You are on page 1of 14

Applied Mathematical Modelling 37 (2013) 7891–7904

Contents lists available at SciVerse ScienceDirect

Applied Mathematical Modelling


journal homepage: www.elsevier.com/locate/apm

Incorporating dust lift-off into a CFD model of a blast furnace


gravity dust-catcher
David Winfield ⇑, Nick Croft 1, Mark Cross 1, David Paddison 2
Postgraduate Engineering Doctorate Research Engineer, Tata Steel Strip Products (TSSP) UK, Swansea University, United Kingdom

a r t i c l e i n f o a b s t r a c t

Article history: A gravity dust-catcher separates a mixture of dusts from the spent top gas flow of a blast
Received 20 December 2011 furnace. These dusts are predominantly made up of limestone, iron ore and coke/coal. As a
Received in revised form 21 April 2013 result of the turbulent gas flow patterns within a dust-catcher, modelling of the flow pat-
Accepted 25 April 2013
tern can be very complex, attributed to the turbulent vortices that can be formed within
Available online 16 May 2013
the main body of the structure. Using data from an experimental prototype test rig, a sim-
ple model to capture the lift-off characteristics of particle lift-off from dust pile surfaces is
Keywords:
created and incorporated into a computational fluid dynamics (CFD) model of the dust-
Dust lift-off
Gravity dust-catcher
catcher.
Lagrangian particle tracking The variation of particle separation performance over a typical blast furnace (BF) opera-
Gas de-dusting tional cycle is analysed. An attempt is made to explain the observed phenomena in terms of
Particle separation particle–fluid interaction. It is found that particle separation efficiency is largely unaffected
Computational fluid dynamics by dust lift-off at low dust-catcher hopper fullness levels, but is significant at higher levels.
It is found that the topography of the dust surface is important when predicting particle
lift-off trends. It is concluded that this is due to the exposure experienced by a given par-
ticle when subjected to a surface velocity.
Ó 2013 Elsevier Inc. All rights reserved.

1. Introduction

Blast furnace dust is a product of the numerous reactions and damage during transport that occurs within the blast
furnace. The dust predominantly contains iron ore, coke and limestone derivatives. Also included are small amounts of zinc,
introduced into the system through the charging of galvanised scrap into the basic oxygen steelmaking (BOS) process. This
BOS dust is then recycled through the sinter plant, where the sinter is charged into the blast furnace. Zinc is classified as a
heavy metal and considered toxic to the environment, this can make blast furnace dust difficult to dispose of. Currently the
dust is disposed of through landfill sites and controlled by tight environmental legislation [1].
Blast furnace dust is made up of particles with diameters ranging from a few micrometres to over 1 mm in size. The large
size distribution is the result of both the damage induced through transport and the reactions that take place within the
furnace between the charge constituents3 and the gaseous components of primarily nitrogen (50%), carbon monoxide
(25%) and carbon dioxide (25%). Sampling takes place in the discharge bin at the bottom of the hopper of the gravity
dust-catcher (primary gas/dust separation), and analyses the size distribution of the separated dust. Clearly, this distribution

⇑ Corresponding author. Tel.: +44 07952055963.


E-mail address: 371961@swansea.ac.uk (D. Winfield).
1
Computational Modelling Group, College of Engineering, Swansea University, Singleton Park, Swansea SA2 8PP, United Kingdom.
2
Senior Engineer, Projects Department, Tata Steel Strip Products UK, Port Talbot Works, Port Talbot, South Wales SA13 2NG, United Kingdom.
3
Inserted via the Paul Wurth top charging system at the top of the blast furnace throat.

0307-904X/$ - see front matter Ó 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.apm.2013.04.042
7892 D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904

is coupled to the granulated coal injection (GCI) rating that the blast furnace operates under. The analysis only considers the size
range distribution of the dust and calculates the percentage of a particular size range present in the dust sample.
Loss from the surface of dust already settled in the dust-catcher hopper is perceived as a major problem by operators and
can contribute to reduced operational efficiencies of the resident dust-catcher and downstream dust cleaning devices such as
the wet scrubber and demister, whilst increasing maintenance costs throughout the gas cleaning plant. Smitham and Nicol [2]
constructed a wind tunnel experiment and analysed the quantity of dust lift-off from the surface of coal stockpiles as a func-
tion/variation of particle size and moisture content. The work demonstrated a close relationship between the mass of coal lost
from the coal bed together with sample moisture content and wind velocity. A simple model was derived to describe the phe-
nomenon of dust lift-off, as well as investigating the lift-off mechanism [2]. Witt et al. [3] have undertaken studies in predict-
ing the dust loss from conveyors using CFD modelling software in a wind tunnel. The work undertaken has shown that a
combination of CFD modelling and experimental analysis is a powerful tool for assessing problems of this nature.
The work reported here is motivated by a requirement at Tata Steel, Port Talbot, to gain a better understanding as to how
dust lift-off occurs in such a large scale industrial process, and in particular to assess the impact of dust lift-off on the effi-
ciency of a blast furnace gravity dust-catcher. The paper is laid out as follows: a brief description of the dust-catcher is fol-
lowed by an overview of the CFD model for predicting the capture of the particulate efficiency together with some core
results. The perceived shortcoming of the CFD model is identified with respect to capturing the effects of dust lift-off and
then an experimental setup and procedure is described to provide data to capture these effects. Finally, the impact of this
data is embedded within the CFD model and the influence of dust lift-off on the separation performance dust-catcher is
considered.

2. Gas cleaning plant layout

No. 4 blast furnace gas cleaning plant (see Fig. 1) contains three main sections for extraction and separation of dust/
particulates from the blast furnace top gas flow. Gas injected into the furnace, produced through numerous complex
chemical reactions is forced out of the gas uptakes (as a result of a net positive pressure in the lower regions of the blast
furnace) and into a down comer section of pipe work before entering a primary particle separation phase which utilises a
gravity dust-catcher.
A dust-catcher can also be substituted for a gas separation cyclone, and indeed in other parts of the world, a cyclone is fa-
voured over a gravity dust-catcher for this phase of the operation. This primary phase of separation expels gas into a wet scrub-
ber where dirty gas is mixed with a water injection stage and passed through two cones under very high pressure, enabling the
majority of the dust laden gas to flow freely to the next separation stage, and the dust to be recovered as a slurry by-product.
Finally, remaining gas is dried in a demister before being recycled back through the system. This recycled gas is used to heat up
incoming clean gas flowing through the hot blast stoves, injected into the furnace creating a continuous process.

2.1. Dust-catcher principles of operation

A dust-catcher relies on gravity alone to separate particulates from the blast furnace top gas flow. Carried by the gas pass-
ing out the top of the furnace, these particulates pass through the down-comer pipe section extending off the top of the blast
furnace, before entering the upper portion of the main body of the dust-catcher. Upon entering, the flow passes through a
divergent trumpet profile which reduces the incoming velocity of the top gas. This flow is then forced to rotate by 180° in
the bottom third of the dust-catcher, causing some of the heavier particles to be deposited in the bottom of the dust-catcher
hopper.

Fig. 1. Left to right; Blast furnace (BF) No. 4 dust-catcher at TSSP UK, 3D computer model of the gas cleaning plant layout at TSSP UK (black arrows signify
direction of gas flow), schematic diagram of BF4 dust-catcher (dimensions in mm).
D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904 7893

The dust-catcher comprises an inlet diameter of 3.15 m, an internal trumpet outlet diameter of 5 m, an outer shell diam-
eter of 10.7 m, and a nominal height of approximately 27 m. Average top gas flow rate is approximately 385  103 Nm3/h.
Inlet temperatures and pressures are 403 K and 1.5 barg respectively. Air density in the computational model is specified
from blast furnace data at 1.420 kg m3. At the inflow boundary the turbulent intensity was set at 3%. The flow is assumed
to be incompressible and isothermal.

3. Mathematical model

3.1. The core governing equations for CFD

The conservation equations for momentum and mass for a 3D flow in an inhomogeneous fluid mixture is expressed in
vector notation as follows:
@
ðquÞ þ r  ðquuÞ ¼ r  ðlruÞ  rp þ S; ð1Þ
@t

@
þ r  ðquÞ ¼ Sm ; ð2Þ
@t
where u is the mixture velocity vector, p is the pressure of the fluid, q is the mixture density, with l being the effective vis-
cosity of the mixture. The source vector in the momentum equation, S, considers the body forces in the fluid such as buoy-
ancy and boundary effects. Sm is representative of any source mass to be considered. Options are available for calculating the
mixture properties of the fluid, where most are based on an arithmetic [4] or harmonic mean [5]. In this instance compo-
nents considered in the solution are weighted by the mass fraction or the component concentration. Mixture density is
calculated from either:

q ¼ Rp Rc mpc qpc ; ð3Þ

or
1 mpc
¼ Rp Rc ; ð4Þ
q qpc
where the summations are over all phases, p, and all of the components, c, of the phase; mpc is defined as the mass fraction of
the cth component of the phase p, with qpc being the value of density for the identical phase-component. For turbulence, the
k–e model of Launder and Spalding [6] refers to the solution of the turbulent kinetic energy (K) equation which is given by:
  
@ qk qmt
þ r  ðqukÞ ¼ r llam þ rk þ qmt G  qe ð5Þ
@ rk
And the dissipation rate (e) equation is as follows:
  
@ qe qmt e e2
þ r  ðqueÞ ¼ r  llam þ re þ C 1 qmt G  C 2 q ; ð6Þ
@ re k k
where the rate of generation of the turbulent kinetic energy, G, is given by:
 2  2  2 !  2  2  2
@u @m @w @u @ m @ m @w @w @ m
G¼2 þ þ þ þ þ þ þ : ð7Þ
@x @y @x @y @x @x @x @y @z

The turbulent viscosity is related to K and e by:


2
k
lt ¼ qC l : ð8Þ
e
The values of empirical constants employed in Eqs. (5)–(8) are:

C l ¼ 0:09; rk ¼ 1:0; re ¼ 1:3; C 1 ¼ 1:44 C 2 ¼ 1:92;


When analysing the turbulent viscosity in the momentum conservation equation, (1) is equal to the sum of both laminar
and turbulent viscosity.
The work documented in this section is all structured within the context of an implicit finite volume discretization. This
scheme considers a fully unstructured mesh with a random mix of elements from both tetrahedral and hexahedral forms.
The implementation has been pursued within the PHYSICA code which targets the solution of highly coupled continuum
physics problems [7]. Cell centred finite volume methods are used, discretising the governing equations together with a
bounded central differencing advection scheme. Resulting linear systems are then solved iteratively using a derivative of
the well known SIMPLEC procedure [8] on fully unstructured meshes containing a mixture of element types [9–12]. With
7894 D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904

a cell centred co-located flow scheme, the Rhie–Chow approximation is utilised in order to prevent checker-boarding of the
generated pressure field [13].

3.2. Particle tracking scheme

The particle-tracking physics were originally developed to simulate liquid metal droplets in an iron converter [11,12],
coupling the influence of a flux of particles to the mean flow characteristics of the fluid through variation in the density pro-
file. The bubble/droplet/particulate (BDP) flow is split into a number of individual packets which each represent an equal
mass, where the total mass of all the individual packets are equal to a required inflow rate. One particle is monitored for
each packet using the conventional Lagrangian method, with the capture of the full particle loading essentially recreated
from the original mass value of each specific particle size fraction.
The particle velocity, Up, is computed from [14]:
@U P
¼ Fðu  U p Þ þ gðqp  qÞ=qp ; ð9Þ
@t
which may be conveniently discretised as:
U p ¼ ðU p þ Su  @tÞ=ð1 þ F  @tÞ;
where

F ¼ 0:75C d qV slip =ðdp qp Þ;

Su ¼ FU þ ðqp  qÞg=qp ;

V slip ¼ jU p  uj: ð10Þ


Subscript ‘‘p’’ indicates particle values, with other values associated with the host fluid. dp represents diameter, and g rep-
resents gravity. The particle time step size, ot, is specified to generate a distinct number of points or steps, which is then used
to track the particle across a given cell or element. Two methods were used to calculate the drag coefficient Cd [14], suitable
for small bubbles that remain spherical in shape. This method, used in many models such as [15], considers the equation:
24:0 0:42
Cd ¼ ð1 þ 0:15Re0:687 Þ þ ; ð11Þ
Re ð1 þ 42;500
1:16 Þ
Re

particle–particle interaction is not considered in the current simulation model as a result of the low loading of particu-
lates within the system (approximately 1–1.5% by mass). As a result of the low particle loading, the system can be assumed
to be one-way coupled.

3.3. Summary of computational details

The simulation model of the dust-catcher was made up of an unstructured mesh ranging from 880  103 to 616  103 and
207  103 to 137  103 elements and nodes respectively, where mesh element and node totals are related to the fullness
level of the dust-catcher hopper (empty to full) during an operational cycle. The varying level of fullness assesses dust-
catcher performance when it is emptied and then gradually filled. Trail simulations demonstrated that the numerical
solution results remain mesh independent at and above these element values, with further simulations conducted to verify this.
The fullness level of the dust-catcher hopper was specified according to a linear (depth of dust) scale between empty (0)
and full (1). Therefore, a fullness level equivalent to a 1=4 is not representative of a 1=4 full volume. Fullness levels of 0, 1=4 , 3/8,
½, 5/8, 3=4 , 7/8, 15/16 and 1, have fullness volumes of 0, 4, 14, 33, 66, 114, 181, 223 and 270 m3 respectively.
A user defined mesh boundary layer was specified to improve the solution resolution in near wall regions, contributing to
higher accuracy, numerical stability and control during simulations. Steady state assumptions together with a false time step
value of 1  103. This value was determined through much experimentation, and was used to ensure the stability of the
simulation during convergence.
The population for the specified particle diameters was set at 8000, ensuring an even spread of particles across the
defined inlet boundary of the dust-catcher. This number of particles provided statistically significant capture data and
was sufficient to ensure results that were effectively independent of particle number. The time step used to track particle
performance in the dust-catcher was a fraction of the local residence time (the ratio of element size to local fluid velocity).
This ensured that particles completed a significant amount of steps/iterations before the distance across the element has
been exceeded, thus helping to increase the resolution of individual particle tracks in the post processing stage.
Individual fullness level solutions were completed to the accuracy available within PHYSICA [7]. The residuals error ratio
used for the simulations was 103 at convergence as the algorithm used in PHYSICA for the computational simulations was
unable to reduce the residual beyond this threshold.
D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904 7895

3.4. Initial simulation results

Initial simulation results [16] (see Fig. 2) shown that dust-catcher performance is relatively inefficient when initially
empty, but with little quantities settled dust, efficiency rises substantially to a peak of around 62% in the mid range
(80–200 m3 fullness volumes). At maximum dust-catcher fullness, efficiency is shown to peak at around 75%. The simulation
results agree well with average TSSP UK plant predictions and are not too dissimilar to manufacturer claims.
Fig. 2 documents the flow field that can be experienced in the dust-catcher. The figure illustrates the development of a
fluid jet in the middle of the trumpet, distinctly separated from the bulk flow in the dust-catcher at low/initial fullness levels
(shown in the left hand side image). The jet flow develops intensity as the hopper fullness levels increase until the core flow
in the trumpet (jet) and that in the bulk volume of the dust-catcher are fully amalgamated, leading to increasingly unsteady
turbulent flow fluctuations in the main body of the dust-catcher.

3.5. Perceived limitations of the existing CFD model

The initial simulations were conducted with particle transport post processed over a converged fluid solution, and depen-
dent on the user specified termination conditions, particles will settle and/or stick to boundary walls where they are termi-
nated. This introduces inaccuracies in the separation profile. The current formulation does not allow for the possibility of a
settled particle being re-entrained into the gas flow if conditions around the particle indicate this might be possible (such as
forces due to drag and momentum, generated from the fluid velocity in the vicinity of the particle).
The phenomenon known as dust lift-off or particle re-entrainment is a form of particle transport that is entirely plausible
considering the fluctuating conditions that can be found around the surface of a settled dust bed.
Although it was quite possible to implement a specific facility in the CFD model to capture the impact of dust lift-off in
general terms following the ideas of Witt et al. [3], its utility would be questionable without specific data to parameterise it
for the ‘dust’ quality in the dust-catcher context. Hence, an experimental prototype was developed in this work to analyse
the lift-off characteristics of a given blast furnace flue dust sample, developing a range of approximations that will provide a
basis for accurately simulating dust behaviour in the furnace. The idea of the work was to generate a lift-off profile for a spec-
ified particle diameter, that when subjected to an explicit flow speed, will generate a precise removal rate of dust into the
flow adjacent to the bed surface. The incorporation of the lift-off data within the CFD model will be covered in a later section
of the paper.

4. Experimental work

A completely polycarbonate test rig (see Fig. 3) was designed so that it contained the main flow features of the resident
dust-catcher at blast furnace 4 (TSSP UK) with respect to the flow characteristics in the neighbourhood of the dust pile in the
hopper. Thus, the test rig possesses; an inflow section, a chamber acting as an expansion, and a constriction leading to
secondary dust separation. An 1100 m3/h duct fan (operating at 15 °C) was used to provide an ample flow rate through
the test rig and to overcome the build up of back pressure generated by the filter. The duct fan was fitted with an analogue

Fig. 2. Left to right; CFD model velocity predictions for an empty dust-catcher (0 m3) [16], for a half full dust-catcher (33 m3) [16], and for a 3=4 full dust-
catcher (114 m3) [16].
7896 D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904

Fig. 3. Prototype dust lift-off model as seen in the engine test cell in Swansea University (inset: swirl reduction vane).

speed controller so that the flow rate through the test rig could be infinitely controlled. The duct fan was connected to a pri-
mary tube 2.5 m in length, specified to be approximately 17 times the diameter (0.15 m) of the inlet of the pipe, thus ensur-
ing complete mixing of the swirling flow from the fan.
In order to reduce swirl levels from the fan, a 0.5 m long swirl reduction vane (see inset Fig. 3) was placed in the inlet of
the tube to reduce swirl in the flow, so that it would be closer to what is expected in the resident dust-catcher. The primary
pipe is connected to a dust-chamber (measuring 0.2  0.2  0.35 m, with a removable, sealable lid) where the specific dust
fractions are placed to a required level. The chamber is connected to an outlet pipe and fed into a stainless steel casing
containing a filter rated at <10 lm.
Using the prototype physical model configuration, horizontal velocities for the fluid flow through the primary pipe were
measured using a pitot static tube. This velocity data is used to track the lift-off velocities in each particle size range for the
lift-off model. The duration for each lift-off experiment was set at 5 min for each specified velocity. This was considered to be
ample time to gather data, as the resident dust-catcher is subjected to constant variations in surface velocity as a result of the
continuous operation of the blast furnace. Time constraints for the experimental runs were not a major issue when consid-
ering particle drying and moisture loss over time. The time scale of the experiment was much smaller than the time scale for
particle drying. In this instance, it is possible to ignore moisture loss from the system.
A diagram of the overall layout is seen in Fig. 3 with a schematic in Fig. 4:

4.1. Experimental variables

In order to gather a large range of data for the lift-off model, flue dust samples were taken from No. 4 blast furnace at TSSP
UK. The samples were obtained from the blast furnace whilst operating under a granulating coal injection (GCI) rating of
190 kg/THM, considered average during a normal operational cycle. The flue dust obtained was dried and separated into frac-
tions using a general sieve and shaker. These fractions were then isolated, and using a KBR press, each sample was formed

Dust
Fan Primary Pipe Chamber

Constriction Pipe Secondary Pipe


Filter

Fig. 4. Schematic view of the prototype dust lift-off model drawn in AutoDesk Inventor™.
D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904 7897

Table 1
Particle properties and size distribution from flue dust analysis from blast furnace 4 dust-catcher [17].

Particle diameter (lm) Density (kg m3) Distribution % Cumulative distribution %


20 2144 0.1 0.1
38 2319 2.1 2.2
75 2340 9.59 11.79
125 2063 18.58 30.37
250 1448 54.15 84.52
355 1231 8.79 93.31
500 1286 6.1 99.41
1180 1086 0.6 100

into small disks, then measured and weighed to obtain fractional densities throughout the sample. The flue dust distribution
is detailed in Table 1.
The data in Table 1 are obtained based on a blast furnace burden composition 39% pellet4, 60% sinter5 and 1%
additive6. In order to obtain the particle size fractions of Table 1, the dust had to be oven dried and then separated. Dust
and fines swept up from the dust chamber were collected by the stainless steel housing and filter. The dust transported from
the chamber was measured (taking into account the mass of the filter bag) using a Salter scale and measured to the nearest
gram. Dust lift-off was not easily visible during the experiments due to the fine size of many particles being investigated.
Larger diameter particles have a lower packing factor than the smaller diameter particles, whilst the smaller diameter
particles have a higher density than larger diameter particles. It was therefore not possible to specify a constant mass of flue
dust to be used in every experiment. This resulted in the mass per given volume for the smaller diameter particles being
higher. Better packed particles are less influenced by the air flowing over them, and are less likely to be subjected to the
forces necessary for dust lift-off.
The velocity used in the experiments was specified from the pressure readings on the pitot static tube. The reading on the
pitot tube was then related to a corresponding fluid velocity at the outer wall of the primary tube, as this is more likely to be
the fluid velocity over the surface of the dust bed (see Fig. 4). By taking this difference into account, particle lift-off trends are
more accurately calculated. The range of velocities that could be generated by the fan was similar to the velocities predicted
by the CFD model along the dust surface over the range of hopper fullness. The relationship between the increase in dust
surface velocity and the increase of dust-catcher hopper fullness was not immediately visible.

4.2. Experimental results

In order to understand the results of varying the air flow over the dust surface, it is necessary to monitor the air flow on its
own. The uniformity of this surface air flow will determine how much turbulence is still retained within the flowing air
stream, and the effect the swirl reduction vane in the primary pipe has had on the swirling air flow created by the duct
fan. For accuracy reasons the mb pressure readings in the pipe were repeated three times and generated an average variance
of 3.75  105 and standard deviation (s.d.) of 5.57  103. A typical reading from the experiments would be around
0.10 mb, so with the current data there is variance 1%. Fig. 5 shows the air flow profile through the primary pipe recorded
just before the flow passes into the dust chamber;
The graph shows that as the air velocity increases the variation in flow velocity across the pipe radius also varies. At low
pipe velocities, the profile across the radius is generally uniform, with little variation between the centre and the edge of the
flow. As the velocity increases, the differential between the centre of the pipe and at the wall becomes larger in magnitude.
The velocity profile at these higher speeds is more parabolic, and is the result of internal skin friction head loss on the pipe
wall when compared to the centre axis. Unfortunately, due to dimensional restrictions of the pitot tube, a complete flow pro-
file across the entire diameter of the primary pipe could not be measured.
However, due to the relative smoothness of most of the recorded profiles, it is concluded that the flow profile will be
relatively symmetrical both sides of the centreline. The results in Fig. 5 shows that the range of velocities encountered on
the dust surface in the experimental test rig are large enough to cover the range of velocities predicted using the CFD analysis.
At all points in all flow profiles in Fig. 5 the Reynolds number of the flow is fully turbulent (fully turbulent after
Re P 2000). This is consistent with the on-site dust-catcher at BF4. In terms of the overall trend seen in Fig. 5, the data is
as expected. Discrepancy is found around fan setting 2 and fan setting 5. In fan setting 2 a decrease in flow speed at a dis-
tance of 50 mm and above is not expected and is attributed to irregular variation in the flow caused by some factor internal
to the experiment. Again, the same explanation is put forward for fan setting 5 where it is unknown why the velocity in the
pipe from 0 mm to 40 mm in the radial direction remains more or less constant. In general, the results in this section provide
an adequate representation of flow through the pipe, and their consistency provides confidence in the results described in
following sections.

4
Iron ore is crushed and ground into a powder so the waste material called gangue can be removed. The remaining iron rich powder is rolled into balls and
fired in a furnace to produce strong marble sized pellets.
5
Produced from fine raw ore, small coke, sand-sized limestone and numerous other steel plant waste materials that contain some iron.
6
A flux that can be pure high calcium limestone, dolomitic limestone containing magnesia, or a blend of the two types of limestone.
7898 D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904

Fig. 5. Variation in air velocity from the centreline to the outer radius of the primary pipe.

4.3. Particle lift-off trends

In order to minimise measurement variability, experiments were repeated three times, and averaged for accuracy. The
results based upon individual particle diameter lift-off performance (see Fig. 6), generate average variance and standard
deviation values of 6.17  102 and 0.217 respectively, indicating a good degree of reproducibility within the results. These
results are based on minimum and maximum recorded values of dust lift off of 0 g and 15 g respectively. The experimental
results for the lift-off prototype yielded some promising trends once the data had been analysed. Fig. 6 shows the variation in
dust lift-off expressed as a rate over a range of flow speeds;
Generally it is shown that as the diameter of the particles decreases, the lift-off rates of these particles increase. Maximum
lift-off rates are present for 20 lm and 38 lm7 diameters, whereas minimum lift-off rates are seen for 1180 lm. This shows
that it is the diameter of the particle (and the resulting surface area) that becomes the over-riding factor dictating particle per-
formance in any given scenario, due to the frictional forces generated around the particle as a result of the surrounding air flow.
Assuming that smaller diameter particles would have higher lift-off rates than larger diameter particles as less force
would be required to overcome the gravitational component, there are unexplained differences between the trends for
125 lm and 75 lm diameter particles in Fig. 6, which have been attributed to the difference in particle density of
2063 kg m3 and 2340 kg m3 for 125 lm and 75 lm diameter particles respectively. Another reason for the difference in
results may be the topography of the samples in the dust chamber.
The irregular topography samples tested had ‘dust piles’ (see Fig. 8) where the piles were created by randomly distribut-
ing flue dust within the polycarbonate chamber. It is thought that utilising some irregularity in the settled dust surface that
this may help with the transport of dust from the dust chamber to the filter. By having these dust piles, it should theoretically
be easier for particles to be lifted off the dust surface. Less particle packing occurs around the top of these piles so more of a
given particle is exposed to the air flow passing over it. Gravity forces should be exceeded by the magnitude of the forces
generated due to skin friction on a particle, promoting lift-off.
Smaller particles have a much greater packing factor than larger particles. Therefore, much less surface area of a given
particle is exposed when air passes in the particles immediate vicinity. This causes gravity forces to become dominant when
generating particle lift off. With this in mind, it is concluded that dust lift-off rates for all particle sizes can be substantially
increased but only if the air flow velocity across the surface of the particle is greatly increased. It is also noted that by cre-
ating a more irregular topography of particles across any given surface, the chances of dust lift-off occurring across that sur-
face should increase.

7
The scale on Fig. 4 only goes up to 0.03 kg/m2/sec to show result trends more clearly. Actual result for 20 lm and 38 lm trend peaks at approximately 0.084
kg/m2/sec.
D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904 7899

Fig. 6. Lift-off performance for the range of particle diameters found in the flue dust particle size distribution of BF4.

4.3.1. Irregular topography analysis


In order to assess the effect an irregular surface topography has on dust lift-off characteristics, the 125 lm particle frac-
tion was subjected to similar surface air velocities as all other experiments. The particle fraction was irregularly distributed
in the dust chamber both in the longitudinal and lateral direction, with irregularities in the surface topography being more
substantial in the longitudinal direction for no other reason than creating the topography to be irregular or random in nature.
Fig. 7 shows the differences between particle performance for the irregular variations in topography;
The results in Fig. 7 suggest that an irregular topography has a substantial effect on the lift-off rate for a given surface air
velocity. It was generally seen that for the entire range of air velocities, lift-off rates appear to be around 220% higher when
compared to a flat dust surface. It must be noted that the 2.2 times multiplier was only what is obtained from the 125 lm
comparison results. This 2.2 factor was not necessarily a factor present if the same set of experiments were performed for
other particle sizes, but can be (cautiously) used for some predictive analysis. Indeed, it is possible that this multiplier may
be higher for large diameter particles, and lower for smaller diameter particles due to the variation in density for the samples
used in the experiments.
The 125 lm particle size fraction was chosen to represent the experimentation of an irregular surface topography due to a
particle diameter and density which is approximately midway through the data values represented in Table 1, and therefore
considered representative.
Figs. 8 and 9 show the differences between the experimental topographies and the resulting topography after the particle
fraction was subjected to the variation in surface air velocity;
Fig. 8 shows the significant difference between the flat dust surface as used in the majority of experiments and the irreg-
ular topography surface. It is found from Fig. 7 that dust lift-off is far more pronounced for the irregular topography than the
flat surface. It is also noted that for the irregular topography tests there was visible particle movement across the surface. The
dust piles created in the dust chamber before the experiment was run were very fluid in their contours i.e. there were no
ridges or sharp profiles present anywhere along the surface. However, after the final speed setting was used, and the exper-
iment was run, there were significant ridge effects and unevenness along the irregular topography surface, shown in detail
and highlighted in Fig. 9.
The left hand image in Fig. 9 was taken halfway through the speed range of experiments (equivalent to approximately
2.5 m s1) and already there are pronounced ridges (highlighted in the white box(s) and drawn to highlight specific con-
tours) in the middle of the image. The right hand image taken after the maximum speed setting was run shows more pro-
nounced ‘wave’ creation at the back of the dust chamber (highlighted).
7900 D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904

Fig. 7. Lift-off performance comparison for the 125 lm, 125 lm irregular topography and complete particle mixture as found in pre-separated flue dust of
BF4.

Fig. 8. Left to right; general flat dust level created for the majority of the experimental runs, irregular topography configuration as used in the 125 lm
topography test.

5. Embedding lift-off data in the cfd model

In order to use the results of Section 4 and create a useable system model for dust lift-off within a computational model,
mathematical models are created describing the lift-off profiles experienced by each particle size fraction in the experiments.
This is undertaken using regression analysis techniques involving a range of functions. After some testing, an equation of the
form y ¼ ð1þBeAAbx Þ/ is applied to the generated lift-off curves, where the constants A, B, b, / are optimised for the particle diam-
eter(s) to be analysed. y is representative of the lift-off rate in kg/m2/sec and x depicts the surface air velocity in m s1. Table
2 gives generalised values for the co-efficient values of the lift-off trend lines in Figs. 6 and 7.
It is found from individual assessment, that a more generalised set of co-efficients can be applied to the range of lift-off
curves shown in Fig. 6, enabling greater simplification for the individual lift-off models to be derived and then applied to the
CFD results of [17]. These generalised co-efficient values are very similar to the flue dust sample which considers the dust in
its raw state from the outlet of the dust-catcher, showing that the variations between the individual particle diameters is
small enough to be attributed to noise.
D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904 7901

Fig. 9. Left to right; resulting longitudinal surface topography details from 125 lm experiment after dust had been subjected to full range of surface
velocities (detail highlighted with white irregular lines), resulting lateral surface topography details from 125 lm experiment after dust had been subjected
to full range of surface velocities (detail highlighted with white irregular lines).

Table 2
Co-efficient values for the lift-off lines of best fit based on Fig. 6 and Fig. 7.

Particle diameter (lm) Exponential equation co-efficients


A B b /
6
20 + 38, 75, 125, 250, 355, 500, 1180 31 21  10 0.06 1
125 topography testing 2.8 21  106 8.76 0.4
Mixture as per flue dust distribution 29.5 21  106 0.06 0.8

The data in Table 2 also illustrates the difference in the co-efficient values between the irregular surface topography and
the relatively flat surface topography (considered for most of the experiments).
The lift-off data used within the computational model is applied as a correction to compensate for dust lift-off as follows:

(a) A series of particle tracks (see Table 1) were calculated using the converged gas flow from a CFD solution [17]. Tracks
were calculated for every degree of fullness considered for the operational cycle (see Section 3.3).
(b) The lift-off data in Table 2, which is applicable to a specific particle diameter, was then used as a corrective tool against
the original post processed particle tracks, computing the mass of each particle size fraction that was ‘lifted off’ the
surface. This mass was dependent on the surface flow speed at that particular fullness level.
(c) The complete data set was then integrated together to produce the particle separation efficiency of the dust-catcher
over the range of considered fullness levels [17].

Some 72,000 particle tracks, 8000 for each fullness level, were required to yield numerically consistent results that were
not significantly changed when increasing the number of particles. The lift-off correction is implemented in the following
way:
Considering part (b) above; if we isolate a 75 lm diameter particle with the dust-catcher hopper at three quarters fullness
(and the corresponding surface velocity that is present at this level), 200 of these ‘dust’ particles are captured by the dust-
catcher. Taking the data from the experimentation phase, a percentage (or rate (by mass)) of dust is lifted off the surface.
If this percentage were e.g. 10% (or the equivalent lift-off rate), then the newly corrected value for the amount of particles
captured by the dust-catcher at this fullness level would then be 200  0.9 = 180. This new value is then used as the data
point with which the new lift-off profile(s) are generated, with comparison made against the original computer simulation
model.
Fig. 10 shows representative diagrams of particle tracks in an empty dust-catcher. It illustrates the relative performance
of two different diameter particles within the flue dust distribution of BF4 dust-catcher. The left hand image shows that the
majority of smaller particles enter the dust-catcher and leave through the gas outlet, following the flow profile. The right
hand image shows the effects of particle diameter on behaviour, showing that larger diameter particles are less influenced
by the flow field around them. They generally sink to the bottom surface (or fullness level) of the dust-catcher hopper where
they remain.

5.1. Particle separation efficiency curves

In order to analyse the data obtained in the experiments, the following section details the separation efficiency profile for
the resident dust-catcher. Fig. 11 contains four extra profiles which are the result of incorporation the correction data
generated from the lift-off experiments into the base case CFD results. The profiles are explained as follows:
7902 D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904

Fig. 10. Left to right; CFD streamline plot for 20 mm particle diameter without incorporating particle lift-off, CFD streamline plot for 500 mm particle
diameter without incorporating particle lift-off.

Fig. 11. Comparison of theoretical CFD results vs. lift-off data results for blast furnace four dust-catcher [17].

CFD only: These results are obtained purely from the CFD flow profile and the particle tracking algorithm as described in
Section 3 of this paper. No experimental data correction is applied.
Linear interpolation on particle capture efficiency: These results are based on using a specified surface velocity (ob-
tained from the CFD only results) and reading the corresponding lift-off rate correction for the required particle diameter
from the graph in Fig. 6.
D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904 7903

Table 3
Average particle separation efficiency percentages for the resident dust-catcher.

Resident dust-catcher
scenario Average efficiency %
Resident (CFD only) 45.82
Linear interpolation on particle capture efficiency 45.75
Linear interpolation on particle capture efficiency with lift-off factor 45.66
Equation interpolation on particle capture efficiency 45.66
Equation interpolation on particle capture efficiency with lift-off factor 45.51

Linear interpolation on particle capture efficiency with lift-off factor: As per ‘Linear interpolation on particle capture
efficiency’ with the lift-off rate multiplied by a factor of 2.2 to account for the surface topography of the dust pile.
Equation interpolation on particle capture efficiency: These results are based on using a specified surface velocity
(obtained from CFD only results), inputting these results into Table 2 to form the line-of-best-fit trend-line, generating a
particle lift-off rate correction for a specific particle diameter based on the lift-off profiles in Fig. 6.
Equation interpolation on particle capture efficiency with lift-off factor: As per ‘Equation interpolation on particle
capture efficiency’ with the lift-off rate multiplied by a factor of 2.2 to account for the surface topography of the dust pile.
Fig. 11 shows the results of the analysis undertaken on the resident dust-catcher on BF4 at TSSP UK using the lift-off equa-
tions documented in Table 2 and correction where necessary. Lift-off data generated from the experiments is applied to the
results obtained from a sister paper [16], where the CFD flow fields are predicted and illustrated in Fig. 2.
As mentioned in Section 4.3.1, there was a marked difference between the lift-off characteristics for a 125 lm diameter
particle and the irregular topography tests for the same particle size. This suggests for a given surface air velocity, the lift-off
rate for an irregular topography is approximately 220% bigger than that for a flat surface. Application of this 220% factor to
the results for all particle sizes for a given surface air velocity generates the results representing particle lift-off for an irreg-
ular surface topography. Fig. 11 shows the comparison results of the interpolated data sets.
The graph in Fig. 11 shows the particle separation efficiency profiles for the on-site dust-catcher. It is shown that dust-
catcher performance begins poorly (26% efficiency), but then begins to increase at a good, steady rate during the mid range
fullness levels of between 80 and 200 m3 with a mid range peak efficiency of 68% being generated. At the upper ranges of
fullness (>200 m3) there is a gradual decrease in the gradient of the efficiency curve, becoming almost flat at the most ex-
treme fullness level that has been simulated, with separation efficiency performance peaking at around 76% from the CFD
modelling analysis. This behaviour is believed to be related to the particle separation mechanism that is documented in [16].
Due to the moisture content (considered extremely low) of the settled dust in the dust-catcher, it has been reported from
operational staff that solidification can occur in the settled dust, creating a cement like substance which affects the removal
of dust from the hopper in the usual manner. As a result, this maximum overall efficiency will probably never be reached,
and it is likely that the dust-catcher hopper would be emptied once it has reached a volumetric fullness level of 220 m3.
With lift-off trends incorporated, Fig. 11 illustrates that the overall particle efficiency rate begins to plateau as hopper
fullness levels exceed 81%, but particle separation trends are more or less identical at levels below this point. This shows
that particle lift-off has an effect on settled dust particles once the surface velocity immediately around the particles exceeds
approximately 3.5 m s1 (or 210 m3 full). Focusing on very high hopper fullness levels, the discrepancy between CFD only
predictions, and the interpolations that consider particle lift-off up to and including the 2.2 multiplier factor that is applied,
are approximately 3.9% at peak particle separation efficiency values. Integration of the trend lines in Fig. 11 generates the
average particle separation efficiency data represented in Table 3.
From Table 3 it is clear that the equation based (CFD only) model provides a good estimate of the particle separation effi-
ciency with a 0.31% variation between the scenarios in Table 3. From operational experience on site at TSSP UK, it is thought
by BF4 operators that the resident dust-catcher is operating at just over 50% efficiency, indicating a good correlation between
the in situ performance and that predicted via computational methods.
As a result of the small differences in Table 3 between CFD only and CFD corrected efficiency ratings for the dust-catcher,
it is thought that the incorporation of the particle lift-off equations into the CFD coding is not necessary in this particular
instance.

6. Conclusion

This paper has detailed how experimental measurements of particle lift-off data may be incorporated into a CFD model to
enable dust lift-off analysis for a sample of blast furnace flue dust obtained from the dust-catcher at blast furnace No. 4 gas
cleaning plant at Tata Steel Strip products (TSSP) UK. The paper examines the lift-off trends over the full particle size range
and uses regression analysis to generate equations to accurately represent the lift-off trends/data.
By applying these lift-off equations as a correction to the particle tracking algorithms within the CFD model, dust-catcher
performance has been modelled more completely, potentially producing more useful predictions of particle separation effi-
ciency through a standard operational cycle on the gas cleaning plant. Having said this, it is clear from Fig. 11 that the impact
7904 D. Winfield et al. / Applied Mathematical Modelling 37 (2013) 7891–7904

of dust lift off from the pile topography within the dust catcher is marginal, except when the dust-catcher is at its largest
degree of fullness.
Maximum efficiency in what is considered to be the general operational range of the dust-catcher is 68%, with an average
efficiency 46%. This compares well to the efficiency estimated on site to be 50%, where a 5% deviation in average efficiency
is considered to be acceptable given the range of uncertainties in the plant operation.
Future work will look to use the lift-off particle tracking algorithm within the CFD model framework to predict the per-
formance of an optimised dust-catcher design for BF4. Particle size fractions may be expanded to provide a more compre-
hensive data set which could then be used for other processes within TSSP UK that deal with dust similar to that within
the blast furnace gas cleaning plant.

References

[1] E.P. England, Water Regulations – Tata Steel Europe Port Talbot Internal Document, 2007.
[2] J.B. Smitham, S.K. Nicol, Physico-chemical principles controlling the emission of dust from coal stockpiles, Powder Technol. 64 (3) (1991) 259–270.
[3] P. Witt, K. Carey, T. Nguyen, Prediction of dust loss from conveyers using computational fluid dynamics modelling, Appl. Math. Modell. 26 (2002) 279–
309.
[4] M. Davis, M. Cross, K.A. Pericleous, P. Schwarz, Mathematical modelling tools for the optimisation of direct smelting processes, Appl. Math. Modell. 22
(2006) 921–940.
[5] K.A. Pericleous, S.N. Drake, An algebraic slip model of PHEONICS for multiphase applications, in: N.C. Markatos et al. (Eds.), Springer-Verlag, NewYork,
1986.
[6] B.E. Launder, D.B. Spalding, The numerical computation of turbulent flows, Comp. Methods Appl. Mech. Eng. 3 (1974) 269–289.
[7] PHYSICA, www.physica.co.uk, (accessed October 2010).
[8] J.P. Van Dormal, G.D. Raithby, Enhancements to the SIMPLE method for predicting incompressible fluid flows, Numer. Heat Transfer 7 (1984) 147–163.
[9] P. Chow, M. Cross, K.A. Pericleous, A natural extension of standard control volume CFD procedures to polygonal unstructured meshes, Appl. Math.
Modell. 20 (1995) 170–183.
[10] N. Croft, K. Pericleous, M. Cross, Numerical Methods in Laminar and Turbulent Flows, in: C. Taylor et al. (Eds.), vol. IX, Pineridge Press, 1995.
[11] T.N. Croft, Unstructured Mesh-finite Volume Algorithms for Swirling, Turbulent, Reacting Flows, Ph.D. Thesis, University of Greenwich, 1998.
[12] M. Cross, T.N. Croft, G. Djambazov, K.A. Pericleous, Computational modelling of bubbles, droplets and particles in metals reduction and refining, Appl.
Math. Modell. 30 (2006) 1445–1458.
[13] C.M. Rhie, W.L. Chow, Numerical study of the turbulent flow past: an aerofoil with trailing edge separation, AIAA J. 21 (1983) 1525–1532.
[14] R. Clift, J.R. Grace, M.E. Weber, Bubbles, Drops and Particles, Academic Press, New York, 1978.
[15] O. Simonin, E. Deutsch, J.P. Minier, Eulerian prediction of the fluid particle correlated motion in turbulent two-phase flows, Appl. Sci. Res. 51 (1993)
275–283.
[16] D. Winfield, D. Paddison, M. Cross, T.N. Croft, Evaluate and model the efficiency of a blast furnace gas cleaning plant, NAFEMS World Congress, Boston,
USA, 23-26 May 2011.
[17] D. Winfield, D. Paddison, M. Cross, T.N. Croft, Geometry optimisation of a blast furnace gravity dust-catcher using computational fluid dynamics
simulation, Chem. Eng. Process. Process Intensif. 62 (2012) 137–144.

You might also like