See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/305696194

Investigation of Coke Collapse in the Blast Furnace Using Mathematical

Modeling and Small Scale Experiments

Article  in  ISIJ International · July 2016

DOI: 10.2355/isijinternational.ISIJINT-2016-114

CITATIONS READS

15 299

2 authors:

Tamoghna Mitra Henrik Saxen

Paul Wurth S.A. Åbo Akademi University
22 PUBLICATIONS   260 CITATIONS    326 PUBLICATIONS   4,969 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

SIMP SC4 "CARBO" - System modelling to optimize low-carbon footprint fuel usage in Finnish metallurgical industry View project

Sustainable steel production View project

All content following this page was uploaded by Henrik Saxen on 23 October 2016.

The user has requested enhancement of the downloaded file.

 ISIJ International, Advance
ISIJ Publication
International,
by Advance
J-STAGE,Publication
DOI: 10.2355/isijinternational.ISIJINT-2016-114
by J-STAGE
ISIJ International, J-Stage Advanced
ISIJ International,
Publication,
ISIJ International,
DOI:
Advance
http://dx.doi.org/10.2355/isijinternational.ISIJINT-2015-@@@
Vol.Publication
56 (2016),ISIJ
No.
by International,
J-Stage
9 Vol. 56 (2016), No. 9, pp. 1–10

Investigation of Coke Collapse in the Blast Furnace Using

Mathematical Modeling and Small Scale Experiments

Tamoghna MITRA* and Henrik SAXÉN

Thermal and Flow Engineering Laboratory, Faculty of Science and Engineering, Åbo Akademi University, Biskopsgatan 8,
FI-20500 Åbo, Finland.
(Received on March 2, 2016; accepted on May 17, 2016; J-STAGE Advance published date: July 28,
2016)

Burden distribution plays a key role in controlling the gas flow conditions inside a blast furnace. The
distribution of ore and coke influences the gas permeability distribution in the lumpy zone and also in the
cohesive zone, where the gas flows mainly through the coke slits. Charging an ore dump on coke can
sometimes cause the coke layer to collapse under the force of the heavier ore particles. This is known as
‘coke collapse’ or ‘coke push’, which results in a higher coke fraction near the center of the furnace than
expected. In this work coke collapse phenomena are evaluated on model charging programs using small
scale experiments and Discrete Element Modeling (DEM). DEM simulations were used to study the extent
of collapse for different charging programs, and experiments were undertaken to verify the results of the
simulations. The slope stability method was used to classify the collapse conditions into no collapse,
impact failure or gravity failure, depending on the stability of the coke layer. The findings were also com-
pared with results from an in-house mathematical model, which was modified to consider the effect of
the collapse on the underlying layer. The corrected mathematical model was found to show results in
general agreement with results from the DEM simulation.

KEY WORDS: ironmaking; blast furnace; burden distribution; coke collapse; factor of safety; Discrete Ele-
ment Modeling.

evaluation of the burden distribution in the blast furnace,

1. Introduction
Burden distribution is the main factor that affects the gas
flow and reduction conditions in the upper part, or “lumpy
zone”, of the blast furnace. The burden distribution condi-
tions are controlled by selecting an appropriate charging
program, i.e., the material, its mass and the position at
which it is charged into the furnace. The charging program
is then repeated to create a layered structure of iron bear-
ing and carbon bearing materials inside the furnace. The
carbon bearing materials are primarily coke while the iron
bearing materials, often collectively referred to as “ore”,
are agglomerated iron ore, in the form of pellet or sinter,
but also smaller quantities of lump ore is used. The ore and
coke have distinct size, shape and gas permeability which
influence the flow of the ascending reducing gas that trav-
els through the bed. It is beneficial to increase the rate of
auxiliary reductants (typically pulverized coal) injected in
the lower parts of the furnace for economic reasons. The
decrease in the coke rate has resulted in thinner coke layers.
Since coke is the more permeable of the two main burden
materials, this change increases the likelihood of operation
irregularities like channeling and obstructed gas flow in the
thin coke slits in the cohesive zone. It has therefore become
increasingly important to gain a deeper understanding of
the formation process of burden layers in the blast furnace.
A number of investigations have been published on the
* Corresponding author: E-mail: tmitra@abo.fi
DOI: http://dx.doi.org/10.2355/isijinternational.ISIJINT-2016-114
based on both numerical and experimental techniques.
These studies focus on different aspects of burden forma-
tion like burden trajectory calculation,1,2) layer formation3,4)
and burden descent.5,6) When ore is charged on coke, the
heavier ore particles may deform and shift the coke layer.
This action is termed “coke push” or “coke collapse”. Coke
collapse has been observed both in numerical studies1,7) and
experimentally.8,9) Various schemes have been suggested to
prevent the collapse, for example creating a coke plateau
near the wall and controlling the particle trajectory.10,11)
Some other schemes have used it to their advantage to
achieve a central working furnace.12,13) It is hard to ascertain
the effect of the charging a layer on the underlying layers
as it is difficult to measure the underlying profile after the
new layer has been charged. Some studies have used mag-
netometer readings14) or image processing techniques8,15) for
experimental verification, but these method have their own
drawbacks: magnetometer is intrusive disturbing the layer
structure and image processing requires a furnace section
with a transparent wall, which induces wall effects. There-
fore some studies have resorted to sophisticated numerical
techniques like Discrete Element Modeling (DEM) for
better understanding of the phenomena involved. DEM can
provide deeper insight on the charging phenomena, but the
computational effort can be excessive because of the large
number of particles involved, resulting in prohibitive simu-
lation time.
Some studies have correlated the coke collapse effect
to the energy of the particle flow,7,16) but these studies
do not consider the shape of burden surface on which the

1 © 2016 ISIJ
 ISIJ International, Advance Publication by J-STAGE
ISIJ International,
ISIJ International,
AdvanceVol.Publication
56 (2016), No.
ore is charged. This is a complication as the profile may
vary strongly depending on the charging program. Some
authors17–19) have utilized the stability theory of slopes to
determine if a layer would be stable upon charging a pellet
layer. This is based on stability theory used extensively in
geological applications and dam constructions.20)
In the present study, a scaled experimental setup has
been used to model some interesting charging programs and
simulated them using an in-house mathematical model21)
and DEM. The occurrence and extent of coke collapse has
been studied for each of the programs. The mathematical
model which did not originally consider the effect of coke
collapse was next modified to consider such effects.
The article has been organized as follows: The next sec-
tion briefly describes the experimental setup. In the subse-
quent section, “Simulation methodology”, the mathematical
model and the DEM setup are described. The section also
describes the coke layer stability model that was used to
identify the stability of a layer and a scheme is outlined how
to modify the layer to reach stability. This is followed by a
section with results and corresponding discussion, where the
results of the charging experiments and the DEM simula-
tions are compared with the predictions by the mathematical
model. Finally, the coke collapse phenomenon is verified
experimentally and a brief stability analysis for different
charging programs is outlined. The last section concludes
the study and provides some guidelines for future work.
2. Experimental Setup
A scaled burden distribution model (1:10 scale), described
in detail in previous studies,4,21) was used for evaluating the
charging programs. The setup consisted of a holding bin
for raw materials, hopper, plunger to control the flow rate,
rotating chute and a cone to collect the charge. The cone
has a throat diameter of 0.625 m and the base was shaped to
mimic the repose angle of iron ore pellet. The particles were
scaled to 1:4 of the true size to avoid inter-particle forces
or dust formation. The coke particles were in the size range
5–20 mm and the pellets were about 3 mm. The chute angle
could be changed in discrete steps of 1 to 11 to allow the
material to be charged as rings at different radial positions:
1 represented the chute angle closest to the furnace center
and chute position 11 was near the wall. The plunger open-
ing was adjusted to create a complete ring of each “dump”.
After charging a layer a profile measurement device was
used for detecting the free burden surface.
3. Simulation Methodology
3.1. Mathematical Model and Stability Theory
A simplified two-dimensional axis-symmetric mathemati-
cal model described in21) was used to simulate burden layer
formation and descent through the blast furnace shaft. The
layers were represented by two line segments whose apex
coincided with the trajectory of burden falling from the
chute. The basic model does not consider the coke collapse
but assumes the earlier layers to be stagnant at charging. The
model was modified to calculate the stability of a coke layer
and to correct the layer structure accordingly.
Coke layer stability theory and burden layer correction:
The stability of a slope may be expressed as a ratio between
the available shear strength (s) and the shear stress (τ) along
a failure plane, referred to as a ‘factor of safety’
s
..................................... (1)
fs =
τ surface and the center of failure surface O. The grid repre-
Theoretically the slope is unstable if the factor of safety
is less than unity. However, studies have experimentally
identified the limit to be closer to 0.8.17) The shape of failure
© 2016 ISIJ
by J-Stage
9
plane may vary but in this study it is assumed to be circular
(Fig. 1, left), which is a necessary for using the “Ordinary
method of slices”20) of factor of safety calculation used in
this study. The shear strength is defined in terms of nor-
mal stress (σ) on the failure plane. Cohesion between the
individual particles is assumed to be absent, because the
particles are large enough. Therefore, the above equation
may be written as
σ tan φ
fs = ................................. (2)
τ
where ϕ is the developed friction angle of the material. Like
in many limit equilibrium methods20) of calculating the fac-
tor of safety for the slip surface, the region above the failure
surface is divided into a finite number of thin vertical slices,
each with a base length of Δln and of unit depth. Figure 1
shows the forces acting on such a slice, n (right) located at
a particular position on the failure surface (left). Wn is the
weight and Fn is the external force on the slice, and αn is
the angle made by the tangent on the failure surface to the
horizontal. In the ordinary method of slices it is assumed
that the inter-slice forces acting in the horizontal direction
cancel each other.
En −1 + En = 0 ................................ (3)
The shear force at the bottom of the slice (Sn) is defined as
Sn = τ n ∆ln .................................. (4)
where τn is the shear stress for slice n. Using Eq. (2) the
above equation may be rewritten as
σ n tan φ
Sn = ∆ln ............................. (5)
fs
where σn is the normal stress on slice n, while fs and ϕ are
assumed to remain constant for all the slices. The region
enclosed by the slip plane is assumed to have the proper-
ties of the coke layer, without evaluating the composition
of parts of other layers which may be included in the
slip region. This assumption is motivated by the fact that
coke layer is the largest fraction constituting the region.
Although, this supposition gives an approximate factor
which depends primarily on the layer surface structure,
yet it is expectedly the most important factor determining
the failure for relatively thicker layers. The approximation
reduces the computational load considerably making it use-
ful for faster evaluation of slope stability, especially useful
for online applications.
Fig. 1. Stability analysis using method of slices. Left: The failure
sents the different positions centers of rotation used for
evaluating the factor of safety. Right: Forces acting on
each slice.
2
 ISIJ International, Advance Publication by J-STAGE
ISIJ International,
ISIJ International,
AdvanceVol.Publication
56 (2016), No.
by J-Stage
9

The normal force (Nn) on the slip surface is given by the center of the furnace, in turn changing the slope of the
layer. The two possible failure modes are indicated in the
N n = σ n ∆ln ................................. (6)
top part of Fig. 3.23) Previous studies have also referred to
From the free body diagram (Fig. 1, right) the normal impact failure as gouge formation.16) Each of these failure
force may also be expressed as modes depends on the shape of the coke layer surface and
the radial position of the pellet stream. The factor of safety
N n = ( Wn + Fn ) cos α n ......................... (7)
of a surface depends on these aspects, and a higher factor
At the critical point of equilibrium, the clockwise and means a stable slope. In this study, the three possibilities of
anti-clockwise moments around O, the center of the slip failure are assessed by estimating the factor of safety of a
surface, are equal, so slope: if the value is high enough then the slope is stable,
for a medium-level value impact failure is taken to occur,
∑ (Wn + Fn ) R sin α n = ∑Sn R ................... (8) while a low value of the safety factor corresponds to a grav-
n n
ity failure. In practice, the impact failure and gravity failure
where R is the radius of curvature for the slip plane. Substi- may occur in combination, but this was neglected in the
tuting the value of Sn from Eq. (5) yields present study to make the model manageable.
σ n tan φ Consideration of coke collapse in the mathematical
∑ (Wn + Fn ) R sin α n = ∑ fs ∆ln R ........... (9) model: Figure 4 shows the schematic diagram of the imple-
n n mentation of the failure modes into the mathematical model.
Using Eqs. (6), (7) and (9) we get The factor of safety is evaluated for a particular coke surface
against a pellet stream entering at a particular position, and
(Wn + Fn ) cos α n tan φ
∑ (Wn + Fn ) R sin α n = ∑ ∆ln R .... (10) is used to determine if the coke slope needs correction. If
n n ∆ln fs
Rearranging the equation, an expression for factor of
safety is obtained as

∑ (Wn + Fn ) cos α n tan φ ................ (11)

fs = n
∑ n (Wn + Fn ) sin α n
Therefore, the stability of a slope formed by granular
particles when an external force is applied on the heap is
a function of magnitude and position of the force acting
on the slope, density of the material forming the slope and
frictional resistance applied by the material.
This factor of safety is evaluated using different trial
points as centers of curvature (intersections of the grid in
left of Fig. 1). The failure plane is assumed to pass through
the intersection of wall and the layer surface.19) Figure 2
shows the contour of calculated factor of safety values for
different centers of rotation. The minimum corresponds to
the critical factor of safety and defines the stability of the
layer surface.
Modes of failure: During the charging process there are
three possibilities. Firstly, the slope does not change signifi-
cantly due to the incoming stream. Secondly, impact failure
may occur when the pellet stream displaces the particles
Fig. 3. Top: Schematic diagram of two types of coke collapse,
but there is no collapse of the layer. Thirdly, gravity failure impact failure and gravity failure.23) Bottom: Correspond-
occurs when the slope is unstable and the particles flow to ing implementation in the present mathematical model.

Fig. 2. Factor of safety contours for different centers of rotation. Fig. 4. Flowchart outlining the implementation of coke collapse
(Online version in color.) in the mathematical model.

3 © 2016 ISIJ
 ISIJ International, Advance Publication by J-STAGE
ISIJ International,
ISIJ International,
AdvanceVol.Publication
56 (2016), No.
by J-Stage
9

the value is lower than a threshold value, impact failure is

taken to occur and the coke apex is only moved towards
the furnace center but the surface would not collapse. For
highly unstable slopes, i.e., very low values of the safety
factor, gravity failure is expected where part of the coke
layer would break away and slide to the center.
The lower part of Fig. 3 schematically illustrates the
correction scheme for a coke layer which results in impact
failure and gravity failure. To correct the coke layer to
accommodate the failure mechanisms the apex of the coke
layer (r, y) is moved towards the center. The radial displace-
ment, Δr, results in a vertical displacement, Δy, which is
calculated from the constraint that the volume of the coke
layer before and after the collapse should be equal. Thus, Fig. 5. Isometric view (left) and orthographic view (right) of the
no change is voidage is taken to occur. In case of impact simulated charging system. (Online version in color.)
failure the apex is displaced by a small amount and the
corresponding factor of safety for the reconstructed layer
Table 1. Charging programs with material type, dump mass and
is calculated. The factor of safety increases as the apex is
chute positions (P: pellets, LC: large coke, SC: small
moved towards the center as shown in Section 4.2 (Fig. 10). coke, CC: center coke).
If the factor reaches the limit of stability the displacement is
stopped, else it is displaced further towards the center and Code Base Layer 1 Layer 2 Layer 3 Layer 4 Layer 5 Layer 6
the factor of safety is recalculated. For gravity failure the
inner (βi) and outer slopes (βo) of the lines describing the Material P LC SC CC P P
upper surface of the coke layer are decreased because of the CP1 Mass (kg) 7.45 3.89 3.89 0.58 16.59 16.59
collapse, where the decrease depends on the displacement
Chute pos. 10 6 4 1 10 10
of the apex. The new slopes βi,new and βo,new are given by
Material P LC SC CC P P
β i,new = ki ∆r β i ............................. (12)
CP2 Mass (kg) 7.45 3.89 3.89 0.58 16.59 16.59
β o,new = ko ∆r β o ............................ (13) Chute pos. 10 8 6 1 10 10
where ki and ko are factors determined empirically and Material P LC SC CC P P
which depend on the dimensions of the furnace. As the
slopes of the lines are known, and are less than the original CP3 Mass (kg) 7.45 3.89 3.89 0.58 16.59 16.59
slope, the layer overflows into the center and the extent of Chute pos. 10 6 9 1 10 10
overflow would depend on how unstable the initial slope is.
Material P LC SC SC CC P P
3.2. Discrete Element Model CP4 Mass (kg) 7.45 3.59 2.33 1.56 0.58 16.59 16.59
Discrete Element Modelling is an explicit numeri-
cal method used for evaluating the transient interactions Chute pos. 10 6 9 3 1 10 10
between solid particles which undergo translational and Material P SC LC CC P P
rotational motion. The motion of the particles are described
CP5 Mass (kg) 7.45 3.89 3.89 0.58 16.59 16.59
mathematically by Newton’s second law. The interactions
on the other hand are described by a spring and dashpot Chute pos. 10 6 4 1 10 10
model which describe the elastic and plastic nature of the
particles, respectively. The governing equations and the
values of different parameters are described in detail in a channel above the chute at the same rate in which they were
previous work.4) charged in the experimental setup, from where they fall
Traditionally, DEM uses spheres to describe the par- down on the chute, slide along it and form a stream which
ticles. Iron oxide pellets are relatively smooth and spheri- hits the burden surface and creates a ring of burden on it.
cal. However, coke particles are highly irregular so they
require special consideration. Therefore coke particles were
described by clumping together multiple spheres in different 4. Results and Discussion
orientations to mimic actual particle templates. Five differ- 4.1. Comparison of Charging Programs
ent particle templates4) were used and they were charged A set of small-scale experiments and corresponding
in equal proportions. The maximum diameter of the coke simulations were undertaken, as listed in Table 1. The
particles for each particle template was 12.5 mm for large charging programs, denoted by CP1–CP5, were designed to
coke, 7.5 mm for small coke and 18 mm for center coke. No study certain basic features of the layer-formation process
size distribution was considered. The pellet particles were and the coke push effect. In the experiments, a thin layer of
represented by spheres of 3 mm diameter. pellets was first charged making the starting burden surface
The simulation setup is described in Fig. 5. DEM simula- profile mimic that of a pellet layer covering the whole sur-
tion is computationally expensive because of a large number face. This base pellet layer also mimics the upper surface
of particles and a maximum time step in the order of 10 − 5 s. of the final pellet layer in the charging program, which in
To reduce the computational burden, a 90° section of the practice is repeated over and over in the real process. This
throat was simulated which reduced the number of particles base layer was not considered in the calculations. Next,
considerably. This introduced artificial wall effects so only two coke dumps were charged, each of two different sizes,
a smaller section in the middle of the simulated region was referred to as large coke (10 mm–16 mm) and small coke
used to report the results. It should also be stressed that (5 mm–10 mm). They were followed by a center-coke
because of this setup the particles have very limited space charge (16 mm–20 mm). Finally, two dumps of pellets were
near the center of the furnace, so the results in this region charged at higher chute angles. The pellet layer covered the
should be disregarded. The particles were generated in a burden surface almost entirely.

© 2016 ISIJ 4
 ISIJ International, Advance Publication by J-STAGE
ISIJ International,
ISIJ International,
AdvanceVol.Publication
56 (2016), No.
by J-Stage
9

In the first charging program, CP1, the small coke and
the large coke were charged at chute positions 6 and 4
respectively. In CP3 the small coke dump was charged at
higher position, while in CP2 both coke sizes were charged
at higher positions compared to CP1. The small coke dump
in CP4 was split into two parts in the ratio of 2:3, and the
small fraction was dumped closer to the center. Finally, for
CP5 the order of charging the coke sizes was interchanged
compared to that of CP1.
For each charging program, a corresponding experiment
was carried out using a small-scale model (see section 2)
along with a corresponding DEM simulation using a furnace
section (subsection 3.2). The mathematical model (subsec-
tion 3.1) was also utilized to predict the burden distribution.
Figure 6 presents the results in a tabular format. The experi-
mental column shows the profile measurements after every
dump. Here, the region in between two consecutive profiles
is assumed to be occupied by the dump of the charged mate-
rial. As this strategy disregards the collapse of the lighter
coke layers upon charging the heavier pellet layer, it treats
the results like the mathematical model, which does not
either consider coke push effects. In cases where the coke
push is significant, e.g., in CP2 and CP3, the experimental
results are very different from those of the DEM simulation
for the state at the end of charging. Furthermore, ‘appar-
ent profiles’ from the DEM simulations have also been
depicted. These are constructed by registering the burden
surface profile after the simulated charging of each dump,

Fig. 6. Burden layers for charging programs CP1–CP5 (cf. Table 1) from experiments, mathematical model, apparent
layers calculated from the DEM simulations and corresponding DEM simulation results. The base pellet layer is
not shown. (Online version in color.)

5 © 2016 ISIJ
 ISIJ International, Advance Publication by J-STAGE
ISIJ International,
ISIJ International,
AdvanceVol.Publication
assuming the region between two such consecutive profiles
be filled by the charged particles in question. Again, it
should be stressed that limited attention should be paid to
the results near the center of the DEM simulations due to the
confined space in this region. This makes the DEM results
unreliable for r < 0.2Rthroat (cf. Fig. 3). Figure 7 presents the
volumetric ore fractions for each of the charging programs,
measured or simulated at different radial positions. These
give an indication of the probable gas flow distribution in
a blast furnace applying the charging programs in question:
regions with low ore fraction (and high coke fraction) results
in higher gas permeability, while a high ore fraction leads
to less gas flow.
The volumetric ore fraction distribution for CP1 shows
higher ore share near the furnace wall, while near the center
the material is mainly coke. In the intermediate region the
ratio is quite constant. The profile measurements and simu-
lated profiles are quite similar. This is so because the pellet
dumps (cf. Fig. 6) mainly form a pocket outside of the ridge
formed by the coke dumps, leading to little shifting of the
coke layers. The effect of the coke collapse in CP2 is clearly
reflected in the profiles, seen as a considerable difference
between the final distribution predicted by DEM and the
other profiles: The coke layer is here moved significantly
to the center of the furnace, and the share of ore near the
wall is significantly higher in the DEM results. CP3 shows
similar, but less pronounced, trends. For both CP2 and CP3
the small coke is dumped at higher chute angles, so the
pellet dump that follows cannot be trapped in the pocket
at the wall like in CP1: The pellet dump overflows the
pocket, breaking away the weakest parts of the lower coke
layer pushing it towards the center. In CP4 the small coke
dump is split into two parts, where the second is charged at
lower chute angles which increases the stability of the coke
layer. Finally CP5 shows a different behavior than CP1,
even though the charging positons are same, since the order
of coke sizes is reversed. The large difference between the
sizes of large coke and the pellet particles induces granular
convection,24) commonly referred to as “brazil nut effect”.
As the pellet particles push some of the larger particles to
the top, this would result in a larger mixed layer. This effect
is difficult to evaluate without DEM simulation.
Fig. 7. Volumetric ratio of ore and coke for each of the model charging programs (cf. Table 1). Because of wall effects
the DEM results are unreliable for < 0.2Rthroat.
© 2016 ISIJ
56 (2016), No.
by J-Stage
9
Figure 8 presents isometric and sectional views of the
burden for CP2 and CP3 at the point when about half of the
first pellet layer has been charged. The coke particles are
colored according to their velocity. In sub-figures (a), (c),
(d), (e), (g) and (h), the pellet particles are colored olive to
distinguish them from the coke particles. In sub-figures (b)
and (f) the pellet particles are rendered invisible, showing
the coke front which is moved by the pellet particles. It is
clear from the figures that both charging programs give rise
to significant coke push effects, but the extent of collapse is
higher in CP3 than in CP2 as the particles which are pushed
have higher velocity. Therefore, unlike CP2 the shifted
particles in CP3 roll to the center of the furnace. The extent
of coke collapse is different in the two cases and CP2 is
classified as impact failure whereas CP3 is gravity failure.
4.2. Revision of the Layers in the Mathematical Model
The results from the mathematical model were seen to
be different from findings of the DEM simulations as the
former fails to consider coke collapse. This may result in
unreliable prediction of the burden distribution for certain
charging programs. Therefore, the slope stability model
was used to predict the occurrence and extent of coke col-
lapse (cf. subsection 3.1). The friction factor for coke is
assumed to be tan(ϕ) = 0.43 and a point force of F = 12 N 25)
is assumed at the point of charging on the coke layer sur-
face. The threshold for critical factor of safety for impact
failure was taken to be 0.8 and that for the gravity failure
was assumed to be 0.6.
Figure 9 shows that the calculated factor of safety for
CP2 and CP3 is very low before charging the pellet layer.
The factor calculated for the coke surface after the pellet has
been charged has become higher as the surface has reached
stability with respect to the force of pellet stream. As the
coke layer was reconstructed by gradually moving the layer
apex towards the center (Fig. 10) it eventually became stable
with respect to the applied external force. The crosses and
circles mark the factors of safety before and after the cor-
rection. It is seen that programs CP2 and CP3 require coke-
layer deformation to make the layer stable, while CP1, CP4
and CP5 are (relatively) stable initially. It is also evident that
distributing the coke layer into two parts (CP4, compared
6
 ISIJ International, Advance Publication by J-STAGE
ISIJ International,
ISIJ International,
AdvanceVol.Publication
56 (2016), No.
by J-Stage
9

Fig. 8. Isometric views for the charging programs CP2 and CP3 with (a, e) pellet particles and the corresponding cross-
sectional view (c, g) while the pellets are being charged. The coke particles are colored according to their veloc-
ity whereas the pellet particles are colored olive. Views (b, f) with “transparent pellets” show the front of coke
particles pushed towards the center of furnace. Views (d, h) show the cross-sectional view at the end of charging.
(Online version in color.)

cal model cross-sections. The corrected results have some

Fig. 9. Calculated factors of safety for DEM before and after
charging the pellet layer, and corresponding values for
mathematical model with and without coke collapse cor-
rection.
to CP3), and charging part of the dump near the center
improves the stability by a great extent. Figure 11 presents
the results from the corrected and uncorrected mathemati-
similarity to the DEM simulation. The volumetric ore frac-
tions show a similar trend near the furnace wall but the trend
is not captured well in the intermediate regions between
center and the wall. In case of CP2, the peak for correc-
tion volumetric fraction near radial distance 0.1 m is sharp,
unlike the DEM simulation which is flatter. This sharpness
is absent in realistic owing to the formation of the mixed
layer of loose coke and pellet. This sharpness would also be
diluted for calculation of actual furnace dimensions. In case
of CP3, as the coke collapses to the center, the repose angle
of the pellet also changes as the lower surface is not stagnant
anymore. Therefore the repose angle should be changed
adequately to match the volumetric ore fraction of DEM. In
this work, the effect of the coke collapse on the pellet repose
angle is not considered. A yellow line in subplot (f) is added
to indicate the pellet layer boundary with lower repose angle
for pellet layer. Visual inspection can confirm that consider-
ing lower repose angle would make the profile similar to the

7 © 2016 ISIJ
 ISIJ International, Advance Publication by J-STAGE
ISIJ International,
ISIJ International,
AdvanceVol.Publication
56 (2016), No.
by J-Stage
9

DEM simulation cross-section (g), which would make the
volumetric ore fraction of the corrected model closer to the
DEM simulation.
4.3. Experimental Verification of the Coke Collapse
Phenomena
Coke collapse was very evident from the DEM simula-
tion but it was difficult to observe experimentally as the
measured burden surface profiles and visual observations
gave little indication about the motion of the coke particles.
Therefore one of the charging programs, CP3, which had
suffered gravity failure and maximum coke collapse, was
repeated with some tracers embedded into the small coke
layer. White colored coke particles were utilized as tracers
that were distinguished from one another by registering their
weight and appearance before the experiment. After normal
charging of the previous layers of the charging program,
the layer of small coke was reconstructed manually on the
basis of the experimental results of Section 4.1 with tracer
particles embedded. The first set of tracers were placed just
above the large coke layer (‘bottom’), the next set halfway
into the layer (‘middle’) and the final ones on the surface
(‘top’) of the layer of small coke, as indicated in the left
part of Fig. 12. The top row of panels (a–c) in Fig. 13 pres-
ents snapshots from above after placing the tracers at each
stage. After this reconstruction, a pellet layer was charged,
cf. Fig. 13(d), which shifted part of the coke layer with the
embedded tracers. The pellet and coke particles were then
gradually and carefully removed using a suction device
without disturbing nearby particles. As a tracer particle
was encountered, a snapshot was taken (Figs. 13(e), 13(f))
and the vertical level of the particle was measured. This
measurement along with the radial and azimuthal position
of the tracer identified using image analysis, gave the final
spatial position of the tracer before and after charging the
pellet layer.
From the right panel of Fig. 12 it is evident that the
tracers have moved towards the center of the furnace. As
expected, the ‘top’ tracers have moved the farthest which is
also confirmed from the isometric view of the tracer posi-
tions in the left panel of Fig. 14. The mean displacement of
the ‘top’ tracers is about four times that in the ‘middle’ and
‘bottom’ (Fig. 14, right). These top tracers are easily swept
by the heavy pellet layer unlike the other ones, which were
constrained by surrounding coke particles. The results of
a DEM simulation of the charging program (Fig. 15) also
yield very similar findings: The particles near the free sur-
face move the farthest towards the center of the furnace. It
can be concluded that the upper portion of the layer travels
farthest to the center, but the extent of this portion should

Fig. 10. Factor of safety for the coke layers in the charging pro-
grams CP1–CP5 as the coke apex is moved towards the
furnace center. The cross and the circle indicate the ini-
tial and the final factor of safety calculated using the cor-
rection scheme.
Fig. 12. Left: Reconstructed profiles from charging experiment
CP3 and corresponding positions of the tracers. Right:
Final position of the tracers after charging the pellet
layer. (Online version in color.)

Fig. 11. Results of mathematical model, without (a, e) and with (b, f) coke surface correction along with the corre-
sponding DEM simulation (c, g) and simulated volumetric fraction of ore (d, h) for charging programs CP2 and
CP3. In subplot (f) the yellow line shows the pellet layer boundary for a lower repose angle. (Online version in
color.)

© 2016 ISIJ 8
 ISIJ International, Advance Publication by J-STAGE
ISIJ International,
ISIJ International,
AdvanceVol.Publication
56 (2016), No.
by J-Stage
9

Fig. 13. Tracers placed at the bottom (a), middle (b) and the top (c) of the layer of small coke. (d) Burden surface after
charging the pellet layer. (e–f) Careful excavation of the burden surface revealing the shifted tracer positions.
(Online version in color.)

Fig. 14. Left: Displacement of the tracer particles between initial and final positions (straight lines) after charging the
first pellet layer. Initial tracer positions at the bottom are indicated by red triangles, in the middle by green
squares and on the top of the layer of small coke by blue circles. Right: Net displacement of the tracer particles
plotted in ascending order. (Online version in color.)

Table 2. Charging program for studying the effect of the large

coke and small coke positions on the stability of coke
layer.

Layer 1 Layer 2 Layer 3 Layer 4 Layer 5

Material LC SC CC P P
Mass, kg 3.89 3.89 0.58, 0.00 16.59 16.59
Chute position 3–9 3–9 1 10 10

analyze the stability of the coke layer. This exercise was

Fig. 15. Streamlines indicating the velocity of selected ‘tracer’
coke particles where the color of the streamline indicates
the velocity of the particle at that position. The selected
particles are shown in blue at their initial positions before
charging pellet layer. (Online version in color.)
depend on the stress exerted by the pellet stream and the
dimension of the particles. Therefore the extent of such slid-
ing behavior would be different in an actual furnace.
4.4. Design of Charging Programs
The results presented above make it possible to predict
the stability of slopes formed by coke on charging pellets
at higher angles using mathematical modeling. A charging
program similar to those analyzed in the previous sections
was chosen with a dump of large coke, a dump of small
coke, a center-coke dump and two pellet dumps. The pellet
dumps were charged at high angles. The chute positions for
the small and the large coke were varied in the range 3–9
(cf. Table 2) and the theory of subsection 3.1 was used to
repeated without the center coke to see the effect of center
coke on the stability of the coke layer. As the small coke
dump was charged more towards the center, i.e., at lower
chute positions, the slope became stable, resulting in higher
factor of safety values (Fig. 16). The results of three of
charging programs analyzed earlier (CP1, CP2 and CP3)
have also been reported in the figure. The figure presents
the safety factors in contour plots for a coke layer with (left)
or without (right) a center coke charge. The center coke is
seen to stabilize the coke layer making it less prone to col-
lapse. Also, except at very large positions of the small coke
dump, the results are quite insensitive to the position of the
large coke, while a clear decrease in stability is noted with
growing position of the small coke dump. Thus, small coke
charged close to the wall is likely to give rise to an unstable
slope of the coke-layer surface and therefore prone to coke
collapse.
5. Conclusions
In this work the coke collapse phenomenon in the blast
furnace has been investigated using five model charging
programs. The charging programs consisted of two coke

9 © 2016 ISIJ
 ISIJ International, Advance Publication by J-STAGE
ISIJ International,
ISIJ International,
AdvanceVol.Publication
Fig. 16. Effect of positions of large and small coke dumps on the factor of stability for programs with (left) and without
(right) a center coke dump. The factor of safety values of the CP1–CP3 (cf. Table 1) are marked in the left fig-
ure. (Online version in color.)
layers charged in different positions, a center coke dump
and two pellet dumps. The pellet dumps were charged at
higher angles which affected the coke layers differently
depending on the charging position of the underlying coke
layers. For each of the charging programs an experiment
was performed using a scaled model and DEM simulation
was carried out. DEM simulation gave a better understand-
ing of the extent of coke collapse in each of the charging
programs, as the mode and extent of collapse was differ-
ent in each case. In cases where significant collapse was
observed they were classified as impact failure or gravity
failure. Experiments suggested that when the pellet stream
hits the coke layer, some coke particles in the topmost layer
obtain much higher velocity than the particles embedded
inside the layer, and therefore they experience the maximum
displacement. It was also observed that instead of charging
the full coke dump at higher angles, charging a fraction of
the dump closer to the center also reduces the chances of
collapse. A mathematical model which was developed in
an earlier study21) was augmented to incorporate the coke
collapse using slope stability theory. After this modifica-
tion, the model was demonstrated to consider the effect
of coke collapse fairly. The effect of coke positions in a
particular set of charging program was finally analyzed. It
was observed that layers formed by charging coke at higher
chute angles were unstable to the pellet stream because the
pocket containing the pellets outside the coke apex was
very small.
Although in this study the collapse has been quantified
well for a scaled model, further work is required to under-
stand if this theory also describes the collapse in full scale.
The findings of the study are still expected to be useful in
determining the stability of coke layers in different charging
programs, thus giving the operator an indication whether
a specific charging program is prone to exhibit collapse
and, if so, also providing guidelines on possible changes to
stabilizing it.
Acknowledgements
The authors would like to acknowledge Dr. Peter Aus-
tin of BlueScope Steel, Australia, for making his research
results on coke collapse, described in Ref. 23), available.
The Graduate School in Chemical Engineering, Tekes and
participating industry in the SIMP program coordinated by
© 2016 ISIJ
View publication stats
56 (2016), No.
by J-Stage
9
FIMECC Ltd. are also acknowledged for financial support.
The discrete element method simulations and analysis were
conducted using the EDEM software provided by DEM
Solutions Ltd., Edinburgh, Scotland.
REFERENCES
1) H. Mio, S. Komatsuki, M. Akashi, A. Shimosaka, Y. Shirakawa, J.
Hidaka, M. Kadowaki, S. Matsuzaki and K. Kunitomo: ISIJ Int., 49
(2009), 479.
2) Y. W. Yu, C. G. Bai, Z. R. Zhang, F. Wang, D. G. Lv and C. Pan:
Ironmaking Steelmaking, 36 (2009), 505.
3) Y. Yu and H. Saxén: Steel Res. Int., 84 (2013), 1018.
4) T. Mitra and H. Saxén: Comp. Part. Mech., (2015), DOI 10.1007/
s40571-015-0084-1.
5) H. Mio, K. Yamamoto, A. Shimosaka, Y. Shirakawa and J. Hidaka:
ISIJ Int., 47 (2007), 1745.
6) M. Icida, M. Takao, K. Kunitomo, S. Matsuzaki, T. Deno and K.
Nishihara: ISIJ Int., 36 (1996), 493.
7) S. Liu, Z. Zhou, K. Dong, A. Yu, J. Tsalapatis and D. Pinson: AIP
Conf. Proc., 1542 (2013), 634.
8) J. Jiminez, J. Mochon, A. Formoso and J. S. Ayala: ISIJ Int., 40
(2000), 114.
9) Y. Kajiwara, T. Jimbo, T. Joko, Y. Aminaga and T. Inada: Trans.
Iron Steel Inst. Jpn., 24 (1984), 799.
10) Y. Matsui, A. Kasai, K. Ito, T. Matsuo, S. Kitayama and N. Nagai:
ISIJ Int., 43 (2003), 1159.
11) P. K. Gupta, A. S. Rao, V. R. Sekhar, M. Ranjan and T. K. Naha:
Ironmaking Steelmaking, 37 (2010), 235.
12) J. J. Poveromo: Iron Steelmak., (1995), 64.
13) M. Nikus, J. Hinnelä and H. Saxén: Steel Res., 68 (1998), 406.
14) Y. Kajiwara, T. Jimbo and T. Sakai: Trans. Iron Steel Inst. Jpn., 23
(1983), 1045.
15) S. H. Gharat and D. V. Khakar: Procedia Eng., 102 (2015), 676.
16) C. K. Ho, S. M. Wu, H. P. Zhu, A. B. Yu and S. T. Tsai: Miner. Eng.,
22 (2009), 986.
17) S. Nag, S. Basu and A. B. Yu: Ironmaking Steelmaking, 36 (2009),
509.
18) S. Wu, C. Fu, C. Liu, J. Xu and M. Kou: J. Iron. Steel Res. Int., 18
(2011), 8.
19) D. S. Tsai, C. C. Lo, J. R. Jeng and C. K. Ho: J. Chin. Inst. Eng., 11
(1988), 199.
20) US Army Corps of Engineers: Eng. Manual EM1110-2-1902, U.S.
Army Corps of Engineers, Washington, DC, (2003).
21) T. Mitra and H. Saxén: Metall. Mater. Trans. B, 45 (2014), 2382.
22) T. Mitra and H. Saxén: Mater. Manuf. Process., 30 (2015), 474.
23) P. Austin and D. Vass: Chemeca 99, Chem. Eng., Inst. of Eng.,
Institution of Engineers, Australia, (1999), 363.
24) J. B. Knight, H. M. Jaeger and S. R. Nagel: Phys. Rev. Lett., 70
(1993), 3728.
25) A. Westerlund: Åbo Akademi University, Turku, (2010), M.Sc. thesis.
10