Computers & Operations Research 37 (2010) 656 -- 667

Contents lists available at ScienceDirect

Computers & Operations Research

journal homepage: w w w . e l s e v i e r . c o m / l o c a t e / c o r

Mathematical modeling and evolutionary algorithm-based approach for integrated

process planning and scheduling
Xinyu Li, Liang Gao ∗ , Xinyu Shao, Chaoyong Zhang, Cuiyu Wang
The State Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan, Hubei, China

A R T I C L E I N F O A B S T R A C T

Available online 21 June 2009 Traditionally, process planning and scheduling were performed sequentially, where scheduling was im-
plemented after process plans had been generated. Considering their complementarity, it is necessary to
Keywords: integrate these two functions more tightly to improve the performance of a manufacturing system greatly.
Integrated process planning and scheduling In this paper, a mathematical model of integrated process planning and scheduling has been formulated.
Scheduling
And, an evolutionary algorithm-based approach has been developed to facilitate the integration and opti-
Mathematical model
mization of these two functions. To improve the optimized performance of the approach, efficient genetic
Evolutionary algorithm
representation and operator schemes have been developed. To verify the feasibility and performance of
the proposed approach, experimental studies have been conducted and comparisons have been made be-
tween this approach and some previous works. The experimental results show that the integrated process
planning and scheduling is necessary and the proposed approach has achieved significant improvement.
© 2009 Elsevier Ltd. All rights reserved.

1. Introduction usually assigned on it without considering the competition for

the resources from other jobs [4]. This may lead to the process
Process planning and scheduling used to link product design and planners favoring to select the desirable machines for each job
manufacturing are two of the most important functions in a man- repeatedly. Therefore, the generated process plans are some-
ufacturing system. A process plan specifies what manufacturing re- what unrealistic and cannot be readily executed on the shop
sources and technical operations/routes are needed to produce a floor for a group of jobs [5]. Accordingly, the resulting optimal
product (a job). The outcome of process planning includes the iden- process plans often become infeasible when they are carried out
tification of machines, tools and fixtures suitable for a job, and the in practice at the later stage.
arrangement of operations for a job. Typically, a job may have one (2) Scheduling plans are often determined after process plans. In the
or more alternative process plans. With the process plans of jobs as scheduling phase, scheduling planners have to consider the de-
inputs, a scheduling task is to schedule the operations of all the jobs termined process plans. Fixed process plans may drive schedul-
on machines while precedence relationships in the process plans are ing plans to end up with severely unbalanced resource load and
satisfied. Although as mentioned above, there is a close relationship create superfluous bottlenecks.
between process planning and scheduling, the integration of them (3) Even though process planners consider the restriction of the cur-
is still a challenge in research and applications [1]. rent resources on the shop floor, the constraints in the process
In traditional approaches, process planning and scheduling were planning phase may have already changed due to the time de-
carried out in a sequential way. Scheduling was conducted after lay between the planning phase and execution phase. This may
the process plan had been generated. Those approaches have be- lead to the infeasibility of the optimized process plan [6]. Inves-
come an obstacle to improve the productivity and responsiveness tigations have shown that 20–30% of the total production plans
of manufacturing systems and to cause the following problems in in a given period have to be rescheduled to adapt to dynamic
particular [2,3]: change in a production environment [3].
(4) In most cases, both for process planning and scheduling, a sin-
(1) In practical manufacturing, process planner plans jobs individu- gle criterion optimization technique is used to determine the
ally. For each job, manufacturing resources on the shop floor are best solution. However, the real production environment is best
represented by considering more than one criterion simulta-
neously [3]. Furthermore, the process planning and scheduling
∗ Corresponding author. Tel.: +86 27 87543871 85; fax: +86 27 87543074. may have conflicting objectives. Process planning emphasizes
E-mail address: [email protected] (L. Gao). the technological requirements of a job, while scheduling

0305-0548/$ - see front matter © 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cor.2009.06.008
 X. Li et al. / Computers & Operations Research 37 (2010) 656 -- 667 657

attaches importance to the timing aspects and resource sharing 2.2. Closed loop process planning
of all jobs. If there is no appropriate coordination, it may create
conflicting problems. The methodology of CLPP is using a dynamic process planning
system with a feedback mechanism. CLPP can be used to gener-
To overcome these problems, there is an increasing need for deep ate real-time process plans by means of a dynamic feedback from
research and application of integrated process planning and schedul- scheduling system. The process planning mechanism generates pro-
ing (IPPS) system. The IPPS introduces significant improvements cess plans based on available resources. Scheduling system provides
to the efficiency of manufacturing through eliminating or reduc- the information about which machines are available on the shop
ing scheduling conflicts, reducing flow-time and work-in-process, floor for an incoming job to the process planning system, so that
improving production resources utilizing and adapting to irregular every plan is feasible with respect to the current availability of pro-
shop floor disturbances [5]. Without IPPS, a true computer integrated duction facilities. This dynamic simulation system can enhance the
manufacturing system (CIMS), which strives to integrate the various real-time, intuition and manipulability of process planning system,
phases of manufacturing in a single comprehensive system, may not and the utilization of alternative process plans.
be effectively realized. Usher and Fernandes [4] divided the dynamic process planning
The remainder of this paper is organized as follows. Section 2 into the static phase and the dynamic phase. Seethaler and Yellowley
introduces a literature survey of IPPS. Problem formulation and [22] presented a dynamic process planning system which can give
mathematical modeling is discussed in Section 3. An evolutionary the process plans based on the feedback of scheduling system.
algorithm (EA)-based approach for IPPS is proposed in Section 4.
Experimental studies and discussions are reported in Section 5. 2.3. Distributed process planning
Section 6 is conclusion.
The methodology of DPP is to perform both the process planning
2. Literature survey and scheduling simultaneously with a hierarchical approach. It di-
vides the process planning and scheduling tasks into two phases. The
In the early studies of CIMS, it had been identified that IPPS is first phase is the initial planning phase. In this phase, the character-
very important to the development of CIMS [7,8]. The preliminary istics of parts and the relationship between the parts are analyzed,
idea of IPPS was first introduced by Chryssolouris and Chan [9,10]. and the primary process plans are determined at this stage as well.
Beckendorff [11] used alternative process plans to improve the flex- The process resources are also evaluated simultaneously. The sec-
ibility of manufacturing systems. Khoshnevis and Chen [12] intro- ond phase is the detailed planning phase. In this phase, the process
duced the concept of dynamic feedback into IPPS. The integration plans are adjusted to the current status of shop floor. The detailed
model which was proposed by Zhang [13] and Larsen [14] extended process plans and scheduling plans are obtained simultaneously.
the concepts of alternative process plans and dynamic feedback and Kempenaers and Pinte [23] demonstrated three modules of the
defined an expression to the methodology of hierarchical approach. collaborative process planning system. Wu and Fuh [24] gave the
Some earlier works of the integration strategy had been summarized integration model of process planning and scheduling in the dis-
in [7]. In the recent years, in the area of IPPS, several models have tributed virtual manufacturing environment. Zhang and Gao [25]
been reported, and they can be classified into three basic models presented a framework of concurrent process planning based on
based on IPPS: nonlinear process planning (NLPP), closed loop pro- Holon. Wang and Song [26] proposed a framework of collaborative
cess planning (CLPP) and distributed process planning (DPP). process planning system supported by a real-time monitoring sys-
tem.
2.1. Nonlinear process planning In this research, a mathematical model of IPPS is proposed based
on the methodology of NLPP. In this model, the process plan of every
The methodology of NLPP is to give all alternative plans for each job and the scheduling plan are generated simultaneously. The se-
part with a rank according to process planning optimization criteria. lection of process plan and generation of scheduling plan are guided
The plan with highest priority is always ready for submission when by the objectives (such as minimizing makespan or cost). And one
the job is required. If the first-priority plan is not suitable for the evolutionary algorithm-based approach is developed to solve this
current shop floor status, the second-priority plan will be provided model.
to the scheduling system.
NLPP is the most basic model of IPPS. Jain [15] proposed a scheme 3. Problem formulation and mathematical modeling
for IPPS that can be implemented in a company with existing process
planning and scheduling departments when multiple process plans 3.1. Problem formulation
for each part type are available. Because the integration methodology
of this model is very simple, most of the current researches on the The IPPS problem can be defined as follows [27]:
integration model focus on the implementation and improvement of
Given a set of N parts which are to be processed on machines with
this model. Kim and Song [16] gave a scheduling system which was
operations including alternative manufacturing resources, select
supported by flexible process plans and based on negotiation. Lee
suitable manufacturing resources and sequence the operations so
and Kim [5] presented the NLPP model which was based on the ge-
as to determine a schedule in which the precedence constraints
netic algorithm (GA). Yang and Parsaei [17] presented a prototype of
among operations can be satisfied and the corresponding objec-
a feature-based alternative process planning system. Kim and Park
tives can be achieved.
[18] used a symbiotic evolutionary algorithm for the integrated pro-
cess planning and job shop scheduling. Li and McMahon [19] used In the manufacturing systems considered in this study, a set of
a simulated annealing (SA)-based approach for IPPS problem. Shao process plans of each part is designed and maintained. The genera-
and Li [20] used a modified genetic algorithm to solve this problem. tion of one scheduling plan and the selection of process plan of each
However, through a number of experimental computations, Usher job from a set of process plans are determined based on the min-
[21] concluded that the advantages gained by increasing the num- imum objectives. If there are N jobs, and each with Gi alternative
ber of alternative process plans for a scheduling system diminishes process plans, the number of possible process plan combinations is
rapidly when the number of the plans reaches a certain level. {G1 × G2 × · · · × Gi × · · · × GN }. This problem is a NP-hard problem.
658 X. Li et al. / Computers & Operations Research 37 (2010) 656 -- 667


Three types of flexibility are considered in process planning 1 the operation oijl precedes the operation opqs on machine k
Yijlpqsk =
[2,28]: operation flexibility, sequencing flexibility and processing 0 otherwise
flexibility [29]. Operation flexibility which is also called routing 
flexibility [30] relates to the possibility of performing one opera- 1 if machine k is selected for oijl
Zijlk =
tion on alternative machines, with possibly distinct processing time 0 otherwise
and costs. Sequencing flexibility is decided by the possibility of
interchanging the sequence of the required operations. Processing Objectives:
flexibility is determined by the possibility of processing the same
manufacturing feature with alternative operations or sequences of (1) Minimizing the makespan which is the completion time of last
operations. Better performance in some criteria can be obtained by operation of all the jobs:
the consideration of these flexibilities [18].
Min makespan = Max{cijlk × Xil × Zijlk }
∀i ∈ [1, N], ∀j ∈ [1, Pil ], ∀l ∈ [1, Gi ], ∀k ∈ [1, M] (1)
3.2. Mathematical modeling
(2) Minimizing the total processing cost:
The mathematical mode of IPPS is defined here. The most pop-
Pil
ular criteria for scheduling include makespan, job tardiness and the 
N 
Min (vijlk × Xil × Zijlk )
balanced level of machine utilization, while manufacturing cost is
i=1 j=1
the major criterion for process planning [31]. In this paper, schedul-
ing is often assumed as job shop scheduling, and the mathematical ∀i ∈ [1, N], ∀j ∈ [1, Pil ], ∀l ∈ [1, Gi ], ∀k ∈ [1, M] (2)
model of IPPS is based on the mixed integer programming model of
(3) Minimizing the lateness:
the job shop scheduling problem (JSP) [32].
In order to solve this problem, the following assumptions are Li = ci − di ; Min Lmax = max{L1 , L2 , . . . , LN } ∀i ∈ [1, N] (3)
made:
(4) Minimizing the total weighted tardiness:

(1) Jobs are independent. Job preemption is not allowed and each Ti = max{ci − di , 0} = max{Li , 0}; Min i Ti ∀i ∈ [1, N] (4)
machine can handle only one job at a time. i
(2) The different operations of one job cannot be processed simul-
taneously. (5) Minimizing the weighted number of tardy jobs:
(3) All jobs and machines are available at time zero simultaneously. 
Min i Ui ∀i ∈ [1, N] (5)
(4) After a job is processed on a machine, it is immediately trans-
i
ported to the next machine on its process, and the transmission
time is assumed to be negligible. (6) Minimizing the total earliness plus the total tardiness:
(5) Setup time for the operations on the machines is independent of  
Ei = max{di − ci , 0}; Min Ei + Ti ∀i ∈ [1, N] (6)
the operation sequence and is included in the processing times.
i i

Based on these assumptions, the mathematical model of IPPS is

Subject to:
described as follows [33]:
The notations used to explain the model are described below:
(1) For the first operation in the alternative process plan l of job i:
N the total number of jobs; (ci1lk × Zi1lk × Xil ) + A(1 − Xil ) ⱖ (ti1lk × Zi1lk × Xil )
M the total number of machines;
di the due date of job i; ∀i ∈ [1, N], ∀l ∈ [1, Gi ], ∀k ∈ [1, M] (7)
i the weight of job i is basically a priority factor, denoting the (2) For the last operation in the alternative process plan l of job i:
importance of job i relative to the other jobs in the system;
Gi the total number of alternative process plans of job i; (ciPil lk × ZiPil lk × Xil ) − A(1 − Xil ) ⱕ makespan
oijl the jth operation in the lth alternative process plan of the job
i; ∀i ∈ [1, N], ∀l ∈ [1, Gi ], ∀k ∈ [1, M] (8)
Pil the number of operation in the lth alternative process plan of (3) The different operations of one job cannot be processed simul-
the job i; taneously:
k the alternative machine corresponding to oijl ;
tijlk the processing time of operation oijl on machine k, tijlk > 0; (cijlk × Zijlk × Xil ) − (ci(j−1)lk1 × Zi(j−1)lk1 × Xil ) + A(1 − Xil )
cijlk the earliest completion time of operation oijl on machine k;
ⱖ (tijlk × Zijlk × Xil )
ci the completion time of job i;
Li the lateness of job i; ∀i ∈ [1, N], ∀j ∈ [1, Pil ], ∀l ∈ [1, Gi ], ∀k, k1 ∈ [1, M] (9)
Ti the tardiness of job i;
(4) Each machine can handle only one job at a time:
Ei the earliness of job i;
vijlk the processing cost of operation oijl on machine k; (cpqsk × Zpqsk × Xps ) − (cijlk × Zijlk × Xil ) + A(1 − Xil )
A a very large positive number;
+ A(1 − Xps ) + A(1 − Yijlpqsk × Zijlk × Zpqsk × Xil × Xps )

1 if ci > di ⱖ (tpqsk × Zpqsk × Xps ) (10)
Ui = the unit penalty of job i
0 otherwise
(cijlk × Zijlk × Xil ) − (cpqsk × Zpqsk × Xps ) + A(1 − Xil ) + A(1 − Xps )

1 the lth alternative process plan of job i is selected + A(Yijlpqsk × Zijlk × Zpqsk × Xil × Xps ) ⱖ (tijlk × Zijlk × Xil )
Xil =
0 otherwise ∀i, p ∈ [1, N], ∀j, q ∈ [1, Pil,ps ], ∀l, s ∈ [1, Gi,p ], ∀k ∈ [1, M] (11)
 X. Li et al. / Computers & Operations Research 37 (2010) 656 -- 667 659

(5) Only one alternative process plan can be selected of job i: chromosome is the alternative process plan string (B-string). The
 positions from 1 to N in B-string represent the jobs from 1 to N.
Xil = 1 ∀i ∈ [1, N] (12) The number in the ith position represents the selected alternative
l process plan of the job i.
The second part of the chromosome is the scheduling plan string
(6) Only one machine for each operation should be selected:
(A-string). In this paper, the scheduling encoding principle is the

M operation-based representation, which is made up of the Genes (the
Zijlk = 1 ∀i ∈ [1, N], ∀j ∈ [1, Pil ], ∀l ∈ [1, Gi ] (13) first number is the job number and the second number is the al-
k=1 ternative machine). This representation uses an unpartitioned per-
mutation with Pil -repetitions of job numbers (the first numbers of
(7) There is only one precedence relation between two operations Genes) [36]. In this representation, each job number occurs Pil times
in a scheduling plan: in the chromosome. By scanning the chromosome from left to right,
the fth occurrence of a job number (the first number of Gene) refers
Yijlpqsk × Xil × Xps × Zijlk × Zpqsk ⱕ 1 (14) to the fth operation in the technological sequence of this job. The
important feature of the operation-based representation is that any
(Yijlpqsk × Xil ) ⱕ (Zijlk × Xil ) (15) permutation of the chromosome can be decoded to a feasible sched-
ule [36].
The first numbers of Genes represent a permutation of the op-
(Yijlpqsk × Xps ) ⱕ (Zpqsk × Xps ) erations. Different appearances of this number in the chromosome
represent different operations of the job, and the sequence of the
∀i, p ∈ [1, N], ∀j, q ∈ [1, Pil,ps ], ∀l, s ∈ [1, Gi,p ], ∀k ∈ [1, M] (16)
appearances of this number is the same with the sequence of the
Pil Gi o −1 operations of the job. The important feature of this representation

N   
pqsk

(Yijlpqsk × Xil × Zijlk ) = Zokm is that all offspring formed by crossover are feasible solutions [36].
i j l ok1 It is assumed that there are N jobs, and qi is the number of opera-
tions of the process plan which has the most operations among all
∀p ∈ [1, N], ∀q ∈ [1, Pil ], ∀s ∈ [1, Gi ], ∀k ∈ [1, M] (17) the alternative process plans of the job i. Then the length of the A-

opqsk −1 string is equal to qi . The number of i (the first number of Genes) in
where ok1 Zokm means the total number of operations before the A-string is equal to the number of operations of the lth selected
opqs on machine k; ok1 means the first operation on machine k; alternative process plan. Based on this principle, the composition el-
opqsk means the current operation on machine k. ements of A-string are determined. If the number of the elements
(8) The completion time of each operation should be either positive 
is less than qi , all the other elements are filled with (0,0). One A-
or zero. string is generated by arraying all the elements randomly. And the
cijlk × Zijlk × Xil ⱖ 0 B-string is generated by selecting the alternative process plan ran-
domly for every job.
∀i ∈ [1, N], ∀j ∈ [1, Pil ], ∀l ∈ [1, Gi ], ∀k ∈ [1, M] (18) The second numbers of Genes represent the alternative machines
of each operation. It denotes the selected machine set of the corre-
The objective functions are equations from (1) to (6). In this
sponding operations of all jobs. Assuming the hth operation in the
paper, the mono-objective problems are considered. It also can
selected lth alternative process plan of job i can be processed by a
use weighted-sums method or pareto strategy to solve the multi-
machine set Silh = {milh1 , milh2 , . . . , milhcilh }. The Gene can be denoted
objective problems. Moreover, the constraints are equations from
as (i, gilh ). The gilh is an integer between 1 and cilh and it means that
(7) to (18). Constraint (9) expresses that different operations of
the hth operation in the lth selected alternative process plan of job
a job are unable to be processed simultaneously. This is the con-
i is assigned to the gilh th machine milhgilh in Silh .
straint of different processes for a job. Constraints (10) and (11)
which are the constraints of machine express that each machine
can handle only one job at a time. The constraints from (7)–(11)
are called disjunctive constraints. Constraint (12) ensures that only 4.2. Initialization and fitness evaluation
one alternative process plan can be selected for each job. Constraint
(13) ensures that only one machine for each operation should be The encoding principle of the A-string in the chromosome in this
selected. Constraints from (14)–(17) ensure that there is only one paper is an operation-based representation. The important feature
precedence relation between two operations in a scheduling plan. of the operation-based representation is that any permutation of the
Constraint (18) ensures that the completion time of each operation chromosome can be decoded to a feasible schedule. It cannot break
should be either positive or zero. The mixed integer programming the constraints on precedence relations of operations [36]. The initial
model of IPPS developed above can be solved by using the mixed population is generated based on the encoding principle.
integer programming methods. JSP has been proved to be one of In this paper, it used makespan as its objective (see Eq. (1)).
the most difficult NP-complete problems [34]. The IPPS problem
which is also known as a NP-complete problem is more complicated
than JSP [35]. For large problems, it is difficult to find perfect solu- 4.3. Genetic operators
tions in reasonable time. Therefore, in this paper, one evolutionary
algorithm-based approach has been developed to solve it. (1) Selection: The tournament selection scheme has been used for
selection operator. In tournament selection, a number of individuals
4. Evolutionary algorithm-based approach for IPPS are selected randomly (dependent on the tournament size, typically
between 2 and 7) from the population and the individual with the
4.1. Representation best fitness is chosen. The tournament selection approach allows
a tradeoff to be made between exploration and exploitation of the
Each chromosome in scheduling population consists of two gene pool [37]. This scheme can modify the selection pressure by
parts with different lengths as shown in Fig. 1. The first part of the changing the tournament size.
660 X. Li et al. / Computers & Operations Research 37 (2010) 656 -- 667

Fig. 1. Chromosome of integration.

(2) Crossover: The procedure of the crossover for scheduling is are the first and fourth elements in B-string of P1 are appended to
described as follows: the same positions in B-string of O2.
Step 1: Select a pair of chromosomes P1 and P2 by the selection Step 3: Secondly, in order to match the B-strings of O1 and O2
scheme and initialize two empty offspring: O1 and O2. and avoid getting unreasonable O1 and O2, the A-strings of P1 and
Step 2: First, the B-strings of O1 and O2 are generated by the P2 are crossovered as follows:
following steps: Step 3.1: The Genes whose the first numbers in A-string of P1
Step 2.1: Compare the B-string of P1 with the B-string of P2, if the equate 2 or 3 (including (0, 0)) are appended to the same positions
element of P1 is the same as P2, record the value and position of this in A-string of O1 and they are deleted in A-string of P1. The Genes
element. This process is repeated until all the elements of B-string whose the first numbers in A-string of P2 equate 2 or 3 (including
have been compared. (0, 0)) are appended to the same positions in A-string of O2 and they
Step 2.2: The recorded elements in B-string of P1 in Step 2.1 are are deleted in A-string of P2.
appended to the same positions in B-string of O1, while the recorded Step 3.2: In this example, n1 =6, n2 =7, n1 < n2 and n2 −n1 =1. For
elements in B-string of P2 in Step 2.1 are appended to the same O1, one (0,0) in O1 is selected randomly and is set to empty, which
positions in B-string of O2. The other elements (they are the different is marked out in A-string of O1 in Fig. 2 (in this example, the fifth
elements between the B-strings of P1 and P2) in B-string of P2 are position in A-string of O1 is selected). Then, the remaining elements
appended to the same positions in B-string of O1, while the other in A-string of P2 are appended to the empty positions in A-string of
elements in B-string of P1 are appended to the same positions in O1 seriatim. For O2, one empty position is selected randomly and is
B-string of O2. filled with (0,0), which is marked out in A-string of O2 in Fig. 2 (in
Step 3: Secondly, in order to match the B-strings of O1 and O2 this example, the 13th position in A-string of O2 is selected). Then,
and avoid getting unreasonable O1 and O2, the A-strings of P1 and the remaining elements in A-string of P1 are appended to the empty
P2 are crossovered as follows: positions in A-string of O2 seriatim.
Step 3.1: If the values of the first numbers of Genes in A-string of Step 4: Then, two valid offspring O1 and O2 are obtained (see
P1 are the same as the recorded positions in B-string, these Genes Fig. 2).
(including (0, 0)) are appended to the same positions in A-string of (3) Mutation: In this paper, three mutation operators have been
O1 and they are deleted in A-string of P1. If the values of the first used. The first one is two-point swapping mutation, the second one
numbers of Genes in A-string of P2 are the same as the recorded is changing one job's alternative process plan, and the third one is
positions in B-string, these Genes (including (0, 0)) are appended to the mutation of alternative machines. In the evolution procedure,
the same positions in A-string of O2 and they are deleted in A-string one operator has been chosen randomly in every generation.
of P2. The procedure of two-point swapping mutation for scheduling is
Step 3.2: Get the numbers of the remaining Genes in A-strings of described as follows:
P1 and P2, n1 and n2 . If n1 ⱖ n2 , for O1, it implies that the number Step 1: Select one chromosome P by the selection scheme.
of empty positions in A-string of O1 is larger than the number of Step 2: Select two points in the A-string of P randomly.
remaining elements in A-string of P2. Therefore, n1 − n2 empty po- Step 3: Generate a new chromosome O by interchanging these
sitions in A-string of O1 are selected randomly and are filled with two elements.
(0, 0). Then, the remaining elements in A-string of P2 are appended The procedure of the second mutation (changing one job's alter-
to the remaining empty positions in A-string of O1 seriatim. For O2, native process plan) for scheduling is described as follows:
n1 ⱖ n2 implies that the number of empty positions in A-string of Step 1: Select one chromosome P by the selection scheme.
O2 is smaller than the number of remaining elements in A-string of Step 2: Select one point in the B-string of P randomly.
P1. So, n1 − n2 (0,0) are selected randomly in A-string of O2 and are Step 3: Change the value of this selected element to another one
set to empty. And then, the remaining elements in A-string of P1 in the selection range (the number of alternative process plans).
are appended to the empty positions in A-string of O2 seriatim; if Step 4: Judge the number of the operations of the selected job's
n1 < n2 , the procedure is reversed. alternative process plan which has been changed. If it increases, a
new A-string of chromosome O is generated by changing the margin
Step 4: Then, two valid offspring O1 and O2 are obtained.
(0, 0)'s which are selected randomly to the job numbers in the A-
One example is showed in Fig. 2.
string of P seriatim; if it decreases, a new A-string of chromosome O
Step 1: Select a pair of chromosomes P1 and P2 by the selection is generated by changing the margin job numbers which are selected
scheme and initialize two empty offspring: O1 and O2. randomly in the A-string of P to (0, 0)'s seriatim.
Step 2: First, the B-strings of O1 and O2 are generated by the The mutation of alternative machines is applied in order to change
following steps: the alternative machine represented. One element in the A-string is
Step 2.1: Compare the B-string of P1 with the B-string of P2, record randomly chosen from the selected individual. Then, this element is
the second and third elements in B-strings of P1 and P2. mutated by altering the machine number (the second number of the
Step 2.2: The second and third elements in B-string of P1 in Step Gene) to another one of the alternative machines in the machine set
2.1 are appended to the same positions in B-string of O1, while the at random.
second and third elements in B-string of P2 in Step 2.1 are appended One example of mutation is showed in Fig. 3. Above the first
to the same positions in B-string of O2. The other elements which broken line, it is an example of two-point swapping mutation, the
are the first and fourth elements in B-string of P2 are appended to selected two points ((4, 1) and (3, 2)) have been marked out, and
the same positions in B-string of O1, while the other elements which O is generated by interchanging (4, 1) and (3, 2). Under the first
 X. Li et al. / Computers & Operations Research 37 (2010) 656 -- 667 661

Fig. 2. Crossover operator.

Fig. 3. Mutation operator.

broken line, it is an example of the mutation of changing one job's Table 1

Alternative process plans of six jobs.
alternative process plan. The selected element 3, which is the fourth
element (for job 4) in B-string of P and has been marked out, has Job Alternative process plans
been changed to 1. Because the number of operations in the first
1 1(10)–3(15)–2(10)–5(20)–4(10)
alternative process plan for job 4 is more than the third one, one 1(10)–3(22)–4(21)–5(12)
(0, 0) which is selected randomly in the A-string of P is changed to 2(10)–3(20)–5(20)–4(15)
(4, 2) in A-string of O. And under the second broken line, it is an
2 1(10)–3(18)–4(12)–5(15)
example of the mutation of alternative machines, the selected point
3(8)–2(12)–1(14)–4(13)–5(8)
(1, 1) in A-string of P has been marked out, and O is generated by 2(10)–4(13)–3(18)–5(14)
changing (1,1) to (1,2). It means that the selected machine of the
third operation of job 1 has changed from the first one to the second 3 3(12)–1(16)–5(10)–4(12)
one in S113 (see Section 4.1). 1(10)–2(8)–3(14)–4(6)–5(10)
2(6)–1(12)–3(12)–4(8)–5(10)

5. Experimental studies and discussions 4 1(6)–3(12)–2(8)–5(12)–4(10)

3(10)–1(8)–2(9)–4(12)–5(10)
2(8)–3(12)–1(6)–5(14)–4(8)
5.1. Example problems and experimental results
5 1(10)–2(15)–4(9)–5(10)
Five experiments are presented to illustrate the effectiveness and 3(10)–2(16)–4(8)–5(8)
4(6)–3(10)–2(8)–1(10)–5(8)
performance of the proposed mathematical model and EA-based
approach for IPPS. We describe the design of the first and second ex- 6 5(6)–2(16)–3(10)–4(10)
amples which are constructed by various jobs with several alterna- 1(9)–2(7)–4(8)–5(8)–3(9)
tive process plans. And the other examples are adopted from some 5(6)–1(10)–2(8)–3(8)–4(9)
other papers. The numbers out parenthesis are the machine numbers; the numbers in parenthesis
In this paper, the makespan is used as the objective. After formu- are the processing time.
lating the mixed integer programming model to represent the ex-
amples as a mathematical model, the solutions are obtained by the
proposed EA-based method. The integration model has been com-
pared with the no integration model. No integration model means set: the size of the population pop_size was 200, total number of
that there is only one process plan for each job (In this paper, each generations max_gen was 50, tournament size b was 2, probability
job selects the first alternative process plan as its process plan in the of reproduction operator pr was 0.1, probability of crossover oper-
no integration model.). ator pc was 0.8, and probability of mutation operator pm was 0.1.
The proposed EA-based approach procedure was coded in C++ Altogether 20 iterations were executed to eliminate the randomness
and implemented on a computer with a 2.0 GHz Core (TM) 2 Duo of the searches in the proposed evolutionary search approach. We
CPU. For experimental comparisons under a same condition, the pa- chose the solutions which appeared the most times. The algorithm
rameters used in the proposed evolutionary search approach were terminated when the number of generations reached to the max_gen.
662 X. Li et al. / Computers & Operations Research 37 (2010) 656 -- 667

5.1.1. Experiment 1 Table 3

Job related-information and alternative operation sequences.
Experiment 1 involves six jobs and five machines. And the alter-
native process plans and processing time are given in Table 1. Job Operation no. Operation for alternative Alternative
Table 2 shows the experimental results, and Fig. 4 illustrates Gantt resources (operation no., operation
machine no., processing sequences
chart of this experiment. The experimental results of Table 2 show
time) considering
that the selected process plans in the integration approach are not precedence
all of the first plans in Table 1, such as, the selected process plan of constraints
job 1 is its third process plan. Comparing the results between the in-
1 O1 ,O2 (O1 , 1, 5), (O1 , 2, 5), (O1 , 4, 6) O1 –O2
tegration model and the no integration model, the selected process (O2 , 2, 5), (O2 , 5, 6)
plans of the jobs between them are different, and the makespan of
scheduling of no integration model is worse than that of the inte-
2 O3 ,O4 (O3 , 3, 6), (O3 , 5, 6) O3 –O4
gration model. (O4 , 2, 5), (O4 , 4, 5), (O4 , 5, 6)

5.1.2. Experiment 2 3 O5 ,O6 ,O7 (O5 , 1, 4), (O5 , 3, 3), (O5 , 4, 5) O5 –O6 –O7
Experiment 2 involves six jobs and five machines. And the alter- (O6 , 1, 8), (O6 , 3, 7), (O6 , 5, 8) O6 –O5 –O7
native process plans and processing time are given in Table 3. In this (O7 , 2, 5)
experiment, each job selects the first alternative process plan as its
process plan in the no integration model. 4 O8 ,O9 ,O10 (O8 , 1, 4), (O8 , 3, 4) O8 –O9 –O10
Table 4 shows the experimental results, and Fig. 5 illustrates Gantt (O9 , 2, 8), (O9 , 3, 7) O8 –O10 –O9
chart of this experiment. Comparing the results between the inte- (O10 , 4, 5), (O10 , 5, 5)
gration model and the no integration model, the selected process
plans of the jobs between them are different, and the makespan of 5 O11 ,O12 ,O13 ,O14 (O11 , 3, 4), (O11 , 5, 3) O11 –O12 –O13 –O14
no integration model is worse than that of the integration model. (O12 , 1, 7), (O12 , 4, 7) O12 –O11 –O13 –O14
(O13 , 1, 8), (O13 , 2, 7), (O13 , 5, 8) O11 –O13 –O12 –O14
(O14 , 1, 5), (O14 , 4, 6)
5.1.3. Experiment 3
Experiment 3 is adopted from Jain [15]. In this experiment, six
6 O15 ,O16 ,O17 ,O18 (O15 , 1, 8), (O15 , 2, 7), (O15 , 4, 8) O15 –O16 –O17 –O18
problems are constructed with eighteen jobs and four machines. (O16 , 3, 5), (O16 , 4, 5) O16 –O15 –O17 –O18
Table 5 shows the experimental results, and Fig. 6 illustrates Gantt (O17 , 2, 8)
chart of the first problem in experiment 3. (O18 , 1, 7), (O18 , 3, 8) , (O15 , 5, 8)
The experimental results of experiment 3 show that the results
of the no integration model are worse than that of the integration

Table 4
Experimental results of experiment 2.
Table 2
Experimental results of experiment 1. Job Alternative process plans

Job Alternative process plans Integration No integration

Integration No integration 1 O1 M4 –O2 M5 O1 M1 –O2 M2

2 O3 M3 –O4 M4 O3 M3 –O4 M4
1 2(10)–3(20)–5(20)–4(15) 1(10)–3(15)–2(10)–5(20)–4(10)
3 O6 M3 –O5 M1 –O7 M2 O5 M3 –O6 M5 –O7 M2
2 2(10)–4(13)–3(18)–5(14) 1(10)–3(18)–4(12)–5(15)
4 O8 M1 –O9 M3 –O10 M5 O8 M1 –O9 M2 –O10 M4
3 3(12)–1(16)–5(10)–4(12) 3(12)–1(16)–5(10)–4(12)
5 O11 M5 –O13 M5 –O12 M1 –O14 M4 O11 M5 –O12 M4 –O13 M5 –O14 M4
4 3(10)–1(8)–2(9)–4(12)–5(10) 1(6)–3(12)–2(8)–5(12)–4(10)
6 O15 M2 –O16 M4 –O17 M2 –O18 M1 O15 M1 –O16 M3 –O17 M2 –O18 M3
5 4(6)–3(10)–2(8)–1(10)–5(8) 1(10)–2(15)–4(9)–5(10)
6 5(6)–1(10)–2(8)–3(8)–4(9) 5(6)–2(16)–3(10)–4(10)
Makespan 27 33
Makespan 92 102 CPU time (s) 3.20 3.20
CPU time (s) 3.23 3.31
O1 means operation 1. M1 means machine 1.

Fig. 4. Gantt chart of experiment 1 (J1, 1 means that the first operation of job 1, makespan = 92).
 X. Li et al. / Computers & Operations Research 37 (2010) 656 -- 667 663

Fig. 5. Gantt chart of experiment 2 (makespan = 27).

Table 5
Experimental results of experiment 3.

Problems Numberof jobs Makespan CPU time (s)

Integration No integration Integration No integration

1 8 520 615 3.28 3.42

2 10 621 831 3.38 3.72
3 12 724 934 3.67 3.91
4 14 809 1004 3.69 4.14
5 16 921 1189 3.73 4.39
6 18 994 1249 4.09 4.69

Fig. 6. Gantt chart of the first problem in experiment 3 (makespan = 520).

model. And the integration model can get better scheduling plans. Using the proposed approach, the best value obtained is 17 com-
This implies that the research on IPPS is necessary. pared with the value of 18 obtained using the heuristic reported by
Nabil [38]. Fig. 7 illustrates Gantt chart of this problem.
5.1.4. Experiment 4

Problem 1. There are two problems in experiment 4. The first prob- Problem 2. In the second problem, the benchmark problem by Nabil
lem is adopted from Nabil [38]. In this problem, there are four jobs is extended to confirm the effectiveness of the proposed method for
and six different machines in the system, and the alternative ma- large-scale and more complex problems. In this problem, eight sizes
chines and processing time are given in Table 6. which contain different jobs and operations are considered. Table 7
664 X. Li et al. / Computers & Operations Research 37 (2010) 656 -- 667

shows the experimental results. Fig. 8 illustrates the Gantt chart of 5.1.5. Experiment 5
size 1 of this problem.
Problem 1. There are two problems in experiment 5. The first prob-
These results reveal that the proposed method can solve the com- lem is adopted from a benchmark problem proposed by Sundaram
plex and large-scale problems effectively. and Fu [39], which is constructed with five jobs and five machines.
Each job undergoes four different operations in a specified order.
Table 6 Alternative machines for processing the parts are given in Table 8,
Experiment data of the Problem 1 in experiment 4.
along with the respective processing time.
Alternative machines

1 2 3 4 5 6 Using the simulated annealing [40], genetic algorithm [41] and

Job 1 this proposed approach (see Fig. 9), the best value obtained of all
O11 2 3 4 / / / the three approaches is 33 compared with the value of 38 obtained
O12 / 3 / 2 4 / using the heuristic algorithm developed in [39]. The advantage of the
O13 1 4 5 / / /
GA and proposed approach compared with the SA approach is the
Job 2 availability of multiple solutions that introduces more flexibility into
O21 3 / 5 / 2 / the IPPS system. In the SA approach, only one solution is produced
O22 4 3 / / 6 / for every execution.
O23 / / 4 / 7 11

Job 3 Problem 2. In the second problem, the benchmark problem by

O31 5 6 / / / /
Sundaram and Fu is extended to confirm the effectiveness of the
O32 / 4 / 3 5 /
O33 / / 13 / 9 12 proposed method for large-scale and more complex problems. The
search space becomes much larger than the benchmark problem
Job 4 by the extension of the setting. Table 9 shows the setting of the
O41 9 / 7 9 / / production demands in which eight terms of production are used
O42 / 6 / 4 / 5
and the experimental results. Fig. 10 illustrates Gantt chart of size
O43 1 / 3 / / 3
1 of Problem 2 in experiment 5.

Fig. 7. Gantt chart of Problem 1 in experiment 4 (makespan = 17).

Table 7
Experimental results of Problem 2 in experiment 4.

Size Jobs Production amount in each job Operations Makespan CPU time (s)

Job 1 Job 2 Job 3 Job 4

1 8 2 2 2 2 24 20 1.27
2 32 10 8 8 6 96 67 9.34
3 64 24 10 20 10 192 136 69.92
4 100 25 25 25 25 300 219 249.83
5 150 30 30 40 50 450 358 1091.44
6 200 50 50 60 40 600 453 2148.69
7 300 80 80 70 70 900 658 7559.00
8 400 100 100 100 100 1200 901 15 328.15
 X. Li et al. / Computers & Operations Research 37 (2010) 656 -- 667 665

Fig. 8. Gantt chart of size 1 of Problem 2 in experiment 4 (makespan = 20).

Table 8
Sundaram and Fu data.

Job Operation 1 Operation 2 Operation 3 Operation 4

1 5(M1), 3(M2) 7(M2) 6(M3) 3(M4), 4(M5)

2 7(M1) 4(M2), 6(M3) 7(M3), 7(M4) 10(M5)
3 4(M1), 5(M2) ,8(M3) 5(M4) 6(M4), 5(M5) 4(M5)
4 2(M2), 6(M3) 8(M3) 3(M3), 8(M4) 7(M4), 4(M5)
5 3(M1), 5(M3) 7(M3) 9(M4), 6(M5) 3(M5)

The numbers in parenthesis are the machine numbers.

Fig. 9. Gantt chart of Problem 1 in experiment 5 (makespan = 33).

Table 9
Setting of production demand changes and experimental results.

Size Jobs Production amount in each job Operations Makespan CPU time (s)

Job 1 Job 2 Job 3 Job 4 Job 5

1 50 10 10 10 10 10 200 252 75.11

2 100 20 20 20 20 20 400 500 592.47
3 100 25 25 25 25 0 400 467 437.78
4 100 15 15 20 25 25 400 521 523.27
5 150 30 25 30 35 30 600 757 1149.67
6 200 40 40 30 50 40 800 1021 4045.25
7 300 60 70 50 50 70 1200 1537 15 404.79
8 400 80 100 80 60 80 1600 2028 42 705.31
666 X. Li et al. / Computers & Operations Research 37 (2010) 656 -- 667

Fig. 10. Gantt chart of size 1 of Problem 2 in experiment 5 (makespan = 252).

These results reveal that the proposed method can solve the com- References
plex and large-scale problems effectively. We also can conclude that
the proposed method is effective in generating the near optimal re- [1] Sugimura N, Hino R, Moriwaki T. Integrated process planning and scheduling
in holonic manufacturing systems. In: Proceedings of IEEE international
sults under various order environments with precedence constraints. symposium on assembly and task planning soft research park, vol. 4, 2001,
And it can solve IPPS problem effectively. Japan, Fukuoka. p. 250–4
[2] Saygin C, Kilic SE. Integrating flexible process plans with scheduling in flexible
manufacturing systems. International Journal of Advanced Manufacturing
5.2. Discussions Technology 1999;15:268–80.
[3] Kumar M, Rajotia S. Integration of scheduling with computer aided process
planning. Journal of Materials Processing Technology 2003;138:297–300.
Overall, the experimental results indicate that the proposed ap- [4] Usher JM, Fernandes KJ. Dynamic process planning—the static phase. Journal
proach is a very effective and more acceptable approach for IPPS, of Materials Processing Technology 1996;61:53–8.
[5] Lee H, Kim SS. Integration of process planning and scheduling using simulation
and the integration model of process planning and scheduling can based genetic algorithms. International Journal of Advanced Manufacturing
get better scheduling plans than no integration model. The reasons Technology 2001;18:586–90.
are as follows. First, the proposed approach considers all the con- [6] Kuhnle H, Braun HJ, Buhring J. Integration of CAPP and PPC—interfusion
manufacturing management. Integrated Manufacturing Systems 1994;5(2):
ditions of process planning and scheduling synthetically. Second, in 21–7.
some experiments, the proposed approach can attain better results [7] Tan W, Khoshnevis B. Integration of process planning and scheduling—a review.
than other previously developed approaches. This means that the Journal of Intelligent Manufacturing 2000;11:51–63.
[8] Kumar M, Rajotia S. Integration of process planning and scheduling in a job
proposed approach has more possibilities to get the best solutions shop environment. International Journal of Advanced Manufacturing Technology
of IPPS problems. 2005;28(1–2):109–16.
[9] Chryssolouris G, Chan S, Cobb W. Decision making on the factory floor:
an integrated approach to process planning and scheduling. Robotics and
Computer-Integrated Manufacturing 1984;1(3–4):315–9.
6. Conclusion [10] Chryssolouris G, Chan S. An integrated approach to process planning and
scheduling. Annals of the CIRP 1985;34(1):413–7.
Considering the complementarity of process planning and [11] Beckendorff U, Kreutzfeldt J, Ullmann W. Reactive workshop scheduling based
on alternative routings. In: Proceedings of a conference on factory automation
scheduling, the research has been conducted to develop a mathe- and information management. Florida: CRC Press; 1991. p. 875–85.
matical model with an evolutionary algorithm-based approach to [12] Khoshnevis B, Chen QM. Integration of process planning and scheduling
facilitate the integration and optimization of these two systems. function. In: Proceedings of IIE integrated systems conference and society
for integrated manufacturing conference. Atlanta: Industrial Engineering &
Process planning and scheduling functions are carried out simulta- Management Press; 1989. p. 415–20.
neously. To improve the optimization performance of the proposed [13] Zhang HC. IPPM-a prototype to integrated process planning and job shop
approach, efficient genetic representation and operator schemes scheduling functions. Annals of the CIRP 1993;42(1):513–7.
[14] Larsen NE. Methods for integration of process planning and production planning.
have been developed. To verify the feasibility of the proposed ap- International Journal of Computer Integrated Manufacturing 1993;6(1–2):
proach, a number of experimental studies have been carried out to 152–62.
compare this approach with other previously developed approaches. [15] Jain A, Jain PK, Singh IP. An integrated scheme for process planning and
scheduling in FMS. International Journal of Advanced Manufacturing Technology
The experimental results show that the proposed approach is very
2006;30:1111–8.
effective for the IPPS problem and achieve better overall optimiza- [16] Kim KH, Song JY, Wang KH. A negotiation based scheduling for items with
tion results. flexible process plans. Computers & Industrial Engineering 1997;33(3–4):
785–8.
With the model developed in this work, it would be possible to
[17] Yang YN, Parsaei HR, Leep HR. A prototype of a feature-based multiple-
increase the efficiency of manufacturing systems. One future work is alternative process planning system with scheduling verification. Computers &
to apply the proposed method to practical manufacturing systems. Industrial Engineering 2001;39:109–24.
The increased use of this will most likely enhance the performances [18] Kim YK, Park K, Ko J. A symbiotic evolutionary algorithm for the integration of
process planning and job shop scheduling. Computers & Operations Research
of future manufacturing systems. 2003;30:1151–71.
[19] Li WD, McMahon CA. A simulated annealing—based optimization approach for
integrated process planning and scheduling. International Journal of Computer
Integrated Manufacturing 2007;20(1):80–95.
Acknowledgments
[20] Shao XY, Li XY, Gao L, Zhang CY. Integration of process planning and scheduling-
—a modified genetic algorithm-based approach. Computers & Operations
The authors would like to thank anonymous referees whose Research 2009;36:2082–96.
comments helped a lot to improve this paper. This research work [21] Usher JM. Evaluating the impact of alternative plans on manufacturing
performance. Computers & Industrial Engineering 2003;45:585–96.
is supported by the National Natural Science Foundation of China [22] Seethaler RJ, Yellowley I. Process control and dynamic process planning.
(NSFC) under Grant no. 50825503, the National High-Tech Research International Journal of Machine Tools & Manufacture 2000;40:239–57.
and Development Program of China (863 Program) under Grant no. [23] Kempenaers J, Pinte J, Detand J. A collaborative process planning and scheduling
system. Advances in Engineering Software 1996;25:3–8.
2007AA04Z107, and the Program for New Century Excellent Talents [24] Wu SH, Fuh JYH, Nee AYC. Concurrent process planning and scheduling in
in University under Grant no. NCET-08-0232. distributed virtual manufacturing. IIE Transactions 2002;34:77–89.
 X. Li et al. / Computers & Operations Research 37 (2010) 656 -- 667 667

[25] Zhang J, Gao L, Chan FTS. A holonic architecture of the concurrent [33] Li XY, Shao XY, Gao L. Optimization of flexible process planning by genetic
integrated process planning system. Journal of Materials Processing Technology programming. International Journal of Advanced Manufacturing Technology
2003;139:267–72. 2008;38:143–53.
[26] Wang LH, Song YJ, Shen WM. Development of a function block designer for [34] Garey EL, Johnson DS, Sethi R. The complexity of flow-shop and job-shop
collaborative process planning. In: Proceedings of CSCWD2005, 2005, Coventry, scheduling. Mathematics of Operations Research 1976;1:117–29.
UK. p. 24–6. [35] Kim KH, Egbelu PJ. A mathematical model for job shop scheduling with
[27] Guo YW, Li WD, Mileham AR, Owen GW. Applications of particle swarm multiple process plan consideration per job. Production Planning & Control
and optimization in integrated process planning and scheduling. Robotics and 1998;9(3):250–9.
Computer-Integrated Manufacturing 2009;25:280–8. [36] Bierwirth C. A generalized permutation approach to job shop scheduling with
[28] Hutchinson GK, Flughoeft KAP. Flexible process plans: their value in genetic algorithms. OR Spektrum 1995;17:87–92.
flexible automation systems. International Journal of Production Research [37] Langdon WB, Qureshi A. Genetic programming—computers using “Natural
1994;32(3):707–19. Selection” to generate programs. Technical report RN/95/76, Gower Street,
[29] Benjaafar S, Ramakrishnan R. Modeling, measurement and evaluation of London WCIE 6BT, UK,; 1995.
sequencing flexibility in manufacturing systems. International Journal of [38] Nabil N, Elsayed EA. Job shop scheduling with alternative machines.
Production Research 1996;34:1195–220. International Journal of Production Research 1990;28(9):1595–609.
[30] Lin YJ, Solberg JJ. Effectiveness of flexible routing control. International Journal [39] Sundaram RM, Fu SS. Process planning and scheduling. Computers & Industrial
of Flexible Manufacturing Systems 1991;3:189–211. Engineering 1988;15(1–4):296–307.
[31] Li WD, Gao L, Li XY, Guo Y. Game theory-based cooperation of process planning [40] Palmer GJ. A simulated annealing approach to integrated production scheduling.
and scheduling. In: Proceeding of CSCWD2008, 2008, Xi'an, China. p. 841–5. Journal of Intelligent Manufacturing 1996;7(3):163–76.
[32] Michael P. Scheduling: theory, algorithm, and systems. second ed., Pearson [41] Morad N, Zalzala A. Genetic algorithms in integrated process planning and
Education Asia Limited and Tsinghua University Press; 2005. scheduling. Journal of Intelligent Manufacturing 1999;10:169–79.