# Application of Intelligent Strategies for Cooperative Manufacturing Planning.

**ABSTRACT** Manufacturing planning is crucial for the quality and efficiency of product development. Process planning and scheduling are the most important and challenging tasks in manufacturing planning. These two processes are usually arranged in a sequential way. Recently, a significant trend is to make the processes to work more concurrently and cooperatively to achieve a globally optimal result. In this paper, several intelligent strategies have been developed to build up cooperative process planning and scheduling (CPPS). Three game theory-based strategies, i.e., Pareto strategy, Nash strategy and Stackelberg strategy, have been introduced to analyze the cooperative integration of the two processes in a systematic way. To address the multiple constraints in CPPS, a fuzzy logic-based analytical hierarchical process (AHP) technique has been applied. Modern heuristic algorithms, including particle swarm optimization (PSO), simulated annealing (SA) and genetic algorithms (GAs), have been developed and applied to CPPS to identify optimal or near-optimal solutions from the vast search space efficiently. Experiments have been conducted and results show the objectives of the research have been achieved.

**0**Bookmarks

**·**

**99**Views

- [Show abstract] [Hide abstract]

**ABSTRACT:**Process planning and scheduling are two of the most important functions in the manufacturing system. Traditionally, process planning and scheduling were regarded as separate tasks performed sequentially, where scheduling was implemented after process plans had been generated. However, their functions are usually complementary. If the two systems can be integrated more tightly, greater performance and higher productivity of manufacturing system can be achieved. In this paper, a new hybrid algorithm (HA) based approach has been developed to facilitate the integration and optimization of these two systems. To improve the optimization performance of the approach, an efficient genetic representation, operator and local search strategy have been developed. Experimental studies have been used to test the performance of the proposed approach and to make comparisons between this approach and some previous works. The results show that the research on integrated process planning and scheduling (IPPS) is necessary and the proposed approach is a promising and very effective method on the research of IPPS.International Journal of Production Economics 01/2010; 126(2):289-298. · 2.08 Impact Factor

Page 1

Application of Intelligent Strategies for Cooperative

Manufacturing Planning

Weidong Li

(Coventry University, Coventry, UK

weidong.li@coventry.ac.uk)

Liang Gao, Xinyu Li

(Huazhong University of Science and Technology, Wuhan, China

gaoliang@mail.hust.edu.cn, lixy@smail.hust.edu.cn)

Abstract: Manufacturing planning is crucial for the quality and efficiency of product

development. Process planning and scheduling are the most important and challenging tasks in

manufacturing planning. These two processes are usually arranged in a sequential way.

Recently, a significant trend is to make the processes to work more concurrently and

cooperatively to achieve a globally optimal result. In this paper, several intelligent strategies

have been developed to build up Cooperative Process Planning and Scheduling (CPPS). Three

Game Theory-based strategies, i.e., Pareto strategy, Nash strategy and Stackelberg strategy,

have been introduced to analyze the cooperative integration of the two processes in a

systematic way. To address the multiple constraints in CPPS, a fuzzy logic-based Analytical

Hierarchical Process (AHP) technique has been applied. Modern heuristic algorithms, including

Particle Swarm Optimization (PSO), Simulated Annealing (SA) and Genetic Algorithms (GAs),

have been developed and applied to CPPS to identify optimal or near-optimal solutions from

the vast search space efficiently. Experiments have been conducted and results show the

objectives of the research have been achieved.

Keywords: Collaborative system, Game Theory, Analytical Hierarchical Process, Particle

Swarm Optimization, Simulated Annealing, Genetic Algorithms

Categories: I.1.2, I.1.4, I.2.1, I.2.4, J.6

1

Introduction

Product development is comprised of various stages, such as conceptual design,

detailed design, prototyping, manufacturing planning, manufacturing and testing, etc.

The task of manufacturing planning is to interpret product models created by a design

team in terms of manufacturing processes, and associate the manufacturing

equipments and resources in shop floors with the interpretation. The major tasks of

manufacturing planning are process planning and scheduling. For a model of design

(e.g., models for vehicles and aircraft), it needs a series of manufacturing operations

(operations in the following content) to make it. For example, to make a hole, it could

include a drilling operation and reaming operation. The task of process planning is

more a product model-oriented. It interprets a model into some detailed operations,

such as primary operations (e.g., forging or casting to generate the rough shape),

rough machining, semi-finish machining, finish machining, surface treatment,

painting, etc. When many models are made together, there is a competition of

Journal of Universal Computer Science, vol. 15, no. 9 (2009), 1907-1923

submitted: 15/8/08, accepted: 24/5/09, appeared: 1/5/09 © J.UCS

Page 2

manufacturing resources such as machines, cutting tools, operators, etc. The task of

scheduling is to assign suitable resources to a batch of models to achieve global

optimization. A good schedule should utilize the full potential of resources while time

and cost should be as short and low as possible. Process planning is usually arranged

prior to scheduling in practice. The two functions could have different objectives and

there might be many routes to choose from, especially when the number of models is

large and the models are complex in terms of geometry and technical specifications.

On the other hand, process planning and scheduling are naturally linked. It will

simplify the multiple decision-making processes and provide a globally optimal

viewpoint if the two functions are well integrated.

In this paper, a novel approach has been developed to establish Collaborative

Process Planning and Scheduling (CPPS) in manufacturing planning. The Game

Theory-based strategies, a fuzzy logic algorithm, and the modern heuristic algorithms,

have been applied to solve the collaborative problem. Experiment results to verify the

effectiveness of the approach are presented.

The rest of the paper is organized as follows. In Section 2, the related work is

reviewed. In Section 3, the CPPS problem is modelled. In Section 4, discussions on

the application of the game theory for the CPPS problem are given. Section 5 presents

the constraint representation and handling. In Section 6, intelligent algorithms to solve

the CPPS problem are introduced. Experiment results are presented in Section 7.

Section 8 concludes the work.

2

Related Work

To develop a collaborative working environment is an important research area in

computer-based applications. To enable it, modern computing and artificial

intelligence technologies have been widely used [Schrum, 06] [Tomek, 01] [Lukosch,

08] [Wurdel, 08].

In the past decade, a few of research works have been reported to integrate

process planning and scheduling to optimize decisions. Some earlier works of the

integration strategy have been summarized in [Tan, 00]. The most recent works are

summarized in [Zhang, 03] [Li, 07] according to two general categories: the

enumerative approach and the simultaneous approach. In the enumerative approach

([Tonshoff, 89] [Huang, 95] [Aldakhilallah, 99] [Sormaz, 03] [Zhang, 03]), multiple

alternative process plans are first generated for each part. A schedule can be

determined by iteratively selecting a suitable process plan from alternative plans of

each part to replace the current plan until a satisfactory performance is achieved. The

simultaneous approach ([Li, 07] [Moon, 02] [Kim, 03] [Yan, 03] [Moon, 05] [Zhang, 05])

is based on the idea of finding a solution from the combined solution space of process

planning and scheduling. In this approach, the process planning and scheduling are

both in dynamic adjustment until specific performance criteria can be satisfied.

Although this approach is more effective and efficient in integrating the two

functions, it also enlarges the solution search space significantly. For this complex

decision-making process, further studies are still required, especially in complex

situations. To minimize the research gap, in this paper, research has been carried out

from the following three aspects:

1908

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 3

• It is imperative to develop a strategy to make the two functions to work together in

a more cooperative way, that is, CPPS. With the CPPS strategy, different

objectives can be prioritized flexibly, and the two functions can be adjusted in a

cooperative way to meet both of the targets. In this research, game theory, which

is the formal study of decision-making processes where several players (e.g.,

functions) make choices that potentially affect the interests of each other, has been

introduced to analyze the cooperation of the functions in a systematic way;

• In practical situations, it might be impossible to satisfy all constraints in a process

plan. For example, a high accuracy hole as a datum surface should be machined

with a high priority according to the primary surfaces constraint, but it may be in

conflict with the constraint of planes prior to holes and slots. Therefore, a fuzzy

logic-based Analytical Hierarchical Process (AHP) technique has been applied to

handle the complex constraints effectively;

• The complexity of manufacturing planning brings forth a vast search space when

identifying good solutions. Three modern heuristic algorithms, i.e., Particle

Swarm Optimization (PSO), Simulated Annealing (SA), and Genetic Algorithm

(GA), have been developed and benchmarked in this research to facilitate the

search process with optimal or near-optimal solutions. Essential performance

criteria, such as makespan, the balanced level of machine utilization, job tardiness

and manufacturing cost, have been defined in the algorithms to address the various

practical requirements.

3

Modelling of CPPS

The CPPS problem can be defined as follows:

•

Given a set of design models, each with a number of operations and set-up plans,

to be processed on a set of manufacturing resources (machines and tools) in a job

shop floor;

•

Alternative process plans and schedules can be generated through process

planning and scheduling flexibility strategies [2]. The processing planning

flexibility refers to the possibility of performing an operation on alternative

machines with alternative tools or set-up plans, and the possibility of

interchanging the sequence in which the operations are executed. The scheduling

flexibility corresponds to the possibility of generating alternative schedules for

jobs by arranging the different sequences of parts to be machined [2];

•

Through selecting suitable manufacturing resources and sequence the operations,

process plans and schedules, in which constraints among operations are satisfied

and pre-defined objectives are achieved, can be generated.

This problem is illustrated in Figure. 1. For example, there are 3 parts that can be

machined by 3, 2 and 3 operations on 3 machines, respectively. For different parts,

there are constraints among the operations to make them (Part1: Oper1 → Oper2 →

Oper3; Part2: Oper4 → Oper5; Part3: Oper6 → Oper7 → Oper8). When all these 8

operations are sequenced as (Oper1 → Oper4 → Oper2 → Oper6 → Oper3 → Oper7

→ Oper8 → Oper5 as shown in Figure 1) and manufacturing resources (machine, tool

and set-ups) are specified, the schedule can be determined accordingly. The CPPS

1909

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 4

problem is to optimize the operation sequence and select the manufacturing resources

so as to achieve the optimal or near-optimal process planning and scheduling

objectives while maintain the manufacturing feasibility with the satisfactory of

constraints.

Figure 1: Illustration of the CPPS problem

The CPPS problem can be modeled as an extension of the operation sequencing

optimization problem relating to a single model [Li, 02] [Guo, 06] into multiple

models with the CPPS objectives. When the process plans of all models are generated

and the manufacturing resources are specified, it is required to determine the schedule

based on this information and calculate the makespan, total tardiness, etc. Here, four

evaluation criteria of the CPPS problem can be calculated as follows.

m

=

(1) Makespan:

)_ ].[(

1

timeAvailablejMachine Max

j=

Makespan

.

(2) Total job tardiness: The due date of a part is denoted as DD , and the completion

moment of the part is denoted as CM . Hence,

_

OtherwiseDDCM

⎩

(3) Balanced level of machine utilization: the Standard Deviation concept is

introduced here to evaluate the balanced machine utilization (assuming there are

m machines, and each machine has n operations).

n

OperationnUtilizatioj Machine

1

0

−

CMthan later isDD if

Tardiness Part

⎨

⎧

=

∑

=

i

=

T Mac

].

i

)_[(]. [

, (

mj

,.., 1

=

)

1910

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 5

m

n Utilizatio

].

jMachine

m

j∑

=

=

1

)[(

χ

∑

=

j

−=

m

n Utilizatio

].

jMachine LevelnUtilizatio

1

2

)[(_

χ

(4) Manufacturing cost for the process plan of a part: In [Guo, 06], the manufacturing

cost associated with the process plan of a part has been defined in terms of machine

utilization, tool utilization, set-up changes, machine changes and tool changes. The

relevant computations are elaborated in [Guo, 06].

4

Applications of Game Theory on CPPS

Game theory is a good tool to analyze the interaction and cooperation of decision

makers with various objectives [Rasmusen, 01] [Xiao, 05]. For example, economists

have used it as a tool to examine the actions of firms in a market. Recently, it has

been applied to some complex engineering problems, such as communications and

networks, power systems, collaborative product design, etc. Game theory consists of a

series of strategies that are applicable for various situations. Here, three popular

strategies in the game theory have been applied to CPPS, i.e., Pareto strategy, Nash

strategy and Stackelberg strategy. The concepts for the three strategies are briefed

below.

•

Pareto strategy. A full cooperative solution between two players. Players in the

game theory can represent a person, a team or a functional module. The strategy

is to combine the objectives of two players as a single goal through weights.

•

Nash strategy. Each player must make a set of decisions that is rational to

him/her by assuming another player’s reaction. If there is an overlap between

these players’ reactions, the result can be selected from the overlap.

•

Stackelberg strategy. A leader-follower solution, which is well suitable for a

situation in which one player dominates the decision-making process.

For the CPPS problem, the objectives of process planning and scheduling need to

be considered from the cooperative point of view to achieve a balanced and overall

target. In many cases, objectives from process planning and scheduling could be

conflicted. For example, a lower manufacturing cost for making a part can be

achieved through the intensive utilization of cheap machines, but it could be

conflicted with the criterion for the balanced utilization of machines. Through

applying the above three game theory-strategies, the solution of CPPS is flexible and

adjustable according to various practical situations and users’ specific requirements.

The application of the Pareto strategy is to combine the objectives from process

planning and scheduling respectively with weights. The strategy is illustrated in

Figure 2(a). The major characteristic of the strategy is that the objective of process

planning is closely associated with that of scheduling. With the combined

consideration, the strategy equals a single level decision-making process so that

iteratively empirical process can be avoided. However, a serious problem is that it is

difficult to determine a reasonable combination weight with engineering meanings.

1911

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 6

Therefore, the strategy is more suitable for the purpose of comparison and trend

studies.

A usual practice to use the Nash strategy in the CPPS problem is to apply the

following procedure to the two functions. Process planning (or scheduling) is invoked

to produce a number of alternative plans with the satisfaction of the process planning

(or scheduling) objective and constraints, from which scheduling (or process

planning) can choose and further decide a group of satisfactory solutions (denoted as

Solutions) according to the scheduling (or process planning) objective. The

overlapped set of the above two Solutions is the final solution of the CPPS problem.

The process is illustrated in Figure 2(b). The strategy is characterized as a more

independent decision-making process for each functional module, and both the

objectives can be considered in a reasonable way. The Nash strategy is the same

effect as the Pareto strategy when the objectives of process planning and scheduling

are harmonious. When the objectives are contradictive, the results of the Nash

strategy is more rational compared with that of the Pareto strategy, which much

depends upon the setting of the weight.

In the application of the Stackelberg strategy to the CPPS problem, for the

dominate function (process planning or scheduling), a number of alternative plans

with the satisfaction of the function’s objective and constraints are generated, from

which another function can choose and further decide a satisfactory solution

(illustrated in Figure 2(c)). This strategy is different from the Nash strategy in that the

latter creates a larger computation space while the computation of the former is

mainly constrained by the dominant function. The characteristic of the Stackelberg

strategy is that it can fully satisfy the most important objective while the minimum

conditions of other objectives can be met. However, the value of one function could

be discounted in another module. For instance, to schedule parts based on generated

process plans sometimes causes some machines to be overloaded to restrict the

capabilities of the machines.

In this research, the CPPS model is equipped with the above three strategies for

users to choose to meet their requirements.

5

Handling of Constraints in CPPS

Manufacturing processes are complex [Kalpakjian, 03]. There are many technical

specifications and requirements. In CPPS, a number of constraints, which arise from

geometric shapes of parts, technical restrictions, best practices, etc., are represented.

A feasible solution of CPPS must comply with the constraints. These constraints can

be summarized below [Ding, 05].

(1) Precedence constraints

•

A parent feature should be processed before its child features.

•

Rough machining operations should be done before semi-finish and finish

machining operations.

•

Primary surfaces should be machined prior to secondary surfaces. Primary

surfaces are usually defined as surfaces with high accuracy or having a high

1912

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 7

impact on the design specifications, such as a datum plane. The rest of the

surfaces are regarded as secondary surfaces, e.g., a threaded hole.

Planes should be machined prior to holes and slots.

Edge cuts should be machined last.

•

•

(2) Successive constraints

•

Features or operations, which can be machined within the same set-up should be

machined successively.

•

Features to be machined with the same cutting tool should be machined

successively.

•

Operations of the same type, such as rough, semi-finish and finish machining,

should be executed successively.

•

Features with similar tolerance requirements should be machined successively on

the same machine tool.

(3) Auxiliary constraints

•

Annealing, normalizing and ageing operations of ferrous metal component should

be arranged before rough machining or between rough and semi-finish

machining.

•

Quenching for ferrous metal workpieces should be arranged between semi-finish

and finish machining or between rough and semi-finish machining if it is

followed by high temperature tempering.

•

Quenching for non-ferrous metals should be arranged between rough and semi-

finish machining or before rough machining.

•

Carburizing would be arranged between semi-finish and finish machining.

To address the complexity of constraints, the AHP technique [Golden, 89], which

specifies a set of fuzzy logic-based numerical weights to represent the relative

importance of the constraints of CPPS with respect to a manufacturing environment,

has been applied to evaluate the satisfaction degree of the constraints. The relevant

computation is depicted below.

PP

S

f(PP+S)

Solutions

PP

S

PP

S

Solutions

PP

S

Solutions

S

PP

(a) Pareto strategy (b) Nash strategy (c) Stackelberg strategy

Figure 2: Illustration of the game theory strategies

PP – Process planning ; S - Scheduling

1913

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 8

Step 1: The constraints are organized in a hierarchy structure, which includes an

overall objective (Level 1), three general constraint groups (Level 2) and

rules under each constraint group (Level 3). This situation is illustrated in

Figure 3. For Level 2, a

33× pair-wise matrix (R0-matrix) is created,

where the number in the ith row and jth column, rij, specifies the relative

importance of the ith group of constraints as compared with the jth group

of constraints. For Level 3, three pair-wise matrices are created for each

group of constraints (R1-matrix (

matrix (

4

Auxiliary constraints). Similarly, the number in the matrix (rij) specifies

the relative importance of rules within each category of constraints. A R-

matrix can be described as:

⎡

i

rr

....

55× ) for Precedence constraints, R2-

4× ) for Succession constraints, and R3-matrix (

33× ) for

⎥

⎦

⎥

⎥

⎥

⎥

⎥

⎤

⎢

⎣

⎢

⎢

⎢

⎢

⎢

=

mm

r

mi

r

m

im

r

.

ii

.

i

.

m

.

r

rr

r

R

..

..

..

..

1

1

1111

where i = 1, 2, …, m (m is the number of groups of constraints in Level 2

or the number of rules for each constraint group in

Level 3),

rii = 1, and

rij = 1/rji.

Evaluating criteria based on a 1-9 scale for the R-matrices, which are used

to indicate the relative importance of two elements, are defined in Table 1.

In order to get more neutral results, a group of experts is invited to fill in

the four R-matrices according to their experience and knowledge.

Step 2:

For instance, considering two rules in the category of Precedence constraints

- Rule 2 and Rule 4:

Rule 2: Primary surfaces should be machined prior to secondary surfaces.

Rule 4: Planes should be machined prior to holes and slots.

From the perspective of an individual expert, if he thinks Rule 2 is much

more important than Rule 4, a weight of '7' is inserted in the juncture cell (r24) of

his filled R1-matrix. On the contrary, the value in the juncture cell (r42) is set to

'1/7'.

Step 3: For Level 2 and Level 3, four weight vectors

to the four R-matrices respectively, are computed. The computation

process consists of the following three steps.

(1) Multiplication ( M ) of all elements in each row of a R-matrix is

computed as:

30

WW

−

, which correspond

∏

=

j

=

n

iji

rM

1

1914

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 9

where

j is the column index of elements, j =1, 2, …, n,

i is the index row of elements, i =1, 2, …, n, and

n is the number of the rows (columns) in a R-matrix.

R

An overview

objective

Precedence

constraints

Succession

constraints

Auxiliary

constraints

R1

1 R1

2 R1

3 R1

4 R1

5 R2

1 R2

2 R2

3 R2

4

R3

1 R3

2 R3

3

5 rules under

precedence

constraints

4 rules under

succession

constraints

3 rules under

auxiliary

constraints

Level 1

Level 2

Level 3

⎥

⎦

⎥

⎥

⎤

⎢

⎣

⎢

⎢

⎡

=

33 3231

23 22 21

1312 11

0

rrr

rrr

rrr

R

⎥

⎦

⎥

⎥

⎥

⎥

⎥

⎤

⎢

⎣

⎢

⎢

⎢

⎢

⎢

⎡

=

5554535251

45 4443 4241

35 3433 3231

25 2423 2221

1514 1312 11

1

rrrrr

rrrrr

rrrrr

rrrrr

rrrr

r

⎥⎥

⎦

⎥

⎥

⎥

⎤

⎢⎢

⎣

⎢

⎢

⎢

⎡

=

44 4342 41

343332 31

2423 2221

141312 11

2

rrrr

rrrr

rrrr

rrrr

R

⎥

⎦

⎥

⎥

⎤

⎢

⎣

⎢

⎢

⎡

=

333231

23 2221

13 1211

3

rrr

rrr

rrr

R

Figure 3: A three-level hierarchy structure for the constraints

Table 1: Evaluation criteria for R-matrices

Definition Intensity of

importance (rij)

Intensity of

importance (rji)

The ith rule and the jth rule

have equal importance

1 1

The ith rule is slightly more

important than the jth rule

3 1/3

The ith rule is more important

than the jth rule

5 1/5

The ith rule is much more

important than the jth rule

7 1/7

The ith rule is absolutely more

important than the jth rule

9 1/9

Intermediate values between

adjacent scale values

2, 4, 6, 8 1/2, 1/4, 1/6, 1/8

1915

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 10

(2) The nth root of M is calculated, that is:

n

ii

Mw =

where i is the row (column) number in a R-matrix, and i =1, 2, …, n.

Therefore, the relative importance weight vector can be built as follows:

n

wwwW

...,,,

21

=

Each element of the weight vector W

through a normalization operation.

) ...,,,(

21

n

www

is finally generated

∑

=

j

=

n

j

i

i

w

1

w

w

For each W , it should be eventually denoted as

individual computation process.

Step 4: There are totally 12 rules defined in this system (5 rules from Precedence

constraints + 4 rules from Succession constraints + 3 rules from Auxiliary

constraints). The element of a total weight vector for each rule -

ttt

www

can be generated as:

30

WW −

according to the

t

W

) ...,,,(

1221

51

1

1

0

51

*

−− =

wwwt

,

41

2

2

0

96

*

−− =

wwwt

,

31

3

3

0

12

−

10

*

−

=

wwwt

Step 5: A series of V-matrices are designed to record the situation of violating

constraints for a process plan. For instance, for Rule k, its V-matrix is

defined as:

⎡

ikk

vv

....

⎥

⎦

⎥

⎥

⎥

⎥

⎥

⎤

⎢

⎣

⎢

⎢

⎢

⎢

⎢

=

knnkni kn

kin

.

kii

.

ki

.

nk

k

vvv

vvv

v

V

..

..

..

.

..

1

1

1111

where n is the number of operations in a process plan,

prior to i Operation if

0

ijji

vv

−=1

.

The value to evaluate the manufacturability of a process plan is

determined.

m

mnn

km

wf

1 11

where m is the total rule number of the constraints (here m =12).

k Ruleobey jOperation prior to i Operation if

k Rule against is jOperation 1

{

=

ij

v

, and

Step 6:

f is finally calculated as:

∑ ∑ ∑

= =

ki

=

=

j

kij

t

v

6

Applications of Modern Heuristic Algorithms

The CPPS problem usually brings forth a vast search space. Conventional algorithms

are often incapable of optimizing non-linear multi-modal functions. To address this

problem effectively, some modern optimization algorithms, such as GA and SA, have

1916

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 11

been developed recently to quickly find a solution in a large search space through

some evolutional or heuristic strategies. In this research, three modern algorithms, i.e.,

PSO, SA and GA, have been applied to facilitate the search process. In [Li, 02] [Guo,

06], the three algorithms have been successfully applied to process planning

optimization problems. Here, the algorithms have been developed further to solve the

CPPS problem. The application of an improved PSO process is explained here for

illustration. More details of SA and GA can refer to [Li, 02] [Li, 07].

A standard PSO algorithm was inspired by the social behavior of bird flocking

and fish schooling [Kennedy, 95]. Three aspects will be considered simultaneously

when an individual fish or bird (particle) makes a decision about where to move: (1)

its current moving direction (velocity) according to the inertia of the movement, (2)

the best position that it has achieved so far, and (3) the best position that its neighbor

particles have achieved so far. In the algorithm, the particles form a swarm and each

particle can be used to represent a potential solution of a problem. In each iteration,

the position and velocity of a particle can be adjusted by the algorithm that takes the

above three considerations into account. After a number of iterations, the whole

swarm will converge at an optimized position in the search space.

A traditional PSO algorithm can be applied to optimize CPPS in the following

steps:

(1) Initialization

• Set the size of a swarm, e.g., the number of particles “Swarm_Size” and the

max number of iterations “Iter_Num”.

• Initialize all the particles (a particle is a CPPS solution) in a swarm.

Calculate the corresponding criteria of the particles (a result is called fitness

here).

• Set the local best particle and the global best particle with the best fitness.

(2) Iterate the following steps until Iter_Num is reached

• For each particle in the swarm, update its velocity and position values.

• Decode the particle into a CPPS solution in terms of new position values and

calculate the fitness of the particle. Update the local best particle and the

global best particle if a lower fitness is achieved.

(3) Decode global best particle to get the optimized solution

However, the traditional PSO algorithm introduced above is still not effective in

resolving the operation sequencing problem. There are two major reasons for it:

•

Due to the inherent mathematical operators, it is difficult for the traditional PSO

algorithm to consider the different arrangements of machines, tools and set-ups

for each operation, and therefore the particle is unable to fully explore the entire

search space.

•

The traditional algorithm usually works well in finding solutions at the early

stage of the search process (the optimization result improves fast), but is less

efficient during the final stage. Due to the loss of diversity in the population, the

particles move quite slowly with low or even zero velocities and this makes it

hard to reach the global best solution. Therefore, the entire swarm is prone to be

trapped in a local optimum from which it is difficult to escape.

1917

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 12

To solve these two problems and enhance the ability of the traditional PSO

algorithm to find the global optimum, new operations, including mutation, crossover

and shift, have been developed and incorporated in an improved PSO algorithm.

Meanwhile, considering the characteristics of the algorithm, the initial values of the

particles have been well planned. Some modification details are depicted below.

(1) New operators in the algorithm

•

Mutation. In this strategy, an operation is first randomly selected in a

particle. From its candidate machining resources (machines, tools, set-ups),

an alternative set (machine, tool, set-up) is then randomly chosen to replace

the current machining resource in the operation.

•

Crossover. Two particles in the swarm are chosen as Parent particles for a

crossover operation. In the crossover, a cutting point is randomly

determined, and each parent particle is separated as left and right parts of the

cutting point. The positions and velocities of the left part of Parent 1 and the

right part of Parent 2 are reorganized to form Child 1. The positions and

velocities of the left part of Parent 2 and the right part of Parent 1 are

reorganized to form Child 2.

•

Shift. This operator is used to exchange the positions and velocities of two

operations in a particle so as to change their relative positions in the particle.

(2) Escape method

•

During the optimization process, if the iteration number of obtaining the

same best fitness is more than 10, then the mutation and shift operations are

applied to the best particle to try to escape from the local optima.

7

Experimental Results

A group of 8 parts taken from [Li, 02] [Guo, 06] have been used for experiments. The

relevant specifications of the parts are given in Table 2. The results of the following

two conditions are taken first to demonstrate the performances of the chosen criteria:

(1) The criteria are manufacturing cost and makespan according to the Pareto

strategy.

(2) The criteria are manufacturing cost and the balanced utilization of machines

according to the Pareto strategy.

All of the results are prone to stabilization after several hundreds of iterations.

Figure 4 indicates clearly that the manufacturing cost and the makespan follow the

similar trends since the reduced numbers of set-ups, machine changes, and tool

changes contribute to both of the lower manufacturing cost and the shorter makespan.

Therefore, the effects of the three game theory strategies are the same. For the

situation with conflicting objectives of CPPS like Figure 5, when the Stackelberg

strategy is applied, the satisfactory results are within the highlighted region A (The

balanced utilization of machines is the leader criteria. Higher values means

unbalanced level) or B (Manufacturing cost is the leader criteria). When the Nash

strategy is applied, the satisfactory results are within the highlighted region C, and

both the objectives, i.e., manufacturing cost and the balanced utilization of machines,

1918

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 13

are discounted. Therefore, the developed method provides the flexibility to choose the

suitable strategy according to the real practical requirement. The three optimization

algorithms have been further compared under the same condition (the above

Condition 1). To make the diagrams clearly, only the makespan has been chosen and

the results are shown in Figure 6.

Table 2: The technical specifications for 8 parts

Part Number of operations Number of constraints

1 7 (9, 9, 27, 8, 8, 9, 36) 11

2 8 (9, 9, 36, 18, 27, 8, 27, 18) 11

3 7 (9, 9, 36, 36, 18, 6, 6) 10

4 9 (9, 9, 27, 6, 36, 36, 6, 18, 18) 18

5 7 (9, 9, 36, 36, 36, 18, 6) 13

6 9 (9, 9, 36, 27, 18, 6, 27, 6, 18) 20

7 5 (9, 27, 27, 18, 9) 5

8 7 (9, 9, 27, 36, 36, 6, 6) 13

Figure 4: Case 1 of applying three strategies

1919

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 14

Figure 5: Case 2 of applying three strategies

Figure 6: Comparisons of three algorithms

It can be observed that all of the approaches can reach good results, while there

are different characteristics due to the inherent mechanisms of the algorithms. The

SA-based algorithm usually takes shorter time to find good solutions but it is vigilant

to its parameters (such as the starting temperature and the cooling parameter) and the

problems to be optimized. The GA- and PSO-based algorithms are slow in finding

good solutions but they are robust for optimization problems. Meanwhile, the SA-

based approach is much “sharper” to find optimal or near-optimal solutions, and the

common shortcoming of the GA- and PSO-based approaches is that they are prone to

1920

Li W., Gao L., Li X.: Application of Intelligent Strategies ...

Page 15

pre-maturity in some cases (converge too early and difficult to find the optimal

solutions).

8

Conclusions

Manufacturing planning, which mainly include process planning and scheduling, is an

important stage in product development. The decision will play a crucial role for the

performance of the final products. Usually, process planning and scheduling are

arranged in a sequential way. With this arrangement, it is difficult to adjust them in a

cooperative way to achieve global optimization. To identify good solutions in

manufacturing planning, in this research, CPPS has been developed. The

contributions of this research include:

• To address CPPS effectively, three game theory-based strategies, i.e., Pareto

strategy, Nash strategy and Stackelberg strategy, have been used to analyze and

facilitate the cooperation of the two processes in a systematic way.

• AHP has been introduced to resolve the multiple constraints in the CPPS problem.

The technique is effective in solving the complex and even conflicting constraints

in manufacturing planning.

• To find optimal or near-optimal solutions from the vast search space efficiently,

modern intelligent algorithms, including PSO, SA and GAs, have been developed

and applied to the CPPS problem. Experiments have been conducted and

computational results have shown the effectiveness of applying these intelligent

strategies. Comparisons have been given to show the characteristics of the

algorithms.

Acknowledgements

This research is supported by an applied research fellowship from the Coventry

University for international collaboration, and the National Basic Research Program

of China (973 Program) under Grant no.2004CB719405, the National High-Tech

Research and Development Program of China (863 Program) under Grant

nos.2007AA04Z107 and 2006AA04Z131 from the Huazhong University of Science

and Technology.

References

[Aldakhilallah, 99] Aldakhilallah, K.A., Ramesh, R.: Computer-Integrated Process Planning

and Scheduling (CIPPS): Intelligent Support for Product Design, Process Planning and Control,

International Journal of Production Research, vol. 37, no. 3 (1999), 481-500.

[Ding, 05] Ding, L., Yue, Y., Ahmet, K., Jackson, M., Parkin, R.: Global optimization of a

feature-based process sequence using GA and ANN techniques, International Journal of

Production Research, vol. 43 (2005), 3247-3272.

[Golden, 89] Golden, B.L., Harker, P.T., Wasil, E.E.: The Analytic Hierarchy Process:

Applications and Studies. Springer-Verlag, 1989.

1921

Li W., Gao L., Li X.: Application of Intelligent Strategies ...