DYNAMIC VISUALISATION OF THE COMBUSTION PROCESSES IN
BOILERS
Marek Gayer
1
, František Hrdlička
2
and Pavel Slavík
3
Department of Computer Science and Engineering
Czech Technical University, Karlovo náměstí 13
121 35 Prague 2
Czech Republic
xgayer@fel.cvut.cz, slavik@cslab.felk.cvut.cz
http://www.cgg.cvut.cz/~xgayer/
ABSTRACT
This paper focuses on the simulation and visualisation of coal combustion in the pulverised coal boilers. It
is important to find optimal boiler configurations (both for the ecological and economical reasons),
determine appropriate combustibles, optimize process of combustion, etc. These tasks are typically solved
using traditional Computational Fluid Dynamics (CFD) methods that are in general computationally very
expensive. Our work is based on a different approach. We use simplified methods for determining
direction and speed of air stream in particular places in the boiler. Further we use simplified methods for
the simulation of combustion processes and heat transfer as well. A particle system is used to simulate
and visualise the behaviour of the coal particles and air streams in voxelized boiler space. We developed
concept of virtual particles – they represent certain amount of coal, air, ash and other materials in a voxel
under investigation.
Keywords: FLUENT, visualisation, fluid, CFD, combustion, pulverized coal
1
Department of Computer Science and Engineering
2
Faculty of Mechanical Engineering, Department of Thermal and Nuclear Power Plants
3
Department of Computer Science and Engineering
1.
INTRODUCTION
This paper describes our current work of
visualisation of combustion processes in pulverized
coal boilers [Hrdl96]. The goal is to improve design
of the boilers - to reduce pollution, find ways of
preparing fuel, determine particle sizes and quantity,
speed etc. In engineering practice, it is very difficult
to investigate the combustion processes of various
kinds of combustibles directly in the boiler. Rather
than constructing real boilers and trying to check
and improve these characteristics „on the fly“,
computers are used to experiment with models of the
boiler.
Therefore the efficient boiler design is
based on simulation models. The models that
simulate combustion processes are of various types.
One type of models deals with simulation and
visualisation of behaviour of flames. These models
use several approaches like cellular automata
[Takai95] or diffusion processes [Stam95]. Models
of this type are more concentrated on the
visualisation part of the combustion process. The
resulting pictures can be used in applications where
the quality of visual effect plays decisive role (e.g.
movies etc.). In our case we investigated the
approach that simulates and visualises the
combustion process from the point of parameters of
the combustion process. These parameters can be:
temperature achieved in various parts of the boiler,
speed of air and gases in the boiler during the
combustion process and similar aspects that would
help the boiler designer during the boiler design.
Some well-known algorithms and
technologies for solving this problem based on CFD
[Anderson95] have been already developed. CFD is
a sophisticated analysis technique. It not only
predicts fluid flow behaviour, but also heat transfers,
mass, phase change (such as in freezing or boiling),
chemical reaction (such as combustion), mechanical
movement (such as an impeller turning), and stress
or deformation of related solid structures (such as a
mast bending in the wind). Using current CFD
methods, we are able to solve only some specific
cases with simplified boundary conditions.
Nevertheless they sufficiently cover our needs.
Slice of the temperature array in a boiler
Figure 1
2.
FLUENT
FLUENT is the most known and respected universal
CFD application for modelling fluid flow and heat
transfer in complex geometries. Currently, FLUENT
is one of the most used professional systems for
CFD.
Geometry modelling in FLUENT is based
on constructing a mesh for object (e.g. boiler).
Supported are 2D and 3D meshes. The second step
(after defining a correct mesh) is to define boundary
conditions – walls, inlets, outlets, and physical
properties and models of used materials and
environments. The FLUENT offers excellent ways
of visualisation of computed results. Various
conditions such as temperature arrays, mass tracks
and heat flux could be displayed (see Fig. 1. and Fig.
2.). Using a special pre-processor – PREPDF – the
system can be applied for solving coal combustion
computation tasks.
We use FLUENT results as reference ones
to verify our results.
Sample visualisation of the vectors of the air stream speed in a boiler
Figure 2
3.
OUR WORK
Current CFD effort is based on solving complex
differential equations (such as the Navier-Stokes
equations). Computation time needed for solving
non-trivial tasks is counted in hours and days, even
on a very powerful system.
In no way, current CFD methods can be
used for dynamic real-time computations and
visualisation with ability to change boundary
conditions online. These real-time simulations are
often needed to determine dynamic characteristics of
the boiler for the transition to the stable state,
synoptically display the flow of fuel and air etc.
The aim of our research is to develop a
much faster system (though less accurate), based on
a completely different approach, than the
computation of the differential equations. It should
even allow dynamic visualisation of the combustion
process in the real-time.
4.
PRINCIPLES
The main principles, by means of which our
methodology enables dynamic visualisation, could
be summarised as follows:
Particle system – is a common method for
visualisation of fuzzy objects (e.g. clouds, water, and
fire) in computer graphics. It is also used for
industrial technology [Rhodes98]. In our case, the
application of particle system represents a real
technological problem. The simulation and
visualisation using particle system is divided into
separated steps. For simplicity and maximum
computation speed, this part and all the other parts
of our system are implemented only in the 2D space.
Pre-calculated vectors of motion - (flow
array) – (See Fig. 3 and Fig. 4) dramatically
increases visualisation and computation speed. The
particles are moved only on the pre-calculated
trajectories. These trajectories are computed only
once at the beginning of simulation. They are
represented as a floating-point data structure – called
Flow array. The size of the flow array in our case is
typically calculated and visualised 32 x 32 elements.
This array divides the area of the boiler into mesh of
squares (voxels in 3D interpretation see Fig. 4).
5.
INPUT
Our system does not depend on some specific tasks
and boiler configurations. We use small flat text files
to describe geometry representation of the boiler and
to configure air and coal jets. This allows us to solve
boilers of different shape. It allows us to develop
some visual editor. It could be useful for
constructing and editing tasks without need of
changing current source codes of our system later
on.
Visualisation of Flow array in the test boiler
Figure 3
Velocity and direction of the vectors of velocity in
the Flow array
Figure 4
6.
FLOW ARRAY GENERATION
There are many ways to obtain flow array. The
classical way is based on the differential equations.
Since we try to reach the maximum simulation
speed, we do not use this approach.
Instead of that, we use isotherm, loose flow
that runs from a circle jet. The air stream flows
through jets to the boiler. The solution of the streams
in a limited area in the boiler is quite complex,
especially for the non-isothermal flow. That is why
we calculate it as an isothermal free stream that
flows from the circle jet. Our solution is based on
the G.N.Abramovič’s idea that can be found in the
[Cihe69].
The air stream forces to move surrounding
air under the influence of the turbulence. This
approach allows us to speed up considerably the
calculation of flow array. The stream can be
considered as a cone. The top angle 2α depends on
the level of the turbulence of the stream in the jet,
see Fig 5.
x
y v
xy
v
x
y
tryska
a=0,07
'
50
26
2
°
=
α
x
Isotherm free stream flowing from the jet
Figure 5
For any distance x from the input of the
stream, the maximal speed in the x-axis and y-axis is
decreasing to zero with the Gauss distribution (see
Fig. 6) described as (Eq. 1):
2
2
2
)
(
2
1
)
(
σ
µ
π
σ
−
−
⋅
=
y
e
y
f
(1)
x
f(y)
µ
Graph of the Gauss distribution
Figure 6
The x-axis speed gradually decreases (see Fig. 7)
and is determinate by the Eq. 2. See Fig. 8 for an
illustration of all the above given approach.
145
,
0
48
,
0
0
0
+
⋅
=
d
x
a
v
v
x
(2)
0,335
6
3
0,5
0
1
0
v
v
x
0
d
x
a
⋅
Curve of the axial speed v
x
Figure 7
Distribution of the speed of the air stream in the
space, see [Faltyn99]
Figure 8
However, there may appear a situation,
when a stream collides with the wall and/or there
may occur a collision with another air stream. In
such a case, other virtual jets are added to handle
this situation and to match the real situation. Detail
description of this situation exceeds the scope of this
paper.
7.
THE PARTICLE SYSTEM AND VIRTUAL
PARTICLES
For our work, the particle system allows us
computation and visualisation of mass elements in
the boiler. The particles displayed and calculated do
not correspond to the real coal particles in the boiler.
Instead of that, they represent some corresponding
mass of coal in the voxel under investigation.
Various particle types represent proper amounts of
air, gas, ash and other materials in the boiler space.
Therefore, we call them virtual particles. Thus, one
virtual coal particle carries many real coal particles.
The quality and speed of simulation and
visualisation could be altered by increasing or
decreasing the amount of these virtual particles.
The movement of the virtual particles is
strongly determined by the flow array. For each
particle, the new x and y position, according to air
speed current in the current voxel is computed. The
magnitude of the speed is time dependent.
There are some factors that could change
their motion. For coal particles, we cannot omit the
force of gravity. The coal particles are attracted to
the bottom of the boiler. The acceleration is
determined by the weight of the particle and by the
surrounding environment. Before moving a particle
to the predicted destination, we must check for
possible collisions with the wall. For each voxel, we
have list of walls the voxel interferes with it. We
first determine, in which voxel the particle is
located, and according to that we check for possible
collisions. If this is the case, the particle track is
mirrored and bounced from the wall (See Fig. 9).
Particles are generated from the jets (usually
installed in the walls of the boiler).
β
β
wall x
Particle before bounce
Particle after bounce
C[x
t
,y
t
]
C[x
t+1
,y
t+1
]
C[x
’
t+1
,y
’
t+1
]
A particle bouncing from the wall
Figure 9
Currently, we are using a simplified model
of particle system, because we ignore collisions
between each single particle, from which the final
motion is calculated. Sample visualisation of our
particle system can be found in Fig. 10
8.
COMBUSTION AND HEAT TRANSFER
Combustion process of the coal particles is in fact a
quite complex problem [Dibble96]. Again we use
some simplifications due to the need for the fast
computation. In each step we compute a temperature
array. It contains weighted average of the particle
temperatures for all the voxels. To start combusting
coal, two conditions must be satisfied: in the voxel
there must be at least some minimal combustion
temperature (which is defined - in our example we
use 300 K), and a proper mass of coal and air
(represented by virtual particles) that is to be burned.
Particles flowing from a jet
Figure 10
Depending on the current temperature,
weight and proportion of the coal, the coal particles
are being burned. For air particles, we just decrease
their appropriate mass. If the mass of the air particle
reaches some minimal value, we remove this particle
from the system. For coal particles, we decrease the
amount of the combustible part of the particle, and
increase the amount of the gas burnt. If the mass of
the combustible part reaches some minimal value,
we assume that the coal particle is burned out and
we change it to the burnt gas particle. This process is
shown in the Fig. 11 and Fig. 12.
The situation in a voxel before the combustion
process start
Figure 11
T = 303K
(above ignition temperature)
t = 0 seconds
Coal particle
Air particle
Partially burned coal particle
A
C
C
C
A
A
A
The situation in a voxel after the end of the
combustion process
Figure 12
Between these processes, depending on
reaction heat transfer, the released heat is transferred
to all the particles, which are present in the current
voxel. Therefore, the temperature of the voxel
increases.
Because of the dynamic processes in the
boiler, the heat is distributed by the moving particles
to the other voxels, thus increasing the temperature
and making possible to start another combustion
reactions. We also count with the heat radiation
between the walls and the mass in the voxels. The
heat transferred from the given surface F during the
time dt is comparable to differences of the
temperatures of the wall and the voxel (power of
four) [Dibble96]. We also need to determine the
coefficient of the radiation C
12
. These ideas are
summarized in Eq. 3.
dt
T
T
C
F
Q
⋅
−
⋅
⋅
=
4
2
4
1
2
.
1
100
100
(3)
We assume, for the sake of the simplicity, that the
temperature of the walls is constant (typically bellow
the minimal combustion temperature). In general, it
is possible to say that our model is based on mutual
reactions of virtual particles of various types in each
voxel in the boiler space.
9.
INITIAL CONDITIONS AND
INITIALIZATION
To start combustion again, there must be already
some appropriate condition in the boiler. It is
necessary, because air and coal, which is coming
through the jets, are not usually heated enough.
Their temperature is under the ignition point. Thus,
there must be a way to start combustion. We assume
in the current implementation, that there are already
some air particles warmed up above the ignition
point, which allows the ignition.
10.
IMPLEMENTATION
All the parts of our system have been implemented
in the standard, ANSI C language. Visualisation is
based on the OpenGL graphics interface.
Windowing interface is maintained by the GLUT
library [Mason 98]. Thanks to this, our system is
easily and fully portable to other systems. No
problems should occur porting to Linux systems or
even SGI workstations, although it was originally
developed on the Windows NT/2000 platform. We
have tested our system modelling a boiler with real
dimensions, characteristics and parameters. The
behaviour and results gained from our system were
well comparable with a situation in a real boiler.
Thus, the current implementation is correct,
although, there are still many things to improve.
11.
VISUALISATION
To maintain the reliable and fast visualisation, our
system uses industry standard OpenGL platform.
Thanks to this, nowadays, our system could be used
on a standard, even a cheap graphics accelerator.
There is no big lack of speed in particle visualisation
even when using 10 000 of particles. Furthermore,
our system uses MGL graphics library on the
backend to maintain easier visualisations of common
OpenGL primitives [Gayer00]. That it is an
OpenGL based library optimised for visualisation of
common 2D graphics and graphical user interface
(supports images and fonts directly). We use it to
easy implement an easy user interface, in common
2D coordinates. Therefore, such tasks are much
easier to program than in a native OpenGL.
The boiler walls and outlets are
approximated by the straight lines. The particles are
displayed using standard OpenGL pixels. For the
examples of our current graphics output, see Fig. 3,
4 and 10. The selected local characteristics in the
voxel, such as the total temperature, mass storage,
the wattage, and heat flux state and/or changes can
be in the real-time visualised (see Fig. 13). The
particle tracks can be easily determined by the fast
particle system animation. Currently, the
characteristics in a voxel are simply visualised by
the quads. Although the quality of the visualisation
could be improved by choosing smaller voxel sizes,
we plan to implement contours to bring smoother
graphics output.
12.
RESULTS
The current research brings promising results. On a
test boiler (dimensions 6m x 13.7m) we have
simulated and visualised combustion processes. We
have discussed the obtained results with the experts
from the Faculty of Mechanical Engineering of CTU
T = 305K
(increased)
t = 0.01 seconds
Coal particle (partially burned)
Air particle (decreased m)
Coal particle transformed to
burned gas particle
A
C
B
C
A
A
with positive response. To compare our results with
current CFD methodology, we used the FLUENT
solver.
Sample visualisation of the total voxel temperature
in the test boiler
Figure 13
The global parameters, which could be
easily compared, match well overall design and
implementation of our ideas, see Table 1.
Parameter
Our system
FLUENT
Average
temperature
1029
o
C
1158
o
C
Outlet temperature
1151
o
C
1384
o
C
Maximum
temperature
2360
o
C
2753
o
C
Average stream
velocity
23 m/s
17 m/s
Average outlet
velocity
28 m/s
21 m/s
Wattage
192 w/m
3
232 w/m
3
Mass total
21.1 kg
21.3 kg
Time needed to
converge solution
12 seconds
7 hours
Global parameters results in the test boiler
Table 1
Next we compared the images of the temperature
and velocity maps, which summarize local
characteristics. We found they are visually similar.
13.
CURRENT IMPLEMENTATION SPEED
The current implementation is very fast. We tested it
on different systems. We measured the number of
the frames (images) which our system was able to
compute and display per second (FPS). On each
system we generated 10.000 particles, and we
measured the FPS value, see table 2.
System
Frames
per sec.
Celeron 300, no hardware OpenGL
accelerator, 64 MB RAM
2
Celeron 400, S3 Savage 4, 128
MB RAM
26
AMD Athlon 1333, nVidia
Geforce 2 MX 400, 256 MB RAM
64
Simulation speed on different systems
Table 2
On an average system (Intel Celeron 400,
128 MB RAM we can reach 26 frames per second
with 5000 particles computed and visualised. Note
that on the Celeron 300 system the speed of the
simulation rapidly decreased due to lack of the
OpenGL accelerator. However, it should not be a
problem, because almost every new computer
system is equipped at least with a cheap graphics
accelerator (such as Savage 4), which is sufficient
for our system.
14.
ADVANTAGES AND DISADVANTAGES
As it is obvious from the previous text, the
main advantage is the speed of computation. The
speed is far beyond of reach offered by the
traditional methods. Thus, the developers of the
boilers could test many configurations and
modifications of the boiler with the immediate
response. This results in the possibility to get a very
good preview of the dynamics of combustion
processes in a boiler. This is not available in the
traditional approaches. Thus, our system could also
be used for experimentations and educational
purposes in the field of the combustion processes.
Although it gives fast and reliable results,
there still would be necessary to test and compare
more deeply the gained results with results gained
from CFD systems. So again, the main advantage
against less accuracy and features offered by the
CFD systems is the speed of computation. The next
advantage is the possibility to visualise the state of
the particles during the combustion process.
15.
FUTURE WORK
The future work will be concentrated on:
•
We plan to implement more accurate heat
distribution.
•
We will develop a methodology for
verification of our results with the CFD
computations.
•
We plan to simulate and monitor additional
characteristics (pressure, turbulence, etc.)
•
Much interest we will probably give to the
computation of the Flow array. While this
is computed only once, with no influence to
the speed of the other computation, there
could be used some more complex
algorithms to compute. We plan to include
influence of the mode of combustion also.
•
Some advantages could be gained through
the visualisation itself. We plan to use
linear approximations to convert attributes
from the singular points to the continual
array. It would be useful for visualisation of
the temperature map.
16.
CONCLUSION
For combustion system visualisation, current
investigation brings an interesting alternative to the
classical CFD applications. We implement a brand
new way based on the fast computation of the flow
arrays. We use simplified model of combustion
process and light-speed visualisation using OpenGL
graphics interface. Current implementation is very
fast even on average systems. The behaviour and
results gained from our system were comparable
with a situation in the real boiler and FLUENT CFD
system. Therefore, our system can be used for
dynamic simulation or preview of dynamic
processes of coal combustion in boilers. This may be
used in the CFD process for the fast and efficient
design for the boilers. The system could also be used
for education purposes in order to give students idea
about the behaviour of boilers under various
conditions.
REFERENCES
[Anderson95]
Anderson, J.: Computational Fluid
Dynamics – The basics with the applications,
McGraw Hill, 1995.
[Dibble96] Dibble, W., Maas U.: Combustion,
Springer, 1996.
[Cihe69]
Cihelka, J.: Vytápění a větrání. SNTL,
Praha, 1969.
[Faltyn99] Faltýn, R.: Diploma work / Diplomová
práce, ČVUT FEL Praha, 1999.
[Gayer00] Gayer, M.: Graphical library MGL,
part of the diploma thesis Programové
knihovny pro grafické akcelerátory (in
Czech),
ČVUT FEL Praha, 2000.
http://sgi.felk.cvut.cz/~xgayer/mgl/
[Hrdl96] Hrdlička,F., Janeba,B.: Fluid
combustion, Energie (Vol 1, pp. 50), 1996.
[Mason98] Mason, W.: OpenGL Architectural
Review Board, Jackie, N., Davis, T., Shreiner,
D.: The OpenGL Programming Guide.
Addison-Wesley, 1998.
[Rhodes98] Rhodes, M.: Introduction to particle
technology, John Wiley & Sons Ltd., 1998.
[Stam95] Stam J., Fiume, L.: Depicting Fire and
Other Gaseous Phenomena Using Diffusion
Processes, Proceedings of SIGGRAPH 95,
Computer Graphics Proceedings, Annual
Conference Series, pp. 129-136, 1995.
[Takai95] Takai, Y., Ecchu, K. Takai, N.: A cellular
automaton model of particle motions and its
applications, The Visual Computer, 11(5), pp.
240-252, 1995.