AbstractSteelmaking and continuous casting is a difficult

process that is manually scheduled due to its overwhelming

number of objectives and restriction factors. This paper
provides a new method of two-stage optimized scheduling for
steelmaking and continuous casting. In the first stage, a
preliminary schedule is established based on the same usage
rate of parallel machines and minimized time conflict. In the
second stage, a linear programming model of elimination time
conflict is established. By solving the two mathematical models,
an operable schedule is produced. The method of two-stage
optimized scheduling has been embedded a scheduling system
which is also developed and successfully applied to a large steel
plant in China. The industrial application has demonstrated
that the proposed method cannot only meet the requirement of
practical production, but also has a wide application in
industry.
I. INTRODUCTION
HE steelmaking and continuous casting (SCC) production
process belongs to hybrid flow shop (HFS). Hybrid Flow
Shop Scheduling (HFSS) is a difficult NP problem [1]. There
are generally two methods that deal with HFSS problems,
namely the classical optimization method and the intelligent
optimization method. The classical optimization methods
include linear programming, integer programming, dynamic
programming, branch and bound method and Lagrangian
Relaxation, etc. The intelligent optimization method uses
expert system, neural network and genetic algorithm etc.
Since the linear programming was first applied to the cost
accounting of steel production by Fabian [2], classical
optimization approach has been applied to solve the
scheduling problem in the 1950s, and this approach was

Manuscript received September 12, 2009. This work was supported by the
National Basic Research Programof Programof China (2009CB320600), the
State Key Programof National Natural Science of China Grant (60534010),
the 111 project (B08015), the National High-tech Program(2007AA041405),
the Funds for Creative Research Groups of China Grant (60521003), and the
Programfor New Century Excellent Talents in University (NCET-05-0294)
in China.
X. Y. Wang is with the Research Center of Automation, Northeastern
University, Shenyang 110006, Liaoning, China, and with the College of
Information and Technology, Qingdao University of Science and
Technology, Qingdao 266042, Shandong, China(bywxy@126.com)
T. Y. Chai is with the Research Center of Automation, Northeastern
University, Shenyang 110006, Liaoning, China(Tychai@mail. neu. edu.cn)
B. L. Zheng is with the Research Center of Automation, Northeastern
University, Shenyang 110006, Liaoning, China(Zblbing@21cn.com)
H. Wang is with the Control SystemCenter, University of Manchester,
Manchester, M601QD, UK and also with Northeastern University, Shenyang
P R China.
further developed in 1960s [3-5]. From the late 1960s,
intensive study of complicated scheduling theory has been
carried out. The rule-based heuristic algorithm became the
main method to solve scheduling problems [6-9], and it was
further improved in the 1970s [10]. In the 1980s, artificial
intelligence became the mainstream of solving scheduling
problems. In the late1980s, large international iron and steel
factories began to utilize computer and artificial intelligence
technology to develop the production control system. For
example, IBM cooperated with J apans Kokan Iron and Steel
Factory and developed the cooperative scheduling system,
which is called Schplan [11]. Korean Gwangyang iron and
steel factory developed the process adjusting and scheduling
system, HIPASS[12]. These systems can automatically
collected important data, optimize each processs time and
routing, and form production schedule, thus ensuring the
rationalization and integration of production management. In
the 1990s, computational intelligence was adopted in the
scheduling theory. Modified genetic algorithm was used to
study the sequencing problem of flow shop.[13] Yu Hai-bin
[14] used neural network to study SCC scheduling problems,
and proposed to solve job-shop scheduling problems with
restricts of starting and completing time. In this approach,
each constraint was described by a restricted neural network,
accordingly adjusted the starting time unqualified with
restricts, leading to a feasible scheduling plan.
The scheduling methods have been continually developed
with the further study for production scheduling. Currently,
the question of how many of these scheduling methods can be
adopted in practical production process was raised in [15].
From the end of 20th century, the efficient scheduling
methods have been studied that could be applied to practical
production processes, where these methods have been
embedded into the scheduling system to reduce the waiting
time, the production cost and increase the competition[16, 17].
In the paper [18, 19], SCC schedule was abstracted as a 3-step
HFS problem, and the scheduling mathematical model was
established and solved by Lagrangian Relaxation and B&B
algorithm [20], Tang Li-xin [21] established SCC scheduling
model based on just-in-time principle, and converted it to a
linear programming model according to the real meaning of
variables. Then the standard linear programming software
package was used to solve the model, and applied to the
production of Shanghai Baosteel.
Two-stage Optimized Scheduling Method and Application for
Steelmaking and Continuous Casting
Xiuying Wang, Tianyou Chai, Fellow, IEEE, Binglin Zheng, Hong Wang
T
2010 American Control Conference
Marriott Waterfront, Baltimore, MD, USA
June 30-July 02, 2010
FrB18.4
978-1-4244-7425-7/10/$26.00 2010 AACC 6109

Since all computational intelligent methods need to be
given an initial solution, such methods can hardly be applied
to practical production processes. This is because different
initial solution would produce different solution speed and
the result, which would influence the stability of the system.
Most papers still use classical optimization methods to solve
practical scheduling problems. Paper [22] had established
intelligent scheduling system (ISS) according to large steel
enterprises need for the scheduling software, where the
architecture and function of the system were introduced in the
paper.
This paper is organized in the following way. The first
section summarizes the relevant research for hybrid flow
shop scheduling method. In Section 2, the SCC scheduling
problem and scheduling strategy of intelligent scheduling
system are described. In section 3, a two-stage optimized
schedule method is given, and finally an application example
is presented in this paper.
II. DESCRIPTION OF SCHEDULING PROBLEM
A. SCC Production Process
SCC Production Process is illustrated in Fig.1. First, raw
materials such as iron ore and dolomite are put into the blast
furnace (BF) and melted, and the liquid iron as the output
from BF is transported to the steelmaking shop where
converters are located. Converters refine liquid iron into
molten steel. The melted steel is poured into ladles that are
transported by crane and trolley to a refining furnace. Special
treatment may be performed in the Ladle-refining equipment
to eliminate impurities from either molten steel or add alloy
ingredients to the molten steel to make high-grads steel.
Finally, casters cast molten steel continuously into billets, and
the billets are rolled into steel products needed by customers.

B. Complexity of SCC Production Process
During the entire iron and steel production process, the
products are being processed and handled at high temperature
with physical and chemical reactions. The production process
is a hybrid process, which has both characteristics of
continual and the discrete phases. The production scheduling
of continual processing phase is relatively easy due to its line
flow, but the discrete processing phase is a multi-stage,
multi-equipment, parallel, asynchronous process. Thus, the
selection of multi-equipment paths is a key during the process
(proven to be the NP-hard problem). Similar to discrete
processing phase, SCC also needs to select among
multi-equipment paths since the production process involves
multiple converters, refining furnaces and continuous casters.
The quality of steel bloom is closely related with the
temperature and ingredient of molten steel arriving at caster,
therefore long wait time between operations is not allowed.
Otherwise, the molten steel has to be reheated which causes
high energy consumption and low quality of steel bloom. As a
result, it is extremely important to control the flow time of
charges. The molten steel must be delivered to caster on time
with prearranged ingredient standard and ideal temperature.
Therefore, production scheduling in hybrid type phase has
been recognized as one of the most difficult industrial
scheduling problems.
Bar Steel
Plate Steel
Seamless Pipe
Cold Rolling
Hot Rolled Coil
Wire Rolling

BF: blast furnace; CF: converter furnace; RF: refining furnace; CC: continuous caster; RM: rolling mill
Fig.1 The SCC production process
Liquid
iron
RH
CAS/KIP
LF
Molten
steel
High-grade
molten steel
High-grade
Moltensteel
High-grade
molten steel
Molten
steel
Molten
steel
Slab
Slab
Slab
Rolling Mill
Rolling Mill
Rolling Mill
CC
CC
CC
CF
CF
CF
BF
Iron Ore
Sintering Ore
Limestone
6110

C. Scheduling Strategy
In order to solve the problems mentioned above, an
intelligent scheduling system (ISS) had been developed [22].
The proposed scheduling strategy of SCC production process
is illustrated in Fig.2. Firstly, according to the cast planning,
optimized goals and constraints, two schedules are generated
by expert system and optimization model method separately.
The better scheme is selected by evaluation criteria
established by the manufacturing process. When schedule
cannot be carried out properly due to the incident disturbing,
the schedule is adjusted by case-based reasoning (CBR)
approach.
1) Optimization Method
SCC production schedule can be described as: based on the
given cast plan M, each cast contains N charges, assign these
N charges to several stations of the J processes, under the
condition of correct joint and casting continuously on the
same cast, determine the job schedule of each machine, and
minimize the objective function (the total flow time, waiting
time between processes).
From the definition of SCC production schedule, it can be
seen that there are two questions needed to be resolved:
(1) Machine selected:
Because each process contains not only one machine, this
requires to determine the charges converter furnace first,
refining furnace second, and continuous caster last. This
means that it needs to select an optimal production path for
the charges. Suppose that one charge needs to pass three
production processes (converter furnace, refining furnace and
continuous caster), each process contains three identical
machines, so there are M1N production paths as shown in
Fig.3. This is an optimization problem to choose which path,
which is needed to be resolved.
(2) Conflict Elimination:
When selecting the production path of the charges, with the
given condition (charges job order and starting casting time
in continuous caster, the processing time on each machine,
and the transportation time between machines), we can
backward infer the charges starting time and completing
time corresponding to the machine in the process of
continuous caster, refining furnace and converter furnace,
and produce the job schedule of charges. However, the two
charges (i and q) in different continuous caster may be
assigned to the same refining machine (LF furnace), and after
backward inferring to the refining machine (LF furnace)
according to the starting casting time of each continuous
caster, the conflict problem between job time will happen. As
such, this is not a feasible scheduling plan. We need to
eliminate this conflict, and produce a schedule that there are
not job time conflicts on any machines.



2) Expert System
The scheduling expert system consists of a knowledge base,
a reasoning machine, a database, the man-machine interface,
a knowledge gained part and an interpreter. A great lot facts
and the rules that can be used to solve scheduling problems
are stored in the knowledge base. The function of reasoning
machine is to select facts and rules related to users problem
from knowledge base and carry out reasoning according to
the certain strategy. Expert system interacts with users
through the man-machine interface. On one hand it accepts
user's information and translates them into the internal form
which the system can process. On the other hand the useful
knowledge generated by reasoning machine from the
knowledge base are sent back to users. The interpreter is that
the scheduling results and proposed problems in scheduling
process are explained rationally by using knowledge from the
knowledge base and facts from the database. The working
principle of expert system is the charge plan etc, so that
information is sent to the reasoning machine by man-machine
interface and the inference machine according to scheduling.
Then the reasoning machine searches mutuality knowledge
from the knowledge base and each fact from the database and
carries out matching and the reasoning. Finally, the reasoning
results (namely the scheduling result) are returned to the users
through the man-machine interface.
3) Evaluation
CF-1
CF-2
CF-M
1

RF-1
RF-2
RF-M
2

CC-1
CC-2
CC-M
m

Fig.3 Workflow of steelmaking and continuous casting process
Optimized Goals
Evaluate
SCC
Production
Process
Case-Based Reasoning
Fig.2 Scheduling Strategy SCC Production Process
Expert
System
Cast Plan
Constraints
Scheduling
Orders
Optimized
Method
Schedule
6111

Because there are two different schedules in the static
scheduling subsystem in terms of how excellent scheme can
be selected in actual production, it is necessary that
reasonable evaluation scheme is set up. The three evaluation
indices are established in system. (1) The total charges
waiting time is minimized for the concerned equipment; (2)
the variance of equipment load is minimized; (3) the waiting
time variance of the charge on the equipment is minimized.
The evaluation functions of three indexes are established, the
better schedule is selected by the fuzzy evaluation method
and the simulation effect. This schedule will be sent to SCC
production process by the scheduling instructions.
4) Case-Based Reasoning
Case-Based Reasoning (CBR) is a scheme with which
production incidents can be solved and then stored in the
case-memory, where the schedulers can retrieve and reuse
them to solve similar case in SCC production process at any
time. SCC production process is complex and shop floor
events are various. To speed up the adjustment, we classified
the previous cases into seven categories such as equipment,
ingredient, and temperature and so on. When shop floor event
occur, the category of the event is determined. Category
descriptors and current status information are input into
computer by scheduler, then the scheme of similar case may
be found in case memory accordingly, Adjustment for new
problem is generated either by the scheme found in case
memory or by adapting an original scheme. If the scheme of
similar case has not been found in case memory, an
adjustment has to be given by scheduler. Every adjustment is
sent to the Gantt simulation editor. If the scheduler satisfies
the simulating result of the scheme and then the adjustment is
sent to SCC production spot. Otherwise the scheme must be
modified and sent to the Gantt simulation editor again until
the result is satisfactory. The successful scheme is stored in
the case memory for future use.
III. TWO-STAGE SCHEDULING METHOD
The scheduling process is divided into two-stages. In the
first stage, a mathematical model of machine assignment is
developed, and the charges are assigned to machines
heuristically to produce a preliminary time table. In the
second stage, other mathematical model of conflict
management is developed and solved by standard linear
programming, with the goal of minimizing cost and ensuring
punctual start of casting.
A. The Mathematical Model of Machine Assignment
1 Notations
Tijk

the processing time of plan i on machine k of
process j;
x
ijk

the starting processing time of plan i on machine
k of process j;

jk
the set of all charges on machine k of process j;
j
q
the set of all machines used for process j in plan
q;
) , , ( k j i
q
f the time conflict function of charge i and q on
machine k of process j;
) , , ( k j i
q
F the set of all charges i conflicting with charge q
on machine k of process j;
*) , , ( k j i
q
F the minimum of all charges i conflicting with
charge q on machine k* of process j;
) , ( k j G the load function on machine k of process j;
2Machine Assignment Model
(a) Machine conflicts function
Assume that
ijk qjk T T
=
, then we can obtain
( , , ) ( , , )
jk
i
F i j k f i j k
q q

(1)
where
,
( , , )
0
ijk ijk qjk ijk qjk ijk
If
x x x x T T
f i j k
q
else
+
=

(2)
(b) Machine load function
( , )
jk
ijk
G j k
T
i
=

(3)
3) Steps of Solution
(a) Basic assumption
The following assumption must be considered before the
schedule is constructed.
The sequences of charges are fixed on casters.
The charges casting continuously on the same caster are
considered as a cast.
There is one cast joining scheduling in each caster at most.
Assuming that the starting casting time in each caster, the
processing time on machines, and the transportation time
between machines were given.
(b) Steps of solution
Step1: From the starting casting time, select the refining
machine according to the refining mode of charges, compute
the conflict function on available refining machines, and
determine the machine with the least conflicts by comparison.
( , , *) min ( , , )
j
q q
k q
F i j k F i j k

=
(4)
Step2: When the least conflict time of two or more
available machines is the same, compute the load value of
available machines and find the machine with the least load
value.
( , ) min
jk
ijk
i
G j k
T

(5)
Step3: When the least load value of two or more machines
is the same, choose the machine with the minimal number,
A preliminary schedule can thus be obtained after above
steps. Since usually there will have time conflicts on some
machines, we also need to eliminate these conflicts in order to
make this preliminary schedule feasible.
B. The Mathematical Model of Machine Conflicts
Eliminated
1Notations
n
N
cast serial number;
cast number n=1,2,N;
6112

i

j
m
n

0 n

S
Si

cc
S

j
M

C1
C2

C3

C4

i
m j jm
Tr
,


ijm
T


Jn


ijm
X

i
jm i
X


serial number of charge;
serial number of process;
serial number of machine;
the set of all charges in the nth cast;
the set of the first charge in the nth cast;
the set of process;
the set of the ith charge pass process
the set of all casters
the set of all machines in jth process;
coefficient of cast break lost penalty for cast k;
coefficient of penalty cost for the waiting time of
chargei after finishing process on machine j;
coefficient of penalty cost for the first charge in the
kth cast starting casting before the time specified;
coefficient of penalty cost for the first charge in the
kth cast starting casting after the time specified;
transportation time from machine m to machine
m in j process;
processing time of charge i on the machine m in jth
process;
specified starting casting time of the first charge in
the nth cast;
starting time of charge i on machine m of jth
process;
starting time of charge i that the immediate charge
i processed on same process j and machine m;

2) Production constraints
For the two contiguous operations for the same charge,
only when the preceding operation has been completed the
next one can be started.
For two contiguous charges processed on the same
machine, only when the preceding charge has been
completed, then the next one can be started.
3) Optimized objective
It is important that the following are achieved.
Wait time is as short as possible form operation to
operation.
Cast break loss penalty is minimum.
Loss penalty that the first charge in each cast cannot be
cast on time is minimum.
4) Decision variable
Use the starting processing time of charge i on machine k
of process j as the decision variable of conflict elimination
linear programming model, and note as
ijk
X , then the
optimized objective function is as follows:

) 4 3 (
) ( 2
) ( ( 1 minZ
, , , 1
1 ,
1 , , ,
1
1
, ,
) 1
1
0
ijm
N
Mj m S j i n
jm
i
m j jm
ijm
Mj m j m S j n i
ijm m i,j
N
n
M m
cc
S j n i
ijm ,jm jm i
N
n
Z C Yi C
Tr T X X C
T Xi X C
cc n
j

=
+
+
+
=

+
=
+ +

+ =
(6)
S.T.
j
ijm ijm
i
jm i
M m Si j n i i T X X

, , , (7)
1
) 1 ( ,
) 1 (
, , ,
+
+
+

+
j j
i
m j jm
ijm ijm m j i
M m M m Si j n i
Tr T X X
(8)
j
jm ijm ijm ijm
S m
cc
S j
n
i
J Z Y X

= +
, ,
0
0
(9)
IV. APPLICATION
A. Introduction of the NO. 1 steelmaking factory
The No. 1 steelmaking factory is a branch of Shanghai
Baoshan Iron and Steelmaking Corporation (Baosteel in short)
that is the largest iron and steel enterprise in China. Baosteel
started to use digital machines and computer systems in a
large scale from the beginning of 1970s when it was founded.
In the middle of 1980s, when the second construction began,
the basic automatic control system (L1); process control
system (L2); production control computer system (L3);
enterprise management system (L4) were imported. However,
in the No. 1 steelmaking factory, the schedule making and
adjusting is still done manually which is not suited for the
realization of CIMS in the factory. Therefore during the third
construction phase, a strategy has been proposed, which is to
embed a model of steelmaking and continuous casting in the
L3 system so as to replace the manual adjustment. Therefore
we developed the intelligent scheduling system on the
requests of Baosteel.
The NO.1 steelmaking factory has majority machines with
the twelve charges to be processed. The initial parameters for
these jobs are given in table I. The route of charges assigned
based on the assignment scheduling model in first phase and
time parameters are presented table II. As the complicated
production flow in China, the system has three converters;
seven refining furnaces which including three RHs, two
CASs, one KIP and one LF and three continuous casters.

Fig4. Three cast plans
6113

B. An Example
Fig.4 shows a screen shot of three casts plan table on
12/10/2005. There are twenty charges need to be arranged,
including seven charges in No.1caster, six charges in No.2
caster and seven charges in No.3 caster.
Fig.5 shows a screen shot of the schedule generated by
the two-stage optimized scheduling method based on the plan
(Fig.4), the horizontal axis represent time; the vertical axis
represent the equipment; the thick line stand for the process
time of each charge on each equipment; the thin diagonal line
represent transportation time of the charge form machine to
the next machine; the vertical broken line is current time of
computer, and the charges cast on same caster is represented
by same color.

C. Application Effect
The application result has indicated that: (1) the time of
constructing a schedule has been significantly decreased; a
schedule with 20 charges is constructed in 30 seconds by the
ISS. (2) the times of on-the-spot adjustment has been reduced.
The number of average adjustment per day reduces to 1/10;
the hit probability of time, ingredient and temperature (are
called three hit probability) from original 54% has been
enhanced to 65.7%. (3) Daily charges have been increased
from 53 to 58, and daily production has been increased by
7%.
V. CONCLUSION
Whether steelmaking, refining and continuous casting can
be well coordinated in production is an important criteria to
evaluate the production and management level of an iron and
steel enterprise. SCC production scheduling is one of the core
technologies in coordinating steelmaking, refining and
continuous casting, and receives much attention from large
steel factories. As a result, it is important to develop an
effective scheduling system to minimize the waste of
manpower and resources. In this paper we have established a
novel two-stage mathematical model which is suitable for
general steel factories scheduling plan. This model is verified
by the production data and the scheduling simulation
platform of a large steel factory. The result shows that this
scheduling method can be successfully applied in the
industry.
VI. REFERENCES
[1] R. Linn and W. Zhang. Hybrid flow shop scheduling: a survey,
Comput Ind Eng, vol. 37, no. 1, pp. 5761, 1999.
[2] T. Fabian. A linear programming model of integrated iron and steel
production, Management Sci. vol. 4, pp. 415449, 1958.
[3] A. S. Manne. On the J ob-Shop Scheduling Problem, Operations
Research, Vol. 8, No. 2, pp. 219-223, 1960.
[4] B. Giffler and G. L. Thompson, Algorithms for Solving Production
Scheduling Problems,Operations Research,Vol. 8, No. 4, pp. 487-503,
1960.
[5] D. S. Palmer. Sequencing jobs through a multi-stage process in the
minimumtotal time-a quick method of obtaining a near optimum,
Operations Research Quarterly, vol. 16, pp. 101-107, 1965.
[6] S. Eilon and R. M. Hodgeson, job Shop Scheduling with Due Dates,
Int. J. Prod. Res, vol. 6, no. 1, pp. 1-13,1967.
[7] S. Eilon and D. J . Cotterill, A Modified SI Rule in J ob Shop
Scheduling, Int. J. Prod. Res. vol. 7, no. 2, pp. 135-145, 1968.
[8] R. H. Hillier, A Simulation Study of Sequencing in Batch Production,
Operations Research. Vol. 19, pp. 389-407, 1968.
[9] C. L. Moodie and S. D. Roberts, Experiments with Priority
Dispatching Rules in a Parallel Processor Shop, Int. J. Prod. Res. vol.
6, no. 4, pp. 303-312, 1968.
[10] S. C. Aggahwai, F . P. Wayman, and B. A. McCARL, An Investigation
of a Cost-based Rule for J ob-shop Scheduling, Int. J. Prod. Res. vol.
11, no. 3, pp. 247-261, 1973.
[11] M. Numao and S. Morishita. Cooperative scheduling and its
application to steelmaking processes, IEEE Transactions on Industrial
Electronics, Electronics. vol. 38, no. 2, pp. 150-155, Apr 1991.
[12] K. R Ryu, O. K. Limand H. K. Lee, The Advanced Production and
Product Scheduling Control Systemat the Kwangyang Works, Proc,
CPC-93 Seoul, Korea, pp. 299-308, 1993.
[13] L. X. Tang and J . Y. Liu, A modified genetic algorithmfor solve flow
shop sequencing problemto minimize mean flow time,.Journal of
Intelligent Manufacturing, vol. 13, no. 1, pp. 61-67, 2002.
[14] H. B. Yu, J . S. Xue and H. B. Wang, A neural network-based approach
to production scheduling, Acta Automatica Sinica, vol. 25, no. 4, pp.
450-456, 1999.
[15] K. N. McKay and V. C. S. Wiers, Unifying the theory and practice of
production scheduling J ournal of Manufacturing Systems, Academic
Research Library, vol. 18, no. 4, pp. 241-255, 1999.
[16] H. L. Pang and Q. Wang and B. L. Zheng, Dist ributive Production
Supervision and Scheduling System for Steel Making Process,
Journal of Northeastern University (Natural Science), vol. 20, no. 6,
pp. 580-582,1999.
[17] Z. Chen and L. Q. Fan, Application of Productive Scheduling System
in Steelmaking-Continuous Casting Process, China Metallurgy, vol.
16, no. 9, pp. 39-44, 2006.
[18] L. X. Tang, P. B. Luh and J . Y. Liu, Steel-making process scheduling
using Lagrangian relaxation, International Journal of Production
Research, vol. 40, no. 1, pp. 55-70, 2002.
[19] H. Xuan and L. X. Tang, Scheduling a hybrid flowshop with batch
production at the last stage, Computer and Operation Research, vol.
34, pp. 2718-2733, 2007.
[20] S. Brah and J . L. Hunsucker, Branch and bound algorithmfor the flow
shop with multiple processors, Eur J Oper Res, vol. 51, no.1, pp.
8899,1991.
[21] L. X. Tang, J .Y. Liu, A.Y. Rong and Z. Yang, A mathematical
programming model for scheduling steelmaking-continuous casting
production, European Journal of Operational Research, vol. 120, pp.
423-435, 2000.
[22] X Y. Wang, T. Y. Chai and B. L. Zheng Intelligent scheduling
software and its application in steeimaking and continuous casting,
Computer Integrated Manufacturing Systems, vol 12, pp1220-1226,
2006.


Fig.5 Gantt chart of schedule
6114