Low-NOx swirl burner

 

Low-NOx burner to be simulated by UZ

 
 
Introduction

 

The test case chosen by UZ is a swirl-type burner. This is a piece of equipment encountered (in different variants) in many Engineering devices in which energy needs to be generated, including domestic boilers, power-station furnaces, industrial ovens and mobile and stationary gas turbines.


Broadly speaking, swirl burners inject a fuel axially into the combustion chamber surrounded by an annular flow of oxidant (air normally) which has, upon injection, certain tangential momentum (‘swirl’). This rotational component, together with the usually divergent geometry of the burner mouth, cause two important effects: firstly, they promote intense mixing between fuel and air, which is important for an efficient and stable combustion, and low emissions; and, secondly, they originate a recirculation region, just at the burner mouth, which traps hot combustion products and acts as a permanent ignition source, hence promoting the stability of the flame. These mechanisms are all illustrated in more detail in the following section.


This basic swirl-burner fluid-dynamics can be modified in multiple ways. For instance, the fuel can be a liquid instead of a gas, and can be either pre-vaporised (as in some modern aero gas turbines) or atomised directly at the burner outlet. For gas burners, the designs can be sophisticated geometrically to achieve a more efficient and cleaner combustion. For instance, the fuel can be partially premixed, and/or injected at several locations. The combustion air can also be injected in sophisticated patterns (eg, two co- or counter-rotating annuli). These sophisticated geometrical designs are the core ingredients of some modern technological variants, such as the ‘fuel-staging’ and ‘air-staging’ concepts.


The need for more efficient and cleaner is also, in the competitive power-generation market, the driving force for the use of CFD for burner design. The well-known benefits of computer modelling of fluid flow are of course also applicable to this field. However, in addition to the geometrical complexity, just mentioned above and further illustrated later, CFD design of combustion equipment has the additional difficulty posed by the modelling of turbulent combustion. Here, the correct modelling of turbulence and chemical reaction, and more importantly of the coupling between them, is crucial for many design goals. The challenge not only defies the abilities of many scientists worldwide, but is also extremely demanding in CPU power. For this reason, these problems are excellent targets for Grid-computing infrastructures.


The system chosen for the current validation exercise is a burner partially designed and built, and comprehensively tested experimentally, at the LITEC combustion laboratory (associated to UZ). The design, performed in the early 90’s, resulted in a burner which produced very low emissions (unburned fuel, CO or NOx) when operating with either natural gas or oil. As a consequence of this local designing and testing, we have been able to retrieve from reports and storage the geometrical features, operating conditions and even experimental data for a number of different geometries and scenarios. The experimental campaign was conducted in the LITEC single-burner, down-fired, cylindrical combustor, which will be described later. This combustor is the aim of the FlowGrid model, with the interior of the burner itself being modelled to provide detailed knowledge of the flow at the burner exit (chamber inlet).


It will be seen that the intricacy of the inlet and the complex physical processes result in problem sizes which make the case a good testbed for grid computing. In respect of geometrical complexity, we have retained most of the geometrical features of the burner. The burner geometry is three-dimensional (two-dimensional reductions are possible for testing at the expense of a loss in realism). Because the geometrical features entail a spread in length scales (eg, secondary-fuel-nozzle diameter vs chamber diameter), the meshing is demanding. Regarding physical models, we have explored several alternatives in order to assess computer times and quality of results. Thus, both k-epsilon and Large-Eddy-Simulation turbulence models have been tried; and the range of combustion models range from the classical and simple eddy-breakup one to the more sophisticated flamelet models. The main guiding principle of this validation exercise is to test the benefits of grid computing, and not to obtain at any expense the best possible results using sophisticated physical models which would require significant time to be implemented in FlowGrid.

 
 
Detailed descriptions

 

As it has been indicated above, the test case for UZ is a low-NOx experimental burner, with a thermal power of 0.35 MW, which is firing steadily into a cylindrical combustor chamber.


The description of the test case below is divided into two parts. In the first part, the geometry of the test case is described. Later, the operating conditions (such as flow rates, inlet temperatures, etc) are defined

.
The next figure shows a schematic of the device. Two zones are distinguished for the purpose of the present report: the burner and the combustion chamber. The burner provides gas and air flows to the combustion chamber, while in the combustion chamber the combustion takes place. The heat released by combustion is partly extracted by the walls of the combustion chamber, which are refrigerated by water, and partly by the exhaust gases which leave the domain at the bottom.


Scheme of the combustor simulated


The next figure shows a detailed scheme of the burner and its dimensions. As indicated in the Introduction, the burner has an advanced design, aimed at producing very low emissions. This translates into a multiplicity of feeding streams, as follows. The primary fuel is injected in the centre of the burner. Around the primary fuel, the primary air is injected. Part of the primary air passes through the swirler, which imparts a rotational component of velocity to this stream. A secondary fuel-stream is injected in the outer part of the burner through several ducts ending in small orifices (2 mm diameter). Between the secondary fuel and the primary air, the inlet of secondary air is located.


Burner schematic.


The next figure shows a schematic of the flame structure originating from this burner. Near the burner mouth, the first combustion zone is located. Here the combustion between the primary gas and primary air takes place. Downstream of this first combustion zone, the combustion of the secondary fuel occurs.


The first combustion zone is characterized by a high availability of air, and a poor mixing of gas and air. Hence, the primary flame is characterised by high temperatures and a high availability of oxygen, which are conditions prone to the formation of NOx.
On the other hand, the second flame features a good mixing between gas and air, due to the longer distance from the injection of the secondary fuel to the combustion zone. Besides, the availability of oxygen is lower due to the dilution with the products from the primary flame. Hence, the second flame has lower temperatures and a lower availability of oxygen. These conditions are conducive to low emissions of NOx.


In summary the first flame produces a strong, stable flame at the expense of a higher NOx formation, while the second flame has a low formation of NOx. To take advantage of the different characteristics of these flames, a third of the gas is used as primary gas, which is burned in a high-formation NOx zone, and the rest is used as secondary gas, which is burned in a low-formation NOx zone. The outcome of this distribution of fuel is significantly-reduced NOx-emissions, compared with conventional burners.


Schematic showing the flame structure


In the next table presents the mass-flow rates for the different streams of air and gas, and their corresponding temperatures.

Stream
Flow rate (Nm3/h)
Temperature (K)
Primary gas
9.28
298
Secondary gas
15.41
298
Primary air
238.05
298
Secondary air
45.34
298

Conditions of operation of the combustor.


This table presents the composition and other properties of the natural gas as fired.

Molar composition (%)
CH4
90.012
C2
8.748
C3
0.709
iC4
0.047
iC5
0.058
nC5
0.005
N2
0.003
CO2
--
PCS (Kcal/Nm3)
10254
PCI (Kcal/Nm3)
9244
Density (Kg/Nm3)
0.7861
Molecular weight
17.567

Composition and other properties for the gas.

Description of relevant parameters to be observed

For the FlowGrid test case, we will not attempt to simulate faithfully every aspect of the flame. This is a difficult task, even with dedicated combustion codes; and would place an excessive emphasis on the provision and evaluation of state-of-the-art physical models, which is certainly not the task of FlowGrid.


The relevant problem parameters to be observed in the FlowGrid test case will be the following:

  1. The system should be able to reproduce broadly the flame structure, and particularly the two distinct combustion zones, as this flame structure is largely the consequence of the burner aerodynamics, rather than the combustion model used. The aerodynamics should be well captured; they are to a great extent determined by the quality and size of the mesh used, and FlowGrid should be capable of affording the user sufficiently-fine meshes to resolve the flow with, at least, classical RANS turbulence models, such as k-epsilon.
  2. The in-flame temperature-patterns should be predicted accurately, since they are indicative of the flame structure and determinant for NOx formation.
  3. Outlet values of O2 are indicative of the degree of combustion, and hence of the efficiency of the burner.
  4. Outlet values of CO and unburned hydrocarbons are also indicative of efficiency; but they are also pollutants, the prediction of which is important for the burner designer.

    Available experimental data

The experimental data for the combustor have been obtained at LITEC, UZ’s associated laboratory for combustion research. As indicated in D8.0, they were obtained prior to the FlowGrid project, and for FlowGrid they have been recovered and appropriately treated so that comparisons with computational results can be readily made.


The experimental data available consists, generally, of velocity profiles for the cold (ie,non-combusting) flow; radial profiles temperature at different axial stations; the concentrations of O2, CO, hydrocarbons and NOx at the outlet of the combustor; and also flame images. Most of this experimental data are available for a wide range of cases corresponding to different operating conditions.


The velocity measurements have been made for cold flow (ie, non-combusting) only. A Prandtl probe, a five-orifice Pitot probe of and a Hubbard probe have been used to measure the velocity. The Prandtl probe measures the velocity only, while the others ones determine the velocity and its direction.


The temperature profiles has been obtained by using thermocouples (Pt/Pt 10 % Rh, type S). The thermocouple wires have diameters between 40 and 70 micron, which limit the error in the measurements to some tens of degrees in the worst case. The temperatures have been measured in radial profiles at 8.9 cm, 16 cm, 23.1 cm, 40.9 cm, 55.1 cm, 72.9 cm, 87.1, 104.9 cm and 136.9 cm from the burner mouth. At each axial station, temperatures are measured approximately every each centimetre in radial direction. Figure 4.5 shows a schematic of these axial stations.



Axial stations for the measurement of temperature
.


The composition of the flue gases is determined by means of a sampling procedure. First, sampling probes are used to extract and quench a flux of gas. Then, the sample is transported and prepared (removal of condensed water and filtration to eliminate particles). And, finally, the composition of the sample is determined by specific gas analysers for each species; for instance, an infrared device is used to measure the concentration of CO. All these processes work in continuous. The next table shows the measured concentrations of different species of the gas at the outlet of the combustor for the case considered for FlowGrid.

 

Composition of the gas
Species
Units
Value
O2
% volume, dry
2.93 %
CO
ppm volume, dry
3.8
HC
ppm volume, dry
0.3
NOx
ppm volume, dry
36.56

Composition of the gas at the outlet of the combustor


The monitoring and visualization of the flame is made with a system that combines cameras, monitors and video recorders. The cameras are situated in a refrigerated device that is separated from the combustion chamber by high-temperature-resistant, transparent glass.

 

 
 
FlowGrid evaluation

 

Aspect: Interactivity
Description: The interaction with the FlowGrid system should be appropriate for the handling of large computational cases, and particularly should allow the user to intervene in the calculation.
Evaluation criteria: The user should be able to monitor, stop and restart the calculation easily, and to selectively retrieve the results.
Acceptance criteria: Features present/absent.

 

Aspect: Accuracy
Description: The FlowGrid calculation should reproduce the main flow features.
Evaluation criteria: Comparison with experimental and computational data as reported earlier in this Section.
Acceptance criteria:
  1. Internal and external recirculation regions of similar location and strength to those found in the experimental data and computational results with third-party software.
  2. Ditto mixing fields, showing the mixing of the two gas streams.
  3. Temperature patterns predicted no worse than the third-party-software predictions for the simpler models, and showing the two combustion zones.


Aspect: Problem size
Description: Ability of the system to handle large datasets and transient calculations, since finer meshes than used for WP2 are needed to accurately capture all the flow features, and the LES calculation is transient in nature.
Evaluation criteria: A case with 5 million cells will be set up, as well as a smaller one (ca. 1 million cells) but transient (eg, LES turbulence model if available in FlowGrid).
Acceptance criteria:

Both cases should run, using faster CPU’s or additional ones, in CPU times no longer than those required in the current facilities for a 900k-cell meshes on Pentium III CPU’s (ie, around 7 days on 6 CPU’s).


Aspect: Problem size
Description: Ability of the system to handle large datasets and transient calculations, since finer meshes than used for WP2 are needed to accurately capture all the flow features, and the LES calculation is transient in nature.
Evaluation criteria: A case with 5 million cells will be set up, as well as a smaller one (ca. 1 million cells) but transient (eg, LES turbulence model if available in FlowGrid).
Acceptance criteria:

Both cases should run, using faster CPU’s or additional ones, in CPU times no longer than those required in the current facilities for a 900k-cell meshes on Pentium III CPU’s (ie, around 7 days on 6 CPU’s).

 

Aspect: Scalability
Description: Good scalability properties with an increasing number of CPU’s for moderate-to-large problem sizes.
Evaluation criteria: A case with approx 5 million cells will be set up and the scalability tested.
Acceptance criteria:

The speedup for such meshes over at least 8 processors should be close to linear, and not significantly worse than it is now for a 900k-cell mesh

 

Aspect: Large-scale problem-solving
Description: The limits to the problem size will be tested.
Evaluation criteria: The largest possible case, over all the available CPU’s in the FlowGrid tested, will be investigated.
Acceptance criteria:

None, given the nature of the exercise.


 

Copyright 2004 FlowGrid Consortium | Please send questions or comments to Norberto.Fueyo@posta.unizar.es, or to any other FlowGrid partner | Last modified on 02/04/2004