to predict the species concentrations in the flame, while ties have to be solved in addition. Since even with today’s super
species mass balances were set up in order to compute the computers resolving turbulent length scales directly results in
resulting flame combustion efficiency. Simulation results showed tremendous effort, Reynolds averaged equations are typically
that incomplete combustion of hydrocarbons may occur at high applied to include the physics of turbulence. Hence, the basic
steam/fuel and air/fuel ratios up to the point where these flames model equations for a fluid in turbulent flow are the Reynolds-
become extinguished. The computational work was greatly averaged Navier–Stokes (RANS) equations. For steady state,
reduced by assuming two-dimensional (2D) axisymmetric these equations are given below:
models. Reynolds numbers of these laboratory scale flames were 2.1. Continuity Equation.
comparable to those for industrial flares. ∇(Fν) ) 0 (1)
CFD has also been used to develop 3D simulations of the
effect of crosswind on a laboratory-scale, turbulent combustion where F is the density of the fluid and νj is its ensemble-averaged
flame.8 The flame was simulated at the exit of a vertical burner velocity vector, defined on a 3D domain.
that was perpendicular to the air flow, a configuration that is 2.2. Momentum Conservation Equation.
relevant to continuous gas flaring in the atmosphere. Simulations
clearly showed that for high momentum flames moderate ∇(Fνν) ) - ∇ p + ∇ (µ(∇ν + (∇ν)T) - Fν′ν′) (2)
velocities may significantly reduce the resulting combustion where ν′ is the turbulent fluctuation of the velocity vector, µ is
efficiency. the dynamic molecular viscosity of the fluid, and F is the
Unfortunately, direct application of CFD to simulate large- pressure. The overbars denote mean values. The Reynolds
scale industrial flares is very difficult. First of all, industrial flares stresses, Fνj′νj′, are extra terms that stem from decomposing
are clearly turbulent, and direct numerical simulation of turbulent solution turbulent variables into the mean and fluctuating
flows is not possible because of the wide range of time and components; these terms must be modeled in order to close eq
length scales. Thus, some type of turbulence model must be 2. A common approach employs the Boussinesq hypothesis to
applied. Second, realistic chemical mechanisms for hydrocarbon relate the Reynolds stresses to the mean velocity gradients:
( )
combustion cannot be described by a single reaction equation.
Such models may include tens of species and hundreds of ∂νi ∂νj 2
-ν′ν′ ) υt + - (υt(∇ν) + k)δij (3)
reactions that are known in detail for only a limited number of ∂xj ∂xi 3
fuels. Hence, some chemistry simplification must be made.
where the Einstein summation notation is being used; that is,
Furthermore, it is necessary to deal with complex turbulence-
δij is the Krönecker delta, νt is the eddy kinematic viscosity,
chemistry interaction due to the sensitivity of reaction rates to
and k is the kinetic energy of turbulence, defined by
local changes.
It is very challenging to perform combustion simulations for 1
large-scale flares. The total number of grids needed to capture k ) ν′iν′i (4)
2
all the combustion details makes the computational work almost
The Boussinesq approach is used in the k-ε model, which
prohibitive for large flares. Employing three-dimensional (3D)
is the turbulence model applied in this work. The advantage of
models significantly increases the computational work. Even
this approach is the relatively low computational cost associated
so, the lack of experimental data for industrial flares makes it
with the computation of the kinematic viscosity. For the k-ε
very difficult to validate potential simulation results. Thus, CFD
model,theeddyviscosityisobtainedfromthePrandtl-Kolmogorov
is restricted to the simulation of wind-tunnel experiments in
relation:
this work, in order to compare model predictions with experi-
mental data. This allows us to validate our results by direct Cµk2
comparison with experimental data. Validated simulation models υt ) (5)
could be used to estimate the actual hydrocarbon emissions from ε
chemical and petrochemical plants. where ε is the rate of turbulent kinetic energy dissipation.
The commercial software Fluent 6.2 is used in this work. As Robustness, economy, and reasonable accuracy for a wide range
a first approach, we have selected the wind-tunnel experiments of turbulent flows explain the popularity of this model in
of low-momentum jet diffusion flames of Johnson and Kostiuk,3 industrial flow and heat transfer simulations.
where natural gas was burned at crosswind velocities ranging 2.3. Energy Conservation Equation. When considering heat
from 1.0 to 11.0 m/s in a 0.0221 m diameter burner. In addition, transfer within the fluid and/or solid regions of the problem,
a new set of wind-tunnel experiments that have been experi- Fluent also solves the energy equation. This equation is given
mentally studied at CANMET Energy Technology Centre, below in a very general form:
Ottawa, are used for simulation of larger scale flares. These
Figure 1. Schematic of a closed-loop wind tunnel facility (all dimension supplementary fans were used to ensure that the plume of
in meters) at the University of Alberta. Reprinted by permission of
Elsevier Science from Johnson and Kostiuk,3 Copyright 2000 by The
combustion products was fully mixed into the wind-tunnel air
Combustion Institute. before sampling. The wind tunnel was sufficiently large that,
during a typical 5-10 min test, the concentration of hydrocar-
bons in the tunnel remained small, and the effects of reburning
were completely negligible.
From a simulation point of view, only the test section around
the burner (e.g., the box that followed the contraction section)
is strictly relevant. The dimensions of this box are 2.44 m in
width by 1.22 m in height by 11.8 m in length. However, a
5 m long box is enough to capture the flame behavior, so
unnecessary computation work could be avoided by reducing
the simulation domain. The floor of the wind tunnel was
constructed with 19 mm thick plywood, while downstream of
the flare the tunnel was covered with 30 gauge aluminum
sheeting to protect it from possible direct flame impingement.
The walls along the test section were primarily Plexiglas. The
ceiling upstream of the flare was constructed with 19 mm thick
plywood, but downstream of the flare the ceiling was made of
19 mm thick ceramic panels that could safely resist the
Figure 2. Sketch of flow structures in a low-momentum jet diffusion accidental impingement of the flame or hot combustion products.
flame in a crosswind.
The diffusion flames were established at the exit of a 24.6
mm o.d. (22.1 mm i.d.) pipe that extended 47 cm into the wind
not considered in this work, so the corresponding term has been
tunnel. The experimental setup also included a 65% blockage
removed from eq 6.
ratio perforated plate “turbulence plug” with 3 mm diameter
2.4. Species Transport Equations. Finally, for reacting holes, which was placed inside the pipe three diameters upstream
systems the species transport equations must be solved. In of the exit. The purpose of this plug was to create velocity
general form, this equation is given by profiles similar to the turbulent pipe flow expected in full-scale
∂ industrial flares, independent of the actual flow velocity in the
(FY ) + ∇ (FνYi) ) - ∇ Ji + Ri (7) laboratory-scale flares. However, computer simulation of that
∂t i
perforated plate is very difficult to perform and may introduce
where Ri is the net rate of production of species i by chemical numerical errors in the solution process. Thus, the turbulence
reaction. Fluent applies the finite volume method to discretize of the fuel gas was adjusted in our simulation to match
and solve the governing flow equations described above. the experimental turbulence intensity measured 5 mm above
the exit plane of the burner tube (see Kostiuk et al.9).
3. Wind Tunnel Configuration For these experiments, the jet exit velocity of the fuel, Vj,
was held approximately constant at 2 m/s, and the crosswind
A set of low-momentum, natural gas diffusion flames located
speed, U, was varied from 1 to 11 m/s. The external cold-flow
in a closed-loop wind tunnel were used for simulation. These
Reynolds number (Re) ranged from 1570 to 17 270 as the
flames were studied experimentally at the University of Alberta
crosswind increased from 1 to 11 m/s. Under these conditions,
and the National Research Council of Canada. Measurement
the flow regime on the outside of the pipe flare could be
of experimental combustion efficiencies was reported by
considered to be in the regime of having a laminar boundary
Johnson and Kostiuk.3 A detailed description of the experi-
layer separation. The turbulent fluctuation in the core flow of
mental setup and results was given as a final report by
the tunnel was found to be consistently less than 0.4% except
Kostiuk et al.9 A schematic representation of the wind tunnel
at low wind speeds (<2 m/s), where the intensity rises to about
is given in Figure 1.
1.8%.
The basic information about the experimental setup is given
below; refer to refs 3 and 9 for a more detailed description. The fuel gas used was sales grade natural gas (95.2% CH4,
The experimental flames were established at the exit of a 2.1% C2H6, 1.7% N2, 0.8% CO2, and 0.2% other, by volume).
burner tube mounted vertically in the wind tunnel and perpen- In order to guarantee ignition, the experiments used a retractable
dicular to the airflow. In the vertical section downstream of the hydrogen jet diffusion flame that was previously ignited by using
flame and in the upper section of the tunnel, a series of a manual high-voltage spark system. Once the hydrogen flame
was correctly positioned, the flow of flare gas was easily ignited.
(9) Kostiuk, L.; Johnson, M.; Thomas, G. Flare research project, final After the flare was ignited, the flow of hydrogen was turned
report, University of Alberta, September, 2004. off and the ignition system was retracted. Ignition can be
CFD on Flare Combustion Systems Energy & Fuels, Vol. 22, No. 3, 2008 1701
Figure 4. Long-exposure photographs of the experimental flame (left side) and simulation results for the temperature contours along the middle
plane (right side). Moving downward, the crosswind velocity is increased from 1.33 m/s (case B) to 11.05 m/s (case G). Reprinted by permission
of Elsevier Science from Johnson and Kostiuk,3 Copyright 2000 by The Combustion Institute.
simulated in Fluent 6.2 by patching a high-temperature region grid. The final 3D grid had 321 040 cells and 990 713 faces,
near the burner tip, which helps to switch on the combustion and it was successfully checked for skewness. The simulations
reactions in the same way as a spark helps ignites an actual were performed by using the 3D segregated solver incorporated
flame. in Fluent. Implicit formulation was chosen; only steady state
solutions were obtained. A second order upwind scheme was
4. Simulation Results used to solve all equations. The SIMPLE algorithm10 was used
For this study, the commercial software Fluent 6.2 was used (10) Patankar, S. V. Numerical Heat Transfer and Fluid Flow; McGraw
for simulation. Gambit 2.0 was used to create the mesh and the Hill: New York, 1980.
1702 Energy & Fuels, Vol. 22, No. 3, 2008 Castiñeira and Edgar
Figure 7. Velocity vector field in the middle plane for cases B and G.
Table 2. Computed Species Mass Flows (10-5 kg/s) and Resultant Combustion Efficiencies for the Simulated Casesa
case (CH4)in (CO2)in (C2H6)in (CH4)out (CO) out (CO2)out (C2H6)out ηc (%)
B 49.947 1.343 2.063 0.00 1.130 142.66 0.00 98.5
C 49.947 1.343 2.063 0.00 0.003 145.18 0.00 100.3
D 49.947 1.343 2.063 0.00 0.002 146.72 0.00 101.4
E 49.947 1.343 2.063 0.00 0.002 141.23 0.00 97.5
F 49.947 1.343 2.063 0.00 2.541 137.94 0.00 95.3
G 49.947 1.343 2.063 0.00 7.317 132.38 0.00 91.4
a Cases C and D predict efficiencies slightly over 100% due to numerical errors.
Figure 9. Representation of the meshed model box for case 3, top view.
Figure 11. Velocity vectors and recirculating region at the middle plane for the CANMET wind-tunnel simulations.
Table 5. Computed Species Mass Flows (10-5 kg/s) and Resultant Combustion Efficiency for the CANMET Wind-Tunnel Simulations
case (CH4)in (CO2)in (C2H6)in (CH4)out (CO)out (CO2)out (C2H6)out ηc experimental (%) ηc simulation (%)
1 380.229 6.782 15.696 0.002 19.183 1075.988 0.000 97.54 97.56
2 380.229 6.782 15.696 0.000 0.000 1095.620 0.000 99.78 99.35
3 253.471 4.522 10.464 0.000 0.000 726.718 0.000 98.71 98.85
of these 3D simulations. Initially, the efficiency is slightly higher N2, 0.62% CO2, and 0.02% C4H10. Table 3 describes some
because more air is entrained into the flame. However, as the physical parameters of the three flames studied in this work,
crosswind velocity grows larger than 5.49 m/s (case E), the including the case number, the run label, the total mass of fuel
combustion efficiency decreases and degrades significantly with sent to the burner, the nominal pipe diameter (Schedule 40),
further increases in the wind speed (cases F and G). Both and the pipe length. The last column lists the experimental
experiments and simulations confirm these results. Notice that combustion efficiencies measured by the authors of the
simulated combustion efficiencies exhibit the same trend as the experiment.
experimental data, with a maximum efficiency located around The variable air conditions in the wind tunnel are given in
4 m/s crosswind velocity. Mismatches between simulations and Table 4, including pressure, temperature, wind speed, and
experiments are on the order of 1-2%, so these CFD models relative humidity.
successfully simulate the wind-tunnel experiments. As observed in Table 3, the pipe dimensions for cases 2 and
A flame bent over with the windward side of the flame 3 are relatively large (the pipe diameter for case 3 is comparable
detached from the stack or a flame base trapped in the to some industrial flare diameters), which makes their computer
recirculating flow in the wake of the stack is not sufficient to simulation more challenging than that of the previous cases.
cause significant inefficiencies, which was also pointed out by The simulation procedure employed was essentially the same
Johnson and Kostiuk.3 In fact, this recirculating zone seems to as that discussed. However, larger simulation domains have to
help stabilize the flame so the resulting combustion efficiency be defined for cases 2 and 3 due to their larger scale. Hence,
is higher than expected at high crosswinds. For high momentum the length of the simulation box was set as 5.0, 6.0, and 7.0 m
flames where flame downwash does not occur, the combustion for cases 1, 2, and 3, respectively. This required a larger number
efficiency may be more sensitive to high crosswinds, as shown of cells in the domain. Moreover, the grid concentration around
by Castiñeira and Edgar.8 the burner was increased for cases 2 and 3 in order to capture
in detail the physics of these flames. A schematic representation
5. Larger-Scale Wind-Tunnel Simulation of the grid model from a top view is presented in Figure 9 for
In this section, we apply CFD for the simulation of a set of case 3.
wind-tunnel experiments that have been experimentally studied The final number of grid cells for cases 1, 2, and 3 were
at the CANMET Energy Technology Centre, Ottawa. These 255 520, 566 720, and 988 960 cells, respectively.
experiments are part of a more extensive study of flare The applied models, parameters, and procedures applied in
combustion systems under crosswind conditions in closed-loop Fluent were very similar to previous simulations. Only boundary
wind tunnels, but only three particular cases are considered here conditions differ significantly in order to fit the new physical
for simulation. conditions for fuel and air. The simulation time increased
The experimental setup was similar to the wind-tunnel significantly due to the large number of cells, especially for
experiments of Johnson and Kostiuk,3 so only relevant differ- case 3. Convergence was achieved after progressively increasing
ences will be given here. First of all, the flare test facility in the under-relaxation factors for density from 0.7 to 1.0.
this case was not closed loop, and the methodology for Figure 10 shows the contour plots for temperature (left side)
measuring efficiency was different. Second, the basic dimensions and CO2 (right side) at the middle plane of the simulation box.
for the box section (i.e., the section after the contraction of Unfortunately, photographs are not available for these experi-
Figure 1) were 8.230 m in length by 1.219 m in width by 1.82 m mental flames. A significant downwash can be observed for
in height. Natural gas was used as a fuel, and its composition
was (by volume) 95.33% CH4, 2.1% C2H6, 0.13% C3H8, 1.8% (18) Gogolek, P. Personal communication. 2005.
1706 Energy & Fuels, Vol. 22, No. 3, 2008 Castiñeira and Edgar
cases 2 and 3, which is likely due to the high crosswind velocity We also simulated larger diameter flares under a crosswind
and the larger burner surface where the flame is attached. that were part of a different set of wind-tunnel experiments. In
The velocity vectors for the three corresponding simulations this case, the computational work increased due to the larger
are displayed in Figure 11. As expected, these vectors reflect simulation domain and larger number of cells. Agreement
how the flow moves downward and becomes trapped in the between the experimental results and the simulations was very
recirculation zone. For case 3, most of the flow exiting the good in all cases.
burner moves to this zone. Computational fluid dynamics is an excellent tool to simulate
Table 5 shows the computed mass flows by Fluent at inlet flares burning in wind-tunnel facilities. Use of simulation in
and outlet boundaries for several species involved in the the study of flare units yields economic savings compared to
chemistry of the flame. The last two columns show the expensive wind-tunnel experiments. Moreover, CFD allows for
experimental combustion efficiencies (ηc) based on eq 9 and the possibility of detailed analysis of species concentration
the simulated combustion efficiencies (ηc) obtained with Fluent. profiles and turbulent flow patterns within the flame, which may
High combustion efficiencies are observed in both experi- not be available experimentally.
ments and simulations in spite of the high crosswinds. This
means that most of the fuel is burned within the recirculating Acknowledgment. The authors thank the Texas Air Research
flow in the wake of the stack. Clearly, the agreement between Center for its support of this research and Dr. P. Gogolek of the
the experimental results and the simulation data is excellent, CANMET Energy Technology Centre, Ottawa, for providing the
and this agreement is even better than in previous wind-tunnel selected data from the Flare Test Facility. They also want to thank
cases. The reason for this may be that turbulence boundary Dr. David Schowalter (Fluent) for his help and recommendations
conditions were known for these CANMET wind-tunnel experi- in implementing the software.
ments. In fact, for the wind-tunnel simulations studied in section
4, it was necessary to add artificial turbulence because of the Nomenclature
perforated plate at the experimental setup, which may introduce A ) area projections over the faces of the zone
some errors in the computations. D ) burner diameter
h ) enthalpy
6. Conclusions bJi ) diffusion flux of species i
k ) kinetic energy of turbulence
CFD analysis has been successfully applied to simulate a set Keff ) effective conductivity
of closed-loop wind-tunnel experiments in order to understand LK ) Kolmogorov length scale
industrial flare behavior and potential gas emissions from these LR ) reaction length scale
units. The laminar flamelet model was shown to be accurate mi ) mass flow rate of species i through a boundary
enough to capture the most important details of the flame. A p ) pressure
thin boundary layer was defined to simulate flame downwash, R ) momentum flux ratio
a phenomenon associated with low-momentum flares under high Ri ) net rate of production of species i by chemical reaction
crosswind. Vj ) jet exit velocity
U) crosswind velocity
Results show that high crosswind velocities affect the
νj ) ensemble-averaged velocity vector
resulting combustion efficiency (in terms of total combustion ν′ ) turbulent fluctuation of the velocity vector
to CO2). However, for low momentum flares (e.g., low gas/ x ) axial position
crosswind ratios) such as these ones, a significant amount of xi ) mass fraction of species i
fuel may be trapped in the recirculating zone near the burner χ ) species mass fraction
wall, which helps stabilize the flame and avoids fuel stripping δij ) Krönecker delta
away from the burner exit without burning. This makes the ε ) rate of turbulent kinetic energy dissipation
resulting combustion efficiencies higher than expected. For high- ηc ) combustion efficiency
momentum flares (e.g., high gas/crosswind ratios), the flame is µ ) dynamic molecular viscosity of the fluid
unlikely to attach to the stack, so lower combustion efficiencies F ) density of the fluid
νt ) eddy kinematic viscosity
are observed even at high crosswinds.8 Both experimental results
and simulations confirm this. EF700545J