Вы находитесь на странице: 1из 13

ORIGINAL ARTICLE

Integrating theory of constraints and particle


swarm optimization in order planning and scheduling
for machine tool production
Li-Lan Liu & Gai-Ping Zhao & Shu-Sheng OUYang &
Ying-Jie Yang
Received: 17 November 2010 / Accepted: 22 March 2011 / Published online: 4 May 2011
# Springer-Verlag London Limited 2011
Abstract Order planning and scheduling has become a
significant challenge in machine tool enterprises, who want
to meet various demands of different customers and make
full use of existing resources in enterprises simultaneously.
Based on the Theory of Constraints, a three-stage order
planning and scheduling solution is proposed to optimize
the whole system performance with bottleneck resources'
capability as the constraints. After the identification of
bottleneck resources, multicriteria priority sequencing is
made with order per-contribution rate, order delivery
urgency, and customer importance as the evaluation criteria,
and the evaluation result deduced from the ideal point
function can decide the production mode of all orders and
products. Then, a PSO-based multiobjective optimization
model is set up with minimizing bottleneck machines'
makespan and minimizing total products' tardiness as the
two objectives. Finally, the proposed solution is applied in
one machine tool enterprise by integrating into Baosight
MES (Manufacturing Execution System) system. In addi-
tion, some comparisons are carried out to evaluate the
proposed PSO optimization method. The comparison with
actual report shows that PSO can satisfy enterprise's needs
better than before; the comparisons with genetic algorithm
and ant colony optimization algorithms indicate that PSO is
more effective than the others because of its faster
convergence rate.
Keywords Order planning and scheduling
.
Priority
evaluation
.
Multiobjective optimization
.
Theory of
constraints
.
PSO
1 Introduction
Machine tool industry is a highly specialize industry whose
core activity is the design and building of tools which shape
metal into precision parts for a variety of mechanical
products [1]. With thousands years of development,
machine tool has been evolved from simple function with
man powered at the beginning to the multifunction with
numerical controlled ones nowadays, such as flexible
manufacturing system (FMS), numerical control (NC) and
computerized numerical control (CNC) machine tool,
multiprocess machining center, etc. [2].
To meet the various requirements, including high
performance, easy to operate, safe and reliable, intelligent,
beautiful and practical, an order of machine tool produc-
tion, a typical kind of small batch sizes, high variety with
Make-To-Order (MTO) as production policy, can be
released to the manufacturing facility only after a firm
demand has been received.
Under this circumstance, customer order management
is necessary for machine tool production management,
who represents the core activity dealing with how and
when to produce, considering customer orders and
material and resource availability while aiming at
minimizing production time and costs, efficiently orga-
nizing the use of resources, and maximizing efficiency in
the production system [3].
The problem under consideration in this paper is referred
to as order planning and scheduling in the order manage-
ment system. In such a problem, we mainly consider a
L.-L. Liu (*)
:
G.-P. Zhao
Shanghai Enhanced Laboratory of Manufacturing Automation and
Robotics, Shanghai University,
Shanghai, China
e-mail: lancy@shu.edu.cn
S.-S. OUYang
:
Y.-J. Yang
Shanghai Baosight Software Co,
Shanghai, China
Int J Adv Manuf Technol (2011) 57:285296
DOI 10.1007/s00170-011-3294-6
facility with m machines, to produce k different product
types for n customer orders, with bottleneck machines and
given production process as constraints. The objective is to
schedule the orders on the machines so as to minimize
makespan and total tardiness of the bottleneck machines, by
means of integrating Theory of Constrains (TOC) and
Particle Swarm Optimizer (PSO) algorithm.
Since the 1950s, academic research has examined the
field of production planning, scheduling, and control from
theoretical, methodological, practical, and empirical aspect
[4]. The concept of customer order scheduling was first
introduced by Julien and Magazine [5], who analyzed the
structure of optimal schedules, provided a dynamic
programming formulation of the problem with two product
types and a given order processing sequence.
Leung and Pinedo [6] proved that the total order
completion time is NP-hard in the sense when facilities
number was over three, and focused on the models with
several objective functions, including total weighted com-
pletion time, maximum lateness the number of orders
shipped late, etc. They [7] studied the problem of
scheduling orders for different product types in a facility
with a number of machines in parallel. And focused on two
cases, one is that each product type can be produced on one
and only one machine which is dedicated to that product
type. The other case is that all machines are identical and
flexible, and each product type can be produced by any one
of the machines.
Yang and Posner [8] studied the problem with two
machines and various objective functions, and proposed
optimal solution for several complex scheduling problems.
Guoqing et al. [9] developed a heuristics to track and
analyze the scheduling problem that each customer order
consists of several jobs of different types, which are to be
processed on m facilities. The objective was to minimize
total weighted completion time.
nc et al. [10] studied the scheduling problem that
customers order several types of products produced on a
single machine. With minimizing the average customer
order flow time as the objective, they compared the
performance of four heuristics, including simulated anneal-
ing, genetic algorithms, tabu search, and ant colony
optimization.
PSO was proposed by Dr. Eberhart R and Dr. James
Kennedy [11], which is a population-based search proce-
dure that could yield global optimum solution. As a new
evolving calculation technology, PSO is a type of intelli-
gent optimization algorithm developed from the observa-
tions of social behavior of animals, such as bird flocking,
fish schooling, swarm intelligence, and the study processes
of human knowledge.
Due to the attractive features of easy implementation,
quick convergence, and few parameters needed, PSO has
been applied in a wide variety of optimization problems
including jobshop/flowshop scheduling problem.
Tasgetiren et al. [12] developed a PSO model for a
permutation flowshop sequencing problem. Liao et al. [13]
extended the PSO based on the discrete version to solve
flowshop scheduling problems. Liu et al. [14] proposed an
effective hybrid PSO for permutation flowshop scheduling
problem with the limited buffers. Jarboui et al. [15]
developed a PSO algorithm for solving the permutation
flowshop scheduling problem, which was an improved
procedure based on simulated annealing. Tasgetiren et al.
[16] used PSO to solve the permutation flowshop schedul-
ing problem with the objectives of minimizing makespan
and the total flow time of jobs.
Lian et al. [17] presented a permutation jobshop
scheduling problem solved by particle swarm optimization.
They introduced some new genetic operators and then
compared the related results with the results obtained by
genetic algorithm (GA) in order to show the efficiency of
their proposed algorithm. Reza et al. [18] presented a
combination of PSO and genetic operators for a multi-
objective jobshop scheduling problem that minimizes the
mean weighted completion time and the sum of the
weighted tardiness/earliness costs. And they proposed a
PSO-SA hybrid algorithm [19] based on PSO and simulat-
ed annealing (SA) for the periodic jobshop scheduling
problem. Yongxian et al. [20] established a jobshop
scheduling model based on PSO and researched the coding
and optimized operation with limited resources. Xia et al.
[21] introduced a hybrid HPSO algorithm based on PSO
and SA to solve the jobshop scheduling problem. Qun et al.
[22] employed the PSO to solve the problem of scheduling
jobs on parallel machines with a total tardiness objective.
Sha et al. [23] constructed a PSO-based multiobjective
algorithm for flowshop scheduling and jobshop scheduling
problem that minimizes makespan, total tardiness, and total
machine idle times.
The concept of the TOC was proposed by Goldratt and
Cox in the book named The Goal [24]. TOC is a
management philosophy that identifies and leverages
constraints (sometimes called critical resource or bottle-
necks) in order to maximize the profit potential of a
company, which means increase throughput, reduce inven-
tory, and cut down on operating expenses. And critical
resources (bottlenecks) play essential roles in determining
throughput and increasing profit [25].
TOC has emerged as an effective management philoso-
phy that has successfully tackled the scheduling problems
in mass manufacturing plants [26]. Zailin et al. [27]
developed an operation and control system based on TOC
for the high-mix/low-volume production of flow manufac-
turing. And identified the bottleneck resources and pro-
posed the management mechanism for the bottlenecks.
286 Int J Adv Manuf Technol (2011) 57:285296
Mabel et al. [28] compared TOC and linear program-
ming (LP) for the production scheduling problem with the
real operations data from a fiber manufacture, and found the
schedule result produced by TOC is very similar to that
generated by LP. Milton et al. [29] proposed a heuristic
algorithm based on TOC and the identification of bottle-
neck to minimize makespan on a flexible flowshop
scheduling problem. Rui et al. [30] developed an intelligent
scheduling system based on TOC. The system integrated a
product model, an operation model, a factory model, and a
knowledge database to make process planning and sched-
uling automatically for ship pipe production. Han-Pang et
al. [31] proposed an online dispatching rescheduling
mechanism for semiconductor manufacturing, by combin-
ing with TOC, GA, and support vector machine (SVM). Li
et al. [32] proposed a TOC combining ant colony
optimization (ACO) sequencing method for the scheduling
of semiconductor manufacturing fabrication lines. The
method identified the bottlenecks in the production line,
and then applied ACO algorithm to obtain the optimized
jobs processing sequence.
2 TOC-based order planning and scheduling
2.1 Problem definition in machine tool production
Machine tool is a typically traditional, discrete manufactur-
ing industry, with MTO as the main production policy. Its
production method can be in batch, in single or small batch.
Manufacturing processes can be classified as casting,
shaping/forming, machining, joining, surface, and assembly
processes, and machining method for the processes can also
be broken down into cutting, abrasive, grinding, and
nontraditional processes [33]. And there are always many
workshops in an enterprise to meet the different require-
ments of machining.
The production planning and scheduling is always
known as the most knowledge-intensive activity in the
machine tool enterprises for the random-coming orders
and complexity in manufacturing process [34]. How to
formulate a more rational production plan and schedule so
as that it not only could meet individual demands of
different customers but also could make full use of
existing resources of enterprises becomes a main concern
for the managers.
This paper focuses on the problem that calculates and
reports weekly or even daily plans and schedules for
workshops, whose machines are limited, machining pro-
cesses are given, and orders reach randomly.
Consider the following scenario of scheduling customer
orders in a workshop. There are N orders with K different
product types should be finished during the time [T
1
,T
2
].
These products are to be processed at M machines with
given machining procedures. Each product can be pro-
cessed on one and only one machine at a time, whereas
each machine can process only one product at a time. The
processing time of each product on each machine is fixed
and known in advance. We formulate the multiobjective
planning and scheduling problem using the following
notation:
& Time range: [T
1
,T
2
], usually week or day;
& Order: O O
i
f g i 1; ; N N is the order number;
& Product type: U U
k
f g k 1; ; K , K is the product
type number belongs to orders;
& Machine: E fE
j
g j 1; ; M , M is the machine
number.
For order O O
i
f g i 1; ; N : T
Ci
is the completion
date, T
Di
is the due date, X
i
is the batch volume, P
i
is the
price, and A
ij
is the workload on machine E
j
.
For product U
k
k 1; ; K , its processing operation
matrix S
UE
in given as:
S
UE
S
kj

k 1; ; K; j 1; ; M 1
S
kj
is the processing number that product U
k
will be
processed on machine E
j
.
The processing time matrix T
UE
is given as:
T
UE
T
kj

k 1; ; K; j 1; ; M 2
T
kj
is the processing time that product U
k
will be
processed on machine E
j
.
2.2 TOC-based planning and scheduling solution
TOC is a management approach that emphasizes continu-
ally improving the performance of manufacturing opera-
tions by improving bottleneck resources [35]. TOC states
that a local optimum is not an optimum at all, and that the
overall system performance is governed by the bottleneck
resource [24]. A bottleneck is defined as a unit in the
manufacturing process that holds down the amount of
products that a factory can produce. A bottleneck may
include a machine or an operator whose capacity limits the
throughput of the whole production process [30].
Based on TOC concept, the order planning and
scheduling problem is thus to find the bottleneck and to
optimize the whole system performance by exploiting the
bottleneck resources thoroughly and perfectly. The decom-
position planning and scheduling procedure is shown in
Fig. 1.
There are three main stages in this procedure:
1. Bottleneck identification
The bottleneck resources can be identified by
comparing the capacity of the resource and workload
Int J Adv Manuf Technol (2011) 57:285296 287
during the time rage [T
1
,T
2
]. For each machine
E
j
j 1; ; M , the workload T
Wj
can be calculated
by:
T
Wj

N
i1

K
k1
A
kj
X
k
3
If T
Wj
T
2
T
1
, machine E
j
(j =1,...,M) is the
bottleneck resource.
2. Multicriteria priority sequencing
If there are bottleneck resources during the time
rage, the priority evaluation and sequencing for the
orders and products are required. In case of the
constraints of bottleneck resources, the over-workload
orders and products will be dealt with by outsourcing
production mode. This stage will be described detailed
in Section 3.
3. PSO-based multiobjective optimization
After sequencing, orders and products that can be
completed in the workshop will reach to this optimiza-
tion stage. With minimizing bottleneck machines'
makespan and minimizing total products' tardiness as
objectives, the plan and schedule will be reported after
optimizing by PSO algorithm. This stage will be
described detailed in Section 4.
3 Multicriteria priority sequencing
3.1 Evaluation criteria
Order per-contribution rate, order delivery urgency, and
customer importance are the three evaluation criteria used
in the paper to sequence the priority of the orders.
1. Order per-contribution rate B
i
If there are bottlenecks E
j
j 1; ; M during the
time range for the order O
i
i 1; ; N , order per-
contribution rate is defined as:
B
i

P
i
A
ij
i 1; ; N; j 1; ; M 4
Otherwise, the order per-contribution rate without
bottleneck resources is defined as:
B
i
P
i
i 1; ; N 5
2. Order delivery urgency P
Ri
Order' due date is always the most concerned criteria
for the MTO machine tool enterprises, for order
O
i
i 1; ; N , order delivery urgency is defined as:
P
Ri

T
2
T
Di

T
2
T
1

6
3. Customer importance C
i
Customer importance is divided into five levels: very
important, important, normal, less important, and unim-
portant, which decided by historical transactions, market
decisions, order batches and volume, and customer'
credibility. C
i
is between [0,1] as shown in Table 1.
3.2 Order priority sequencing based on ideal point method
Ideal point method (IPM) is applied in the paper to
calculate the final evaluation function result for each order
based on the three criteria.
Ideal point is defined as the optimal value of the
functions, and it is one of the methods to solve multicriteria
problem [36]. For multicriteria evaluation function Fx
Table 1 Customer importance range
Customer importance C
i
Very important C
i
=[0.8,1.0]
Important C
i
=[0.6,0.8]
Normal C
i
=[0.4,0.6]
Less important C
i
=[0.2,0.4]
Unimportant C
i
=[0,0.2]
Fig. 1 TOC-based order planning and scheduling procedure
288 Int J Adv Manuf Technol (2011) 57:285296
f
1
x; ; f
k
x
T
k 1; ; K , its ideal point can be
defined as F

f

1
; f

k
_ _
T
k 1; ; K , f

k
is the ideal
point of f
k
(x). Then, the multicriteria evaluation result can
be calculated by jjFx F

jj !min Fx F

_
_
_
_
.
In the case of order priority sequencing, we define the
evaluation function F
i
x B
i
; P
Ri
; C
i
f g for each order,
and B
i
, P
Ri
, C
i
can be obtained by Eq. (4)/(5), Eq. (6), and
Table 1. Suppose the ideal point is F

i
B

i
; P
R

i
; C

i
_ _
,
B

i
; P
R

i
; C

i
is the ideal point of B
i
, P
Ri
, C
i
separately. Then,
the priority evaluation function can be defined as:
min F
i
x F
i
x F

i
_
_
_
_
_
_

B
i
B

i
_ _
2
P
Ri
P
R

i
_ _
2
C
i
C

i
_ _
2
_
7
With Eq. (7), we get the evaluation result for every order
O O
i
f g i 1; ; N . Order priority sequence is sorted
by the evaluation result, the larger the result, the more prior
the order.
4 PSO-based multiobjective optimization
PSO is an evolutionary technique [11] for solving uncon-
strained continuous optimization problems. In PSO, each
potential solution is a point in the search space and may be
regarded as a particle. Initially, a bunch of particles are
randomly created and set into motion through the multidi-
mensional problem space. In their movement, the particles
have memory, and each particle adjusts its position based
on its own experience as well as the experience of a
neighboring particle by utilizing the best position encoun-
tered by itself and its neighbors. Particles cooperate to
determine the best position (solution) in the search space
(solution space).
4.1 Objectives
For the optimization of order planning and scheduling
problem, we set minimizing bottleneck machines' make-
span and minimizing total products' tardiness as the two
objectives.
1. Minimizing bottleneck machines' makespan T
min
T
min
min max
1jM
max
1kK
T
kj
_ _ _ _
8
2. Minimizing total products' tardiness T
L
T
L
min

K
k1
max 0; T
Ck
T
Dk

_ _
9
3. Evaluation function
Ideal point method is also used here for the evaluation
function of the two objectives.
Evaluation function and its ideal point are defined as
Fx T
min
; T
L
f g, F

min
; T

L
_ _
. And the evaluation
result can be deducted with:
minFx

T
min
T

min

2
T
L
T

2
_
10
4.2 Planning and scheduling steps based on PSO
Step 1 Initialization. Initialize a population of particles
with random positions and velocities on D
dimensions in the search space, and each particle
represents a candidate position (i.e., solution). W is
the inertia factor, C
1
is the particle acceleration
coefficients, C
2
is the population acceleration
coefficients, and C
1
and C
2
are two positive
constants. Pe
num
is the size of population, and
r
max
is the max iteration counter. A particle is
considered as a point in a D-dimensional space,
and its status is characterized according to its
position and velocity.
Step 2 Fitness. Measure the fitness of each particle in the
population, and evaluate the objective function of
all particles. Get the objective value of T
min
and T
L
for particle i at current position X
i
, calculate the
objective function F X
i
, and set the objective
value for them.
Step 3 Update the best value of each particle. Set
P
i best
as the best value for particle i, and get its
current position value as P
i
r
. If P
i
r
> P
i best
, then
P
i best
P
i
r
.
Step 4 Update the best value of the population. Map the
position of each particle into the solution space
and evaluate its fitness value according to the
desired optimization fitness function. The best
value of the population is g
best
, g
best
min P
i best
,
i 1; 2; :::; Pe
num
.
Step 5 Termination. Set r r 1, if r < r
max
, go to Step
6. Otherwise, stop the algorithm and report the
parameters for the products and machines, such as,
the start time/end time for each product, the
processing time of each machine, the max make-
span of the bottleneck machines, and each prod-
ucts tardiness and total tardiness.
Step 6 Update. Update the velocity and position of each
particle. Update according to Eq. (11).
V
ij
t 1 W

V
ij
t c
1

R
1

P
ij
t X
ij
t
_ _
c
2

R
2

P
gj
t X
ij
t
_ _
X
ij
t 1
X
ij
t V
ij
t 1
11
Int J Adv Manuf Technol (2011) 57:285296 289
X
ij
(t),X
ij
t 1 is the position of t and t +1 generation of
particle i. V
ij
t,V
ij
t 1 is the velocity of t and t +1
generation of particle i. R
1
, R
2
are two distinct random
numbers drawn out from the range [0,1].
W is the inertia factor which plays a key role in the
avoidance of rushing the PSO into a local optima, in order
to balance the search performance between global and local
exploration and exploitation, and then result in lesser
iterations on the average to find an optimal solution
efficiently. W is defined as Eq. (12), so that it can decrease
linearly during the run time.
W W
max

W
max
W
min
r
max

r 12
W
max
, W
min
are the maximum value and minimum value
of inertia weight, and r, r
max
are the current iteration
counter and the max iteration counter.
Go to Step 2 after updating the velocity and position.
5 Applications in machine tool production
This planning and scheduling solution is integrated into the
Baosight MES (Manufacturing Execution System) software
system, and is applied in a machine tool enterprise in
Shanghai, China. And some case studies have been carried
out to verify the effectiveness of the proposed solution.
We take the orders in a week from May 31 to June 6,
2009 as the application case, and should give an optimized
plan and schedule report automatically after calculating.
The parameters are given as: Orders number is 5,
O O
i
f g i 1; ; 5 ; products number is 5, U U
k
f g
k 1; ; 5 ; machines is 8, E E
j
_ _
j 1; ; 8 . The
parameters of the orders and products are shown in Table 2.
The processing time for the products in each machine is given
in Table 3.
By Eq. (3), we calculate the workload for each machine,
and identify the bottleneck machines for these orders, as
shown in Table 4.
From Table 4, machine E
2
and machine E
8
are the
bottleneck resource for this week's orders.
5.1 Priority sequencing
The orders should be sequenced because there are bottle-
neck resources according to Fig. 1.
With machine E
2
and E
8
as the bottleneck, value of order
per-contribution rate B
i
, Order delivery urgency P
Ri
, and
Customer importance C
i
can be calculated by Eq. (4), Eq.
(6), and Table 1.
The results for orders O O
i
f g i 1; ; 5
are listed as: B 0:89; 1:56; 0:54; 0:41; 1:32 f g, P
R

0; 0:4; 0:2; 0:2; 0:4 f g, C 0:4; 0:3; 0:8; 0:2; 0:9 f g
From the definition of the ideal point, the ideal point for
B, P
R
, and C should be B
*
=1.56, P
R

0, C
*
=1.0.
According to Eq. (7), we calculate the priority evaluation
function for each order, and the evaluation result is shown
in Table 5.
As shown in Table 5, the evaluation results for products
U U
k
f g k 1; ; 5 are listed as F(x)={0.89, 0.6, 1.3,
1.18, 0.85, 1.10}, and the minimum value is 0.6 for product
U
3
of order O
2
, so order O
2
will be outsourced based on the
constraints of E
2
and E
8
. The rest orders and products can
be completed in the enterprise.
5.2 Plan and schedule optimization
According to Eq. (1) and Eq. (2), constructing the
processing operation matrix S
UE
and the processing time
matrix T
UE
for the products with production mode is
normal in Table 5.
Table 2 Orders and parameters
Customer Order Product Price Volume Due date
001 O
1
U
1
50 22 09.6.6
U
2
63 15 09.6.6
002 O
2
U
3
29 28 09.6.4
003 O
3
U
4
18 18 09.6.5
004 O
4
U
5
43 27 09.6.5
005 O
5
U
3
29 2 09.6.4
Table 3 Products' processing time (unit, hour)
Product U
1
U
2
U
3
U
4
U
5
Machine
Forging E
1
1.32 0.81 1.1 0.71 0.56
Turing E
2
1.44 1.08 1.39 1.63 1.46
Drilling E
3
0.57 0.91 0.79 0.82 0.62
Broaching E
4
0.38 0.12 0.24 0.11 0.18
Hobbing E
5
0.74 1.52 0.61 1.38 0.62
Shaving E
6
0.2 0.75 0.52 0.33 0.48
Milling E
7
0.14 0.25 0.35 0.11 0.21
Grinding E
8
0.93 0.67 0.88 0.72 0.54
Table 4 Bottleneck resources (unit, hour)
E
1
E
2
E
3
E
4
E
5
E
6
E
7
E
8
Capability 105 120 85 28 90 58 28 80
Workload 102 158 81 24 88 50 24 85
Balance 3 38 4 4 35 33 4 5
Bottleneck No Yes No No No No No Yes
290 Int J Adv Manuf Technol (2011) 57:285296
For the products U U
k
f g k 1; ; 5 , and the
machines E E
j
_ _
j 1; ; 8 , S
UE
and T
UE
are:
S
58

1 3 2 4 5 6 7 8
2 3 1 6 7 5 8 4
1 4 5 7 6 2 3 8
3 1 2 5 4 8 6 7
2 1 3 6 7 5 4 8
_

_
_

_
T
58

1742 752 1900 501 976 264 184 1227
972 819 729 1125 225 1368 603 108
132 28 73 42 62 166 94 105
885 766 1760 1490 118 777 356 118
2365 907 1004 777 340 1004 291 874
_

_
_

_
According to the steps of PSO-based optimization, the
application progress is shown in Fig. 2. The parameters are
initialized as: dimensions of the search space D=5,
population size P
enum
=10, particle acceleration coefficients
C
1
=2, population acceleration coefficients C
2
=2, and max
iteration counter r
max
=20. Inertia factor W is calculated by
Eq. (12), which W
max
=0.9, W
min
=0.4, which means W is
decreased linearly from about 0.9 to 0.4. And the ideal
point for the two objectives, minimizing bottleneck
machines' makespan T
min
, minimizing total products'
tardiness T
L
, are given as: T

min
10000, T

L
100.
After 20 times iteration, the parameters for each product
are reported detailed as shown in Table 6. According to the
report, the Gantt of the scheduling result is shown in Fig. 3.
The values of optimization objectives and evaluation
function are listed in Table 7.
According to Table 7, with May 31, 2009 as the start time
for these orders, the tardiness for products U U
k
f gk
1; ; 5 are separately 1,697m, 1,270m, 1,403m, 686m,
4,011m, and T
L
=9,067; the makespan for bottleneck
machines E
2
, E
8
are 7,268m, 12,651m, and T
min
=12,651;
then, the evaluation function value F(x)=9,350 is calculated
by Eq. (10).
Then, the planning and scheduling solution gives a
suggestion that set May 28, 2009 as the start time so that it
can get better evaluation function result. The values of the
optimization objectives and evaluation function are also shown
in Table 7: the tardiness for every product U
k
(k=1,...,5) is 0m,
and T
L
=0; the makespan for bottleneck machines E
2
, E
8
are
7,268, 12,651m, and T
min
=12,651; then, the evaluation
function value F(x)=2,652, which is 71.6% better than the
value that the start time is May 31, 2009.
6 Comparisons
6.1 Comparison with actual application
In order to demonstrate the effectiveness of this order
planning and scheduling solution, comparison is made with
the actual plan and schedule report that the enterprise used
at that time. The comparison values of the optimization
objectives and evaluation function are shown in Table 8.
According to Table 8, the actual results enterprise used
are: the tardiness for products U = {U
k
} k 1; ; 5 are
870m, 1,947m, 4,824m, 1,444m, 4,552m, and T
L
=13,637;
the makespan for bottleneck machines E
2
, E
8
are 9,767m,
13,192 m, and T
L
=13,192; then, the evaluation function
value F(x)=13,908.
The comparison of products' tardiness is shown in
Fig. 4a, the comparison of bottleneck machines' makespan
is shown in Fig. 4b, and the comparison of objectives' value
is shown in Fig. 4c.
From Table 8 and Fig. 4, the result obtained from our
solution is much better than before: the tardiness of the
products reduces 33.5%, the makespan of bottleneck
machines reduces 4.1%, and the value of evaluation
function reduces 32.8%.
6.2 Comparison with GA and ACO algorithm
From the state of art, GA, ACO, and PSO are the three
main algorithms for the optimization of multiobjective
scheduling problem in these days. To illustrate the
efficiency of our PSO-based algorithm, we compare the
results with GA and ACO.
1. GA
GA was firstly invented by John Holland [37] and
his associates in 1975. It is inspired by the mechanism
of natural selection, the rule of the fittest will survive,
Customer Order Product Price Volume Due date Evaluation result Mode
001 O
1
U
1
50 22 09.6.6 0.89 Normal
U
2
63 15 09.6.6 0.6 Normal
002 O
2
U
3
29 28 09.6.4 1.30 Outsourcing
003 O
3
U
4
18 18 09.6.5 1.18 Normal
004 O
4
U
5
43 27 09.6.5 0.85 Normal
005 O
5
U
3
29 2 09.6.4 1.10 Normal
Table 5 Evaluation result of
order priority
Int J Adv Manuf Technol (2011) 57:285296 291
where stronger individuals are likely to be the winners in
a competing environment. Throughout the genetic evolu-
tion, starting from a population of chromosomes, some
fitter chromosomes tend to yield good quality offspring,
and this means better solutions to a problem [38].
The parameters and execution process of GA algo-
rithm are as follows:
Step 1 Initialize parameters: chromosome crossover
probability P
c1
=0.9, gene crossover probabil-
ity P
c2
=0.6, chromosome mutation probability
P
m
=0.1, randomized cross-bit C=10, gene
population size P
num
=10, maximum number
of iterations R
max
=20. The initial population is
randomly generated.
Step 2 Calculate the fitness for individuals in the current
population. Calculate T
min
, T
L
, and F(X
i
) to each
chromosome X
i
(i=1,...,P
num
).
Step 3 Evaluate each individual by its fitness, and
select the optimal chromosome X
best
from the
c ur r e nt popul a t i on, t he n F X
best

min F X
i
i 1; :::; P
num
.
Step 4 Judge the termination condition. R R 1, if
R < R
max
, then go to Step 5; else terminate
search and get the value of the parameters for
all orders.
Step 5 Crossover and mutation
5.1 Crossover. Select chromosomes with probability
P
c
1 from the population, crossover them with
the best individual X
best
, then the two best
individuals are selected as the child solutions.
5.2 Mutation. Select chromosomes with probability
P
m
from the population generate a child chro-
mosome by exchange the genes of randomly
selected individuals.
5.3 If the termination condition is satisfied, the
algorithm ends; else, go to Step 2.
Table 6 Report of order planning and scheduling (unit, minute)
Machine Process plan and schedule of each product Makespan
E
1
Sched U
3
U
1
U
2
U
4
U
5
4,349
Time 0 132 132 1,874 1,874 2,603 2,676 3,442 3,442 4,349
E
2
Sched U
2
U
5
U
4
U
1
U
3
7,268
Time 0 972 972 3,337 3,442 5,202 5,202 7,102 7,102 7,268
E
3
Sched U
2
U
4
U
1
U
5
U
3
7,362
Time 972 1,791 1,791 2,676 2,676 3,428 4,349 5,353 7,268 7,362
E
4
Sched U
3
U
1
U
4
U
5
U
2
11,350
Time 132 160 7,102 7,603 7,603 7,721 10,951 11,242 11,242 11,350
E
5
Sched U
3
U
4
U
1
U
2
U
5
10,951
Time 160 233 5,210 6,700 7,603 8,579 8,579 9,947 9,947 10,951
E
6
Sched U
2
U
5
U
3
U
1
U
4
9,208
Time 2,603 3,728 5,353 6,130 6,130 6,192 8,579 8,843 8,852 9,208
E
7
Sched U
3
U
2
U
5
U
1
U
4
9,326
Time 233 275 3,728 3,953 6,130 6,470 8,843 9,027 9,208 9,326
E
8
Sched U
4
U
3
U
2
U
1
U
5
12,651
Time 7,721 8,498 8,498 8,603 9,947 10,550 10,550 11,777 11,777 12,651
Fig. 2 Application progress of PSO optimization
292 Int J Adv Manuf Technol (2011) 57:285296
2. ACO
ACO is a class of optimization algorithms modeled
on the actions of ants developed by Dorigo [39] in
1999. ACO is useful in problems that need to find paths
to goals, and has been successfully applied to several
NP-hard optimization problems, such as the Traveling
Salesman Problem (TSP), Quadratic Assignment Prob-
lem (QAP), Vehicle Routing Problem (VRP), Job-Shop
Scheduling Problem (JSP), Flow-Shop Scheduling
Problem (FSP), etc. [40].
The parameters and execution process of ACO
algorithm are as follows:
Step 1 Initializing parameters. The number of ants K=20,
the maximum number of iteration T
max
=5, the
initial pheromone C
0
=0. For each ant
k(k =1,...,K), G
k
is the list of all search
processes, S
k
is the list of next search processes,
and L
tabuk
is the list of searched processes. The
ants are distributed randomly in G
k
at first.
Step 2 Searching solutions for each ant. Ant k(k=1,..., K)
selects its next process in S
k
with the highest
probability P
ij
(t) according to Eq. (13). Which, i,
j are two different processes in the search space,
C
ij
(t) is the pheromone between (i, j) at time t.
The selected process is added to L
tabuk
and
deleted from S
k
at the same time. This step is
repeated until S
k
is empty, and L
tabuk
is the
solution obtained by ant k.
p
ij
t
t
ij
t

j
t
ij
t
13
Step 3 Calculating parameters and objective values
includes T
min
, T
L
and F(X
i
).
Fig. 3 The Gantt of scheduling
Table 7 Value of optimization objectives and evaluation function (unit, minute)
Start time: May 31, 2009
Products T
L
Bottleneck machines T
min
F(x)
Makespan U
1
U
2
U
3
U
4
U
5
Makespan E
2
E
8
11,777 11,350 8,603 9,326 12,651 9,067 12,651 9,350
Tardiness 1,697 1,270 1,403 686 4,011 7,268 12,651
Start time: May 28, 2009
Products T
L
Bottleneck machines T
min
F(x)
Makespan U
1
U
2
U
3
U
4
U
5
0 Makespan E
2
E
8
11,777 11,350 8,603 9,326 12,651 12,651 2,652
Tardiness 0 0 0 0 0 7,268 12,651
Int J Adv Manuf Technol (2011) 57:285296 293
Step 4 Updating pheromone. Update the pheromone
values according to Eq. (14). Then repeat Step
2 and Step 3. Which =0.5 is the remained
pheromone parameter, t
k
ij
1 i s t he
remained C
ij
(t) for ant k.
t
ij
t n r

t
ij
t

m
k1
t
k
ij
14
Step 5 If the termination condition is satisfied, the
algorithm ends; else, go to Step 2.
3. Results and comparison of PSO, GA, and ACO. The
computational results by GA and ACO are given in
Table 9. The results of GA solution generated are: the
tardiness for products U = {U
k
} (k = 1,...,5) are
2,323 m, 1,896m, 395m, 0m, 4,637m, and T
L
=9,521;
the makespan for bottleneck machines E
2
, E
8
are 7,163
m, 13,277m, and T
min
=13,277; then, the evaluation
function value F(x)=9,720. The results of ACO solution
generated are: the tardiness for products U={U
k
}
(k=1,...,5) are 2,927m, 2,778m, 580m, 2,245m, 634m,
and T
L
=9,164; the makespan for bottleneck machines
E
2
, E
8
are 9,815m, 13,007m, and T
min
=13,007; then, the
evaluation function value F(x)=9,549.
The comparisons are shown in Fig. 5. Figure 5a is the
comparison of products' tardiness, Fig. 5b is the compar-
ison of bottleneck machines' makespan, and the comparison
of objectives' value is shown in Fig. 5c.
From Table 9 and Fig. 5, comparing the results obtained
from PSO and GA solution: the tardiness of the products
reduces 1.99%, the makespan of bottleneck machines
reduces 4.7%, and the value of evaluation function reduces
3.8%. Comparing the results obtained from PSO and ACO
solution: the tardiness of the products reduces 1.1%, the
makespan of bottleneck machines reduces 2.7%, and the
value of evaluation function reduces 2.1%. The results and
comparisons show that there is no significant priority
among PSO, GA, and ACO solutions, and each one can
find the near optimal result for this scheduling problem;
however, the proposed PSO solution is more effective than
the others because of its faster convergence rate.
7 Conclusions
With MTO as production policy, machine tool enterprises
are concerned with the questions on how to make full use
of enterprise resources, to adapt to the individual demands
of customers, and to improve the benefits accordingly.
Consequently, order planning and scheduling becomes one
of the key tasks in the operation of enterprises, which aim
at giving a report about plan and schedule for all the orders
and products weekly or daily.
In order to solve this NP-hard planning and schedul-
ing optimization problem, based on TOC concept, a
Fig. 4 Comparison with actual result
Table 8 Values of comparison with actual result (unit, minute)
Products T
L
Bottleneck machines T
min
F(x)
Tardiness U
1
U
2
U
3
U
4
U
5
Makespan E
2
E
8
PSO Result 1,697 1,270 1,403 686 4,011 9,067 7,268 12,651 12,651 9,350
Actual result 870 1,947 4,824 1,444 4,552 13,637 9,767 13,192 13,192 13,908
294 Int J Adv Manuf Technol (2011) 57:285296
three-stage order planning and scheduling solution is
proposed in this paper to optimize the whole system
performance by exploiting the bottleneck resources
thoroughly and perfectly.
Bottleneck resources are identified after reading, clus-
tering, and analyzing all the orders and products in this
solution at the first stage. Then, with bottleneck resources
as the constraints, order per-contribution rate B
i
, order
delivery urgency P
Ri
, and customer importance C
i
as the
evaluation criteria, multicriteria priority sequencing is made
based on the IPM for all orders and products. The
sequencing evaluation function value will decide the
production mode for each order and product.
After that, a PSO-based multiobjective optimization
model is set up at the third stage. Planning and scheduling
steps based on PSO are described detailed with minimizing
bottleneck machines' makespan T
min
and minimizing total
products' tardiness T
L
as the two objectives.
Finally, this order planning and scheduling solution is
applied in one machine tool enterprise by integrating into
Baosight MES (Manufacturing Execution System) system.
The comparison with actual application report shows that
the plan and schedule generated by this solution is much
better than before. The comparisons with GA and ACO
algorithms indicate that there is no significant priority
among PSO, GA, and ACO solutions, but PSO is more
effective than the others because of its faster convergence
rate.
More studies will be undertaken in the near future: the
efficiency of the proposed solution should be examined
more carefully, and a greater range of applications will be
tried in discrete manufacturing enterprises.
Acknowledgments This work is supported by the National Natural
Science Foundation of China (NSFC) under Grant no. 50805089, and
the Shanghai Science and Technology Committee (STCSM) under
Grant no. 09DZ1122500. The authors are grateful for the financial
supports, and also, we would like to thank Professor Kesheng Wang
from Norwegian University of Science and Technology (NTNU) for
his helpful discussions and suggestions.
References
1. Greis NP (1995) Technology adoption, production design, and
process change: a case study in the machine tool industry. IEEE
Trans Eng Manage 42(3):192202
2. de Barr E, Finstp BSC (1972) Trends in machine tool research.
Prod Eng 6:213218
3. Arianna Alfieri, Tullio Tolio, Marcello Urgo (2010) A project
scheduling approach to production and material requirement
planning in manufacturing-to order environments. Journal of
Intelligent Manufacturing. Springer Online
4. Sprague LG, Ritzmann LP, Krajewski L (1990) Production
planning, inventory management and scheduling: spanning the
boundaries. Manag Decis Econ 11(5):297315
5. Julien FM, Magazine MJ (1990) Scheduling customer orders: an
alternative production scheduling approach. J Manuf Oper 3:177
199
Table 9 Values of comparison with GA and ACO (unit, minute)
Products T
L
Bottleneck machines T
min
F(x)
Tardiness U
1
U
2
U
3
U
4
U
5
Makespan E
2
E
8
PSO result 1,697 1,270 1,403 686 4,011 9,067 7,268 12,651 12,651 9,350
GA result 2,323 1,896 395 0 4,637 9,251 7,163 13,277 13,277 9,720
ACO result 2,927 2,778 580 2,245 634 9,164 9,815 13,007 13,007 9,549
5000
T
a
r
d
i
n
e
s
s
4500
4000
3500
3000
2500
2000
1500
1000
500
0
0
U1 U2
E2 E8
U3 U4 U5
Products
Bottleneck machines
Objectives
GA
PSO
ACO
GA
PSO
ACO
GA
PSO
ACO
M
a
k
e
s
p
a
n
12000
10000
8000
6000
4000
2000
14000
12000
10000
8000
6000
4000
2000
0
V
a
l
u
e
s
TL Tmin F(x)
a
b
c
Fig. 5 Comparison with GA and ACO
Int J Adv Manuf Technol (2011) 57:285296 295
6. Leung JY-T, Li H, Pinedo M (2005) Order scheduling models: an
overview. In: Kendall G, Burke E, Petrovic S, Gendreau M (eds.)
Multidisciplinary scheduling 3756
7. Leung JY-T, Li H, Pinedo M (2008) Scheduling orders on either
dedicated or flexible machines in parallel to minimize total
weighted completion time. Ann Oper Res 159:107123
8. Yang J (2005) The complexity of customer order scheduling
problems on parallel machines. Comput Oper Res 32:19211939
9. Guoqing Wang TC, Cheng E (2007) Customer order scheduling to
minimize total weighted completion time. Int J Manag Sci
35:623626
10. Hazir , Gnalay Y, Erel E (2008) Customer order scheduling
problem: a comparative metaheuristics study. Int J Adv Manuf
Technol 37:589598
11. Kennedy J, Eberhart RC (1995) Particle swarm optimization. Proc
IEEE Int Conf Neural Networks, Piscataway, NJ, USA 4:1942
1948
12. Tasgetiren MF, Liang YC, Sevkli M, Gencyilmaz G (2007) A
particle swarm optimization algorithm for makespan and total
flowtime minimization in permutation flowshop sequencing
problem. Eur J Oper Res 177(3):19301947
13. Liao CJ, Tseng CT, Luarn P (2007) A discrete version of particle
swarm optimization for flowshop scheduling problems. Comput
Oper Res 34(10):30993111
14. Liu B, Wang L, Jin YH (2008) An effective hybrid PSO-based
algorithm for flow shop scheduling with limited buffers. Comput
Oper Res 35(9):27912806
15. Jarboui B, Ibrahim S, Siarry P, Rebai A (2008) A combinational
particle swarm optimisation for solving permutation flowshop
problems. Comput Ind Eng 54:526538
16. Tasgetiren MF, LiangYC SevkliM, Gencyilmaz G (2007) A
particle swarm optimization algorithm for makespan and total
flowtime minimization in the permutation flowshop sequencing
problem. Eur J Oper Res 177:19301947
17. Lian Z, Jiao B, Gu X (2006) A similar particle swarm
optimization algorithm for job-shop scheduling to minimize
makespan. Appl Math Comput 183:10081017
18. Reza Tavakkoli-Moghaddam, Mozhgan Azarkish, Azar
Sadeghnejad-Barkousaraie (2010) Solving a multi-objective job
shop scheduling problem with sequence-dependent setup times by
a Pareto archive PSO combined with genetic operators and VNS.
Int J Adv Manuf Technol 8, online
19. Amin Jamili, Mohammad Ali Shafia, Reza Tavakkoli-
Moghaddam (2010) A hybrid algorithm based on particle swarm
optimization and simulated annealing for a periodic job shop
scheduling problem. Int J Adv Manuf Technol 9, published online
20. Yongxian L, Xiaotian L, Jinfu Z (2008) Research on job-shop
scheduling optimization method with limited resources. Int J Adv
Manuf Technol 38:386392
21. Xia WJ, Wu ZM (2006) A hybrid particle swarm optimization
approach for the job-shop scheduling problem. Int J Adv Manuf
Technol 29:360366
22. Niu Q, Zhou T, Wang L (2010) A hybrid particle swarm
optimization for parallel machine total tardiness scheduling. Int J
Adv Manuf Technol 49:723739
23. Sha DY, Lin H-H (2009) A particle swarm optimization for multi-
objective flowshop scheduling. Int J Adv Manuf Technol 45:749758
24. Goldratt EM, Cox J (1992) The goal. Second revised edition.
North River Press, Croton-on Hudson, NY
25. Wei C, Liu PH, Tsai YC (2002) Resource-constrained project
management using enhanced theory of constraint. Int J Project
Manage 20(7):561567
26. Singh RK, Prakash A, Kumar S et al (2006) Psychoclonal based
approach to solve a TOC product mix decision problem. Int J Adv
Manuf Technol 29(1112):11941202
27. Guan Z, Peng Y, Ma L, Zhang C, Li P (2008) Operation and control of
flow manufacturing based on constraints management for high-mix/
low-volume production. Front Mech Eng China 3(4):454461
28. Qiu M, Fredendall L, Zhu Z (2002) Production scheduling: TOC
or LP? Manuf Eng 8:190195
29. Milton J. Acero-Dominguez, Carlos D. Patermina-Arboleda (2004)
Scheduling Jobs on a K-stage flexible flowshop using a TOC-based
(Bottleneck) Procedure. Proceedings of the 2004 Systems and
Information Engineering Design Symposium 295298
30. Li R, Hamada K, Shimozori T (2010) Development of a theory of
constraints based scheduling system for ship piping production. J
Shanghai Jiaotong Univ (Sci) 15(3):354362
31. Han-Pang Huang and Tien-Ying Chen (2006) A new approach to on-
line rescheduling for a semiconductor foundry fab. 2006 IEEE
international conference on systems, man, and cybernetics 47274732
32. Li Li, Pan Gu, Fei Qiao, Ying Wu (2010) ATOC combining ACO
sequencing method for the semiconductor manufacturing fabrica-
tion. Proceedings of the 8
th
World Congress on Intelligent Control
and Automation 46024607
33. Newman ST, Nassehi A (2009) Machine tool capability profile for
intelligent process planning. CIRP Annals-Manuf Technol
58:421424
34. Crama Y (1997) Combinatorial optimization models for produc-
tion scheduling in automated manufacturing systems. Eur J Oper
Res 99:136153
35. Gupta M (2003) Constraints management-recent advances and
practices. Int J Prod Res 41(4):647659
36. Balbas M, Ballve P, Guerra J (2001) Density theorems for ideal
points in vector optimization. Eur J Oper Res 133:260266
37. Holland JH (1975) Adaptation in natural and artificial systems: an
introductory analysis with applications to biology, control, and
artificial intelligence. MIT Press, London
38. Man KF, Tang KS, Kwong S (1996) Genetic algorithms: concepts
and applications. IEEE Trans Ind Electron 43:519532
39. E. Bonabeau, M. Dorigo, and G. Theraulaz (1999) Swarm
intelligence: from natural evolution to artificial systems. Oxford
University Press
40. Marco Dorigo and Thomas Sttzle (2004) Ant colony optimiza-
tion. The MIT Press, Cambridge, London, ISBN 0-262-04219-3
296 Int J Adv Manuf Technol (2011) 57:285296
Copyright of International Journal of Advanced Manufacturing Technology is the property of Springer Science
& Business Media B.V. and its content may not be copied or emailed to multiple sites or posted to a listserv
without the copyright holder's express written permission. However, users may print, download, or email
articles for individual use.

Вам также может понравиться