Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Powder Technology 286 (2015) 722–731

Contents lists available at ScienceDirect

Powder Technology

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

Heat transfer in indirect heated rotary drums filled with monodisperse


spheres: Comparison of experiments with DEM simulations
H. Komossa a,⁎, S. Wirtz a, V. Scherer a, F. Herz b, E. Specht b
a
Department of Energy Plant Technology (LEAT), Ruhr-University Bochum, Universitätsstraße 150, 44780 Bochum, Germany
b
Institute of Fluid Dynamics and Thermodynamics, Otto von Guericke University Magdeburg, Universitätsplatz 2, 39106 Magdeburg, Germany

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

Article history: Numerical simulations with the discrete element method (DEM) and corresponding experimental investigations
Received 4 May 2015 were carried out to understand and to quantify the heat transfer in indirect heated rotating drums. Monodisperse
Received in revised form 2 July 2015 glass spheres (diameter 2 mm) were used and the bulk movement was kept within the rolling motion mode
Accepted 15 July 2015
(rotational speed between 1 and 9 rpm). The focus is on the heat transfer between the covered wall and the
Available online 23 July 2015
particles in contact with this wall, as well as between the particles on the free bed surface and the adjacent
Keywords:
fluid. Radiative heat transfer has been neglected due to the low maximum temperature within the system
Discrete element method (474 K). Effective heat transfer coefficients for the heat fluxes mentioned were derived from the DEM simula-
Numerical simulation tions, considering the actual particle velocities on the free bed surface and on the wall.
Experimental investigation The particle movement and the heat transfer resulting from the simulations show good agreement with the
Rotary drum experiments in general and thus allow the calculation of the effective heat transfer coefficients for the range of
Contact heat transfer parameters considered in the current study.
© 2015 Published by Elsevier B.V.

1. Introduction Fundamental investigations of the heat transfer within packed and


stirred beds were performed [3,27], early simulations were carried out
Heating or cooling of particulates is an essential step in many industri- [4,26] and analytical models of the material movement were developed
al processes dealing with granular materials. Depending on the properties [5–7] and extended to macroscopic engineering descriptions of the heat
of the material or details of the process envisaged, different equipment is transfer in rotary kilns [8]. The major drawback of approaches approxi-
employed. Especially in the basic materials and minerals processing in- mating the bulk material as a continuum is the assumption of continu-
dustry, where large and continuous throughputs of bulk material are ous temperature distributions within the material. If larger particles
common, rotary kilns are the device of choice. In these kilns heat transfer are considered and the thermal inertia of the particles gains importance,
may be performed by direct heating as in cement kilns, exposing the only a discrete, particle based description provides the information
product to a flame, or—for more sensitive materials—by indirect heating needed, i.e. to ensure that the particles are exposed to a given tempera-
where the hot wall of the drum supplies the heat to the granular material. ture range for a closely controlled time. The mechanism controlling the
The heat transfer in rotary kilns strongly depends on the combined conductive heat transfer in a mechanically agitated bulk, leading to an
transport of mass and heat within the granular material. While the de- “apparent”, time averaged effective conductivity, is the transient heat
sign parameters (diameter, length and inclination) roughly determine conduction during the collisions or enduring contacts of the particles.
the residence time for a given throughput, the operational parameters Zhou et al. [8] investigated this detail with the finite element method
(e.g. rotational speed, filling degree) in conjunction with the material and concluded that the respective analytical models overestimate the
properties (particle size, particle shape, dynamic angle of repose) con- heat transfer. A theoretical study to determine the effective thermal
trol the three-dimensional movement of the solid material and thus in- conductivity of a stationary granular material using DEM can be found
fluence residence times in different regions or contact times with in Zhang et al. [9]. Coupled DEM/CFD simulations of rotary kilns are re-
heating surfaces. ported from [10] with the focus on the influence of the interstitial gas.
Due to its industrial relevance the heat transfer to the bulk material However, no comparison with experiments is given. The influence of
in rotating drums has already been investigated for a long time [1,2]. the drum wall geometry (wavy drum wall) on the heat conduction
within the bulk was investigated in a DEM study by Gui and Fan [11]
neglecting convective effects. The wavy drum walls are found to en-
⁎ Corresponding author. Tel.: +49 234 3226373; fax: +49 234 3214803. hance the heat transfer inside the drum. This study was also based on
E-mail address: komossa@leat.ruhr-uni-bochum.de (H. Komossa). simulations only. The current study compares DEM simulations with

http://dx.doi.org/10.1016/j.powtec.2015.07.022
0032-5910/© 2015 Published by Elsevier B.V.
H. Komossa et al. / Powder Technology 286 (2015) 722–731 723

corresponding experiments and therefore allows a statement how reli- The particle velocity (particle velocity on the free surface of the bed
ably DEM is able to predict indirect heat transfer in rotating drums. as well as the slip velocity between particles and drum wall) is recorded
Note that it has already been shown in [6] that the mechanical be- by a video camera which allows the determination of particle move-
haviour of the bed can be reliably reproduced by DEM for the current ment relative to the drum wall. Note that these measurements were un-
experimental configuration. The simulated dynamic angle of repose, dertaken within an acrylic glass drum lined with sandpaper to provide
the thickness of the active layer and the particle velocity on both the the same roughness as in the sandblasted steel drum. Same mechanical
bed surface and the drum wall were compared with measurements behaviour was verified by a comparison of the dynamic angle of repose
for different Froude numbers. It has been shown that the particle move- in both drums. A small sandpaper-free gap (30 × 30 mm) of the acrylic
ment in rotating drums can be described by DEM simulations with good glass wall allows recording the relative velocity between bulk and drum
accuracy in the rolling and slumping bed motion. The present work ex- (slip).
tends the prior results by corresponding heat transfer experiments and The Froude number used in the following to characterize the indi-
simulations. vidual experiments is defined as the ratio of centrifugal force to gravita-
tional force:

2. Experimental setup
ω2  R 2πn
Fr ¼ with ω ¼ ð1Þ
g 60
The test rig shown in Fig. 1 was used for the measurements. The
drum consists of two steel tubes with a length of 500 mm (cold part)
and 350 mm (heated part), with a wall thickness of t = 2 mm and an with rotational frequency n, drum radius r and gravitational force g.
inner diameter of D = 300 mm. The inclination angle of the drum is The drum was driven by an electric motor with 1.5 kW and controlled
zero. Batch experiments are carried out in the heated part of the drum by a frequency converter to obtain constant rotational speeds. Rotational
filled with bulk material. The inner wall of the heated drum was speeds between 1 and 9 rpm were used in the current experiments to en-
sandblasted to provide a greater roughness. sure the rolling motion mode. The corresponding Froude numbers are
The left part of the drum was heated by an electrical heating band on Fr = 1.7 × 10−4 and 1.4 × 10−2, respectively. U = 0 rpm (static bed) is
the outside of the steel drum and insulated against the other drum by a used as a reference case.
closed split wall made of fibre alumino-silicate-glass coated with silicone. Monodisperse glass spheres with dP = 2 mm were used as bed ma-
The unheated drum provides space for the measurement instrumenta- terial. The material properties of the glass spheres are shown in Table 1.
tion of the heated drum. On the inner wall of the heated drum a friction The filling degree in the experiments was f = 15%.
thermocouple was installed to measure the wall temperature and to con-
trol the power supply to the heating band. A slip ring at one end of the 3. Heat transfer within DEM
drum ensured energy supply of the rotating heating band. A measuring
rod with eight thermocouples (5 mm distance between the thermocou- In a discrete, particle based description the movement of each single
ples, the first thermocouple being 5 mm from the drum wall) was particle in the bulk is described by simultaneous integration of
installed in the bulk to measure the bulk temperature at eight radial po- Newton's, Euler's and heat conduction equations in the individual parti-
sitions. Note that a separate thermocouple was located outside the bulk cles while incorporating all mechanical and thermal interactions among
(centrically above the free bed surface) in the gas phase to measure gas the particles, the wall and the free surfaces [12–14]. In this study the dis-
phase temperature approximately 5 mm above the bed surface. The heat- crete element code of LEAT [15–18] was used. The particles are consid-
ed drum is closed by a steel plate on the left front end. The steel plate has ered as soft spheres, the linear spring dashpot model is used for normal
a central orifice of 150 mm diameter which is open to the surrounding air. and tangential forces. To allow particle rolling translational as well as ro-
According to the manufacturer, the uncertainty of the used thermocou- tational motion are resolved. Further details on the mechanical models
ples is 0.75% of the measured value or at least 2.5 K. The uncertainty of used may be found in [19].
the instrumentation is approximately 1.0 K and the uncertainty by heat Depending on the material pairing (particle–wall and particle–
conduction along the measuring rod can be estimated to be below particle) the coefficients of friction, rolling and restitution have dif-
0.5 K. The analysis showed a maximal uncertainty of 5.2 K. Corresponding ferent values. The required coefficients for the two pairings were de-
error bars are complemented in the figures. termined experimentally [17] and are summarised in Table 2.

Fig. 1. Assembly of the experimental rig.


724 H. Komossa et al. / Powder Technology 286 (2015) 722–731

Table 1 Table 3
Material properties of investigated glass spheres (experiment and simulation). Simulation parameters.

Properties Characteristic value Unit Parameter Characteristic value Unit

Densitya 2500 kg/m3 Simulation time step 2.2 × 10−5 s


Surface hardness (Mohs)a 6 – Collision stiffness 100,000 N/m
Young's modulusa 7000 N/m3 Damping coefficient 0.04 kg/s
Poisson ratiob 0.2 – Particle diameter 0.002 m
Specific heat capacityb 720 J/(kg K) Maximum number of particles 65.000 –
Thermal conductivityb 1.06 W/(m K)
a
Manufacturer's data.
b
Literature data.

The required thermal constriction resistance Rc is computed as:


The coefficient of rolling friction was calculated from measured
rolling speeds following the energy balance procedure of Kuchling 1
Rc ¼ ð3Þ
[20]. The coefficient of restitution was obtained from impact experi- 2  khm  r c
ments with a high speed camera. The glass spheres were fixed on a
board to measure the static and dynamic coefficient of friction. For with the contact radius rc of the circular contact zone:
this purpose the weight-loaded board was pulled over a surface of
0   113
glass (representing the material combination of the contact between 3  1−θ2hm  j F n j  r hm
two spheres) and sandpaper (contact between sphere and wall of the rc ¼ @ A: ð4Þ
2  Ey;hm
drum). The maximum force FH and the dynamic frictional force FG
were measured with a spring scale. All measurements were repeated
10 times, the results and the associated standard deviation are shown Fn is the normal force vector acting on the two contacting particles
in Table 2. (or on the wall and the particle in contact with this wall). Note that
In contrast to the experiments, where the drum had a fixed length of the value of Young's modulus used in Eq. (4) is the actual material
350 mm, the length of the drum has been restricted to 40 mm (20 times value, which is different from the value used in the DEM soft-sphere
the particle diameter) in the simulations. It has been checked that the contact model. Ey,hm, rhm, θhm and khm are the harmonic mean values of
influence of the sidewalls on the result in the centre plane is sufficiently Young's modulus, sphere radius, Poisson's ratio and thermal conductiv-
small at a length of 20 times the particle diameter. At a filling degree of ity of the two particles in contact calculated by the following equation:
f = 15% the number of particles is between 2500 and 65,000 in the sim-
ulations with a length of 40 mm (Table 3). Xi  X j
X hm ¼ 2  with X ∈ Ey ; r; θ; k: ð5Þ
For the current study the following heat transfer mechanisms have Xi þ X j
been considered:
1) Thermal conduction within solid particle For rj ➔ ∞ the equations can also be used to calculate the contact area
2) Thermal conduction through the contact surfaces of two particles between a particle and a wall.
3) Thermal conduction through the contact surfaces of a particle and a The stagnant gas zone resistance (which is assumed to be stagnant)
wall in the void around the particle contact area is obtained from
4) Thermal conduction through the fluid film near the contact surface
lg
of two particles and a particle and the drum wall Rg ¼ ð6Þ
kg  Ag
5) Heat transfer by convection from a gas surrounding the packed bed
to the bed
with the exposed surface Ag and the average distance lg:
Radiant heat transfer is neglected due to the low temperatures with-
in the bed of max. 474 K. Ag ¼ 2  π  r 2 −π  r 2c ; ð7Þ
Modelling of contact heat transfer between the particles (and a wall
 π
and the particles) and conduction through the interstitial gas (mecha-
r 2hm  1−
nism 2) and 4)) are based on the approach described by Vargas and Mc- lg ¼ 4 : ð8Þ
Carthy [21]. Here the elasticity theory of Hertz [22] is used to determine r hm −r c
the (small) contact area between two particles where conduction be-
tween the two solids occurs. The two heat fluxes, since they occur in For the computation of the temperature distribution within the par-
parallel, are combined from the resistances as: ticles the unidimensional heat conduction equation is solved numerical-
ly for each particle. This is based on a radial discretisation, thus
  implicitly assuming spherical symmetry and averaging the heat fluxes
1 1  
Q i; j ¼ þ  T surf ; j −T surf ;i : ð2Þ over the individual particle surface. Simultaneous integration of the
Rc Rg
heat conduction equation for all moving particles results in a tempera-
ture field information combining the effects of heat conduction on
both particle and drum scale. Eq. (9) shows the simplified differential
equation which has to be solved [23].
Table 2
2
Parameters used in the simulations. 1 ∂ 1 ∂T
 ðr  T Þ ¼  ð9Þ
Material Glass–glass Glass–steel Standard r ∂r2 ψ ∂t
deviation

Dynamic coefficient of friction [–] 0.1966 0.254 0.023 The boundary condition on the free bed surface is defined by the
Static coefficient of friction [–] 0.231 0.735 0.031 convective heat exchange with the particles on the surface of the so
Coefficient of rolling friction [m] 0.62 × 10−4 1.90 × 10−4 5.67–10−5 called “active layer” [2], where the particles continuously move down-
Coefficient of restitution [–] 0.904 0.768 0.071
hill, as caused by the dynamic angle of repose (see Fig. 2).
H. Komossa et al. / Powder Technology 286 (2015) 722–731 725

with Ra being:

3  
li  g  T surf ;i −T fluid
Ra ¼ Pr  Gr ¼ Pr  ð14Þ
ν2  T fluid

and Nuforced,i single particle by Gnielinski [25] with the following


equation:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Nuforced;i ¼ 2 þ Nu2 forced;lam;i þ Nu2 forced; turb;i ð15Þ

with
pffiffiffiffiffiffi p ffiffiffiffiffi
Nuforced;lam;i ¼ 0:664  Re  Pr ð16Þ
3

and

0:037  Re0:8  Pr
Nuforced;turb;i ¼ : ð17Þ
1 þ 2:443  Re−0:1  ðPr 2=3 −1Þ

The determination of the Nusselt number requires the knowledge of


the Reynolds number Re and thus the relative velocity between particle
and surrounding gas. In the current context the surrounding fluid is as-
sumed to be stagnant and the gas velocity corresponds to the particle
Fig. 2. Heat fluxes in a rotary kiln. velocities as known from the DEM simulation.

Due to the partially open left side wall of the drum convection be- 3.1. Effective heat transfer coefficients
tween drum and surroundings can occur. Therefore the fluid above
the active layer has a lower temperature than the bulk in the drum. For an indirect heated rotary drum the heat transfer coefficient be-
Thus the heat loss of the i'th particles on the free bed surface can be de- tween the covered wall and the particles in contact with this wall is of
fined as: great importance. With DEM simulations an effective heat transfer coef-
ficient can be derived which reflects the influences like particle sizes,
   drum diameter or rotational speed. This heat transfer coefficient can
Q i;∞ ¼ α i  Ai  T Fluid −T surf ;i ð10Þ
be used later on in continuum models for drum layout. Because contin-
uum models always assume a uniform temperature within the bulk all
where Tfluid is the fluid temperature, measured by the thermocouple
further definitions are based on a mean bulk temperature TS (although
above the free surface as mentioned section experimental set-up, Tsurf,i
local temperatures in DEM would be available).
is the surface temperature of the particle, and Ai is the surface area of
For the definition of the effective heat transfer coefficient the follow-
the particle exposed to the fluid. For simplicity it has been assumed
ing three heat fluxes in a rotary kiln (see Fig. 2) must be balanced:
that half of the sphere surface area is exposed to the fluid. αi is the
heat transfer coefficient. 

Q WS Conductive heat flux wall-bulk


The heat transfer can either be calculated by combining DEM simu- 

QS Heat flux absorbed by the bulk


lation with a CFD code which increases the computational effort signif- 

Q FBS Convective heat flux gas-bulk


icantly [10]. Therefore in the current simulations the following
simplifying approach has been chosen. Convective heat transfer is calcu-
lated by Nusselt number correlations: The conductive heat flux between the covered wall and the particles
at this wall is calculated as follows:
Numixed;i  λi  X 

αi ¼ ð11Þ Q WS ¼ α WS  ðT W −T S Þ  A ¼ Q i;wall ð18Þ


li

where TW is the wall temperature, and TS is the mean bulk temperature.


with the thermal conductivity of the fluid λi, and the characteristic
TS is calculated from the DEM simulations as the mean value of the par-
length li which is equivalent to the particle diameter. Note that for
ticles temperatures averaged over the particle volume. A is the wall sur-
such a simplified approach TFluid above the particle bed in Eq. (10)
face area covered with particles and αWS is the effective heat transfer
must be known from measurements (or has to be defined based on rea-
coefficient (wall–bulk) of interest here.
sonable assumptions) which is not required for a coupled DEM/CFD
The heat flux absorbed by the bulk is defined as:
simulation.
Since in the case of the experiment mixed convection (overlap of  dT S X 

free and forced convection) can be expected the approach of Churchill Q S ¼ m  cP  ¼ Q s;i ð19Þ
dt
[24] was chosen. Numixed,i is then calculated as follows:
where m is the mass of all particles, cP is the heat capacity of the bulk
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
and dTS/dt is the temporal change of the mean bulk temperature.
Numixed;i ¼ 3 Nu3 forced;i þ Nu3 free;i ð12Þ
The convective heat transfer is defined as follows:
 X 
with Nufree,i for the flow around a single particle (Raithby [25]): Q FBS ¼ α FBS  A FBS  ðT Fluid −T S Þ ¼ Q i;∞ ð20Þ

 
14 
Pr with Q i;∞ from Eq. (10). αFBS denotes the total convective heat transfer
Nufree;i ¼ 0:56   Ra þ 2 ð13Þ
0:846 þ Pr coefficient between bulk and fluid.
726 H. Komossa et al. / Powder Technology 286 (2015) 722–731

The resulting effective heat transfer coefficient between wall and


bulk αWS is consequently defined as follows:

dT S
m  cP  −α FBS  A FBS  ðT Fluid −T S Þ
α WS ¼ dt : ð21Þ
ðT W −T S Þ  A

4. Results and discussion

As a reference situation the heat transfer without any influence of


the bed movement with a static bulk is included. Note that the temper-
atures of the wall and the air just above the particle layer (fluid temper-
ature) were measured in the experiments and used as time-dependent
boundary conditions for the DEM simulations.
Fig. 4. Frequency distribution of particle velocity at different time points in a non-steady
As a first general insight into the development of the three dominat- state (numbers represent the time in s).
ing heat fluxes (Eqs. (18), (19), (20)) over time, a representative DEM
simulation is depicted in Fig. 3 (D = 300 mm, dP = 2 mm, f = 15%,
U = 1 rpm). The heat flux absorbed by the bulk rises quickly over
time and has the same value as the conductive heat flux from the wall After start of the rotation the frequency distribution of this velocity fluc-
to the bulk in the first 200 s, the heat loss at the surface (convective tuates (points 1, 2 and 3) until the steady state is reached after approx-
heat flux) is nearly zero at this time and increases during the further imately four seconds (point 4).
course up to a maximum value of 5.5 W while the two other heat fluxes Evaluating the frequency distributions of the particle velocity in the
decrease. As can be seen, the heat loss at the free surface can be steady state regime (values in the time span from t = 10 to t = 20 s)
neglected in the first 2–3 min, since the temperature difference be- leads to the results presented in Fig. 5. For increasing rotational speed,
tween bulk and fluid is almost zero. However, it is necessary to include the mean value of the distribution rises accordingly and the distribution
the heat loss for a correct calculation of the bed temperature when time broadens. Note that comparisons with measurements are not given
is progressing. Especially for the steady state situation it significantly in- here, but can be found in [6] and show good agreement.
fluences the terminal temperature of the bulk as will be shown later.
As Eq. (21) shows, the convective heat transfer coefficient bulk-fluid
αFBS directly has influence on the effective heat transfer coefficient wall- 4.3. Effective heat transfer coefficient bulk–fluid
bulk αWS. Therefore, in the following, the convective heat transfer coef-
ficient and its dependence on particle velocity at the bulk surface are In Fig. 6 the frequency distribution of the effective heat transfer coef-
analysed first before discussing details of the temporal evolution of ficient based on the particle velocities presented above is shown. Anal-
bed temperature and bulk-wall heat transfer coefficient. ogously to the particle velocity distribution, the mean value of the
distribution presented increases with increasing rotational speed and
4.1. Effective heat transfer between bulk and fluid the distribution broadens. The mean value at a rotational speed of
U = 3 rpm is αFBS = 46.2 W/(m2 K) and at a rotational speed of
4.2. Particle velocity on the free bed surface 9 rpm αFBS = 54.3 W/(m2 K), the increase is 17.5%. Note that in case
The total heat flux between the bed surface and the adjacent fluid of a static bulk—where only free convection is present—the heat transfer
consists of free convection and forced convection. The Nusselt number coefficient has a value of αFBS = 39.3 W/(m2 K). The influence of forced
calculation for forced convection as presented in chapter 3 requires convection is thus relatively small.
the particle velocity of each individual particle which is known from
the DEM simulations. 4.4. Temporal evolution of bed temperature
Fig. 4 depicts the frequency distribution of the magnitude of the par-
ticle velocity on the free bed surface at different times for a rotational Fig. 7 shows the temperature evolution in the static bulk and in the
speed of U = 3 rpm, therefore the particle velocity of every single par- moving bulk for different rotational frequencies at t = 25 min.
ticle at one time step was saved within DEM and the relative frequency It can clearly be seen that the drum wall is heating the bulk and how
was afterwards calculated and plotted over particle velocity. At the be- the temperature profile from the wall to the inner bulk develops. The
ginning the drum is in rest and the bulk is correspondingly not moving.

Fig. 5. Relative frequency of particle velocity at the free bed surface for different rotational
Fig. 3. Relevant heat fluxes (DEM). speeds.
H. Komossa et al. / Powder Technology 286 (2015) 722–731 727

Fig. 6. Relative frequency of convective heat transfer coefficient at the free bed surface.

higher the number of revolutions the higher the bulk core temperature, Fig. 8. Heat conduction into the static bulk.
an indication that particle mixing is accelerating bulk heating.
A comparison with experiments for the static bulk (u = 0 rpm, filling
degree 15%) is given in Fig. 8. The outer wall was heated to a maximum
temperature of TWall = 474.3 K. T8-simulated (close to free bed surface) is initially higher than T5-
The wall is heated with 1.5 kW and the temperature continuously simulated (further within the bed) which indicates that initially heat
increases from 293 K to 443 K after 100 min. The thermocouples T1 to is transferred from the gas phase to the bulk. When time is progressing
T8 measure the temperature in the bulk at different levels, T1 is located T8 becomes lower than T5 (see for example at 6000 s) which means that
close to the drum wall and T8 is located near the bed surface (but within heat is then transferred form the bulk to the gas phase. The same effect
the bed). The simulated particle temperatures at these designated levels is present in the experiments.
show a very good agreement with the measured temperatures from the Fig. 9 (U = 1 rpm) and Fig. 10 (U = 3 rpm) show the temperature
experiment. The maximum deviation is 9.6% at thermocouple T5 after a evolution of the heated wall, the thermocouples T1 and T8 and the air
time of 17 min and decreases with time. Steady state has still not been atop the particle bed. The maximum temperature difference between
reached after 17 min. T1 and T8 is ΔT = 24.3 K at U = 1 rpm and is only ΔT = 7.6 K at
Note that the fluid inside the drum is heated faster initially by the U = 3 rpm, thus increasing the rotational frequency reduces the tem-
hot drum wall than by the bulk surface. This is why the temperature perature gradient in the bulk.

Fig. 7. Particle temperature evolution at t = 25 min for the different rotational frequencies.
728 H. Komossa et al. / Powder Technology 286 (2015) 722–731

Fig. 9. Temperature evolution of the moving bulk at U = 1 rpm.


Fig. 11. Temperature evolution with and without convective boundary conditions.

During heating of the bulk the simulated temperatures are slightly


higher than the temperatures measured in the experiments (T1, T8) set to the same values, as measured in the experiments at U = 1 rpm. By
but the deviations are small; maximum deviation 5.8%. The deviations doing so it is possible to compare the simulations with each other based
may stem from the fact, that the thermocouples in the moving bulk on the same boundary conditions and the only value varied is the rota-
cannot actually measure the particle temperatures but rather reflect tional speed of the drum in the simulations. With increasing rotational
some mean value of particle surface temperature and interstitial air speed, the mean temperature of the bulk increases. The largest step oc-
temperature next to the thermocouple. Differing therefrom the values curs between the bulk in rest and U = 1 rpm while a further increase of
extracted from the DEM simulations are particle surface temperatures the speed of rotation leads to diminishing additional temperature
and thus expected to be higher. However, the correspondence between differences.
the DEM simulations and the experiments is very good. Note that the The temperature difference between bulk in rest and U = 1 rpm has,
small deviation of measured and calculated temperatures in steady for example, a value of ΔT = 27.7 K after a time of t = 2000 s. The
state (t N 2500 s) is a good indication that the assumptions made for difference between U = 1 rpm and 3 rpm is ΔT = 4.5 K and between
the convection from gas to bulk at the free surface are reasonable be- U = 3 rpm and 9 rpm ΔT = 1.9 K. This clearly shows that high heat
cause otherwise a remaining final temperature deviation would occur. transfer rates are already obtained at fairly low rotational frequencies
This holds true for U = 1 and 3 rpm as well as for the static bed, (provided that the rolling mode can be established) and that any addi-
respectively. tional gain requires progressive effort.
To further highlight the importance of correct convective boundary
conditions Fig. 11 shows the temperature evolution with and without
(adiabatic bed surface) convective boundary conditions. As expected, 4.5. Effective heat transfer wall–bulk
for small times, the differences are small as convective heat flux is low
(as has been shown in Fig. 3). However, with increasing time the tem- 4.6. Particle velocity wall (slip)
perature difference rises to a maximum value of T = 12 K. Macroscopic models for the particle movement in drums commonly
This also shows the importance of correct convective boundary con- assume no slip conditions between the drum wall and the adjacent par-
ditions for a precise calculation of the heat fluxes and the resulting effec- ticle layer. This is a crucial assumption because macroscopic correlations
tive heat transfer coefficients wall-bulk within the drum. For a (Eq. (22)) for the effective heat transfer coefficient wall–bulk typically
rotational speed of U = 1 rpm the resulting effective heat transfer coef- depend on the contact time. This can be seen in Eq. (22) which shows
ficients are αWS = 95.9 W/(m2 K) without convective boundary condi- the correlation of Li et al. (2005) which is an extension of the approach
tions and αWS = 71.2 W/(m2 K) with convective boundary conditions. by Sullivan et al. (1975) [7].
The heat loss at the free bed surface reduces the effective heat transfer
coefficient between wall and bulk. For further details on heat transfer
coefficient wall-bulk see Section 4.3.
In Fig. 12 the evolution of the simulated mean bulk temperatures TS 1
α WS ¼ ð22Þ
(average of all particles in the rotary kiln) are plotted for different rota- Γ  dP 1
þ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
tional speeds. In all five simulations wall and air temperature have been λG ρS  cp;S  λS
2
π  t Contact

Fig. 10. Temperature evolution of the moving bulk at U = 3 rpm. Fig. 12. Temperature evolution at different rotational frequencies.
H. Komossa et al. / Powder Technology 286 (2015) 722–731 729

As can be seen, there is a good agreement between the simulations


and the experiments. The particle movement relative to the wall in-
creases continuously with increasing the Froude number and the parti-
cle rotation (rolling) shows a high level from the beginning and
increases only slightly, indicating that the particles at the wall are rolling
along the wall. The slight increase of the particle rotation leads to the as-
sumption that the particles are not rolling faster with increasing Froude
number though they are slipping stronger. The mean relative slip differs
between 45.5% at U = 1 rpm and 27.9% at U = 3 rpm.
These results demonstrate that the assumption of no slip conditions
between drum wall and adjacent particle layer in macroscopic models
(Eq. (22) for example) for particle movement in drums is not accurate
and that the slip has to be taken into account.

4.7. Effective heat transfer coefficient wall–bulk


Fig. 13. Frequency distribution of particle velocity at the wall.

In the following the effective heat transfer coefficient between the


with covered wall and the moving bulk is determined, again for the
experiments and the simulations. The results, covering a range of rota-
γ tional speeds from U = 0 rpm to 3 rpm are shown in Fig. 15. The effective
t Contact ¼ : ð23Þ heat transfer coefficient in the stationary bulk turns out with an average
πn
value of about αWS = 25.4 W/(m2 K) (experiment) and αWS = 28.5 W/
(m2 K) (DEM simulation), which is a deviation of 10.9%. If the bulk is
The contact time will be smaller for no slip conditions compared to moving, these deviations between the mean values reduce with increas-
particle slip. Since time-dependent trajectories of all particles are avail- ing rotational speed from 8.6% at U = 1 rpm to 0.7% at U = 3 rpm. The
able from the DEM simulations, all particles close to the wall can be error bars denote the deviations from the mean values. The larger scatter
identified and the velocity component tangential to the wall (slip) can with increasing rotational speed of measuring points around the mean
be computed. values results from the fact that the wall temperature was measured by
The simulation result as a function of rotational speed is shown in a friction thermocouple with direct contact to the moving wall. With
Fig. 13 for the centre plane of the rotating drum (middle of the drum). higher rotational speed this thermocouple does not stick sufficiently
The frequency distribution is chosen to present the tangential compo- steady to the rough wall which is why these results were excluded from
nent of the particle velocities and to compare these values with the ve- further analysis. Nevertheless, these results show that the prediction of
locity resulting from the solid body rotation of the drum wall, sketched the heat transfer coefficients with DEM is satisfactory.
as vertical lines in Fig. 13. In Fig. 16 the effective heat transfer coefficients obtained from the
The rotational frequency of the drum in the simulation with the DEM simulations and the relative velocity between the particles next
smallest Froude number (Fr = 1.68 × 10−4) is ω = 0.105 rad/s indicat- to the wall and the wall (slip) are plotted over the Froude number. Like
ed by the vertical line. The particle velocities differ from the rotational the temperature, the heat transfer initially increases rapidly until a linear
frequency and show a skew distribution positioned at an average increase with the Froude number is established. The effective heat
value of ωP = 0.035 rad/s. Thus, there is a fairly distributed slip between transfer coefficient rises from αWS = 28.5 W/(m2 K) at U = 0 rpm to
the particles and the wall. The slip increases with the Froude number αWS = 103.9 W/(m2 K) at 3 rpm and αWS = 127.9 W/(m2 K) at 9 rpm
(rotational speed of the drum) from 0.048 rad/s (Fr1 = 1.68 × 10−4) which corresponds to increases of about 265% and 349% with respect
to 0.336 rad/s (Fr4 = 1.36 × 10−2) and in addition the frequency distri- to the initial value. The quality of mechanical mixing is not increasing lin-
bution broadens considerably. early, because of the rising slip and therefore the effective heat transfer is
In Fig. 14 the mean angular velocities at the drum wall in the DEM not increasing linearly as well.
simulation (transparent drum wall) for the different Froude numbers As shown here, the influence of rotational frequency is much more
are compared with experimental angular velocities measured at the pronounced on the effective heat transfer coefficient between covered
transparent drum wall. They are plotted together with the wall surface wall and bulk as between bulk and fluid. That is why the bulk is heating
velocity as resulting from the solid body rotation of the drum wall. up faster with increasing rotational speed.

Fig. 14. Slip velocity and rotational frequency around particle axis. Fig. 15. Comparison of effective heat transfer coefficients (experiment and DEM).
730 H. Komossa et al. / Powder Technology 286 (2015) 722–731

When correcting the contact time with the simulated contact time (in-
cluding slip) the agreement improves.

Nomenclature
Latin symbols


Q FBS [W] Convective heat flux gas–bulk




QS [W] Heat flux absorbed by the bulk




Q WS;ε [W] Radiation wall–bulk




Q WS [W] Conductive heat flux wall–bulk


Ag [m2] Exposed surface
Asurf [m2] Particle surface area
Ey [N/mm2] Young's modulus
FG [N] Dynamic frictional force
FH [N] Static frictional force
Fig. 16. Effective heat transfer plotted over Froude number. Fn [N] Normal force vector
Rc [K/W] Thermal constriction resistance
Rg [K/W] Stagnant gas zone resistance
cp [J/(kg K)] Specific heat capacity
Finally Fig. 16 shows a comparison of the simulated effective heat
dp [m] Particle diameter
transfer coefficient wall-bulk in comparison to the correlation of Li keff [W(m K)] Thermal conductivity (lateral mixing)
using the standard value of 0.085 for gamma as well. As can be seen lg [m] Average distance length
the Li correlation is in general good agreement with simulations and rc [m] Contact radius of the circular contact zone
f [%] Filling degree
the trend of the curve is well reproduced. However effective heat trans-
Gr [–] Grashof number
fer coefficients are slightly overpredicted by Li's correlation. When l [m] Overflow length
correcting the contact time in the Li correlation with the simulated con- Nu [–] Nusselt number
tact time (including slip) the agreement becomes very good. Note that Pr [–] Prandtl number
Li has found that using Γ = 0.085 as standard value as suggested by Sul- Ra [–] Rayleigh number
D [m] Drum diameter
livan does not reproduces measurements with sufficient accuracy. Li
Q [W] Heat transfer rate
suggested to gamma might be variable (suggested values between T [K] Temperature
0.096 and 0.198). As a future extension of Li's correlation it might be V [m3] Particle volume
more promising to adapt tContact. A dependency of tContact could for ex- g [m/s2] Acceleration of gravity vector
k [W/(m K)] Thermal conductivity
ample be derived from DEM simulations.
m [kg] Mass
q [W/m2] Heat flux
5. Summary and conclusion r [m] Particle radius
t [mm] Reactor tube wall thickness
In the current study the heat transfer in indirect heated rotary v [m/s] Velocity

kilns was experimentally investigated and compared with DEM simula- Greek symbols
tions. Bed material were monodisperse glass spheres with a diameter of αWS [W/(m2 K)] Effective heat transfer coefficient wall–bulk
dP = 2 mm in a drum of D = 300 mm. Rotational speed was varied from βi [m/s] Diffusional mass transfer coefficient of species i
τ [kg/(m s2)] Stress tensor
1 to 9 rpm with 0 rpm as a reference case.
αBF [W/(m2 K)] Effective heat transfer coefficient bulk–fluid
The heat transfer model implemented in the DEM-code was initially γ [–] Filling angle
verified by comparing simulations and measurements of the heating of λ [W/(m K)] Thermal conductivity
a static bulk in the rotary kiln. The simulation of a static bed allows dif- ε [–] Emissivity
ferentiating between conductive heat transfer and effects which stem θ [–] Poisson's ration
μ [Pa s] Dynamic viscosity
from particle movement. With a maximum deviation of 9.6% between
ν [m2/s] Kinematic viscosity
simulated and measured bed temperatures the validity of the heat con- ρ [kg/m3] Density
duction model implemented in the DEM-code was shown. σ [W/(m2 K4)] Stefan–Boltzmann–Constant (σ = 5.67 ⋅ 10−8)
In the current study the convective heat transfer coefficient between τ [–] Catalyst tortuosity
free bed surface and surrounding gas were incorporated by Nusselt ψ [m2/s] Thermal diffusivity
ϕ [–] Porosity/void fraction
number correlations for free and forced convection. The calculation of
forced convection needs the knowledge of the relative velocity between
Acknowledgement
particle and gas phase. By assuming that the gas phase is stagnant the
problem reduces to the detection of the particle velocity. Frequency dis-
The current study has been funded by the German Federation of In-
tributions of particle velocities at the free bed surface were extracted
dustrial Research Associations (AiF) within the project AiF 17133 BG/2
from DEM simulations. It could be shown that the mean value of the dis-
and by the German Research Foundation (DFG) within the project
tribution rises and the distribution broadens for increasing rotational
SCHE322/10-1. The authors would like to acknowledge the generous
speeds. In case of a static bulk the heat transfer coefficient has a value
support.
of αFBS = 39.3 W/(m2 K) and increases by only 38.1% (9 rpm). The influ-
ence of the forced convection is thus relatively small.
In a further step the effective heat transfer coefficient between the References
covered wall and the moving bulk in dependence on the mechanical ag- [1] W.C. Saeman, T.R. Mitchell, Analysis of rotary dryer and cooler performance, Chem.
itation of the bulk has been analysed. The results clearly showed that Eng. Prog. 50 (9) (1954) 467–475.
[2] G.W.J. Wes, A.A.H. Drinkenburg, S. Stemerding, Heat transfer in a horizontal rotary
high heat transfer rates are already obtained at fairly low rotational fre-
drum reactor, Powder Technol. 13 (1976) 185–192.
quencies (provided that the rolling mode can be established) and that [3] E.U. Schlünder, Heat transfer to packed and stirred beds from the surface of im-
any additional gain required progressive effort. The prediction of the ef- mersed bodies, Chem. Eng. Process. 18 (1984) 31–53.
fective heat transfer coefficients with DEM is satisfactory. A comparison [4] A. Sass, Simulation of the heat-transfer phenomena in a rotary kiln, Ind. Eng. Chem.
Process Des. Dev. 6 (1986) 532–535.
of the simulation results with Li's correlation showed that the effective [5] X. Liu, J. Zhan, E. Specht, Y. Shi, F. Herz, Analytical solution for the axial solids trans-
heat transfer coefficients are slightly overpredicted by the correlation. port in rotary kilns, Chem. Eng. Sci. 64 (2009) 428–431.
H. Komossa et al. / Powder Technology 286 (2015) 722–731 731

[6] H. Komossa, S. Wirtz, V. Scherer, F. Herz, E. Specht, Transversal bed motion in rotat- [17] D. Höhner, S. Wirtz, V. Scherer, Experimental and numerical investigation on the in-
ing drums using spherical particles: comparison of experiments with DEM simula- fluence of particle shape and shape approximation on hopper discharge using the
tions, Powder Technol. 264 (2014) 96–104. discrete element method, Powder Technol. 235 (2013) 614–627.
[7] F. Herz, I. Mitov, E. Specht, R. Stanev, Experimental study of the contact heat transfer [18] H. Kruggel-Emden, Analysis and improvement of the time-driven discrete element
coefficient between the covered wall and solid bed in rotary drums, Chem. Eng. Sci. method(PhD Thesis) Shaker Verlag, 2008.
82 (2012) 312–318. [19] F. Sudbrock, E. Simsek, S. Wirtz, V. Scherer, An experimental analysis on mixing and
[8] J.H. Zhou, A.B. Yu, M. Horio, Finite element modeling of the transient heat conduc- stoking of monodisperse spheres on a grate, Powder Technol. 198 (2010) 29–37.
tion between colliding particles, Chem. Eng. J. 139 (2008) 510–516. [20] H. Kuchling, Taschenbuch der Physik, Fachbuchverlag Leipzig, 17, Auflage, 2004.
[9] H.W. Zhang, Q. Zhou, H.L. Xing, H. Mulhaus, A DEM study on the effective thermal [21] W.L. Vargas, J. McCarthy, Conductivity of granular media with stagnant interstitial
conductivity of granular assemblies, Powder Technol. 205 (2011) 172–183. fluids via thermal particle dynamics simulation, Int. J. Heat Mass Transf. 45 (2002)
[10] D. Shi, W.L. Vargas, J.J. McCarthy, Heat transfer in rotary kilns with interstitial gases, 4847–4856.
Chem. Eng. Sci. 63 (2008) 4506–4516. [22] H. Hertz, Über die Berührung fester elastischer Körper, J. Für Die Reine Und Angew.
[11] N. Gui, J. Fan, Numerical study of heat conduction of granular particles in rotating Mech. 92 (1881) 156–171.
wavy drums, Int. J. Heat Mass Transf. 84 (2015) 740–751. [23] B. Brosch, Erweiterung der Diskreten Elemente Methode zur Simulation
[12] P.A. Cundall, O.D.L. Strack, A discrete numerical model for granular assemblies, bewegter und reagierender Feststoffschütttungen mit der Anwendung auf
Geotechnique 29 (1979) 47–65. Rostfeuerungssysteme(Dissertation) Shaker Verlag, 2012.
[13] A. Di Renzo, F.P. Di Maio, Comparison of contact-force models for the simulation of [24] S.W. Churchill, A comprehensive correlating equation for laminar, assisting, forced
collision in DEM-based granular flow codes, Chem. Eng. Sci. 59 (2004) 525–541. and free convection, AIChE J 23 (1977) 10–16.
[14] F.P. Di Maio, A. Di Renzo, Analytical solution for the problem of frictional-elastic colli- [25] e.V VDI, VDI-Wärmeatlas, Springer Verlag, 2013.
sions of spherical particles using the linear model, Chem. Eng. Sci. 59 (2004) [26] W.L. Vargas, J.J. McCarthy, Heat conduction in granular materials, AIChE J 47 (2001)
3461–3475. 1052–1059.
[15] H. Kruggel-Emden, E. Simsek, S. Rickelt, S. Wirtz, V. Scherer, Review and extension [27] S. Yagi, D. Kunii, Studies on effective thermal conductivities in packed beds, AIChE J 3
of normal force models for the Discrete Element Method, Powder Technol. 171 (1957) 373–381.
(2007) 157–173.
[16] H. Kruggel-Emden, S. Wirtz, V. Scherer, A study on tangential force laws applicable
to the discrete element method (DEM) for materials with viscoelastic or plastic be-
haviour, Chem. Eng. Sci. 63 (2008) 1523–1541.

You might also like