MS - Thesis - FURNACE HEATING
MS - Thesis - FURNACE HEATING
MS - Thesis - FURNACE HEATING
A Thesis in
Industrial Engineering
by
Masahito Kominami
August 2015
The thesis of Masahito Kominami was reviewed and approved* by the following:
Robert C. Voigt
Professor of Industrial and Manufacturing Engineering
Thesis Adviser
Harriet B. Nembhard
Professor and Interim Department Head of Industrial and Manufacturing Engineering
ii
ABSTRACT
The cost of operating reheating furnaces, used for heating mainly billets or blooms in
steel rolling mills is quite large. Therefore, reduction of reheating costs is one of the major
challenges in rolling mills. The reheating furnaces are usually controlled manually by
operators who must respond to changes in downstream rolling conditions. Their reheating
furnace control is not consistent and has been observed to depend on operator characteristics,
experiences or skills.
In many cases, steel billet lots are small, requiring various types of billets/blooms
with different specifications to be heated in a furnace at the same time. This means that it is
hard to find the optimal heating conditions due to changes in product mix. Additionally, once
operational troubles happen at downstream rolling operations, unexpected stoppages are
caused. The operators of furnaces are then required to adjust reheating furnace temperatures
so that billet/bloom overheating does not occur. It is also difficult to re-establish steady-state
reheating condition after the stoppages, because the bulk temperature of the billets/blooms,
can be quite different than the observed billet/bloom surface temperature. Therefore, the
operators have to rely on their experience when making furnace adjustment during and after
stoppages.
In this research, a billet simulation model for a walking hearth type reheating furnace
was created and an optimization method for economical operation is proposed. The
simulation model employs a three dimensional (3-D) difference method and a dynamic
programming methodology developed in Matlab. Also, the thermal radiation view factor
from bricks inside furnaces to billets/blooms was calculated dynamically. The hearth
temperature was approximated using the simulated bottom face temperature of billets.
In the optimization method, the extraction temperatures of billets are predicted for
current operating conditions. Based on the result, the furnace temperature in each zone of the
furnace is controlled. The major feature of this control strategy is having two policies. One is
targeting the zone and the time period where billets temperatures can be controlled
effectively in changing furnace temperature set points, considering heating and cooling delay
and updating the feasible region dynamically. The other is prioritizing the zones for
iii
increasing furnace temperature. It was first zone 3, then zone 2, then zone1 and finally zone
4, considering the differences in heat transmission efficiency.
The final goal of this thesis is to develop an optimization method that can find an
optimal solution for furnace temperature control within 10 [min]. This goal was achieved by
developing a 2-D billet temperature simulation model, selecting appropriate time increments
and mesh size, setting amplifier and lower limiter for temperature increments in optimization,
and selective billet tracking for optimization for billet temperature increments.
iv
TABLE OF CONTENTS
v
2.4.1. Heat transfer .......................................................................................................... 20
2.4.2. Heat transfer from gas to billets/blooms ............................................................... 21
2.4.3. Heat transfer coefficient .................................................................................... 21
2.5. Thermal conduction ................................................................................................ 22
vi
3.4. Effect of thermal conductivity on center temperature ............................................ 58
vii
4.4. Overall Control Performance ................................................................................ 101
5.2. Insight for better furnace structure based on simulation results ........................... 120
5.3. Limitation of this research and further research recommendations ...................... 121
APPENDIX E. View-factor calculation from small plate to parallel plate with off-set. ...... 138
viii
LIST OF TABLES
ix
LIST OF FIGURES
x
Figure 3-13. Simulated temperature difference of each component for different modeling
time increments. ...................................................................................................................... 55
Figure 3-14. Computation time for various simulation conditions ......................................... 56
Figure 3-15. Computation time for various time increments up to 4920 [sec] (=82 [min]). .. 57
Figure 3-16. Relationship between temperature and thermal conductivity for various steels 58
Figure 3-17. Temperature differences for steel with various thermal conductivities. ............ 60
Figure 3-18. Comparison in rolling load between a billet with satisfactory center temperature
and a billet with unsatisfactory center temperature. ............................................................... 63
Figure 4-1. Comparison of the impact of an increase or a decrease in furnace temperature on
billet temperature changes in the various reheating furnace zones. ........................................ 68
Figure 4-2. Constraint example illustration. ........................................................................... 72
Figure 4-3. Feasible region after consolidating constraints. ................................................... 73
Figure 4-4. Upper limits for descretized variables.................................................................. 73
Figure 4-5. Main optimization steps. ...................................................................................... 75
Figure 4-6. Relationship between an increase of furnace temperature and the resultant
increase in billet center temperature. ...................................................................................... 77
Figure 4-7. Influence range of each overheating level. .......................................................... 83
Figure 4-8. Converting the updated heat pattern to a discrete expression. ............................. 85
Figure 4-9. Updated feasible region of furnace temperatures................................................. 86
Figure 4-10. Prolongation of heating and cooling phases....................................................... 87
Figure 4-11. Updated discrete lower limits for the variables.................................................. 87
Figure 4-12. Obtained heat patterns for each zone before final treatment. ............................. 91
Figure 4-13. Obtained optimal heat patterns for each zone after final treatment. .................. 91
Figure 4-14. Improvement of ∆Tex for each billet after optimization. .................................... 92
Figure 4-15. dTa history of each iteration. .............................................................................. 92
Figure 4-16. Relationship between dTa and ∆Tex for low furnace temperature. ..................... 94
Figure 4-17. Relationship between dTa and ∆Tex for high furnace temperature. .................... 95
Figure 4-18. Comparison of computation time based on the number of tracked billets. ........ 97
Figure 4-19. Average ∆Tex and minimum ∆Tex for the various cases. ................................... 98
Figure 4-20. Computation time comparison for various amplifiers and lower limiters. ........ 99
Figure 4-21. Average of ∆Tex and ±1σ range for various amplifiers and lower limiters. ..... 100
xi
Figure 4-22. Total over-heat for 85 billets for various amplifiers and lower limiters. ......... 100
Figure 4-23. Obtained optimal heat patterns for each zone. ................................................. 101
Figure 4-24. Difference in heat pattern between lower limiter 0 and 10 [K] (1). ................. 102
Figure 4-25. Difference in heat pattern between lower limiter 0 and 10 [K] (2). ................. 103
Figure 4-26. Change of ∆Tex before and after optimization. ................................................ 104
Figure 4-27. Comparison of ∆Tex between lower limiter 0 and 10 [K]. ............................... 105
Figure 4-28. Average ∆Tex and minimum ∆Tex for different lower limiter conditions. ....... 105
Figure 4-29. Heat pattern differences for various initial furnace temperatures (1). ............. 107
Figure 4-30. Heat pattern differences for various initial furnace temperatures (2). ............. 108
Figure 4-31. Average ∆Tex and minimum ∆Tex for various initial furnace temperatures. .... 109
Figure 4-32. Computation time and number of iterations for various initial furnace
temperatures. ......................................................................................................................... 109
Figure 4-33. Heat pattern of billets with high goal temperatures (1).................................... 111
Figure 4-34. Heat pattern of billets with high goal temperatures (2).................................... 112
Figure 4-35. Computation time and number of iterations for a case with high goal
temperature billets. ................................................................................................................ 113
Figure 4-36. Average ∆Tex and minimum ∆Tex of a case with high goal temperature billets.
............................................................................................................................................... 113
Figure 4-37. Change of ∆Tex before and after optimization in a case having high goal
temperature billets. ................................................................................................................ 114
Figure 4-38. Change of ∆Tex before and after optimization of tracked billets for a case having
high goal temperature billets. ................................................................................................ 114
Figure 4-39. Change of ∆Tex before and after optimization for a case having high goal
temperature billets with shifting the tracked billets. ............................................................. 115
Figure 4-40. Average ∆Tex and minimum ∆Tex for a case having high goal temperature billets
with shifting the tracked billets. ............................................................................................ 115
Figure B-1. Thermal radiation from small area dA1 to hemisphere. .................................... 124
Figure D-1. Positional relation of two perpendicular plates. ................................................ 136
Figure E-1. Positional relation of two parallel plates. .......................................................... 138
Figure E-2. View factor between parallel plates with off set. .............................................. 143
Figure E-3. View factor between parallel plates without off set. ......................................... 144
xii
ACKNOWLEDGEMENT
I would like to appreciate my sponsor for all the supports to my study in The
Pennsylvania State University.
I would like to appreciate Dr. Robert C. Voigt for his continued support throughout
my project and Dr. Enrique del Castillo for his greatly helpful suggestions in my project.
At last, I would like to thank my wife and my son for their patience and their
supports.
xiii
Chapter 1. INTRODUCTION
1
1.1. Background
2
1.1.3. Primary causes of non-optimal furnace operation
Once an operational trouble occurs downstream of the reheating furnace, the expected
time for fixing the trouble is announced by the operators who are responsible for getting the
rolling operations back on line. Based on this expected delay time, the operator of a furnace
lowers the furnace temperature to minimize fuel cost and prevent billets/blooms from
overheating. The extent of the temperature change from excess time in the reheating furnace
is dependent on the operator’s experience, personality and preference. If the temperatures of
billets/blooms at extraction are not high enough, another operational trouble is caused.
Therefore, most operators tend to set the temperature higher than necessary to avoid
subsequent rolling issues. After extracting, operators adjust the furnace temperature based on
the temperature measured by radiation thermometers equipped in a rolling line. This
inevitable conservative action of operators leads to larger reheating energy costs. This is
exacerbated by the fact which the operators cannot know the inside temperature of
billets/blooms and predict the temperature at extraction precisely. It is difficult to estimate
the bulk temperature of all billets/blooms in regular operation, and is even more difficult to
estimate during non-steady state conditions, though it can be measured using thermocouples
by experiments [3], [4].
3
proposed methods in which the furnace temperature is controlled in such a way the products
temperature follow their ideal trajectories [10], [11], [12]. However, in practice, product
temperature does not need to follow an ideal trajectory and may in fact undergo many
acceptable trajectories. This makes the heating pattern more flexible and fuel cost becomes
lower as a result. Also, Yang and Lu proposed an optimization model for slabs using
dynamic programming [13]. However, it gives only stationary optimal set points of each
zone. Therefore, optimization methods which can respond to dynamical condition changes
and minimize the fuel cost without using trajectories are to be developed for further energy
savings in real furnace operation for billets.
4
1.2. Objective of this research
In this research, there are two main objectives. The first objective is to develop a
simulation model of billet temperature considering the real time changes in thermal
conditions, including their thermal properties and thermal radiation view factors. The second
goal is to develop a practical furnace control optimization method that responds to real time
non-steady state condition changes in the operation of a reheating furnace in rolling mills
without using trajectories.
By applying these simulation model and control methods to the real operation of
reheating furnaces, reheating fuel costs can be minimized and the loss caused by operators’
differences and conservative actions can also be minimized.
5
Chapter 2. MODELING REHEATING FURNACE
6
2.1. Line and reheating furnace performances
In wire rod mills, billets are usually reheated up to about 1273 [K]. Those billets are
subsequently rolled by multiple rolling mills. Since the front end and the tail end of the
billets are unstable in quality, they are cut off by an on-line crop shear. After passing through
the final mill, the wire is formed into rings by a laying head. Then, it is fed to a reforming tub
through a cooling conveyor and those rings are reformed into a coil. The cooling rate can be
controlled at various rates on the conveyor to obtain the required mechanical property.
The chosen rolling speed is determined by the rate limiting performance among the
rolling machines and operational conditions. Also, the time interval between billets is
decided by the rate limiting performance among all the machines in the line and operational
condition as well. For example, if the cooling conveyor cannot feed rings quickly, and the
next wire comes without enough interval, those wires would collide each other. To avoid
such conflicts, a long enough interval between billets must be chosen. If intervals are short
and the holding time of billets in the furnace becomes too short, their bulk temperature would
not be high enough for rolling. In this case, a stoppage is scheduled to further heat the billets
before extracting them from the reheating furnace for avoiding downstream troubles.
7
2.1.2. Reheating furnace
In rolling mills, two types of reheating furnaces are mostly used -- walking-hearth
type and walking-beam type furnaces (more common). Figure 2-2 shows the motion of a
walking-hearth type reheating furnace. The walking hearths lift up all of the billets inside the
furnace at the same time and move them forward. Then, they are dropped down to the lower
limit position. At this point, all the billets are supported by the stationary hearths. The
walking hearths then move backward and return to the original position. Walking-beam type
furnaces employ the same mechanism for feeding billets, but billets are supported by beams
instead of hearths. In this research, a walking-hearth type furnace was considered, because of
the geometric complexity.
Reheating furnaces usually have multiple zones, preheating zones, a heating zone
and a soaking zone. The soaking zone is to homogenize the temperature from the surface to
the center of a billet. The temperature set point in the soaking zone is usually lower than that
of the heating zone. Furnace zones are segmented by dividing walls. A typical reheating
furnace structure is shown in figure 2-3. Because of the dividing walls, the furnace
temperature can be controlled independently for each zone.
The billet temperature is dependent on the furnace temperature of each zone and the
billet holding time in the furnace. The holding time is affected by many factors, including
rolling speed, furnace performance in cyclic motion, regular intervals between billets,
expected stoppages and unexpected stoppages.
Billet/Bloom
Walking
hearth
Stationary
hearth
8
Billet Ceiling
9
2.1.4. Cycle time of walking hearth furnaces
The furnace cycle time is defined as (2.1). The cycles of billets/blooms inside a
furnace are determined by the extracting conditions.
t cT = t cw + t cs ⋯ (2.1)
W
t r + t rv = + t rv when a billet/bloom is rolled
t cw ={ vw
t v when there is no billet/bloom to be rolled
where
vw = ρvf Af
t v = t cu + t cf + t cd + t cb + t sv
trv and tsv are adjusted by the operators and the specified operational conditions.
10
2.1.5. Positions of billets/blooms
An example of furnace temperature trends in a typical furnace zone and the furnace
temperature that a billet/bloom experiences are shown in figure 2-4. The furnace temperature
of each zone always changes and they are mostly different, and are independent unless the
temperature gap is quite large. Since the furnace temperature cycle that a billet/bloom
experiences depends on the time and the zone where it stays, it is important to track the
positions of all the billets/blooms in a furnace when modeling billet/bloom temperature.
The initial positions of billet/bloom i, Ii, before charging can be expressed in (2.2).
where
Their positions after the ncth cycle are obtained using (2.3).
Pi,nc = Ii + nc × K ⋯ (2.3)
By finding the number of cycles at time t, the positions of all of the billets/blooms in
the reheating furnace are obtained. In order to obtain the number of cycles at time t, expected
intervals and stoppages must be known. From a schedule table of stoppages, the times of
stoppages tb,nc can be estimated just before the ncth cycle is carried out. Using (2.1), the
cumulative time when the ncth cycle is completed can be calculated by (2.4).
11
nc
t cm,nc = ∑ (t cT,k + t b,k ) ⋯ (2.4)
k=1
where
Pi,t = Ii + nc × K ⋯ (2.6)
An example of the relationship between time and the number of cycles is shown in
table 2-1. If t=20 [min], the number of strokes nc is 0. If t=33 [min], the number of strokes is
2.
In a later chapter, another type of holding time will be discussed. Let the time period
described in this section be called the computational time period, tcom.
12
Furnace temperature
Zone 1
Furnace temperature
Zone 2
Furnace temperature
Zone 3
Furnace temperature
Zone 4
Time
Billet/bloom experiencing
Furnace temperature
Time
Figure 2-4. Furnace temperature trend and billet/bloom furnace temperature experience.
13
Table 2-1. Example of the relationship between the number of cycles and furnace travel
distance.
2 1 2 0 24 1× SK
3 0 1 10 35 2× SK
⁞ ⁞ ⁞ ⁞ ⁞ ⁞
nc-2 0 1 0 ⁞ (nc-3) × SK
nc-1 1 2 10 ⁞ (nc-2) × SK
nc 1 2 0 ⁞ (nc-1) × SK
14
2.2. Heat balance inside the furnace
15
Tc
qrad,bi-1→bi qrad,bi→bi+1
Tbi-1 Tbi Tbi+1
TH
The unit length of the heating time period is defined as follows (2.7).
tL
tp = [sec] ⋯ (2.7)
s
where
16
t L : Furnace holding time of the last billet/bloom in the considered range
Testimated
Q̇total = ρV ∫ Cp (T) dT [W] ⋯ (2. 8)
Tcurrent
where
V: Volume [m3 ]
If the time period is short, Cp can be approximated as a function of Tcurrent . In this case,
the equation (2.8) can be rewritten as (2.9).
Q̇total
→ Testimated = Tcurrent + ⋯ (2. 9)
ρV ∙ f(Tcurrent )
17
2.3. Thermal radiation between billet/bloom and furnace walls
where
Q̇rad : Total heat from the furnace bricks to the billets/blooms by radiation [W]
σ: Stefan-Boltzmann constant 5.670373 × 10−8 [Wm−2 K −4 ]
Ab : Surface area of the billet/bloom [m2 ]
ϕwb (Fwb , Fbw , εw , εb ): Radiation coefficient
Fwb : View factor from the furnace bricks to the billets
Fbw : View factor from the billets to fthe urnace bricks
εw : Emissivity of the furnace bricks
εb : Radiation absorption rate of the billets
2.3.2. Emissivity
Emissivity indicates how much of thermal energy the surface of a material can emit
or absorb. It ranges from 0 to 1.0. If it is 0, it implies that the material is a black body. Also,
it is known that polished metal has emissivity values close to 0. The emissivity of oxide steel,
appropriate for steel heated in air furnaces, is approximately 0.9 [16].
18
2.3.3. View-factor
The view-factor indicates how much radiation can reach geometrically from one
surface to another surface. It is defined by (2.11) [15], [17]. See appendix A for the details on
the use of view-factors.
1 cos φ1 cos φ2
F12 = ∫ ∫ dA1 dA2 ⋯ (2. 11)
A1 A1 A2 πr 2
19
2.4. Heat transfer between billets/blooms, the furnace atmosphere
and the hearths
q̇ L = hL (T1 − T2 ) ⋯ ( 2. 12)
where
When T1 and T2 do not depend on location, (that is, temperatures are uniform) the
total transferred heat through area A is calculated as (2.13).
When A is constant,
20
2.4.2. Heat transfer from gas to billets/blooms
Since some types of gas emit radiation when they are combusted, billets/blooms are
heated by heat transfer and thermal radiation from the combusted gas in the furnace at the
same time [15], [18]. However, gas is not ‘distinct in shape’. It is difficult to calculate view
factors between billets/blooms surface and the furnace gas. Accordingly, the total heat from
gas to a billet/bloom was calculated by (2.14), defining μgb as the rate which heat is
transmitted to one billet/bloom by thermal radiation.
where
Tg : Gas temperature
Tb : Billet/bloom temperature
In this research, it was assumed for simplicity that the combusted furnace gas
generates a non-luminous flame, so that it does not simultaneously emit thermal radiation and
qgb can be expressed only by its heat transfer term.
21
2.5. Thermal conduction
ΔT
J = λgradT = λ ⋯ ( 2. 15)
d
where
J: Transmitted heat from one portion to another portion within the billet
/bloom [Wm−2 ]
Using equation (2.15), transmitted heat from adjacent portions of a billet by thermal
conduction is calculated as (2.16).
ΔT
Q̇ cond = λA ⋯ (2. 16)
d
where
22
2.6. Thermal Properties of Materials
where
23
2.6.2. Emissivity/Absorption rate
To know how much heat is transmitted by radiation, the radiation coefficient must be
estimated in (2.10). It is a function of view factor and emissivity [17], [18]. Additionally,
radiosity must be considered to calculate the radiation coefficient; because emission,
absorption or permeation and reflection occur in radiation. For simple geometries, such as
plates in parallel, radiation coefficients are easily calculated. However, it is hard to calculate
the values in complex systems such as furnaces. In this research, the effect of radiosity was
included in the emissivity for simplicity.
where
Emissivity and absorption are usually handled together. Substances have their own
values. They are affected by the surface conditions, such as smoothness, shape and
composition. These properties should be found for applying to simulation models in advance.
In this research, it was assumed that the emissivity of bricks and billets/blooms were
constant.
24
2.7. Furnace Modeling
(ℓ,m,n)
n
C D E
m
B I F
A H G
ℓ
(1,1,1)
25
2.7.2. Heat balance modeling in each component.
The surface of billets/blooms receives heat through thermal radiation and heat
transfer, while the inside of billets/blooms is heated by thermal conduction. Figure 2-7
illustrates the model of the billet/bloom heat balance. The component subscripts correspond
to those in figure 2-6. Also, the detailed heat transfer calculations are shown in Appendix C.
qcond,(x-1,m,1) qcond,(ℓ-1,m,1)
qcond,(2,m,1) qcond,(x+1,m,1) qrad,Hb+qrad,cb
+qrad,wb+qrad,bi→bi+1
+qtran,gb
qrad,Hb+qrad,cb
+qrad,wb+qrad,bi-1→bi
+qtran,gb qcond,(x-1,y,1) qcond,(ℓ-1,y,1)
qcond,(2,y,1) qcond,(x+1,y,1) qrad,Hb+qrad,cb
+qrad,wb+qrad,bi→bi+1
+qtran,gb
qcond,(x-1,1,1) qcond,(ℓ-1,1,1)
qcond,(2,1,1) qcond,(x+1,1,1)
qrad,Hb+qrad,cb qrad,Hb+qrad,cb
+qrad,wb +qrad,wb+qrad,bi→bi+1
+qrad,bi-1→bi +qtran,gb
+qtran,gb qtran,Hb
qrad,Hb+qrad,cb qrad,Hb+qrad,cb qrad,Hb+qrad,cb
+qrad,wb+qtran,gb +qrad,wb+qtran,gb +qrad,wb+qtran,gb
Component A (1,1,1) Component H (x,1,1) Component G (ℓ,1,1)
26
2.7.3. View factors from furnace walls, hearths and ceiling to a mesh
Each mesh receives thermal radiation from different regions of the walls, the hearths
or the ceiling. The hatched area in figure 2-8 shows the considered furnace regions that emit
thermal radiation to small areas on different faces. Based on these configurations, the view
factors were calculated for every billet face. (Appendix E and F). At the ends of the billets,
the thermal radiation from the hearths is transferred over beyond the dividing walls, because
there are open spaces under the dividing walls. In this research, it was assumed for simplicity
that the temperature at the dividing walls is uniform and the brick’s temperature in a zone
where a billet is located is used for the calculation as a representative temperature. In
practice, the furnace temperature of each zone is different and the hearth temperature of each
zone is expected to be different. The effects by this simplified furnace temperature
approximation at the dividing walls are not expected to be significant.
When billets are charged continuously, the distance to the adjacent billet is constant.
However, if billets are not charged continuously, the distance between billets can vary. In
this simulation, the distance to the adjacent billet at both downstream and upstream sides is
determined by their initial positions and is always tracked. The regional range of the hearth to
be considered can be easily found from the distance. However, the regional ranges of the
ceiling, the dividing walls and the side walls must be calculated geometrically. Figure 2-9
indicates the geometric relationship between the ranges and the distance to the adjacent billet.
When θ ≤ ϕ, the whole range of the ceiling and the dividing walls are effective as the areas
which emit thermal radiation to the targeted mesh of a billet. On the other hand, when θ > ϕ,
the thermal radiation region depends on the position of the targeted mesh. The effective areas
are determined in (2.21) and (2.22).
Hc
when θ > 𝜙, Hwe = Hw − Lw tan θ, Lce = ⋯ (2.21)
tan θ
27
where
28
Front end and tail end
Upper face
Upstream side
Downstream side
29
θ
ϕ
θ
ϕ
ϕ θ
θ
ϕ
30
2.7.4. Heat transmission between billets/blooms and the hearths
The front ends of the billets/blooms are aligned on the same line as shown in figure 2-
3. However, the tail ends may not be aligned on the same line because the length of the
billets/blooms varies. At the tail ends, the length of the billets/blooms affects the heat
transmission between the billets/blooms and the hearths. If the difference of the heat
transmission is considered in the model, the calculation, especially for the temperature of the
hearths, becomes more complex. Also, it lengthens the calculation time. Therefore, it was
assumed in this research that, for the calculation of the hearths temperature only, the length
of the billets/blooms was constant by employing a representative length.
Also, the furnace temperature inside the furnace can fluctuate due to various factors
such as an imbalance in burner performance, differences in the billets/blooms length, etc..
Thus, the difference of the furnace temperature in axis z should be considered in the model.
This can be expressed as a linear model, as follows (2.23).
TaF,j,t − TaT,j,t
Ta,j,z,t = (Lu z + dF ) + TaF,j,t ⋯ (2. 23)
WH
where
TaF,j,t : Furnace temperature at the front side wall in zone j during time period t [K]
TaF,j,t : Furnace temperature at the tail side wall in zone j during time period t [K]
dF : Set Distance between the front side wall and the front end of billets/blooms [m]
31
2.7.5. Local temperature of the hearth
The local temperature of the hearths can vary. For simplicity, the hearth temperature
was assumed to be uniform and it was divided into two different temperatures. One is the
main zone in which billets/blooms are loaded regularly. The temperature of this area is
affected by the heat transfer from billets/blooms when they are loaded, by thermal radiation
from the furnace ceiling and the side walls, and by heat transfer from combustion gas when
billets/blooms are not loaded. The other temperature is in the reserved zones where
billets/blooms are not placed on regularly. It is assumed that the temperature of this area
follows the furnace temperature in the same zone and it reaches temperature immediately
when the furnace temperature changes, because heat transmission by thermal conduction is
small in the bricks which have high thermal insulation performance and the surface
temperature of the bricks changes quickly..
For calculating transmitted heat between the main zones and billets/blooms, the
average bottom face temperature of the billet/bloom was used. This bottom face temperature
varies for every billet/bloom and also changes in real time. Therefore, the temperature of the
main zones must be computed every time period. This can influence the temperature whether
or not there is a billet/bloom on the main lot in zone j during time t.
where
hgh : Heat transfer coefficient from the combustion gas to the billet
32
2.7.6. Interaction between billets/blooms
In reheating furnaces multiple billets/blooms are heated at the same time. These
billets/blooms can have different temperatures thoroughly and locally. Therefore,
temperature interaction by thermal radiation is expected between the billets/blooms. Since
they are finely meshed for the calculation of local temperature, the computation time
becomes quite large when the interaction of each pair of components is considered. Thus, the
effect of billet-to-billet interaction was first computed to investigate how much this
interaction affects the temperature change. Figure 2-10 shows the image of thermal radiation
from each portion with different temperature of a billet to a portion of another billet. Table 2-
2 shows the computational condition to evaluate the interaction. Figure 2-11 shows the
assumption for this evaluation in temperature difference between two billets.
33
1200
1100
1000
900
Temperature [C°]
800
700
600
500
400
300
200
0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 54 57 60 63 66 69 72 75 78 81 84 87 90
Holding time [min]
400
380
360
340
Temperature [C°]
320
300
280
260
240
220
200
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Holding time [min]
34
In figure 2-12, the result of the simulation is shown. As the distance between the
components increases, the extent of temperature interaction becomes smaller. When the
considered range in z is up to 400-500mm, the cumulative temperature increase was about
0.25 °C and almost saturated. Even if the range of ±500 mm in z is considered, the
temperature increase would only be about 0.5 °C. Consequently, the effect of the interaction
by thermal radiation between billets/blooms is negligible in this furnace operation. It was
assumed in this case that the holding time was 90 minutes and the temperature gap with
adjacent billet was 10 [K]. If much larger temperature gap between adjacent billets is
induced, the influence of this interaction might be necessary to be considered.
0.30
Temperature increase [C°]
0.25
0.20
0.15
0.10
0.05
0.00
0.0-0.1 0.1-0.2 0.2-0.3 0.3-0.4 0.4-0.5 0.5-0.6 0.6-0.7 0.7-0.8 0.8-0.9 0.9-1.0
Position of components from the end of billet/bloom [m]
35
Chapter 3. SIMULATION OF THE MODEL
36
3.1. Billet/Bloom initial orders and their parameters
Using the model that was created in chapter 2, the temperature of billets in the
reheating furnace was simulated. Then, the performance and the reasonability of this model
were evaluated.
3 2
C𝑝,𝑖|Tb≤750℃ = 𝑎𝑠𝑏 Tb,(𝑥,𝑦,𝑧),𝑖 + 𝑏𝑠𝑏 Tb,(𝑥,𝑦,𝑧),𝑖 + 𝑐𝑠𝑏 Tb,(𝑥,𝑦,𝑧),𝑖 + 𝑑𝑠𝑏 ⋯ (3. 1)
3 2
C𝑝,𝑖|Tb≥750℃ = 𝑎𝑠𝑎 Tb,(𝑥,𝑦,𝑧),𝑖 + 𝑏𝑠𝑎 Tb,(𝑥,𝑦,𝑧),𝑖 + 𝑐𝑠𝑎 Tb,(𝑥,𝑦,𝑧),𝑖 + 𝑑𝑠𝑎 ⋯ (3. 2)
3 2
λ𝑖 = 𝑎𝑐 Tb,(𝑥,𝑦,𝑧),𝑖 + 𝑏𝑐 Tb,(𝑥,𝑦,𝑧),𝑖 + 𝑐𝑐 Tb,(𝑥,𝑦,𝑧),𝑖 + 𝑑𝑐 ⋯ (3. 3)
37
The emissivity, the heat transfer coefficient against combusted gas and the heat
conductance against the hearth were assumed constant. The coefficient and the values which
were used for the simulation are summarized in table 3-1. Additionally, the other data for
simulation, such as the furnace structure and the operational condition, are shown in table 3-
3.
Table 3-2. Specification of the computer and operating system used for simulation.
Category Specification/version
Software Matlab R2012b
CPU Intel(R) Core(TM)2 Quad CPU Q9400 2.66GHz
Memory Installed memory (RAM) 4.00GB (3.87GB usable)
System type 64-bit Operating System
38
Table 3-3. Specification, parameters and conditions for the simulation.
39
3.2. Performance of the simulation model
Table 3-4. Simulation properties and operational conditions for the simulation.
Time increments Unit mesh size Billet/bloom Furnace holding Each zone
[sec] [mm× mm× mm] number time [min] temperature
1 Constant and
0.5 11×11×50 82
Charged firstly uniform
Moving direction
Downstream Upstream
#7 #8 #9
#4 #5 #6
#1 #2 #3
z Front end
x
40
Down Up
stream stream
#7 #8 #9
#4 #5 #6
y #1 #2 #3
z Front end
x
Figure 3-2. Simulated temperature trend at the billet front end (z=1).
41
Down Up
stream stream
#7 #8 #9
#4 #5 #6
y #1 #2 #3
z Front end
x
Figure 3-3. Simulated temperature trend at the middle in the billet length.
42
At the front end, the temperature of the upper corners, #7 and #9, were saturated in
zone 4. However, even at upper corners, the temperature does not reach the furnace
temperature of zone 4. Whereas the front end receives thermal radiation from the side-wall of
the furnace, the middle section does not. As a result, the billet temperature is much lower in
the middle than at the front end.
The temperature of billets was simulated for three different cases in the furnace
temperature of each zone. The result is shown in figure 3-4 and figure 3-5. Case 1 is the case
which zone 1 and zone 3 have relatively high temperature. Case 2 is the average case among
the three cases. Case 3 is the case where the temperature of zone 1 is lower and that of zone 3
is higher. In both of the upper corner at downstream side and the center, the temperature in
case 3 was the lowest until it reached zone 3 due to the lower temperature in zone 1. After
reaching zone 3, the temperature in case 3 exceeds that in case 2, because the furnace
temperature in case 3 is higher than that in case 2.
Consequently, this simulation indicates that this model can respond to furnace
temperature changes well.
1260
1235 1239
1240 1229 1233 1233
1226
Simulated billet temperature [◦C]
1220
1160
1134 1138
1140 1131
1120 1124
1118
1120
1100
#1 #2 #3 #4 #5 #6 #7 #8 #9
Figure 3-4. Simulated temperatures of each portion in the middle section at extraction.
43
Furnace temperature [K]
Zone 1 Zone 2 Zone3 Zone4
Case 1 1223 1223 1373 1333
Case 2 1223 1273 1323 1333
Case 3 1173 1273 1373 1333
1400
1300
At portion #7
1200
Simulated billet temperature [K]
1100
1000
Down Up
900 stream stream
800 #7 #8 #9
700
600
500 #4 #5 #6
400
300 y #1 #2 #3
200 Front end
z
100 x
0
1021
1361
1701
2041
2381
2721
3061
3401
3741
4081
4421
4761
5101
5441
5781
6121
6461
6801
7141
7481
7821
8161
8501
8841
9181
9521
1
341
681
Time [×0.5sec]
1400
1300 At portion #5
1200
Simulated billet temperature [K]
1100
1000
900
800
700
600
500
400
300
200
100
0
1021
1361
1701
2041
2381
2721
3061
3401
3741
4081
4421
4761
5101
5441
5781
6121
6461
6801
7141
7481
7821
8161
8501
8841
9181
9521
1
341
681
Time [×0.5sec]
44
3.2.2. Difference of simulated sectional temperature
Figure 3-6 indicates the temperature differences at extraction in different sectional
portions of the billet. In the front end of the billet (z=1), the temperature of the upper corners
was the highest. On the other hand, the temperature of the center and the bottom face shows
the lowest. In the middle section, the upper corners show the highest temperature and the
center and the bottom face shows the lowest as well. The tendency that the temperature of the
corners is higher and the temperature of the center and the bottom face is lower, was the
same for each furnace analysis condition.
The temperature change in the longitudinal direction z is shown in figure 3-7. In most
portions, the temperature is the minimum at around 6400 [mm]. Billets are aligned to the
front end and the distance between the sidewall of the furnace and the front end is 200 [mm]
in this model. The width of the furnace is 13,000 [mm]. The minimum temperature position
corresponds to the middle of the furnace width. This seems to indicate that the thermal
radiation is the smallest at the middle of the furnace wide, since it is the farthest from the
both sidewalls of the furnace. At the front end and the tail end, the temperatures are much
higher than other regions. They receive relatively intense thermal radiation from the
sidewalls of the furnace in addition to the ceiling and the hearths. Therefore, these regions
are much higher in temperature than other sections. Compared to the temperature at the front
end, the temperature at the tail end is lower. This is caused by the difference in the distance
from the each sidewall. While the distance to the front end is 200 [mm], the distance to the
45
tail end is 2,800 [mm]. The reheating furnace model expresses these thermal relationships
well.
1,320 1,312
1,310 1,305 1,306
1,299 1,299
1,300
1,290
1,278
1,280
1,270 1,264
1,260
1,250
1,240
1,230
#1 #2 #3 #4 #5 #6 #7 #8 #9
Front end (z=1)
Portion in section
1,260
1,239
1,233
Simulated temperature at extraction [K]
1,240
1,220
1,160
1,138
1,140
1,124
1,120
1,100
1,080
1,060
#1 #2 #3 #4 #5 #6 #7 #8 #9
Middle (z=50)
Portion in section
Figure 3-6. Difference in simulated billet component temperature at extraction along the
length of the billet.
46
Figure 3-7. Simulated component temperature difference at extraction in the billet
longitudinal direction.
47
3.2.3. Heat transmission in billet longitudinal direction
Billets/blooms are usually long products. The possibility of reducing the dimension of
the model was investigated to simplify the model and shorten the computation time, as
Steinboeck et al. proposed a 1-D slab temperature simulation model [21].
The total transmitted heat through thermal conduction is shown in figure 3-8 and the
total transmitted heat through thermal conduction in z is shown in 3-8. The operational
condition is indicated in table 3-5 and the unit mesh size is 55×55×100 [mm× mm× mm].
According to the result in figure 3-9, up to 700 [mm] from the front end and the tale end,
large heat is transmitted in direction z. However, in the region of more than 700 [mm], the
heat became much smaller. In figure 3-10, the rate of the total transmitted heat in z to the
total transmitted heat is shown. Transmitted heat in z occupies large percentages up to 700
[mm]. On the other hand, the heat does not contribute an increase of the temperature in the
region of more than 700 [mm]. If the furnace temperature is higher than that in this
simulation, the region, which the heat transmission in z should be considered, might be
longer than 700 [mm]. If only the middle section is used for the simulation, the transmitted
heat in z is negligible and the model can be reduced to 2-dimension from 3-dimension.
48
1800
1600
1400
Total transmitted heat [kJ]
1200
1000
800
600
400
200
0
13
17
21
25
29
33
37
41
45
49
53
57
61
65
69
73
77
81
85
89
93
97
1
5
9
#1 #2 #3 #4 #5
#6 #7 #8 #9
1800
Total transmitted heat in direction z [kJ]
1600
1400
1200
1000
800
600
400
200
0
13
17
21
25
29
33
37
41
45
49
53
57
61
65
69
73
77
81
85
89
93
97
1
5
9
#1 #2 #3 #4 #5
#6 #7 #8 #9
Figure 3-9. Total transmitted heat in the longitudinal direction until extraction.
49
35%
Rate of transmitted heat in z to total transmitted
30%
25%
20%
heat
15%
10%
5%
0%
13
17
21
25
29
33
37
41
45
49
53
57
61
65
69
73
77
81
85
89
93
97
5
1
#1 #2 #3 #4 #5 #6 #7 #8 #9
Figure 3-10. Rate of transmitted heat in the z direction to total transmitted heat until
extraction.
50
3.3. Selection of appropriate mesh size
The temperature of each sectional portion is simulated and compared at various sizes
of the unit mesh. The result is shown in figure 3-11. At the front end of a billet, the bottom
face temperature, #2, and the center temperature, #5, had relatively large variation in 11 ×11
[mm×mm] to 55×55 [mm×mm]. The largest gap was 11 [K] at #2. On the other hand, they
had relatively small gap 2 to 5 [K] in the middle section. This means that even if 55×55
[mm×mm] is chosen as the mesh size, the difference from 11 ×11 [mm×mm] in the center
temperature is about 11 [K] at the front end and about 5 [K] at the middle section. By
leveling or compensating, the center temperature in the middle section can be used for
simulation with little error.
51
Increments Furnace temperature [K] Furnace holding
time [sec] Zone 1 Zone 2 Zone3 Zone4 time [min]
1 1173 1273 1373 1333 82
1,340
1,320
Simulated temperature [K]
1,300
1,280
1,260
1,240
1,220
1,200
1,180
1,160
1,140
1,120
#1 #2 #3 #4 #5 #6 #7 #8 #9
Front end (z=1)
Simulated positions
1,340
1,320
Simulated temperature [K]
1,300
1,280
1,260
1,240
1,220
1,200
1,180
1,160
1,140
1,120
#1 #2 #3 #4 #5 #6 #7 #8 #9
Middle (z=50)
Simulated positions
Figure 3-11. Relationship between simulated temperature and unit mesh size.
52
Increments Furnace temperature [K] Furnace holding
time [sec] Zone 1 Zone 2 Zone3 Zone4 time [min]
1 1173 1273 1373 1333 82
1,360
1,350
Simulated temperature [K]
1,340
1,330 1,325 1,323
1,318 1,316
1,320 1,313 1,315
1,310
1,310 1,306 1,305
1,302 1,301
1,300 1,297 1,297
1,293
1,286 1,288
1,290
1,280
1,280
1,268
1,270
1,260
#1 #2 #3 #4 #5 #6 #7 #8 #9
Front end (z=1)
Portions in section
1,220 1,2171,217
1,210
1,210
1,210
1,200
Simulated temperature [K]
53
3.3.2. Mesh size and time increments
In dynamic programming, time is dealt with as discrete time. That is, until time
moves to the next time period, the current condition is maintained. During each modeled time
increment, the temperature of a unit mesh does not change. If the size of unit mesh becomes
smaller, the volume and the surface area also become smaller. However, the volume and the
surface area do not become smaller at the same rate. Heat by thermal radiation is received
through the surface of the mesh. As a result, the heat by thermal radiation, which the unit
mesh receives per unit time and per volume, becomes larger by choosing smaller mesh sizes.
Therefore, shorter time increments must be chosen when a smaller unit mesh size is selected.
Otherwise, the computation will diverge. In table 3-6, convergence properties were
investigated for four mesh sizes and various modeling time increments. For mesh size 55×
55×100 [mm×mm×mm], the time increments for which computation were completed without
divergence was up to 82 [sec]. In figure 3-13, the simulated temperature in 55×55×100
[mm×mm×mm] mesh sizes and various time increments is shown. At tp=82 [sec], the
temperature was significantly larger than for the others. Even if it is converged, shorter time
increments should be selected for more reasonable modeling estimates.
Table 3-6. Model convergence (C) and divergence (D) for different mesh sizes and modeling
time increments.
54
Mesh size Furnace temperature [K] Furnace holding
[mm×mm×mm] Zone 1 Zone 2 Zone3 Zone4 time [min]
55×55×100 1173 1273 1373 1333 82
1340
Simulate temperature at extraction [K]
1320
1300
1280
1260
1240
#1 #2 #3 #4 #5 #6 #7 #8 #9
Front end (z=1)
1240
Simulate temperature at extraction [K]
1220
1200
1180
1160
1140
1120
#1 #2 #3 #4 #5 #6 #7 #8 #9
Middle (z=50)
Figure 3-13. Simulated temperature difference of each component for different modeling
time increments.
55
3.3.3. Mesh, time increments and computation time
In optimization programming, many iterations are usually carried out. However, if the
total computation time is longer than prediction time range in the simulation, the next
computation cannot be done in real time before the next billets/blooms, which are scheduled
in the future out of the prediction range, are charged. In this case, this simulation cannot be
used for real time control of furnaces. Since unit mesh size and time increments influence the
computation time significantly, the extent of this influence was investigated more
thoroughly.
In figure 3-14, the effect of size change in unit mesh on the computation time is
shown for tp=1 and tp=2.5. As the unit mesh size becomes large, the computation time
decreases nonlinearly. At tp=2.5 in unit mesh size 55×55×100 [mm×mm×mm], the
computation time was 42 [sec]. In this case, if there are 20 billets in a furnace and 10
iterations are needed, the total computation time will be 42×20×10=8400 [sec], ≈140 [min].
Assuming the furnace holding time is 82 [min] and the 20 billets are charged continuously
every 1.5[min], the prediction time range becomes 112 [min]. Therefore, the computation
time exceeds the prediction time range. In this case, it is necessary to shorten the
computation time if it is to be used as part of a real-time control strategy.
500 500
373
400 400
277
300 300
- -
0 20 40 60 0 20 40 60
Sectional size of unit mesh [mm] Sectional size of unit mesh [mm]
56
In figure 3-15, the effect of time increments on the computation time is illustrated. In
33×33×100 [mm×mm×mm], the computation time decreased to 13 [sec] at tp=30 [sec]. In
55×55×100 [mm×mm×mm], the computation time was saturated at tp=30 [sec] and it was
around 11 [sec]. When the computation time is 11 [sec], the total computation time in the
case shown above will be about 11 [sec] × 20 billets × 10 iterations =2200 [sec] and it is
shorter than the prediction time range. In this case, about 30 times iterations are affordable in
this simple estimation. However, for more iterations, it is necessary to shorten the
computation time. It will be discussed in later chapter about optimization whether the
computation should be shortened more or not.
140 126
Size=33×33×100
120
Computation time [sec]
100
80
59
60
32
40
21
13
20
-
0 5 10 15 20 25 30 35
Increments time [sec]
140
100
80
60
42
40 25
18
11 12 10 12 10
20
-
0 20 40 60 80 100 120 140
Increments time [sec]
Figure 3-15. Computation time for various time increments up to 4920 [sec] (=82 [min]).
57
3.4. Effect of thermal conductivity on center temperature
90
80
Thermal conductivity [W/(m·K)]
70
60
50
40
30
20
10
0
0 200 400 600 800 1000 1200 1400
Temperature [K]
Figure 3-16. Relationship between temperature and thermal conductivity for various steels
58
3.4.2. Impact of thermal conductivity on billet temperature
The result of the simulation is shown in figure 3-17. The computational condition is
shown in table 3-7. At the upper corner, #7, the temperature of steel 3 was the lowest and that
of steel 2 was the highest. The gap between those 2 types of steel was approximately 20 [K]
in the middle. Furthermore, at the center, #5, the highest was in steel 3 and the lowest was in
steel 2. The gap was also approximately 20 [K] in the middle. In steel with high thermal
conductivity, heat from the upper corner to the center is transmitted greater distances. As a
result, the temperature of the upper corner in steel 3 was lower than that in steel 2, while the
temperature of the center in steel 3 was higher than that in steel 2.
From these results, a 1 [W/(m·K)] difference in thermal conductivity can make about
a 0.8 [K] difference in the estimated billet temperature at extraction. Considering the
optimization of furnace operation, the accuracy of the model should be within 1 [W/(m·K)].
59
1340
53
69
85
13
17
21
25
29
33
37
41
45
49
57
61
65
73
77
81
89
93
97
1
5
9
Mesh number in z
1300
Simulated temperature at extraction [K]
1280
1260
1240
1220
1200
1180
1160
1140
1120
1100
13
17
21
25
29
33
37
41
45
49
53
57
61
65
69
73
77
81
85
89
93
97
1
5
9
Mesh number in z
(b) #5 - center
Figure 3-17. Temperature differences for steel with various thermal conductivities.
60
3.5. Parameters selection for optimization
The role of reheating furnaces is raising the billets/blooms temperature high enough
for effective rolling. This ‘high enough’ level is determined considering the following
factors.
a) The rolling load of each rolling machine in a line must be under the upper limit of
their specifications which are different for each machine. If the billet bulk
temperature is below the minimum level, rolling machines are damaged or the rollers
are broken. Insufficient bulk temperature leads to other operational trouble, for
instance biting failure in rolling. These troubles cause serious production loss,
especially in steel companies operating for 24 hours/day. The rolling load of each
machine, which is correlated to the billet temperature [23], is sometimes measured
using root mean square (RMS) in their motors which indicates the load on the motors
per unit time.
b) After rolling, when the steel reaches the cooling process in a rolling line, it must have
enough residual temperature to guarantee its quality as a start temperature [24]. If the
temperature is below a threshold temperature, the products are graded down or
rejected and the rolling line is shut down. Then, billets/blooms in a furnace are
reheated until their temperature reaches enough level to re-start the rolling line.
61
In a furnace, assuming that the furnace temperature is always higher than billets
temperature, once the billet center temperature has reached the required level, it can be said
that the temperature of the other portions of the billet are also sufficient. When the surface
temperature and the center temperature of billets are sufficient, the rolling load at the first
rolling mill and at a mill after several mills is under the upper limits. However, when the
surface temperature is satisfactory and the center temperature is unsatisfactory, the rolling
load at the first rolling mill might be under the upper limit because the surface with
satisfactory temperature of billets is the portion of the billet that is primarily rolled. But, the
rolling load at mills after several mills would exceed their upper limits if the billet center is
cold. An example image describing this situation is shown in figure 3-18. The minimum
center temperature of billets is in the middle of the furnace width. Consequently, the center
temperature in the middle of the furnace width should be focused on for reheating furnace
control optimization. In this case, the 3-D control model can be reduced to 2-D control
model.
62
#1 rolling mill #6 rolling mill
Rolling load
Billet temperature Billet temperature
Rolling load
Figure 3-18. Comparison in rolling load between a billet with satisfactory center temperature
and a billet with unsatisfactory center temperature.
63
3.5.2. Selection of model and parameters for reheating furnace control
The computation time of the simulation model is one of the major issues in adapting
its use for the optimization and control of the furnace operation. Although many
computational iterations must be done to obtain an optimal modeling result, the computation
must be done within certain limited time.
In slab simulation, when the dimension of the model was reduced from 2-D to 1-D,
the computation time was shortened dramatically [21]. Also, it is said that 3-D is not
appropriate for billet temperature simulation due to the large computation [25].
Based on the discussion so far, the following condition indicated in table 3-8 was
selected for the optimization, considering the reasonability of the simulated temperature and
computation time constraints.
Table 3-8. Selected model and parameters for furnace control optimization.
55×55
2 dimension model 30 ≈10
(3×3)
64
Chapter 4. OPTIMIZATION OF FURNACE OPERATION
65
4.1. Optimization Problem
A simulation model for billets temperature has been introduced. In this chapter, a
furnace control optimization method exploiting the simulation model will be proposed. The
full implementation and use of the detailed reheating furnace billet temperature model,
discussed in the previous chapter, takes longer than one minute computation time to obtain a
complete solution. The goal of this chapter is to develop a simplified method to obtain a
near-optimal solution more quickly that can be used for direct control of the furnace
temperature.
The first policy is that the furnace zone and the time period where the total gap
between the goal temperatures and the simulated temperatures of billets is the largest are
targeted for changing set-point furnace temperature.
The second policy is that the priority for increasing furnace temperature is first zone
3, then zone 2, then zone1 and finally zone 4. This prioritization is based on the concept that
heating in zone 3 is likely to be the most efficient. If the furnace temperature in zone 1 is
increased, the temperature of billets in zone 1 increases during certain times. This reduces the
heat transmission after the certain time, since the gap between the billets temperature and the
furnace temperature afterwards in zones 2-4 becomes smaller. This means that earlier action
is not likely to be efficient. Zone 4 is typically referred to as the soaking zone. The main role
of this zone is homogenizing the temperature throughout the billet. In this zone, the furnace
66
temperature is set at lower level than that of zone 3 to avoid surface overheating. The upper
limit of the temperature in this zone is relatively low and this zone is not selected for
increasing the billets temperature. Zone 3 is referred to as the heating zone and the furnace
temperature set point is the highest among the four zones. In this high temperature
environment, thermal radiation has the largest impact on billet temperature. The impact of an
increase of the furnace temperature in zone 3 is expected to be most significant, because the
transmitted heat via thermal radiation is controlled by the fourth power of the furnace
temperature and the billets temperature. As the furnace temperature is increased, the effective
heat transmission to the billet will increase greatly. A comparison of the billet center
temperature increase at the extraction of a billet when the furnace temperature of any zone is
increased 1 [K] from the standard temperature is shown in figure 4-1 (a). Figure 4-1 (b)
shows a comparison of the billet center temperature decrease at extraction of a billet when
the furnace temperature of any zone is decreased 1 [K] from the standard temperature. As a
result, zone 3 had the largest increase and decrease, because the furnace temperature was the
highest. Considered the upper limit temperature constraints for the furnace temperature in
zone 4, the control priority when the furnace temperature is increased is first zone 3,
followed by zone 2, then zone 1 and finally zone 4. However, even if there are billets in zone
3, when the gaps from their goal temperatures are small, choosing zone 3 is not preferable
because effective heat transmission rates in this zone will be small. Overall, the target zone
and time period will be selected by balancing policy 1 and policy 2 in this control method.
67
2.0 1500 2.0 1500
1423
-∆Tex [K]
∆Tex [K]
(a) 1 [K] increase from low temperature (b) 1 [K] decrease from high temperature
Once the priority is decided, minimizing the difference between the estimated billet
temperature and their goal temperature is chosen as the objective function for this problem.
I I
where
i: Billet number
68
In this simulation model, the maximum number of billets that can fit in the furnace at
the same time is 70. For a full billet load, the whole existing schedule should be considered
to minimize the fuel cost. Billets are continuously charged after the 70th billet is charged. If
the time horizon is infinity, the number of the decision variables becomes infinity, because it
is dependent on the number of the zones and the time periods. Therefore, the time horizon
must be limited. Since the temperatures and the specifications of the billets after the 70th
billet can affect the heat pattern of the 70th billet, the simulation must be extended beyond
just the first 70 billets. If the goal computation time for optimization is 10 [min], the number
of the billets to be increased can be minimized by running the simulation every 10 [min]. In
this research, an extra 15 billets, which is the maximum number of billets zone 3 and zone 4
can hold at the same time, was chosen as the modeling estimation range, considering the goal
computation time. That is, 85 billets are tracked in this optimization.
The same types of billets are usually scheduled for rolling at the same timing for
higher efficiency in downstream rolling operations. Let the group of the same billet type be
called a lot. Since each zone heats multiple billets at the same time, it is not necessary to fully
track all billets for control optimization. For example, by tracking the first billet and the last
billet in a lot and optimizing their heat pattern, the heat pattern of the other billets in the same
lot is also optimized (assuming that the furnace hearths themselves are at their steady state
operating temperature). If the number of the billets in a lot is large, the first billets or the last
billets in a lot are in different furnace zones. To avoid initial and final billet temperature
deviations, every certain number of billets in a lot, billets should be tracked in addition to the
first and the last billets. In this research, it was investigated every how many billets should be
tracked, balancing the computation time and the accuracy loss in the later section.
69
4.1.2. Decision variables
The controllable parameter for optimization is the setting furnace temperature of each
zone during a time period as follows.
where
As the length of the time period decreases, the number of variables increases because
the time horizon is constant and the computation time increases. Additionally, there is a delay
in heating and cooling the furnace temperature. Even if the furnace temperature set point is
changed, there is a delay before the furnace temperature reaches the new set point
temperature. The length of this time period must be estimated. In this research, 5 [min] was
chosen as the length of time period ta for a furnace zone temperature set point change to
result in an actual furnace temperature change. In practice, the length of the time period
should be appropriately adjusted based on actual furnace operating performance data.
Although the length of the time period was set at 5 [min], the length of the time
period used for computing the temperature of billets should be shorter to obtain more precise
simulation. In chapter 3, the modeling time increments were evaluated comprehensively. A
modeling time period, of tp=30 [sec], can be used for the estimation of billet temperature.
4.1.3. Constraints
Key constraints in reheating furnace operation are the upper limits of the furnace
temperature in each zone. They are constrained by both furnace specification limits and
upper billet temperature restriction for guaranteeing billets quality. Furnace zone
specification temperature limits are constant but the upper limits required for billets quality
depend on what types of steel being heated in zone j during tp. Since the time period has 5
[min] length, the minimum upper limits among the billets in zone j during ta should be
70
selected as the upper limits in j during ta. Furnace zone lower limits are not restricted, but are
only limited by the goal temperatures of the billets in the furnace.
The heating rate and cooling rate of the furnace temperature are other
constraints. Even if the furnace temperature in zone j during time t is under the upper limit
after an increase, the temperature cannot be increased from time t-1 to t instantaneously.
There is a delay until the temperature reaches the set temperature in heating and in cooling.
Since the temperature during heating or cooling must not exceed the upper limit, these rates
become the constraints. For simplicity, it was assumed in this study that the furnace
temperature rises and declines linearly during set point changes. The rates for this simulation
are shown in table 4-1. Other more complex heat-up and cool-down assumptions could be
made; however, they are not expected to have a significant effect on the modeled billet
temperatures because furnace temperature set point changes during operation are very small.
From these constraints, the feasible region for this problem can be determined. Each
constraint is illustrated in figure 4-2 and the feasible region after consolidating the constraints
is shown in figure 4-3. When the variables in these simulations (the furnace temperature in
zone j during t) are changed, the temperature is assumed to be constant during ta in this
simulation. Thus, the discrete upper limits for variables were generated from the feasible
region. Figure 4-4 illustrates this discrete upper limit approach.
71
min(TUL,i,j,t)
i
Temperature
ta
Time
Time
Time
72
Temperature
Time
Time
73
4.2. Optimization method
Figure 4-5 shows the main steps in this optimization method. Every iteration calls for
the billet temperature simulation program based on the 2-D dynamic programming model
outlined in chapter 2. The key concept of this proposed control method is how to optimize
the simulation model, shortening the computation time with the smallest loss in billet
temperature prediction accuracy. The major characteristics of this model are the following
five functions. Each step will be described individually in later sections.
1. Effective targeting of a furnace zone and a time period for temperature change
adjustments
2. Classified searching for efficient temperature change recommendations
3. Dynamic updating of feasible solution region
4. Amplifier and lower limiter of temperature change adjustments
5. Selective tracking of billets for simulation iterations
74
Step 1 Set initial atmosphere temperature of each zone
75
4.2.2. Determining the initial solution
For the initial solution of this problem, the current furnace temperature, Ta,j,t, is used.
Also, current simulated billet temperatures are input into the established model, when the
simulation is carried out continuously and cyclically. The temperature of billet/bloom i at
extraction, Tex,i, is simulated for the initial state. Based on the target extraction temperatures
for each billet/bloom, Tp,i. The differences between Tex,i and Tp,i are calculated in (4.2).
If ∆Tex,i is positive, this implies that the billet i will be over-heated. On the other
hand, if ∆Tex,i is negative, this means that it will be under-heated. Over-heating is acceptable
in practice though it is not ideal, whereas under-heating is unacceptable since it causes
operational troubles. Therefore, the furnace variables are updated as negative ∆Tex,i becomes
non negative. An increase of some variables may affect the extraction temperature of billets
with positive ∆Tex,i, because it is possible that billets with positive ∆Tex,i and other billets
with negative ∆Tex,i exist in the same furnace zone at the same time
76
4.2.3. Unit increment of furnace temperature
Targeted zones and time periods are selected as discussed previously. However, it is a
difficult to estimate how much the furnace zone temperature should be increased, because it
is difficult to estimate the effect of an increase of furnace temperature on the final extraction
temperature of billets analytically. Therefore, a simulation model to estimate billets
temperature is employed. To know the extent of the effects of changing the set point furnace
temperature, the simulation has to be run. In figure 4-6, a sample calculation result is shown.
It indicates how much the sectional center billet temperatures increase by increasing the
temperature in zone j by 1 [K], 25 [K], 50 [K]. The billet material properties are the same as
those used previously in chapter 3. According to this result, the change in billet temperature
is not likely to exceed the increment of the furnace temperature change. Therefore, ∆Tex can
be effectively used as the unit increment of furnace temperature.
60
An icrease of simulated center temperature at
50
extraction of a billet ∆Tex [K]
40
30
20
10
0
0 10 20 30 40 50 60
An increasae of atmosphere temperature ∆T a [K]
Figure 4-6. Relationship between an increase of furnace temperature and the resultant
increase in billet center temperature.
77
4.2.4. Determination of the schedule matrix and the upper limit of temperature change
Using pi,tcom from chapter 2, the zone where the billet i is staying during the
computational time period tcom can be found. Schedule matrix is defined as Ai,j,t.
The schedules of billets with ∆Tex,i≥0 and billets with ∆Tex,i<0 are separately defined
in (4.4) and (4.5)
To find the upper limit of the temperature increment, C+,i,j,t and C-,i,j,t can be defined by (4.6)
and (4.7).
where
e1,NN : defined as a unit matrix with 1 × NN in size and all the elements are 1.
78
The upper limit of the increase of the variable was decided by taking the minus of the
maximum C-,i,j,t among i.
In preparation for finding the effective zone and time period on an increase in billet
temperatures, the number of billets which have scheduled matrices with positive ∆Tex,i is
counted by (4.9).
Similarly, the number of billets which have scheduled matrices with negative ∆Tex,i is
counted by (4.10).
79
4.2.5. Effective zone and time period targeting for estimating billet temperature changes
Initially, the most effective zone and time period for an increase in billet temperature
was evaluated from the first policy.
To find the most effective time period for reducing ∆Tex,i, an operational time period
for variables, ta, had to be employed. Initially a unit length of the time period, tu, 5 [min] was
selected, considering practical furnace operation. During this computation, no furnace control
action can be taken. Also, during the first operational period, the furnace temperature cannot
be increased because of the initial temperature adjustment calculation needed to reach the set
temperature at the first operational period. At this point in time, the computation time would
be 10 [min] and the unit length of the operational time period is 5 [min]. The total 15 [min]
initial time period has to be masked for the variables.
te
where
t s = t fin + t a × u + 1
t e = t fin + (t a + 1) × u
tu
u=
tp
80
The upper limit of variable increment is shown in (4.12).
The score, Scnj,ta, in each zone at each operational time period is defined as (4.13) to
evaluate whether zone j and time period ta are the most effective or not.
Selecting zone j and time ta with the highest score directly equals the first policy.
In addition to the first policy, the second policy is reflected in (4.13) by exploiting the
weights of each furnace zone. Since the priority is zone 3, zone 2, zone 1 and zone 4, the
weights, wj, should be decided keeping this order, i.e. zone 3 = 1.5, zone 2 = 1.3, zone 3 =
1.1 and zone 4 =1.0. The scores of (4.13) are multiplied by wj as (4.14).
Hence, the zone and the operational time period holding the maximum weighted score
becomes the candidate for an increase in a variable.
81
influenced time range is dependent on the amount of the temperature increase. If the increase
is large, the time range becomes wide. Even if ∆TaUL,j,t𝑎 is large, it is not always the most
efficient to increase the temperature up to ∆TaUL,j,t𝑎 , because other time periods might be
affected by increasing temperature. This implies that, if there are billets with positive ∆Tex
after and before the targeted period, those billets are heated more even though they have
already achieved their goal temperature. Therefore, not only billets with negative ∆Tex, but
also billets with positive ∆Tex must be evaluated for their extent of the influence by
increasing temperature.
Ideally, the delay effect should be evaluated for all levels of temperature increase.
However, this leads to a large increase in the computation load. To avoid this, five levels, TL,
are considered in this model. The first level is ≤25 [K] (=T1), the second level is ≤50 [K]
(=T2), the third level is ≤75 [K] (=T3), the fourth level is ≤100 [K] (=T4), and the fifth level is
>100 [K] (=T5). Each influence range is illustrated in figure 4-7.
Similarly with Nnmj,t, the number of billets with positive ∆Tex, during ta is counted
using (4.15) to evaluate the extent of over-heating.
te
82
Temperature
Ea,j,ta
Ea,j,ta-1 Ea,j,ta+1
Level1 Initial temperature
Time
Temperature
Level2
Time
Temperature
Level3
Time
Temperature
Level4
Time
83
For the next step, the extent of the influence is calculated geometrically. Let the area
in zone j during ta be Ea,j,ta and the affected range of operational time period be from trs to tre.
tre
tre
This is the score when the furnace temperature in zone j during t is increased ∆TaUL,j,ta . By
substituting ∆TaUL,j,ta with TL if all the temperature levels are lower than ∆TaUL,j,ta , each
score can be calculated. The temperature with the minimum score is selected as an increase
of furnace temperature shown in (4.19). Let the targeted zone be j=jtar, and operational time
period be ta=ttar for further discussion. Updated set furnace temperature can be obtained by
(4.20).
Using this new furnace temperature, the temperatures at the extraction of billets are
recalculated, and ∆Tex,i is updated. After increasing the furnace temperature, the temperature
during heating or cooling has a gradient. However, when the time period is chosen for an
increase of temperature for later iterations, the values must be constant within the operational
time period. Thus, the updated heat pattern is expressed in a discrete way as indicated in
figure 4-8 for the next iteration.
84
Temperature
Time
Time
This process is iterated until the minimum ∆Tex,i ≥0 or any Sct,j,ta >0. When some
billet i has a negative ∆Tex,i but any Sct,j,ta >0, Scnw,j,ta is used as the score to decide the target
instead of Sct,j,ta . From this point, an increase in furnace temperature always leads to
overheating other billets. Also, it is necessary to run the iteration until the minimum ∆Tex,i ≥0
or any Scn,j,ta =0. If all the furnace temperatures in zone j during ta holding the billets with
∆Tex,i <0, reach their upper limits and the minimum ∆Tex,i <0, this means that there is no
solution under the current furnace condition, and the furnace holding time has to be extended.
After this increase phase finishes, a decrease phase starts as the next step.
85
4.2.7. Updating the feasible region
After the increase phase finishes, the feasible region is updated before the decrease
phase. The temperature of some billets reaches their goal temperature by an increase in
furnace temperature. If the furnace temperature during holding them is decreased, the
temperatures of the billets go below their goal temperatures again. However, there is a case
when these billets are confined with the billets that were initially overheated. The
temperature of these billets should be low. The billets that initially reached their goal
temperature and the billets that reached their goal temperature after the increase phase should
be clearly distinguished. Referring to the heating schedule of the billets that reach their goal
temperature after the increase phase, the furnace temperature in zone j during computational
time period t holding the billets is fixed as the new lower limit of the feasible region. Figure
4-6 shows the procedure to update the feasible region after N increase phase.
When a lower limit of furnace temperature is initially set, the feasible region is
updated again. Considering heating and cooling rates, the line indicating their lower limits
during heating or cooling is prolonged until it crosses the lower limit as shown in figure 4-10.
Also, this line is expressed for variables in a discrete manner shown in figure 4-11.
Temperature
Lower limit
Time
86
Temperature
Lower limit
Time
Time
87
4.2.8. Decrease phase
After the feasible region is updated, the decrease phase starts. The basic control
strategy is the same as the increase phase. The upper limit of a decrease of furnace
temperature is expressed by (4.21).
Also, the upper limit for the operational time period of the variables is expressed by (4.22).
tre
tre
88
Find j and ta holding the maximum Sco,j,ta
When there is long space before the initial charged billet in the simulation, the
temperature of the first billet becomes much higher than other billets at extraction because
there are no obstacles in front of the first billet and it receives the most intense thermal
radiation. In this case, dTa is decided from the temperature of the first billet, due to its high
score. To avoid this, the first billet should be removed from the billet candidates for a
decrease of furnace temperature.
In the increase phase, when all Sct,j,ta becomes positive, Scnw,j,ta is used instead of
Sct,j,ta . However, in the decrease phase, the temperature at extraction is not allowed to be
lower than the goal temperature. Therefore, once all Sct,j,ta becomes negative, and the
decrease phase finishes.
89
4.2.10. Initial performance check
By calculating the developed optimization programming, an optimal solution was
found for an example case. The computational condition is shown in table 4-2. Figure 4-12
illustrates the obtained heat pattern of each zone over the entire reheating cycle by this
optimization programming before final treatment. Figure 4-12 illustrates optimal heat pattern
after final treatment.
Figure 4-8 shows a change from the initial ∆Tex to ∆Tex after optimization. All the
initial ∆Tex had negative values. After optimization, all the billets had higher temperature
than their goal temperature. In front of the first billet, there was no billet in the furnace. Thus,
it receives more thermal radiation than other billets, especially from the hearths, so that the
temperature was much higher than that of the second billet.
The average of initial ∆Tex was -23.14 [K] and the average of ∆Tex after optimization
was 2.52 [K]. A large improvement in ∆Tex by this optimization was confirmed.
However, the computation time was 134,532 [sec] ≈ 37.37 [hour]. This is too long for
practical control of reheating operations. The main reason of this long computation time is
that the increment of furnace temperature becomes smaller as the computation is iterated and
∆Tex becomes smaller. Figure 4-15 shows the ∆Tex history for iterations. After 143 iterations,
dTa always became under 1 [K]. This implies that the solutions are almost the same after 143
iterations since the increment is quite small. Even if the computation stops at 143 iterations,
it will take about 1.8 [hour] instead of 37.37 [hour]. Therefore, it is necessary to shorten the
computation time even further for practical operation.
90
1450
1400
Atmosphere temperature [K]
1350
1300
1250
1200
1150
1100
1050
1000
14
27
40
53
66
79
92
1
105
118
131
144
157
170
183
196
209
222
235
248
261
274
287
300
313
326
339
352
365
378
Time [×30sec]
Figure 4-12. Obtained heat patterns for each zone before final treatment.
1450
1400
Atmosphere temperature [K]
1350
1300
1250
1200
1150
1100
1050
1000
14
27
40
53
66
79
92
1
105
118
131
144
157
170
183
196
209
222
235
248
261
274
287
300
313
326
339
352
365
378
Time [×30sec]
Figure 4-13. Obtained optimal heat patterns for each zone after final treatment.
91
dTa [K] ∆Tex [K]
0
5
10
15
20
25
30
35
40
-60
-40
-20
0
20
40
60
1
#1
91
181 #4
271 #7
361 #10
451 #13
541
#16
631
721 #19
811 #22
901 #25
991 #28
1081 #31
1171
1261 #34
#37
92
1351
#40
∆Tex after optimization
1441
1531 #43
1621 #46
1711
Billet number
#49
Number of iterations
1801
1891 #52
1981 #55
Initial ∆Tex
2071 #58
2161
2791 #79
2881
#82
#85
4.3. Shortening computation time
In figure 4-16, the gradient of the line in zone 1was the smallest. When dTa=100 [K],
∆Tex=28.5 [K]. The gradient was about 3.5. This indicates that there is an opportunity to
amplify 3.5 to ∆Tex for an increase of furnace temperature. Hence, it is possible to boost dTa
by amplifying for the shorter computation time. The gradient also changes if the furnace
temperature changes, as shown previously. Therefore, this amplifying level also must be
carefully chosen by balancing the computation time and the over-heat or under-heat
thresholds.
93
Goal temperature Initial furnace temperature [K] Weights
of all the billets [K] Zone 1 Zone 2 Zone 3 Zone 4 Zone 1 Zone 2 Zone 3 Zone 4
1133 1173 1273 1373 1333 1.2 1.4 1.5 1.0
50
45
40
35
30
∆Tex [K]
25
20
15
10
5
0
0 20 40 60 80 100 120
∆Ta [K]
15
14
13
12
11
10
9
∆Tex [K]
8
7
6
5
4
3
2
1
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
∆Ta [K]
Figure 4-16. Relationship between dTa and ∆Tex for low furnace temperature.
94
Goal temperature Initial furnace temperature [K] Weights
of all the billets [K] Zone 1 Zone 2 Zone 3 Zone 4 Zone 1 Zone 2 Zone 3 Zone 4
1133 1323 1373 1423 1333 1.2 1.4 1.5 1.0
50
45
40
35
30
-∆Tex [K]
25
20
15
10
0
0 20 40 60 80 100 120
∆Ta [K]
15
14
13
12
11
10
9
∆Tex [K]
8
7
6
5
4
3
2
1
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
∆Ta [K]
Figure 4-17. Relationship between dTa and ∆Tex for high furnace temperature.
95
4.3.2. Selective billet tracking
The temperature of all billets in the furnace has been simulated so far. However, if
their reheating specification is the same, it might be possible to remove some billets from the
simulated billets. Selecting tracked billets appropriately can shorten the computation time in
the optimization.
When tracked billets are selected, the hearth temperature is of concern. The
temperatures of the hearths are affected by the temperature of the entering billets. This
implies that all the temperature of billets must be simulated. If tracked billets are selected and
the temperatures of the other billets are not simulated, the hearth temperature cannot be
correctly calculated. However, the computation of all the billets temperature takes hours. To
overcome this difficulty, the bottom face temperature of the billets which are not simulated in
optimization was linearly approximated using the temperature of foregoing billets which are
tracked. From the position of the tracked billets, the bottom face temperature of unfocused
billets is calculated by (4.21).
Tfb − Tofb
Tub = Tfb − × (Pfb − Pub ) ⋯ (4.21)
Pfb
where
96
4.3.3. Effects of selective tracking, amplifying and lower limiter
To find appropriate parameters, the effects of selective tracking, amplifying and
lower limiter on the computation time and the accuracy were investigated.
Figure 4-18 shows the comparison of the computation time in 6 different cases to
determine the appropriate number of billets to be tracked. The differences of the cases are
how many billets are tracked and fully simulated for their temperature. When all of the billets
were tracked and the operation was optimized by this method, the computation time was
2460 [sec] with amplifier 3 and a lower limiter of 5 [K]. This computational simplification
exceeds the goal computation time. Considering a case that it takes extra time to complete
computing, case 3 is a candidate for selective billets tracking.
3,000
2,460
2,500
Computation time [sec]
2,000
1,500
1,000 798
Goal computation time: 600 [sec] 545
500 359
188
33
-
3 11 15 20 30 85
Case 1 Case 2 Case 3 Case 4 Case 5 Case 6
The number of tracked billets in simulation
Figure 4-18. Comparison of computation time based on the number of tracked billets.
97
Figure 4-19 shows the average of ∆Tex and the minimum of ∆Tex in 6 cases. The error
bars indicated ±1σ. Case 6 has relatively high average temperature, but the minimum is
positive while the minimum for the other cases is negative. This implies that all the billets
satisfy the goal temperature in case 6, whereas some billets are over-heated or under-heated
in the other cases. If there is an acceptable range of ∆Tex, such as ±3 [K] or ±5 [K], the other
cases, except case 1, can be used, because the averages are not different from that in case 6.
Now, since case 6 is not realistic due to the computation time, case 3 is the most appropriate
in this condition. In later discussion, case 3 is employed for this reason.
10.0
(1.8)
(5.0)
(10.0)
(15.0)
(20.0)
(25.0) -23.2
Average Min
Figure 4-19. Average ∆Tex and minimum ∆Tex for the various cases.
98
Figure 4-20 shows a comparison of the computation time for various lower limits and
amplifiers. If there is no lower limiter, the effective computation time becomes 2106 [sec]
with amplifier 3, even if tracked billets are selected. When 5 [K] or 10 [K] are chosen as the
lower limiter, the computation time was dramatically reduced, and all the computation times
were within 10 [min]. On the other hand, the effect of amplifier on the reduction of
computation time is smaller than that of lower limiter. Figure 4-21 illustrates the average of
∆Tex and ±1σ range in various amplifier and lower limiters. Also, figure 4-22 shows the total
over-heat for various amplifier and lower limiters. The total over-heat was obtained by
multiplying the specific heat of billets to each ∆Tex and taking the summation. As a result,
when lower limiter 10 [K] and amplifier 1.5 to 2.0, the total over-heat was higher, and the
average of ∆Tex was higher as well. When the lower limiter is 5 [K], or when the lower
limiter 10 [K] and the amplifier 1, the total heat, the average of ∆Tex and the computation
time are well balanced. The combination of the lower limiter 10 [K] and the amplifier 2.5 to
3 has a potential for high total over-heat, because higher values were observed in amplifier
1.5 and 2.0. In summary, 10 [K] as the lower limiter, 1 as the amplifier (no amplifier) and
case 3 as selective tracking were selected for control optimization in this thesis.
2,000
1,500
1,000
462 444 393 364 359 364
500 309 289 257 213
-
3 1 1.5 2 2.5 3 1 1.5 2 2.5 3
Lower Lower limiter 5 [K] Lower limiter 10 [K]
limiter 0
[K]
Amplifier
Figure 4-20. Computation time comparison for various amplifiers and lower limiters.
99
14.00
12.00
10.00
Average of ∆Tex [K]
8.00
6.00
4.00
2.00
-
(2.00)
(4.00)
3 1 1.5 2 2.5 3 1 1.5 2 2.5 3
Lower Lower limiter 5 [K] Lower limiter 10 [K]
limiter
0 [K]
Amplifier
Figure 4-21. Average of ∆Tex and ±1σ range for various amplifiers and lower limiters.
1,000
902
900
800 718
Total over-heat [kJ]
Figure 4-22. Total over-heat for 85 billets for various amplifiers and lower limiters.
100
4.4. Overall Control Performance
1450
1400
Atmosphere temperature [K]
1350
1300
1250
1200
1150
1100
1050
1000
99
15
29
43
57
71
85
1
113
127
141
155
169
183
197
211
225
239
253
267
281
295
309
323
337
351
365
379
Time [×30sec]
101
1450
105
118
131
144
157
170
183
196
209
222
235
248
261
274
287
300
313
326
339
352
365
378
Time [×30sec]
(a) Zone 1
1450
Optimal atmosphere temperature [K]
1400
1350
1300
1250
1200
1150
1100
1050
1000
14
40
66
92
27
53
79
1
118
144
105
131
157
170
183
196
209
222
235
248
261
274
287
300
313
326
339
352
365
378
Time [×30sec]
(b) Zone 2
Figure 4-24. Difference in heat pattern between lower limiter 0 and 10 [K] (1).
102
1450
Optimal atmosphere temperature [K]
1400
1350
1300
1250
1200
1150
1100
1050
1000
15
29
43
57
71
85
99
1
127
197
113
141
155
169
183
211
225
239
253
267
281
295
309
323
337
351
365
379
Time [×30sec]
(c) Zone 3
1450
Optimal atmosphere temperature [K]
1400
1350
1300
1250
1200
1150
1100
1050
1000
15
29
43
57
71
85
99
1
225
323
113
127
141
155
169
183
197
211
239
253
267
281
295
309
337
351
365
379
Time [×30sec]
(d) Zone 4
Figure 4-25. Difference in heat pattern between lower limiter 0 and 10 [K] (2).
103
Figure 4-26 shows a change from the initial ∆Tex to ∆Tex after optimization in each
billet. Some billets have negative values, but they are small enough to be ignored. Figure 4-
27 shows a comparison of ∆Tex between the two cases. In the case of lower limiter 10 [K],
∆Tex is slightly higher than those of the case of no lower limiter. The average of ∆Tex and the
minimum of ∆Tex are shown in figure 4-28. Based on this result, the temperature difference
was 1.4 [K]. The computation time is shortened by sacrificing this amount of heat under this
operational condition. If this amount of heat is not acceptable, lower limiter must be reduced
or the number of tracked billets should be increased.
60
50
40
30
20
10
∆Tex [K]
0
-10
-20
-30
-40
-50
-60
#1 #4 #7 #10#13#16#19#22#25#28#31#34#37#40#43#46#49#52#55#58#61#64#67#70#73#76#79#82#85
104
60
50
40
∆Tex [K]
30
20
10
0
#4
#1
#7
#73
#10
#13
#16
#19
#22
#25
#28
#31
#34
#37
#40
#43
#46
#49
#52
#55
#58
#61
#64
#67
#70
#76
#79
#82
#85
-10
Billet number
12.0
10.0
8.0
Temperature [K]
6.0
3.9
4.0
2.5
2.0
-
Amplifier 1, Lower limter 0 [K] Amplifier 1, Lower limter 10 [K]
(2.0) -0.7
-1.6
(4.0)
Figure 4-28. Average ∆Tex and minimum ∆Tex for different lower limiter conditions.
105
4.4.2. Effects of initial furnace temperature
When the initial furnace temperature is set at different level, the optimal heat pattern
might be different. In this section, the performance of this model for various initial furnace
temperatures was evaluated. Table 4-4 indicates four cases with different initial furnace
temperatures.
Figure 4-29 and 4-30 illustrate the optimized heat patterns by this model. Each case
had a different optimal heat pattern. In zone 1 in figure 4-29, the temperature in case 1 and
case 2 decreased from their initial temperature, whereas there was no change in case 3 and
case 4. This implies that the optimal heat pattern is changed slightly if the initial furnace
temperature is different. In figure 4-31, the average of ∆Tex and the minimum of ∆Tex were
compared in the four cases. Case 1 and case 2 showed high average and high minimum
temperatures. On the other hand, case 3 and case 4 showed low average and low minimum
temperatures. In case 3 and case 4, they experience increase phase, that is, the initial ∆Tex are
both negative. Therefore, once ∆Tex enters the acceptable range, the computation stops even
if the value is negative. This algorithm seems to cause the low minimum temperature in case
3 and case 4. Conversely, in case 1 and case 2, ∆Tex starts from positive values. Hence, it is
thought that the average and the minimum were relatively high.
Figure 4-32 shows a comparison of the computation time. Since the optimal heat
pattern was close to the initial furnace temperature in case 3 and case 4, these cases had
shorter computation time.
Overall, because safety in operation is a high priority, the initial furnace temperature
should be set at a slightly high level. If the cost is the priority, then it should be set at a
slightly low level.
106
(a) Zone 1
(b) Zone 2
Figure 4-29. Heat pattern differences for various initial furnace temperatures (1).
107
(c) Zone 3
(d) Zone 4
Figure 4-30. Heat pattern differences for various initial furnace temperatures (2).
108
15.0
13.0
11.0
9.0
Temperature [K]
7.0 6.3
5.6
5.0 3.7 3.9
3.0
1.0 0.12
(5.0)
Figure 4-31. Average ∆Tex and minimum ∆Tex for various initial furnace temperatures.
800 80
700 70
600 62 62 60
546
Computation time [sec]
535
Number of iterations
500 50
400 39 364 40
333 38
300 30
200 20
100 10
0 0
Case 1 Case 2 Case 3 Case 4
Figure 4-32. Computation time and number of iterations for various initial furnace
temperatures.
109
4.4.3. Effects of inserting billets with higher goal temperatures
The case that all the billets have the same specifications has been evaluated so far.
The optimization performance was further investigated in a case which 22 billets with higher
goal temperature, 1213 [K] which is 80 [K] higher, are inserted as the 19th to 40th billets.
Figure 4-33 and 4-34 show the resultant heat pattern. Because of the high goal
temperatures, the temperature in zone 1 and zone 2 showed higher temperature when the
billets entered those zones. Also, figure 4-35 shows the computation time. Since the optimal
heat pattern became further from the initial furnace temperature by inserting billets with a
higher goal temperature, the computation time became longer.
Figure 4-36 indicates the average and the minimum of ∆Tex. Compared to a case that
all the billets have the same goal temperature, the average became higher. Also, the
minimum was -11.3 [K] – a value greatly lower than their goal temperature. Figure 4-36
shows the ∆Tex change of each billet. According to this result, the first 3 billets and the last 3
billets among the billets with high goal temperature had large negative values. In figure 4-37,
the ∆Tex change of tracked billets in the optimization. All ∆Tex were positive. From these
data, the first 3 and the last 3 billets with high goal temperature were not tracked in this
optimization. Therefore, it can be said that they had large negative values in ∆Tex.
Now, to track those billets, the inserting positions were shifted 3 billets later in the
next trial. In this case, the positions of the inserted billets are from 22th to 43th billet. Then,
the obtained ∆Tex of each billet is shown in figure 4-39. All the values became positive. Also,
in figure 4-40, the average and the minimum are indicated. After shifting the billets, the
average was increased and the minimum was positive as mentioned above.
110
during the rolling schedule, consolidating the same type of billets in rolling schedules is
important to reduce this energy loss.
1450
Simulated atmosphere temperature [K]
1400
1350
1300
1250
1200
1150
1100
1050
1000
49
13
25
37
61
73
85
97
1
109
121
133
145
157
169
181
193
205
217
229
241
253
265
277
289
301
313
325
337
349
361
373
385
Time [×30sec]
(a) Zone 1
1450
Simulated atmosphere temperature [K]
1400
1350
1300
1250
1200
1150
1100
1050
1000
97
13
25
37
49
61
73
85
1
109
121
133
145
157
169
181
193
205
217
229
241
253
265
277
289
301
313
325
337
349
361
373
385
Time [×30sec]
(b) Zone 2
Figure 4-33. Heat pattern of billets with high goal temperatures (1).
111
1450
1350
1300
1250
1200
1150
1100
1050
1000
12
23
34
45
56
67
78
89
1
100
111
122
133
144
155
166
177
188
199
210
221
232
243
254
265
276
287
298
309
320
331
342
353
364
375
386
Time [×30sec]
(c) Zone 3
1450
Simulated atmosphere temperature [K]
1400
1350
1300
1250
1200
1150
1100
1050
1000
12
23
34
45
56
67
78
89
1
100
111
122
133
144
155
166
177
188
199
210
221
232
243
254
265
276
287
298
309
320
331
342
353
364
375
386
Time [×30sec]
(d) Zone 4
Figure 4-34. Heat pattern of billets with high goal temperatures (2).
112
800 80
700 70
600 60
Computation time [sec]
478 51
Number of iterations
500 50
38
364
400 40
300 30
200 20
100 10
0 0
Inserted high goal temperature billets Only same type of billets
Figure 4-35. Computation time and number of iterations for a case with high goal
temperature billets.
40
30
20
Temperature [K]
14.2
10
3.9
0
Inserted high goal temperature billets Only same type of billets
-1.6
-10
-11.3
-20
Figure 4-36. Average ∆Tex and minimum ∆Tex of a case with high goal temperature billets.
113
100
80
60
40
20
∆Tex,i [K]
0 #1
#4
#7
#10
#13
#16
#19
#22
#25
#28
#31
#34
#37
#40
#43
#46
#49
#52
#55
#58
#61
#64
#67
#70
#73
#76
#79
#82
#85
-20
-40
-60
-80
-100
-120
-140
Figure 4-37. Change of ∆Tex before and after optimization in a case having high goal
temperature billets.
100
80
60
40
20
∆Tex,i [K]
0
#1 #2 #8 #15 #22 #29 #36 #43 #50 #57 #64 #70 #71 #78 #85
-20
-40
-60
-80
-100
-120
-140
Figure 4-38. Change of ∆Tex before and after optimization of tracked billets for a case having
high goal temperature billets.
114
100
80
60
40
20
∆Tex,i [K]
0
#1
#4
#7
#10
#13
#16
#19
#22
#25
#28
#31
#34
#37
#40
#43
#46
#49
#52
#55
#58
#61
#64
#67
#70
#73
#76
#79
#82
#85
-20
-40
-60
-80
-100
-120
-140
Figure 4-39. Change of ∆Tex before and after optimization for a case having high goal
temperature billets with shifting the tracked billets.
60
50
40
Temperature [K]
30
24
20
14.2
10
3.9
0.1
0
Inserted high goal temperature Only same type of billets After shifting
-10 billets -1.6
-11.3
-20
Figure 4-40. Average ∆Tex and minimum ∆Tex for a case having high goal temperature billets
with shifting the tracked billets.
115
4.4.4. Initial control action when unexpected stoppage occur
When unexpected stoppages, such as operational troubles, occur, the operator who is
responsible for the stoppage announces the expected time to fix. Unless the time is short, the
furnace temperature of a furnace should be reduced temporally to avoid billet overheating.
Just after the announcement, this simulation starts to find the optimal furnace temperature
based on the operators’ stoppage time estimate. During computing, the heat to the reheating
furnace is stopped. From this, the billet temperature is expected to decline during computing.
Then, the predictively reached temperature after the computation time is determined as the
initial temperature for optimization.
116
Chapter 5. CONCLUSION
117
5.1. Conclusion summary
In this research, an improved method for reheating furnace operation in rolling mills
has been proposed in order to achieve the most economical operation under various
circumstances. Since furnace temperature is not stable and is not maintained at the same
level, a dynamic programming was developed to estimate billets temperature. However, the
computation time for the initial complex model was the difficulty for real time operation.
This research also has focused on how to shorten the computation time without large loss in
temperature prediction accuracy allowing the model to be used for real time reheating
furnace temperature control.
A simulation model for billet temperature was created using Matlab software and a
commercial personal computer. The major features of this model are;
The most important temperature estimate for the accurate control of billet temperature
is the sectional center temperature in the middle of furnace width, which is the lowest
temperature of a billet and is the bottle neck in terms of rolling load during subsequent
rolling.
When the sectional center temperature of billets is considered, the heat transmission
in longitudinal direction of the billet can be considered negligible. Hence, the dimension of
the control model can be reduced from 3-D to 2-D. Also, 55 [mm] × 55 [mm] as the unit
118
mesh size of a billet in the model and 30 [sec] as time increments were selected, balancing
the computation time and the temperature prediction accuracy.
Exploiting this model, an optimization method was developed. This method utilized
two policies.
1. The zone and the timing, where the total gap between goal temperature and simulated
center temperature of billets at extraction is the largest, were selected to increase or
decrease the furnace temperature as a priority.
2. The priority in increasing furnace temperature is first zone 3, then zone 2, then zone1
and finally zone 4, because of the differences in heat transfer efficiency in the various
furnace zones.
1. Effective targeting of a furnace zone and a time period for temperature change
adjustments
2. Classified searching for efficient temperature change recommendations
3. Dynamic updating of feasible solution region
4. Amplifier and lower limiter of temperature change adjustments
5. Selective tracking of billets for simulation iterations
119
5.2. Insight for better furnace structure based on simulation results
From the discussion so far, some insights for improved future furnace specification
and operation were obtained.
There is a case that billets with high goal temperature and billets with low goal
temperature are scheduled at close timing. This might cause violations of the upper limit of
furnace temperature of billets with low goal temperature or failing to reach goal temperature
for billets with high goal temperature. Thus, this rolling order should be rescheduled. For
economically better operation, not only optimizing the operation, but also scheduling the
rolling order is quite important. By combining the optimization method proposed in this
thesis and the optimization of rolling schedule as Fujii et al. proposed [26], further
improvement in reheating furnace operation is expected.
120
5.3. Limitation of this research and further research
recommendations
For simplicity, it was assumed that the hearth temperature and other wall temperature
were uniform in the various zones. However, the temperature will not be uniform in practice.
Local billets temperature was calculated every mesh. Similarly, the local temperature of
furnace bricks also should be computed by the difference method for further improvement of
this simulation model. Also, if there are temperature differences in opposite faces of a billet,
the whole shape of a billet may be deformed because of the thermal expansion difference.
Since this deformation influences the state of heat transmission, this deformation also should
be taken into consideration in the model.
The sectional center temperature of billets was used in this thesis because it can be
the representative temperature indicating subsequent rolling load. In practice, the rolling load
is affected by other factors, such as roller gap, roller diameter, sectional size variance of
billets and so forth. To separate these from the effect of the sectional center temperature on
rolling load, these effects on rolling load should be investigated.
Heating and cooling rates in furnace temperature were assumed constant for
simplicity. However, these rates are influenced by other factors, for instance the number of
billets heated in the same zone, their temperature, brick temperature, burners’ performance
and so on. Those rates should be investigated in detail for future model refinement.
Additionally, a constraint of furnace temperature difference between zones was ignored in
121
this thesis. However, if there is huge difference between adjacent zones, the atmosphere with
higher temperature flows into the adjacent zone, because there is an open space under the
dividing wall between them. As a result, the atmosphere with the lower temperature fails to
maintain the optimal temperature. To improve this model, the restriction in temperature
difference between zones should be installed into the constraints of this model. The weights
which were used to decide the priority of zones in this thesis should also be correspondingly
adjusted based on the performance of real furnaces.
In the near future, the validity of these model and improved control method will have
to be verified on the production floor.
122
APPENDIX A. Dimensions of model furnace.
Lzc(1)
Lzc(2)
Wth
Lzc(3)
Hbw
/Hfw
Lzc(4)
df
WF Hc
123
APPENDIX B. General calculation of view-factor.
Generalized equation for finding view factors is obtained by the way shown in [14].
dφ
dA2
dω
dA1
r=1
dθ
θ
𝑖 𝑑 2 𝑄̇
=
𝑑𝐴1 cos 𝜑 𝑑𝜔
where
dω = sin φ dφdθ
π
2π
2
E = ∫ dφ ∫ i cos φ sin θ dφ
0 0
where
124
When i is uniform for all the emitting directions, the integral above is computed simply.
E = πi
Since E = σεT 4 ,
σεT 4
i=
π
i1 d2 Q̇1
=
dA1 cos φ1 dω
dA2 cos φ2
dω =
l2
cos φ1 cos φ2
→ d2 Q̇1 = i1 dA1 dA2
l2
cos φ1 cos φ2
→ d2 Q̇2 = i2 dA1 dA2
l2
where
cos φ1 cos φ2
d2 Q̇ = d2 Q̇1 − d2 Q̇ 2 = (i1 − i2 ) dA1 dA2
l2
cos φ1 cos φ2
→ d2 Q̇ = σ(T14 − T24 ) dA1 dA2
πl2
Then,
cos φ1 cos φ2
→ Fd12 = ∫ dA2
A2 πl2
1 1 cos φ1 cos φ2
F12 = ∫ Fd12 dA1 = ∫ ∫ dA1 dA2
A1 A1 A1 A1 A2 πl2
125
APPENDIX C. Heat transmission calculation.
When z=1
πr
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t + q rad,bi+1→bi,t ) + Lu × r
2
× (q cond,(1,2,1),t + q cond,(2,1,1),t )
πr 2
+ (q cond,(1,1,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )
4
Li πr 2
= ρi × ( × ) × ci × (Ti,(1,1,1),t+1 − Ti,(1,1,1),t )
n 4
→ Ti,(1,1,1),t+1 = Ti,(1,1,1),t
4 πrLu
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
πr 2 L u ρi ci 2
+ q rad,bi+1→bi,t ) + rLu (q cond,(1,2,1),t + q cond,(2,1,1),t )
πr 2
+ (q cond,(1,1,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )}
4
where
Li
Lu = ∶ Unit length of the mesh in z axis
n
wi π
Li = {Hb Wb − 4 (r 2 − 4 r 2 )} : Length of billet/bloom i
ρi
126
Heat balance at component B
Wb − 2r
Lu × × (q trans,bH + q cond,(x,2,1),t ) + Lu × r × (q cond,(x−1,1,1),t + q cond,(x+1,1,1),t )
ℓ−2
Wb − 2r
+ × r × (q cond,(x,1,2),t + q rad,bHe,t + q rad,wb,t )
ℓ−2
Wb − 2r
= ρi × (Lu × × r) × ci × (Ti,(x,1,1),t+1 − Ti,(x,1,1),t )
ℓ−2
→ Ti,(x,1,1),t+1 = Ti,(x,1,1),t
ℓ−2 Lu (Wb − 2r)
+ { (q trans,bH + q cond,(x,2,1),t )
rLu (Hb − 2r)ρi ci ℓ−2
+ rLu (q cond,(x−1,1,1),t + q cond,(x+1,1,1),t )
r(Wb − 2r)
+ (q cond,(x,1,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )}
ℓ−2
πr
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t − q rad,bi→bi−1,t ) + Lu × r
2
× (q cond,(ℓ,2,1),t + q cond,(ℓ−1,1,1),t )
πr 2
+ (q cond,(ℓ,1,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )
4
πr 2
= ρi × (Lu × ) × ci × (Ti,(ℓ,1,1),t+1 − Ti,(ℓ,1,1),t )
4
→ Ti,(ℓ,1,1),t+1 = Ti,(ℓ,1,1),t
4 πrLu
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
πr 2 L u ρi ci 2
− q rad,bi→bi−1,t ) + rLu (q cond,(ℓ,2,1),t + q cond,(ℓ−1,1,1),t )
πr 2
+ (q cond,(ℓ,1,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )}
4
127
Heat balance at component D
Hb − 2r
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t + q rad,bi+1→bi,t + q cond,(2,y,1),t )
m−2
Hb − 2r
+ Lu × r × (q cond,(1,y+1,1),t + q cond,(1,y−1,1),t ) + ×r
m−2
× (q cond,(1,y,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )
Hb − 2r
= ρi × (Lu × × r) × ci × (Ti,(1,y,1),t+1 − Ti,(1,y,1),t )
m−2
→ Ti,(1,y,1),t+1 = Ti,(1,y,1),t
m−2 Lu (Hb − 2r)
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
rLu (Hb − 2r)ρi ci m−2
+ q rad,bi+1→bi,t + q cond,(2,y,1),t ) + rLu (q cond,(1,y+1,1),t + q cond,(1,y−1,1),t )
r(Hb − 2r)
+ (q cond,(1,y,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )}
m−2
Wb − 2r Hb − 2r
Lu × × (q cond,(x,y−1,1) + q cond,(x,y+1,1),t ) + Lu ×
ℓ−2 m−2
Wb − 2r Hb − 2r
× (q cond,(x−1,y,1),t + q cond,(x+1,y,1),t ) + ×
ℓ−2 m−2
× (q cond,(x,y,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )
Wb − 2r Hb − 2r
= ρi × (Lu × × ) × ci × (Ti,(x,y,1),t+1 − Ti,(x,y,1),t )
ℓ−2 m−2
→ Ti,(x,y,1),t+1 = Ti,(x,y,1),t
(ℓ − 2)(m − 2) Lu (Wb − 2r)
+ { (q cond,(x,y−1,1)
Lu (Wb − 2r)(Hb − 2r)ρi ci ℓ−2
Lu (Hb − 2r)
+ q cond,(x,y+1,1),t ) + (q cond,(x−1,y,1),t + q cond,(x+1,y,1),t )
m−2
(Wb − 2r)(Hb − 2r)
+ (q cond,(x,y,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t
(ℓ − 2)(m − 2)
+ q tran,gb,t )}
128
Heat balance at component F
Hb − 2r
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t − q rad,bi→bi−1,t
m−2
Hb − 2r
+ q cond,(ℓ−1,y,1),t ) + Lu × r × (q cond,(ℓ,y+1,1),t + q cond,(ℓ,y−1,1),t ) +
m−2
× r × (q cond,(ℓ,y,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )
Hb − 2r
= ρi × (Lu × × r) × ci × (Ti,(ℓ,y,1),t+1 − Ti,(ℓ,y,1),t )
m−2
→ Ti,(ℓ,y,1),t+1 = Ti,(ℓ,y,1),t
m−2 Lu (Hb − 2r)
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
rLu (Hb − 2r)ρi ci m−2
− q rad,bi→bi−1,t + q cond,(ℓ−1,y,1),t ) + rLu (q cond,(ℓ,y+1,1),t + q cond,(ℓ,y−1,1),t )
r(Hb − 2r)
+ (q cond,(ℓ,y,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )}
m−2
πr
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t + q rad,bi+1→bi,t ) + Lu × r
2
× (q cond,(1,m−1,1),t + q cond,(2,m,1),t )
πr 2
+ (q cond,(1,m,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )
4
πr 2
= ρi × (Lu × ) × ci × (Ti,(1,m,1),t+1 − Ti,(1,m,1),t )
4
→ Ti,(1,m,1),t+1 = Ti,(1,m,1),t
4 πrLu
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
πr 2 L u ρi ci 2
+ q rad,bi+1→bi,t ) + rLu (q cond,(1,m−1,1),t + q cond,(2,m,1),t )
πr 2
+ (q cond,(1,m,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )}
4
129
Heat balance at component H
Wb − 2r
Lu × × (q rad,cb,t + q rad,wb,t + q tran,gb,t + q cond,(x,m−1,1),t ) + Lu × r
ℓ−2
Wb − 2r
× (q cond,(x−1,m,1),t + q cond,(x+1,m,1),t ) + ×r
ℓ−2
× (q cond,(x,m,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )
Wb − 2r
= ρi × (Lu × × r) × ci × (Ti,(x,m,1),t+1 − Ti,(x,m,1),t )
ℓ−2
→ Ti,(x,m,1),t+1 = Ti,(x,m,1),t
ℓ−2 Lu (Wb − 2r)
+ { (q rad,cb,t + q rad,wb,t + q tran,gb,t
rLu (Hb − 2r)ρi ci ℓ−2
+ q cond,(x,m−1,1),t ) + rLu (q cond,(x−1,m,1),t + q cond,(x+1,m,1),t )
r(Wb − 2r)
+ (q cond,(x,m,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )}
ℓ−2
πr
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t − q rad,bi→bi−1,t ) + Lu × r
2
× (q cond,(ℓ,m−1,1),t + q cond,(ℓ−1,m,1),t )
πr 2
+ (q cond,(ℓ,m,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )
4
πr 2
= ρi × (Lu × ) × ci × (Ti,(ℓ,m,1),t+1 − Ti,(ℓ,m,1),t )
4
→ Ti,(ℓ,m,1),t+1 = Ti,(ℓ,m,1),t
4 πrLu
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
πr 2 L u ρi ci 2
− q rad,bi→bi−1,t ) + rLu (q cond,(ℓ,m−1,1),t + q cond,(ℓ−1,m,1),t )
πr 2
+ (q cond,(ℓ,m,2),t + q rad,Heb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t )}
4
130
When z=z
πr
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t + q rad,bi+1→bi,t ) + Lu × r
2
πr 2
× (q cond,(1,2,z),t + q cond,(2,1,z),t ) + (q cond,(1,1,z+1),t + q cond,(1,1,z−1) )
4
πr 2
= ρi × (Lu × ) × ci × (Ti,(1,1,z),t+1 − Ti,(1,1,z),t )
4
→ Ti,(1,1,z),t+1 = Ti,(1,1,z),t
4 πrLu
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
πr 2 L u ρi ci 2
+ q rad,bi+1→bi,t ) + rLu (q cond,(1,2,z),t + q cond,(2,1,z),t )
πr 2
+ (q cond,(1,1,z+1),t + q cond,(1,1,z−1) )}
4
Wb − 2r
Lu × × (q trans,bH + q cond,(x,2,z),t ) + Lu × r × (q cond,(x−1,1,z),t + q cond,(x+1,1,z),t )
ℓ−2
Wb − 2r
+ × r × (q cond,(x,1,z+1),t + q cond,(x,1,z−1),t )
ℓ−2
Wb − 2r
= ρi × (Lu × × r) × ci × (Ti,(x,1,z),t+1 − Ti,(x,1,z),t )
ℓ−2
→ Ti,(x,1,z),t+1 = Ti,(x,1,z),t
ℓ−2 Lu (Wb − 2r)
+ { (q cond,bH + q cond,(x,2,z),t )
rLu (Hb − 2r)ρi ci ℓ−2
+ rLu (q cond,(x−1,1,z),t + q cond,(x+1,1,z),t )
r(Wb − 2r)
+ (q cond,(x,1,z+1),t + q cond,(x,1,z−1),t )}
ℓ−2
131
Heat balance at component C
πr
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t − q rad,bi→bi−1,t ) + Lu × r
2
πr 2
× (q cond,(ℓ,2,z),t + q cond,(ℓ−1,1,z),t ) + (q cond,(ℓ,1,z+1),t + q cond,(ℓ,1,z−1),t )
4
πr 2
= ρi × (Lu × ) × ci × (Ti,(ℓ,1,z),t+1 − Ti,(ℓ,1,z),t )
4
→ Ti,(ℓ,1,z),t+1 = Ti,(ℓ,1,z),t
4 πrLu
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
πr 2 L u ρi ci 2
− q rad,bi→bi−1,t ) + rLu (q cond,(ℓ,2,z),t + q cond,(ℓ−1,1,z),t )
πr 2
+ (q cond,(ℓ,1,z+1),t + q cond,(ℓ,1,z−1),t )}
4
Hb − 2r
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t + q rad,bi+1→bi,t + q cond,(2,y,z),t )
m−2
Hb − 2r
+ Lu × r × (q cond,(1,y+1,z),t + q cond,(1,y−1,z),t ) + ×r
m−2
× (q cond,(1,y,z+1),t + q cond,(1,y,z−1) )
Hb − 2r
= ρi × (Lu × × r) × ci × (Ti,(1,y,z),t+1 − Ti,(1,y,z),t )
m−2
→ Ti,(1,y,z),t+1 = Ti,(1,y,z),t
m−2 Lu (Hb − 2r)
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
rLu (Hb − 2r)ρi ci m−2
+ q rad,bi+1→bi,t + q cond,(2,y,z),t ) + rLu (q cond,(1,y+1,z),t + q cond,(1,y−1,z),t )
r(Hb − 2r)
+ (q cond,(1,y,z+1),t + q cond,(1,y,z−1) )}
m−2
132
Heat balance at component E
Wb − 2r Hb − 2r
Lu × × (q cond,(x,y−1,z) + q cond,(x,y+1,z),t ) + Lu ×
ℓ−2 m−2
Wb − 2r Hb − 2r
× (q cond,(x−1,y,z),t + q cond,(x+1,y,z),t ) + ×
ℓ−2 m−2
× (q cond,(x,y,z+1),t + q cond,(x,y,z−1),t )
Wb − 2r Hb − 2r
= ρi × (Lu × × ) × ci × (Ti,(x,y,z),t+1 − Ti,(x,y,z),t )
ℓ−2 m−2
→ Ti,(x,y,z),t+1 = Ti,(x,y,z),t
(ℓ − 2)(m − 2) Lu (Wb − 2r)
+ { (q cond,(x,y−1,z) + q cond,(x,y+1,z),t )
Lu (Wb − 2r)(Hb − 2r)ρi ci ℓ−2
Lu (Hb − 2r)
+ (q cond,(x−1,y,z),t + q cond,(x+1,y,z),t )
m−2
(Wb − 2r)(Hb − 2r)
+ (q cond,(x,y,z+1),t + q cond,(x,y,z−1),t )}
(ℓ − 2)(m − 2)
Hb − 2r
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t − q rad,bi→bi−1,t
m−2
Hb − 2r
+ q cond,(ℓ−1,y,z),t ) + Lu × r × (q cond,(ℓ,y+1,z),t + q cond,(ℓ,y−1,z),t ) +
m−2
× r × (q cond,(ℓ,y,z+1),t + q cond,(ℓ,y,z−1),t )
Hb − 2r
= ρi × (Lu × × r) × ci × (Ti,(ℓ,y,z),t+1 − Ti,(ℓ,y,z),t )
m−2
→ Ti,(ℓ,y,z),t+1 = Ti,(ℓ,y,z),t
m−2 Lu (Hb − 2r)
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
rLu (Hb − 2r)ρi ci m−2
− q rad,bi→bi−1,t + q cond,(ℓ−1,y,z),t ) + rLu (q cond,(ℓ,y+1,z),t + q cond,(ℓ,y−1,z),t )
r(Hb − 2r)
+ (q cond,(ℓ,y,z+1),t + q cond,(ℓ,y,z−1),t )}
m−2
133
Heat balance at component G
πr
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t + q rad,bi+1→bi,t ) + Lu × r
2
πr 2
× (q cond,(1,m−1,z),t + q cond,(2,m,z),t ) + (q cond,(1,m,z+1),t + q cond,(1,m,z−1) )
4
πr 2
= ρi × (Lu × ) × ci × (Ti,(1,m,z),t+1 − Ti,(1,m,z),t )
4
→ Ti,(1,m,z),t+1 = Ti,(1,m,z),t
4 πrLu
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
πr 2 L u ρi ci 2
+ q rad,bi+1→bi,t ) + rLu (q cond,(1,m−1,z),t + q cond,(2,m,z),t )
πr 2
+ (q cond,(1,m,z+1),t + q cond,(1,m,z−1) )}
4
Wb − 2r
Lu × × (q rad,cb,t + q rad,wb,t + q tran,gb,t + q cond,(x,m−1,z),t ) + Lu × r
ℓ−2
Wb − 2r
× (q cond,(x−1,m,z),t + q cond,(x+1,m,z),t ) + ×r
ℓ−2
× (q cond,(x,m,z+1),t + q cond,(x,m,z−1) )
Wb − 2r
= ρi × (Lu × × r) × ci × (Ti,(x,m,z),t+1 − Ti,(x,m,z),t )
ℓ−2
→ Ti,(x,m,z),t+1 = Ti,(x,m,z),t
ℓ−2 Lu (Wb − 2r)
+ { (q rad,cb,t + q rad,wb,t + q tran,gb,t
rLu (Hb − 2r)ρi ci ℓ−2
+ q cond,(x,m−1,z),t ) + rLu (q cond,(x−1,m,z),t + q cond,(x+1,m,z),t )
r(Wb − 2r)
+ (q cond,(x,m,z+1),t + q cond,(x,m,z−1) )}
ℓ−2
134
Heat balance at component I
πr
Lu × × (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t − q rad,bi→bi−1,t ) + Lu × r
2
× (q cond,(ℓ,m−1,z),t + q cond,(ℓ−1,m,z),t )
πr 2
+ (q cond,(ℓ,m,z+1),t + q cond,(ℓ,m,z−1) )
4
πr 2
= ρi × (Lu × ) × ci × (Ti,(ℓ,m,z),t+1 − Ti,(ℓ,m,z),t )
4
→ Ti,(ℓ,m,z),t+1 = Ti,(ℓ,m,z),t
4 πrLu
+ { (q rad,Hb,t + q rad,cb,t + q rad,wb,t + q tran,gb,t
πr 2 L u ρi ci 2
− q rad,bi→bi−1,t ) + rLu (q cond,(ℓ,m−1,z),t + q cond,(ℓ−1,m,z),t )
πr 2
+ (q cond,(ℓ,m,z+1),t + q cond,(ℓ,m,z−1),t )}
4
135
APPENDIX D. View-factor calculation of perpendicular plates.
A5
A3
A6
A1
A4
A2
(A1 + A2 + A3 )F(1+3+5),(2+4+6)
= A1 (F12 + F14 + F16 ) + A3 (F32 + F34 + F36 ) + A5 (F52 + F54 + F56 )
A1 F14 = A3 F32
A5 F54 = A3 F36
A1 F16 = A5 F52
1
→ A1 F14 = A3 F32 = {(A1 + A3 )F(1+3),(2+4) − A1 F12 − A3 F34 }
2
1
→ A5 F54 = A3 F36 = {(A5 + A3 )F(3+5),(4+6) − A5 F56 − A3 F34 }
2
136
(A1 + A2 + A3 )F(1+3+5),(2+4+6)
= A1 (F12 + 2F16 ) + {(A1 + A3 )F(1+3),(2+4) − A1 F12 − A3 F34 } + A3 F34
+ {(A5 + A3 )F(3+5),(4+6) − A5 F56 − A3 F34 } + A5 F56
1
A1 F16 = [(A1 + A2 + A3 )F(1+3+5),(2+4+6) − A1 F12 − A3 F34 − A5 F56
2
− {(A1 + A3 )F(1+3),(2+4) − A1 F12 − A3 F34 }
− {(A5 + A3 )F(3+5),(4+6) − A5 F56 − A3 F34 }]
1
A1 F16 = {(A1 + A2 + A3 )F(1+3+5),(2+4+6) + A3 F34 − (A1 + A3 )F(1+3),(2+4)
2
− (A5 + A3 )F(3+5),(4+6) }
137
APPENDIX E. View-factor calculation from small plate to
parallel plate with off-set.
W
H
A1
φ1
d φ2
dA2
From (2.5),
1 cos φ1 cos φ2
F12 = ∫ ∫ dA1 dA2 ⋯ (2.5)
A1 A1 A2 πr 2
Now, since dA2 is small, dA2 is constant. Also, since φ1 =φ2 when two plates are parallel, cos
φ1 =cosφ2.
138
dA2 cos 2 φ1
F12 = ∫ dA1
A1 A1 πr 2
d
Since cosφ1 = r and r = √𝑥 2 +y 2 + d2 ,
dA2 d2 1 dA2 d2 1
F12 = ∫ dA1 = ∫ dA
A1 A1 πr 4 πA1 A1 (𝑥 2 +y 2 + d2 )2 1
dA2 d2 𝐻 𝑊 1
→ F12 = ∫ ∫ dxdy
πA1 0 0 (𝑥 2 +y 2 + d2 )2
1
∫ = In
(x 2 + a2 )𝑛
1 x
In = { + (2n − 3)In−1 }
2(n − 1)a2 (x 2 + a2 )n−1
1 x
I1 = tan−1
a a
1 x 1 x 1 x
I2 = 2
{ 2 2
+ I1 } = 2 { 2 2
+ tan−1 }
2a x + a 2a x + a a a
Let a2 be y2+d2,
𝑊 W
1 1
∫ dx = ∫ du
0 {𝑥 2 +y 2 + d2 } 2
0 (x 2 +a2 )2
1 x 1 −1
x W
= [ 2{ 2 + tan }]
2a x + a2 a a 0
1 W 1 −1
W
= ( + tan )
2a2 W 2 + a2 a a
1 𝑊 1 −1
W
= ( + tan )
2(y 2 + d2 ) 𝑊 2 + y 2 + d2 √y 2 + d2 √y 2 + d2
139
Therefore,
dA2 d2 𝐻 1 𝑊 1 W
F12 = ∫ { 2 2
( 2 2 2
+ tan−1 )} dy
πA1 0 2(y + d ) 𝑊 + y + d √y 2 + d2 √y 2 + d2
H H
dA2 d2 W 1 W
= {∫ 2 2 2 2 2
dy + ∫ tan−1 dy}
2πA1 0 (y + d )(W + y + d ) 0 √y 2 + d2 (y 2 + d2 ) √y 2 + d2
𝐻
W 1 𝐻 1 1
∫ 2 2 2 2 2
𝑑𝑦 = ∫ ( 2 2
− 2 ) 𝑑𝑦
0 (y + d )(W + y + d ) 𝑊 0 y +d y + W 2 + d2
H
1 1 y 1 y
= [ tan−1 − tan−1 ]
W d d √W 2 + d2 √W 2 + d2 0
1 H 1 H
= tan−1 − tan−1
Wd d W√W 2 + d2 √W 2 + d2
Let y be d tan θ.
dy d y d
= , sin θ = , cos θ =
dθ (cos θ)2 √y 2 + d2 √y 2 + d2
H
1 W
∫ tan−1 𝑑𝑦
0 √y 2 + d2 (y 2 + d2 ) √y 2 + d2
H
tan−1
d 1 W d
=∫ tan−1 ( ) 2
dθ
0 d3 √1 + (tan θ)2 (1 + (tan θ)2 ) d√1 + (tan θ)2 (cos θ)
H
tan−1
d cos θ W cos θ
=∫ tan−1 ( ) dθ
0 d2 d
sin θ ′ cos θ
f′ = ( 2 ) = 2
d d
140
W cos θ
g = tan−1 ( )
d
H
tan−1
d cos θ W cos θ
∫ tan−1 ( ) dθ
0 d2 d
−1 H
sin θ W cos θ tan
d sin θ W cos θ ′
= 2 tan−1 ( ) −∫ {tan −1
( )} dθ
d d 0 d2 d
W cos θ
Let be t.
d
dt −W sin θ
=
dθ d
W sin θ
−1
W cos θ ′ −
d
{tan ( )} =
d W cos θ 2
1+( )
d
H
tan−1
d cos θ W cos θ
∫ tan−1 ( ) dθ
0 d2 d
−1 H W sin θ
sin θ W cos θ tan
d sin θ −
= 2 tan−1 ( ) −∫ d dθ
d d 0 d2 W cos θ 2
1+( )
d
𝐻 −1 H
𝑦 −1
W 𝑊 tan d (sin 𝜃)2
=[ tan ( )] + ∫ dθ
d2 √𝑦 2 + 𝑑2 √𝑦 2 + 𝑑 2 0 𝑑 0 d2 + (W cos 𝜃)2
𝑦2
𝐻
𝐻 W 𝑊 + 𝑑2 𝑑 𝑦2
= tan−1 ( )+ ∫ 2 2 dy
d2 √𝐻 2 + 𝑑2 √𝐻 2 + 𝑑 2 𝑑 0 2 𝑊 𝑑 𝑦 + 𝑑2
2
𝑑 + 2
𝑦 + 𝑑2
𝐻
𝐻 W 𝑦2
= tan−1 ( )+𝑊∫ 2 2 2 2 2 2 2 2
dy
d2 √𝐻 2 + 𝑑 2 √𝐻 2 + 𝑑2 0 𝑑 (𝑦 + 𝑑 ) + 𝑊 𝑑 (𝑦 + 𝑑 )
𝐻 W 𝑊 𝐻 𝑦2
= tan−1 ( )+ 2∫ dy
d2 √𝐻 2 + 𝑑 2 √𝐻 2 + 𝑑2 𝑑 0 (𝑦 2 + 𝑑 2 )(𝑦 2 + 𝑑2 + 𝑊 2 )
141
𝐻 −1
W 𝑊 𝐻 𝑦2 1 1
= tan ( )+ 2∫ ( − ) dy
d2 √𝐻 2 + 𝑑 2 √𝐻 2 + 𝑑 2 𝑑 0 𝑊 2 𝑦 2 + 𝑑2 𝑦 2 + 𝑑2 + 𝑊 2
𝐻
𝐻 W 1 𝑑2 𝑑2 + 𝑊 2
= tan−1 ( )+ ∫ {(1 − ) − (1 − )} dy
d2 √𝐻 2 + 𝑑2 √𝐻 2 + 𝑑 2 𝑊𝑑 2 0 𝑦 2 + 𝑑2 𝑦 2 + 𝑑2 + 𝑊 2
𝐻
𝐻 W 1 𝑑2 + 𝑊 2 𝑑2
= tan−1 ( )+ ∫ { − } dy
d2 √𝐻 2 + 𝑑 2 √𝐻 2 + 𝑑 2 𝑊𝑑 2 0 𝑦 2 + 𝑑2 + 𝑊 2 𝑦 2 + 𝑑2
𝐻 W
= tan−1 ( )
d2 √𝐻 2 + 𝑑2 √𝐻 2 + 𝑑2
𝐻 𝐻
1 2 2)
1 2
1
+ 2
{(𝑑 + 𝑊 ∫ 2 𝑑𝑦 − 𝑑 ∫ 2 2
dy}
𝑊𝑑 0 𝑦 2 + (√𝑑 2 + 𝑊 2 ) 0 𝑦 +𝑑
𝐻 W
= tan−1 ( )
d2 √𝐻 2 + 𝑑 2 √𝐻 2 + 𝑑 2
1 1 𝐻 1 𝐻
+ 2
{(𝑑2 + 𝑊 2 ) ( tan−1 ) − 𝑑 2 ( tan−1 )}
𝑊𝑑 2
√𝑑 + 𝑊 2 2
√𝑑 + 𝑊 2 𝑑 𝑑
𝐻 W √𝑑 2 + 𝑊 2 𝐻 1 𝐻
= tan−1 ( )+ 2
tan−1 − tan−1
d2 √𝐻 2 + 𝑑 2 √𝐻 2 + 𝑑 2 𝑊𝑑 √𝑑 2 + 𝑊 2 𝑊𝑑 𝑑
Hence,
dA2 d2 1 H 1 H
F12 = [ tan−1 − tan−1
2πA1 Wd d W√W 2 + d2 √W 2 + d2
H W √d2 + W 2 H
+ tan−1 ( )+ 2
tan−1
d2 √H 2 + d2 √H 2 + d2 Wd √d2 + W 2
1 H
− tan−1 ]
Wd d
dA2 d2 W H H W
= [ tan−1 + tan−1 ( )]
2πA1 d2 √W 2 + d2 √W 2 + d2 d2 √H 2 + d2 √H 2 + d2
dA2 1 H 1 W
= { tan−1 + tan−1 ( )}
2π 𝐻√W 2 + d2 √W 2 + d2 W√H 2 + d2 √H 2 + d2
142
𝐴1 1 H H W W
𝐹21 = 𝐹12 = { tan−1 + tan−1 ( )}
𝑑𝐴2 2π √W 2 + d2 √W 2 + d2 √H 2 + d2 √H 2 + d2
W
A5 H
A3 A6
dA2
A4
y0
d
x0
Figure E-2. View factor between parallel plates with off set.
dA2 1 y0 1 x0
F12 = { tan−1 + tan−1 ( )}
2π y0 √x0 2 + d2 √x0 2 + d2 x0 √y0 2 + d2 √y0 2 + d2
dA2 1 H − y0
+ { tan−1
2π (H − y0 )√x0 2 + d2 √x0 2 + d2
1 x0
+ tan−1 ( )}
x0 √(H − y0 )2 + d2 √(H − y0 )2 + d2
143
dA2 1 H − y0
+ { tan−1
2π (H − y0 )√(W − x0 )2 + d2 √(W − x0 )2 + d2
1 W − x0
+ tan−1 ( )}
(W − x0 )√(H − y0 )2 + d2 √(H − y0 )2 + d2
dA2 1 y0
= { tan−1
2π y0 √(W − x0 )2 + d2 √(W − x0 )2 + d2
1 W − x0
+ tan−1 ( )}
(W − x0 )√y0 2 + d2 √y0 2 + d2
A2
W
H
A4 A5
A3
A6
A1
Figure E-3. View factor between parallel plates without off set.
1 H H W W
𝐹12 = { tan−1 + tan−1 ( )}
2π √W 2 + d2 √W 2 + d2 √H 2 + d2 √H 2 + d2
144
BIBLIOGRAPHY
[1] Y. Wada, “On the Minimum Specific Heat Consumption of the ReheatingFurnace in
the Iron and Steeel Industry and a Calculation of It,” Tetsu to Hagane, vol. 64, no. 13,
pp. 1988-1995, 1978.
[2] Y. Takamura, "Suggestion of Effective Energy Using for Heating Furnace," Denki
Seikou, vol. 84, no. 2, pp. 79-88, 2013.
[3] A. Jaklic, F. Vode and T. Kolenko, "Online simulation model of the slab-reheating
process in a pusher-type furnace," Applied Thermal Engineering, vol. 27, pp. 1105-
1114, 2007.
[5] S. H. Han and D. Chang, "Optimum residence time analysis for a walking beam type
reheating furnace," International Journal of Heat and Mass Transfer, vol. 55, pp. 4079-
4087, 2012.
[6] S. K. Dubey and P. Srinivasan, "Steel billet reheat simulation with growth of oxide
layer and investigation on zone temperature sensitivity," Journal of Mechanical Science
and Technology, vol. 28, no. 3, pp. 1113-1124, 2014.
[7] A. Jaklic, B. Glogovac, T. Kolenko and B. Zupancic, "A simulation of heat transfer
during billet transport," Applied Thermal Engineering, vol. 22, pp. 873-883, 2002.
[8] D. Wild, T. Meurer and A. Kugi, “Modeling and experimental model validation for a
pusher-type reheating furnace,” Mathmatical and Computer Modeling of Dynamical
Systems, vol. 15, no. 3, pp. 209-232, 2009.
145
[9] K. Watanabe, “A Mathematical Model of Continuous Billet Heating Furnace,” Tetsu to
Hagane, vol. 62, no. 10, pp. 1329-1338, 1976.
[10] N. Yoshitani, Y. Naganuma and T. Yanai, "Optimal slab heating control for reheating
furnaces," in American Control Conference, IEEE, Boston, 1991.
[11] N. Yoshitani, T. Ueyama and M. Usui, "Optimal Slab Heating Control with
Temperature Trajectory Optimization," IEEE, pp. 1567-1572, 1994.
[13] Y. Y. Yang and Y. Z. Lu, "Dynamic Model Based Optimization Control for Reheating
Furnaces," Computers in Industry, vol. 10, pp. 11-20, 1988.
[14] Division and Committee of high temperature strength (Koonkyodo bumon iinkai),
“Results of collecting documents about physical property in metals (Kinzoku zairyou
no butsurijosuu ni kansuru siryou syuusyuukekka ni tsuite),” Materials (Zairyou), vol.
17, no. 174, pp. 264-268, 1968.
[15] T. Aihara, Heat transfer engineering (Dennetsu kogaku), Tokyo: Tatsuji Yoshino, 1994.
[16] J. R. Singham, "Tables of emissivity of surfaces," Int. J. Heat Mass Transfer, vol. 5,
pp. 67-76, 1962.
[17] H. C. Hottel, Heat Transmission (3rd ed.), McAdams, W. H. , Eds., New York,
Toronto, London: McGRAW-HILL BOOK COMPANY, INC., 1954.
[18] P. Mullinger and B. Jenkins, Industrial and Process Furnaces (2nd ed.), Oxford:
Elsevier Ltd., 2014.
146
[20] A. Fujibayashi, M. Iguchi, T. Ishii and M. Nakaseko, “Heat Transfer Analysis with the
Latent Heat of Phase Transformation — Heat Transfer Measurement of Steel Plate
Cooling —,” Jikken Rikigaku, vol. 11, no. 4, pp. 367-370, 2011.
[21] A. Steinboeck, D. Wild, T. Kiefer and A. Kugi, "A fast simulation method for 1D heat
conduction," Mathematics and Computers in Simulation, vol. 82, no. 3, pp. 392-403,
2011.
[23] F. Asano and H. Kawaguchi, “Direct Rolling Technology of Steel Bars,” KOBE STEEL
ENGINEERING REPORTS, vol. 58, no. 2, pp. 2-6, 2008.
[24] J. Yanagimoto, "Numerical Analysis for the Prediction of Microstructure after Hot
Forming of Structural Metals," Materials Transactions, vol. 50, no. 7, pp. 1620-1625,
2009.
[25] Y. J. Jang and S. W. Kim, "An Estimation of a Billet Temperature during Reheating
Furnace Operation," International Journal of Control, Automation, and Systems, vol. 5,
no. 1, pp. 43-50, 2007.
147