Flow Shop Scheduling

Production Scheduling

P.C. Chang, IEM, YZU.

Definitions
Contains m different machines.
Each job consists m operators in different
machine.
The flow of work is unidirectional.
Machines in a flow shop = 1,2,.,m
The operations of job i , (i,1) (i,2) (i ,3)..(i, m)
Not processed by machine k , P( i , k) = 0

Production Scheduling

P.C. Chang, IEM, YZU.

Flow Shop Scheduling

Baker p.136

The processing sequence on each machine are all the same.

2 3 1 5 4

1
2..

2 3 1 5 4 Flow shop

1 3 2 4 5 Job shop

...

n! - flow shop permutation schedule

n!.n! .n! - Job shop
(n!) m
(n!)
or k
m

Production Scheduling

k : constraint
( routing problem)

P.C. Chang, IEM, YZU.

Workflow in a flow shop

Type 1.

Baker p.137

Input
Machine
1

Machine
2

Machine
3

Machine
M-1

Machine
M

output

Type 2.
Input

Input

Input

Machine
1

Machine
2

Machine
3

output

output

output

Production Scheduling

Input

Input

Machine
M-1

Machine
M

output

output

P.C. Chang, IEM, YZU.

Johnsons Rule

Baker p.142

Step1 : Find mini {t i1 ,ti2 }

Step 2a : If the min t requires machine 1, place the job in the first available position
in sequence. go to step3.
Step 2b : If the min t requires machine 2, place the job in the first available position
in sequence. go to step3.
Step3 : Remove the assigned job from consideration and return to step 1 until all
positions in sequence are filled.

Note:
Johnsons rule can find an optimum with two machines
Flow shop problem for makespan problem.

Production Scheduling

P.C. Chang, IEM, YZU.

Ex.
j
tj1
tj2

1
3
6

2
5
2

3
1
2

4
6
6

5
7
5

Stage

Min tjk

Assignment

Partial Schedule

1,2,3,4,5

t31

3=[1]

3xxxx

1,2,4,5

t22

2=[5]

3xxx2

1,4,5

t11

1=[2]

31xx2

4,5

t52

5=[4]

31x52

t11

4=[3]

31452

Production Scheduling

P.C. Chang, IEM, YZU.

Ex.

M1 3
M2

1
3

4
1

5
4

2
5

2
24

The makespan is 24

Production Scheduling

P.C. Chang, IEM, YZU.

The B&B for Makespan Problem

The Ignall-Schrage Algorithm

(Baker p.149)

- A lower bound on the makespan associated with any

completion of the corresponding partial sequence is
obtained by considering the work remaining on each
machine. To illustrate the procedure for m=3.
For a given partial sequence , let
q1= the latest completion time on machine 1 among jobs in .
q2= the latest completion time on machine 2 among jobs in .
q3= the latest completion time on machine 3 among jobs in .
The amount of processing yet required of machine 1 is t j1
j '

Production Scheduling

P.C. Chang, IEM, YZU.

The Ignall-Schrage Algorithm

In the most favorable situation, the last job
1) Encounters no delay between the completion of one operation
and the start of its direct successor, and
2) Has the minimal sum (tj2+tj3) among jobs j belongs to
{t j2 t j3 }
Hence one lower bound on the makespan is b1 q1 j' t j1 min
j '
{t j3 }
A second lower bound on machine 2 is b 2 q 2 j' t j2 min
j '

A lower bound on machine 3 is b 3 q 3 j' t j3

The lower bound proposed by Ignall and Schrage is B max{b1 , b 2 , b 3 }

Production Scheduling

P.C. Chang, IEM, YZU.

The Ignall-Schrage Algorithm

Job in
M1
q1
M2

tk2

.
.

q2

M3

tk1

.
.

q3
M2

.
.

q2
M3
q3
Production Scheduling

tk3

.
.

.
.
10

b1

tk2
tk3
b2
P.C. Chang, IEM, YZU.

Ex. B&B
m=3
j

tj1

11

10

tj2

12

tj3 10

13

1 5

6 min 9 13
12 2

Production Scheduling

For the first node: =1

q1 t 11 3
q 2 t11 t12 7
q 3 t11 t12 t 13 17
The lower bound
b1 3 28 6 37
b 2 7 22 2 31
b 3 17 20 37
B max(37,31,37) 37

11

P.C. Chang, IEM, YZU.

Ex.

2
1

Partial
Sequence

( q1 , q2 , q3 )

(b1,b2,b3)

1xxx

( 3 , 7 , 17 )

( 37 , 31 ,
37 )

37

2xxx

( 11 , 12 , 17 )

( 45 , 39 ,
42 )

45

3xxx

( 7 , 16 , 29 )

( 37 , 35 ,
46 )

46

4xxx

( 10 , 22 , 24 )

( 37 , 41 ,
52 )

52

12xx

( 14 , 15 , 22 )

( 45 , 38 ,
37 )

45

13xx

( 10 , 19 , 32 b) q (
37, 34
t j1, min{39
t j2 t j3 }
1
1
39
j '
j)'

14
14xx
2

9 13
( 37 , 40 ,
45
14 45
(7 ) 10) min

15
12

1
2
132x
( 45 , 36 ,
45
17 ( 21
22 , 22 , 37 ) 45
39 )

Production Scheduling

( 13 , 25 , 27 )

12

P.C. Chang, IEM, YZU.

Ex. B&B
P0

1xxx

2xxx

B=
37

12xx

13xx

B=
45

B=
45

4xxx
B=
46

B=
52

14xx
B=
39

132x

3xxx

B=
45

134x
B=
45

Production Scheduling

B=
39

13

P.C. Chang, IEM, YZU.

Refined Bounds
The use of q2 in the calculation of b2 ignores the possibility that
the starting time of job j on the machine 2 may be constrained
by commitments on machine1. Hence:
Modification1: consider idle time

q' 2 max

q'3 max

q 2 , q1 min[ t j1 ]
j '

q 3 , q 2 min[ t j2 ], q1 min[ t j1 t j2 ]

j '
j '

Production Scheduling

14

P.C. Chang, IEM, YZU.

Refined Bounds
Modification2: (McMahon and Burton)

b 4 q1 max t k1 t k 2 t k 3 min t j1 , t j3
k '
j '

j k

b 5 q ' 2 max t k 2 t k 3 min t j2 , t j3

k '
j '

B' max{B, b 4 , b 5 }

Previous : Machine-based bound

Modification2 : Job-based bound
Production Scheduling

15

P.C. Chang, IEM, YZU.

Refined Bounds
Obviously, B>=B, This means that the combination
of machine-based and job-based bounds represented
by B will lead to a more efficient search of the
branching tree in the sense that fewer nodes will be
created.

Production Scheduling

16

P.C. Chang, IEM, YZU.

Hw.
Consider the following four-job three-machine problem
j

a.

b.

tj1 13

26

tj2

12

tj3 12 16

Find the min makespan using the basic Ignall-Schrage

algorithm. Count the nodes generated by the branching
process.
Find the min makespan using the modified algorithm.

Production Scheduling

17

P.C. Chang, IEM, YZU.

Heuristic Approaches
Traditional B&B:
The computational requirements will be severe for large
problems
Even for relatively small problems, there is no guarantee
that the solution can be obtained quickly,
Heuristic Approaches
can obtain solutions to large problems with limited
computational effort.
Computational requirements are predictable for problem of
a given size.
Production Scheduling

18

P.C. Chang, IEM, YZU.

Palmer
Palmer proposed the calculation of a slope index, sj, for
each job.
s j (m 1) t j,m (m 3) t j, m 1 (m 5) t j, m 2 (m 3) t j, 2 (m 1) t j,1

Then a permutation schedule is constructed using the

job ordering
s[1] s[ 2] s[ n ]

Production Scheduling

19

P.C. Chang, IEM, YZU.

Gupta
Gupta thought a transitive job ordering in the form of follows
that would produce good schedules. Where
sj

ej
min{t j1 t j2 , t j2 t j3 }

Where
1 if t j1 t j3
ej
1 if t j1 t j3

Production Scheduling

20

P.C. Chang, IEM, YZU.

Gupta
Generalizing from this structure, Gupta proposed that for m>3,
the job index to be calculated is
sj

ej
min {t jk t j, k 1}

1 k m 1

Where
1 if t j1 t jm
ej
1 if t j1 t jm

Production Scheduling

21

P.C. Chang, IEM, YZU.

CDS
Its strength lies in two properties:
1.It use Johnsons rule in a heuristic fashion
2.It generally creates several schedules from which a best
schedule can be chosen.
The CDS algorithm corresponds to a multistage use if
Johnsons rule applied to a new problem, derived from the
original, with processing times t ' j1 and t ' j2 . At stage 1,
t ' j1 t j1 and t ' j2 t jm

Production Scheduling

22

P.C. Chang, IEM, YZU.

CDS
In other words, Johnsons rule is applied to the first and mth
operations and intermediate operations are ignored. At stage
2,
t' t t
and t ' t t
j1

j1

j2

j2

jm

j, m 1

That is, Johnsons rule is applied to the sums of the first two
and last two operation processing times. In general at stage i,
i

t ' j1 t jk
k 1

and t ' j2 t j, m k 1

Production Scheduling

k 1

23

P.C. Chang, IEM, YZU.

Ex.

tj1

tj2

tj3

Palmer: s j m 1 t j 3 m 1 t j1 2t j 3 2t j1

s1 8 s2 6 s3 4 s4 2 s5 2
3 5 4 2 1 M 37

1
1
1
s2
s3
10
2
12
5 3 4 1 2 M 36

Gupta: s1

s4

1
1
s5
13
11

CDS: 3-5-4-1-2 M=35

Production Scheduling

24

P.C. Chang, IEM, YZU.

HW.

tj1

11

tj2

11

tj3

13 10

Let {1,3}
1. Use Ignall-Schrage & McMahon-Burton
2
2
b
,
b
,
b
,
b
,
b
of
P
,
P
to solve 1 2 3 4 5
13 xxx
31 xxx
2. Use Palmer, Gupta, CDS to solve this
problem.
Production Scheduling

25

P.C. Chang, IEM, YZU.