Académique Documents
Professionnel Documents
Culture Documents
Summary:
An industrial cyclone, which is used for dust separation from a hot gas mixture, is studied by means of
numerical simulation. Because of high gas temperatures (about 1075 K), the inner surfaces of the
cyclone separator and the inlet pipe have to be lined with refractory bricks, which cause high wall
roughness. The virtual wall model is used to simulate this roughness, and a new model extension
accounting for joints between the bricks is proposed and validated by dedicated experiments.
As the cyclone inlet mass load is relatively low (0.01 kg/kg), the Eulerian-Lagrangian approach is used
for the simulations. The continuous phase conservation equations are solved in the Eulerian reference
frame on a fixed grid, whereas the discrete phase properties are resolved by computing particle
trajectories through the computational domain in a Lagrangian reference frame.
Besides an inlet particle diameter spectrum ranging from 1 µm to 100 µm, also different particle
materials with different particle densities have to be considered. The measurements and the analytical
Muschelknautz model are compared to the numerical simulations regarding pressure loss, particle
spectra at the outlet and the fractional separation efficiency.
The Muschelknautz model shows some difficulties with the kind of wall roughness considered here
(especially joints between the bricks). Additionally, this method was not designed for different particle
densities simultaneously.
Nevertheless, comparisons show a good agreement between numerical simulations, measurements
and theory.
Keywords:
3.5
Rep = 30
3
2.5
CL
Rubinow, Keller
2
Rep = 50
1.5
1
Rep = 100
0.5
Rep = 140
0
-1 0 1 2
10 10 10 10
Ω
Fig. 2: Rotational lift coefficient C LM according to the measurements of Oesterlé and Bui Dinh, [7],
and analytical results of Rubinow and Keller, [9]
Thereby, two approaches are possible: in the top-down strategy, one can provide appropriate
distributions for particle’s translational and rotational velocities and the rebound angle after an
encounter with a brick joint. Many parameters play a large role for the creation of such distributions:
- particle’s velocity and rotation before the impact,
- angle of impact relative to the wall,
- joint dimensions (width, depth),
- material combination of particle and wall (restitution and friction coefficients),
- wall roughness etc.
Besides of the parameters listed above, a particle in a joint is usually not facing only single, but
multiple collisions. Very comprehensive measurement and simulation series would therefore be
needed in order to cover the parameter area accordingly and to collect sufficient data for a reasonable
statistical description of the joint occurrence. For this reason, the top-down strategy is refused.
The brick joint model developed in the frame of this work is rather based on the bottom-up strategy:
each encounter of a particle with a joint is simulated by several wall collisions. The probability of an
impact is defined by the joint and brick dimensions (Fig. 3),
a
Pjo int = . (2.6)
a+c
c
Fig. 3: Brick joints
Thus, the decision whether a particle is going to impact a joint or not is made upon a random number.
If this case arises, it can be assumed that every point of impact along the joint width has an equal
probability, compare big dots in Fig. 4 a.
In the first step of a particle-joint encounter (Fig. 4 a) it has to be determined which side of the joint the
particle is going to collide with first. For this purpose, the joint aspect ratio is needed (Fig. 3),
The geometry of the hot gas cyclone is shown in Fig. 5 a. Cyclone dimensions (see Fig. 1) are listed
below:
- Total height htot = 11.8 m
- Height of the cylindrical part hcyl = 3.8 m
- Height of the conical part hc = htot − hcyl = 8 m
- Cyclone radius ra = 1.8 m
- Radius of the vortex finder ri = 0.72 m
- Height of the vortex finder hi = 2.9 m
- Radius at the lower outlet r3 = 0.77 m
- Cone angle ϕ = 0.128 rad = 7.3°
- Inlet height he = 2.15 m
- Inlet width b = 0 .6 m
A gas mixture flows through the inlet pipe and enters the cyclone with the mass flow rate of
m& g = 30.88 kg/s. The gas temperature at the inlet amounts Tg = 1075 K, and the gas density is
ρ g = 1.31 kg/m³.
The particulate phase is composed of two different materials which occur in equal parts. The particle
densities are ρ p1 = 1100 kg/m³ and ρ p 2 = 2800 kg/m³. For the analytical cyclone dimensioning
according to Muschelknautz an average particle density of ρ p = 1950 kg/m³ is assumed, whereas in
numerical simulations it is possible to account for different particle densities.
Simulations are carried out on a numerical grid with 230 000 hexahedral cells. For the turbulence
modelling, the Reynolds Stress Model (RSM) is used. In contrast to the one- and two-equation
models, RSM does not use the turbulent viscosity hypothesis. This turbulence model rather solves the
transport equations for the Reynolds stresses ui'u 'j (averaged velocity fluctuations) and for the
turbulent dissipation rate ε. This requires the solution of seven additional transport equations, which
d p (µm) 1 2 4 7 10 20 50 100
Percent by weight (%) 1 2 3 4 10 31 35 14
m& p (kg/s) 0.0015 0.0031 0.0046 0.0062 0.015 0.0479 0.054 0.0216
Table 1: Simulated particle diameters and mass flow rates
4 Results
Tangential and axial velocity profiles on the intersection lines between vertical plane and planes 02÷05
(see Fig. 5 b) are shown in Fig. 6 and Fig. 7. Tangential velocities show the typical characteristic for a
cyclone separator. The largest tangential velocity lies under the wall of the vortex finder. The velocity
decreases almost linearly toward the cyclone centre (solid body rotation). Beyond the region of strong
swirling flow and toward the cyclone’s outer wall, tangential velocity decreases in a manner of the
loss-free rotation (Rankine vortex). Due to increased wall roughness and the influence of joints the
movement of solid particles becomes more irregular thus decreasing the gas tangential velocity (Fig.
6).
Axial velocities rather show an asymmetrical behaviour, especially in the lower cyclone part. This
points out that the inner cyclone swirl is not completely concentric. It rather meanders and spins
relatively slowly around the cyclone axis. Bigger wall roughness increases the positive axial velocities
(under the vortex finder). This can be explained through the attenuation of the inner swirl.
One can generally say that larger wall roughness intensifies the axial flow inside the cyclone
(downwards at the outer wall, upwards in the cyclone core). The swirling flow is however weakened,
and this makes a negative influence on cyclone’s separation efficiency. By the weakening of the eddy
inside the cyclone the pressure loss also sinks.
Fig. 8 shows the particle concentration (kg/m³) at the cyclone inlet. The arrows mark the secondary
flow of the gaseous phase. In Fig. 8 a, there are no particles in the simulation, so only secondary flow
vectors are drawn. The solid particles distribution at the cyclone entrance is very important for the
general separation behaviour. For this reason the cyclone is simulated including its long inlet pipe (Fig.
5 a). As is shown in Fig. 8, the previous history can very well affect the particle behaviour.
Higher discrete phase concentration near the cyclone’s upper wall results from the inlet pipe
previously mentioned. Smooth walls from the simulation in Fig. 8 b do not change much in this
circumstance, the particles enter the cyclone at the conceivably most unfavourable position, i.e. at the
upper wall and on the inner side of the inlet, near the vortex finder. In the simulations with rough walls
and joints, respectively, the particle concentration is displaced towards the outer wall because of
increased particle dispersion. Also, the intense secondary flow is alleviated.
35 35
30 30
25 25
utan
utan
20 20
15 15
10 10
5 5
0 0
-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5
ra ra
25 25
utan
20
utan
20
15 15
10 10
5 5
0 0
-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5
ra ra
Plane 02 Plane 03
15 15
10 10
5 5
uax
uax
0 0
-5 -5
-10 -10
-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5
ra ra
5 5
uax
uax
0 0
-5 -5
-10 -10
-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5
ra ra
0.9
0.8
0.7
0.6
*100%
0.5
0.4
0.3
Measurement, inlet
0.2 Measurement, outlet
Muschelknautz
0.1 Simulation without joints
Simulation with joints
0
-6 -5 -4 -3
10 10 10 10
dp
Fig. 9: Particle spectra: measurement, Muschelknautz method and simulations
The reason for this lies in the weakening of the inner cyclone swirl through very high mean wall
roughness (which is specified in order to account for the joints). With lower wall roughness (no joints),
the overall separation efficiency (and the pressure loss, too) would increase, the outlet particle
spectrum would be shifted towards smaller diameter (to the left in Fig. 9).
The results of numerical simulations (Fig. 9, Table 2) predict rather good cyclone performance. Apart
from the unsatisfactory description of the particle-turbulence interaction (Random Walk Model in
Fluent), the causes can also be sought after in other disadvantages of the Discrete Phase Model:
relatively small number of simulated particle trajectories per particle diameter, density and time step
(due to computing time), very limited possibilities of consideration of particle relaxation times (due to
the dimensions of the simulation domain and the number of trajectories) etc.
5 Conclusion
A brick-lined hot gas cyclone separator is investigated by means of numerical simulation. Thereby
different models accounting for wall collisions, brick joints and rotation of discrete particles are
implemented and tested. The comparison is made to the experimental measurements and the
modelling method by Muschelknautz. Both, the numerical simulations and the analytical method, have
their pros and contras.
The simulations, on the one hand, are well capable of calculating the wall collisions and the brick
joints. There is, however, the need for considering the particle-particle interactions, even for such
small mass loading ratios as is the case in this work. Thus the four-way coupling mechanism is
unavoidable for cyclone separators in order to resolve the particle strands properly.
The analytical modelling method, on the other hand, does not consider the influence of the particle
behaviour before the cyclone separator and the way in which the particles enter the cyclone. It is
further not capable of accounting for brick joints inside the cyclone body. Instead of joints, the mean
wall roughness can be increased, which has a negative influence on the overall separation efficiency.
However, the analytical method accounts for the formation of ropes and strands and is therefore an
essential tool for the comparison with the numerical methods.
Further improvements of numerical models, e.g. the four-way coupling mechanism and the particle-
turbulence interaction, are planned for the future. If this succeeds, the numerical simulations are
believed to be clearly superior to any analytical modelling methods.