Int J Adv Manuf Technol (2012) 59:273–287

DOI 10.1007/s00170-011-3488-y

ORIGINAL ARTICLE

A hybrid genetic algorithm for integrated process planning

and scheduling problem with precedence constraints
M. R. Amin-Naseri & Ahmad J. Afshari

Received: 24 October 2010 / Accepted: 20 June 2011 / Published online: 7 July 2011
# Springer-Verlag London Limited 2011

Abstract Process planning and scheduling are two important 1 Introduction

functions in a modern manufacturing system. Although
integrating decisions related to these functions gives rise to a Modern manufacturing environment is characterized by
hard combinatorial problem, due to the impressive improve- dynamic customer demands, increased product variety,
ment in system performance which is resulted through this demand for reduced production cycles, intense competitions
integration, developing effective methods to solve this across the globe, and increased use of automation, which
problem is of great theoretical and practical importance. In highlights the importance of adopting new technologies and
this research, after formulating the integrated process planning systems to deal with this environment. Computer-integrated
and scheduling problem as a mathematical program, we manufacturing (CIM) is the most effective tool to deal with
propose a hybrid genetic algorithm (GA) for the problem. In these challenges [1]. CIM is defined as an architecture for
the proposed algorithm, problem-specific genetic operators integrating the applications and information flow of product
are designed to enhance the global search power of GA. Also, design, process planning, production planning, and manu-
a local search procedure has been incorporated into the GA to facturing processes [2]. The main function of this system is
improve the performance of the algorithm. The model the integration of manufacturing activities from order to
considers precedence relations among job operations, based delivery, and computer-aided process planning (CAPP) as
on which feasible process plans for each job can be one of the CIM components plays a key role in achieving
represented implicitly. A novel neighborhood function, full integration.
considering the constraints of a flexible job shop environment Process planning is a manufacturing system function which
and nonlinear precedence relations among operations, is prepares detailed operation instructions to transform an
presented to speed up the local search process. In experimental engineering design to a final part [3]. The output of process
study, the performance of the proposed algorithm has been planning is the information it gives about manufacturing
evaluated based on a number of problems adopted from the processes and their parameters, operation sequence,
literature. The experimental results demonstrate the efficiency machines, fixtures, and tools required to manufacture the
of the proposed algorithm to find optimal or near-optimal part. CAPP systems which are developed to automate process
solutions. planning serve as the critical link between engineering design
and manufacturing.
Keywords Integrated process planning and scheduling . In the traditional approach, the manufacturing process is
Process planning . Job shop scheduling . Hybrid genetic composed of a number of activities such as design,
algorithm . Precedence relations planning, manufacturing, and distribution which are viewed
as discrete stages and performed sequentially. However,
decisions taken in one stage affect decisions in the
M. R. Amin-Naseri (*) : A. J. Afshari subsequent stages. The integration of different functions
Industrial Engineering Department, Faculty of Engineering,
can introduce significant improvements in the efficiency of
Tarbiat Modares University,
Tehran 14115-143, Iran the manufacturing facilities through a reduction in produc-
e-mail: [email protected] tion costs, shorter lead times, an increase in resource
274 Int J Adv Manuf Technol (2012) 59:273–287

utilization, and an improvement of the ability to respond 4. IPPS can effectively contribute to the realization of
rapidly to market changes [4]. modern manufacturing concepts and paradigms which
During the past two decades, much research work has have been developed to enhance the flexibility, produc-
been carried out about integrating process planning with tivity, and agility of manufacturing systems. Several
other manufacturing processes. While most of these studies have addressed the implementation of IPPS idea
research works have been focused on integrating process in a modern manufacturing framework such as concurrent
planning with design function, little attention has been engineering [10], holonic manufacturing system [11],
given to the integration of process planning and down- CIM [1], and flexible manufacturing system (FMS) [7].
stream production functions, which is equally likely to
Although several approaches have been proposed to
yield high benefits [5].
integrate process planning and scheduling due to the
Production scheduling is a manufacturing function that
complexity of the problem and its huge solution space,
allocates the operations to available resources and deter-
more efficient algorithms have to be developed. Also, most
mines the starting and completion time of each operation
of the developed algorithms ignore alternative operation
while some criteria, such as makespan, job tardiness, and
sequences generated based on precedence relations among
manufacturing cost, are optimized [1]. Since scheduling and
the operations of an order, which is the case in many of the
process planning both involve the assignment of resources
real manufacturing environments [12].
and sequencing of operations, they are highly interrelated
The objective of this research was to develop an
and their integration can greatly improve the productivity of
algorithm for IPPS problem which minimizes the total
the manufacturing system. Traditionally, process planning
processing time by considering alternative sequences of
and scheduling tasks are performed sequentially, and
operations with precedence constraints. For this purpose,
decisions taken at the process planning stage constrain the
first, the sequence and operation flexibilities are introduced
alternatives that might be used to make improvements
into the job shop manufacturing setting by which the IPPS
during scheduling [6]. The integration of these two
problem addressed in this paper is defined. The resulting
functions has several advantages, including:
problem is called “flexible job shop scheduling problem
with alternative process plans” (FJSP-APP) hereafter. Then,
1. Performance measures of a manufacturing system such
a genetic local search algorithm is proposed in which
as throughput time, resource utilization, work-in-
problem-specific genetic operators are designed to enhance
process inventory, average flow time, and average
the global search ability. Also, a local search procedure
tardiness can be improved significantly [5]. This is
based on a graph model is incorporated into the genetic
due to the increase in solution space of the problem
algorithm (GA) to improve the quality of solutions.
which increases the complexity of the problem too.
The remainder of this paper is organized as follows.
2. The dynamic nature of real manufacturing systems
Following a literature review in Section 2, the problem
which is caused by several events such as machine
statement and mathematical model is presented in Section 3.
breakdown, new job arrival, job cancellation, and
A hybrid GA for the problem is proposed and exemplified
adding new machine can make a plan infeasible.
in Section 4. Experimental results are discussed in
Consequently, frequent alteration of the plans due to
Section 5. Finally, Section 6 gives some concluding
the changed status of the shop floor is required. The
remarks and future research directions.
integration of process planning with scheduling reduces
the time delay between production planning and plan
execution, and thus, a more realistic plan without
frequent need for modification is generated [7]. 2 Review of related works
3. Process planning emphasizes the technological require-
ments of a job with no regard for the competition that Since the work of Chryssolouris et al. [13] who addressed the
exists between jobs for the resources, while scheduling problem of integrating process planning with scheduling for
involves the timing aspects and resource sharing of all the first time, several studies have been done in this area. A
jobs [8]. Therefore, these functions may have different comprehensive review of earlier research can be found in [5].
and sometimes conflicting objectives. Integrated Most of the recent works have used artificial intelligence-
process planning and scheduling problem (IPPS) based heuristics for optimization.
offers the possibility of considering these conflicting Kim et al. [14] presented a symbiotic GA for the
criteria simultaneously, which can result in produc- integration of process planning and scheduling in a job
tion cost reduction, bottleneck elimination, balanced shop problem. They assumed three kinds of flexibility, i.e.,
level of machine utilization, and improved facility operation, sequencing, and processing flexibility for each
productivity [9]. job, and also suggested a simple measure for the complexity
Int J Adv Manuf Technol (2012) 59:273–287 275

of each problem based on these flexibilities. Baykasoglu and Some researchers have studied IPPS under the supply
Ozbakir [15] addressed the integration of process planning chain management environment in the context of advanced
and scheduling problem with two objectives: total flow time planning and scheduling system (APS). Lee et al. [22]
and total cost of process plans. In the proposed model, the focused on a scheduling problem with alternative machines
generic process plan is represented as a grammar, dispatching for each operation, which include outsourcing machines
rules are applied for sequencing operations, and an optimiz- and alternative sequences of operations. They developed a
ing algorithm based on multi-objective tabu search is GA-based model to determine the optimal schedule with a
employed to generate alternative solutions. minimal makespan, given the due date of each order. The
Li and McMahon [16] employed a simulated annealing- APS problem in manufacturing supply chain was also
based approach for the integrated process planning and studied by Moon et al. [23]. Alternative resources and
scheduling problem. They examined the proposed approach sequences, sequence-dependent setup, and transportation
with several objective functions including makespan, ma- times were considered in the model, and an adaptive GA
chine utilization, job tardiness, and manufacturing costs. Guo was proposed to optimize the model. Moon et al. [12]
et al. [3] developed a modified particle swarm optimization presented a GA to solve the IPPS problem in a supply
algorithm to solve the IPPS problem. Alternative machines, chain. The proposed model considers alternative resources
tools, and manufacturing plans, were taken into account in and process plans, with precedence constraints, and
the model with the goal of optimizing makespan, total cost, minimizes the makespan.
and balanced machine utilization separately. A number of researchers adopted the multi-agent approach
Li et al. [9] suggested an alternative formulation for for IPPS. Shen et al. [24] presented a review paper on agent-
IPPS and proposed an evolutionary-based model with based manufacturing process planning, scheduling, and their
proper genetic operators to solve the problem. Shao et al. integration. More recently, Wong et al., [25] developed a
[17] suggested a two-phase approach to optimize process multi-agent system (MAS) for integrating process planning
planning and scheduling simultaneously. In the first step, with reactive scheduling in job shop environments. Leung et
considering production time as an objective, three to five al. [26] developed an agent-based system in which ant
near-optimal process plans for each job are determined. colony optimization was incorporated into the MAS. The
Then, the scheduling problem (using the selected process artificial ants were implemented as software agents and
plans) is optimized. Two objective functions are considered improved the computational ability of the proposed system.
in this step: minimizing makespan and balancing level of Ey et al. [27] proposed a knowledge-based algorithm for a
machine utilization. For both phases, modified GAs have dynamic job shop scheduling problem with routing and
been developed. sequencing flexibilities. It their approach, a colored Petri net
Li et al. [18] proposed a GA to solve the problem of was used for modeling the manufacturing system, a fuzzy
integrated process planning and scheduling. They applied a reasoning scheme was employed to incorporate expert
local search strategy based on tabu search to improve the knowledge, and reinforcement learning techniques combined
performance of the algorithm. Candido et al. [19] included with a fuzzy clustering method were used to refine the initial
a number of additional constraints such as alternative knowledge.
processing plans for each job, alternative resources for A few studies focused on mathematical programming
each operation, possibility of overlapping of successive methods to solve IPPS. Tan and Khoshnevis [28] presented
operations, sequence-dependent setup times, and non-zero a linearized polynomial mixed integer programming model
ready times in the job shop scheduling problem to bring it (PMIPM) for the integration of process planning and
closer to the real-world scheduling problems. They pre- scheduling problem. They first formulated the problem as
sented a robust procedure by incorporating a local search a PMIPM. Then, using proper linearization techniques, they
strategy into the GA to solve the problem. transformed it into an equivalent linear model. Sherali et al.
Kis [20] studied an extension of a job shop problem that [29] considered a job shop scheduling problem in a FMS
arises in advanced process planning systems. In this environment with alternate routing combination for each
problem, each job has alternative routings which can be job. The problem was formulated as a binary mixed integer
represented as acyclic graphs. Two heuristic approaches, model, and regarding the structure of the model, two
based on tabu search and GA, were proposed for the heuristics based on implicit enumeration and Lagrangian
problem to minimize the makespan. In Mati and Xie [21], relaxation were proposed for medium- and large-sized
the authors addressed a generalization of the two-job shop problems. Özgüven et al. [30] developed a mixed integer
scheduling in which one of the jobs has routing flexibility. linear programming (MILP) model for exact solutions of
They developed an exact algorithm based on a geometric flexible job shop scheduling problem and extended the
approach for the two-job shop problem which is applicable model to include alternative process plans. The proposed
to any regular objective function. model is efficient only for small-sized problems.
276 Int J Adv Manuf Technol (2012) 59:273–287

It can be seen through a literature review that most of the an extension of classic job shop scheduling problem (JSP). A
employed approaches are based on evolutionary algorithms. FJSP, like a JSP, consists of N jobs, each of which consists of a
This is mainly due to the NP complexity of the IPPS problem number of operations that have to be performed in a unique,
[3] and the efficiency of evolutionary algorithms to solve hard linear sequence to complete the job. There are M machines on
combinatorial optimization problems. While several studies the shop floor and each operation can be processed on one
have utilized the searching ability of GAs to solve the machine out of a given subset of these machines, whereas in a
integration problem, there is a paucity of studies investigating JSP, each operation has only one feasible machine. Solving a
the incorporation of efficient local search operators into this FJSP involves two sub-problems: assignment of each
kind of algorithms. In this research, a hybrid GA has been operation to a feasible machine and ordering of the operations
developed for the IPPS problem in which the exploration on the assigned machine.
power of evolutionary approach is combined with the Considering FJSP-APP, there are multiple feasible
intensification capability of an ad hoc local search procedure. sequences for the operations of each job. In this model,
Also, in most of the previous works which used a GA- which is a more realistic and practical version of FJSP,
based approach for IPPS (e.g., [9, 12, 17, 18]), genetic sequence flexibility is incorporated into the problem.
representation schemes and genetic operators were Sequence flexibility occurs when an operation can have
designed with the assumption that all possible process more than one predecessor or successor [6]. Precedence
plans were generated or known in advance. In the proposed relations among operations reflect the technological restric-
method, there is no need to generate all possible process tions and machining requirements to meet the needed
plans explicitly. Rather, precedence relations between specifications in a part. In the proposed algorithm, we use
operations of an order are assumed to be known and the precedence graph to represent nonlinear precedence
represented through a precedence graph. Considering this relations among operations based on which all feasible
property, the genetic operators, representation scheme, and process plans can be produced. Data for a sample FJSP-
local search strategies are designed so that the proposed APP problem with three jobs and four machines are shown
hybrid GA can efficiently generate quality solutions. in Table 1. This sample problem is used in Section 4 to
illustrate the proposed hybrid GA.
The objective of the model was to minimize the
3 Problem description makespan. This objective is important in scheduling
problems with flexible machines since it not only enforces
3.1 Statement of the problem the jobs to be completed in the shortest overall time period
but also distributes the workload evenly among all the
The FJSP-APP problem is an extension of the well-known available machines [23]. The FJSP-APP addressed in this
flexible job shop scheduling problem (FJSP), which is in turn paper can be formally stated as follows:

Table 1 Data for a sample FJSP-APP problem

Processing time on
alternative machines

Operation M1 M2 M3 M4 Precedence Graph Feasible process plans

O11 10 - 8 12
O11
O11 O12 O13 O14
O12 5 6 9 -
Job 1

O13 O14
O13 10 15 - 8
O12 O11 O13 O14
O12
O14 - 3 5 -

O21 8 - 10 6
Job 2

O21 O22 O21 O22

O22 12 7 5 9

O31 10 - 12 - O31 O32 O33

O31 O32
Job 3

O32 5 6 12 6 O31 O33 O32

O33
O33 8 9 - 9 O33 O31 O32
Int J Adv Manuf Technol (2012) 59:273–287 277

Given a set of n orders or jobs each of which consists a ptijk Processing time if operation j of job i is to performed
number of operations and are to be processed using m on machine k
machines, with alternative operation sequences for jobs and H A very large positive number
alternative resources for operations, determine a suitable
machine for each operation, routing plan for each job, and Decision variables:
the overall schedule which satisfies the precedence con-
straints and is optimal with respect to the minimization of stijk Starting time of operation j of job i on machine k
the makespan criterion [23]. ctijk Completion time of operation j of job i on
machine k
3.2 Mathematical model xijk 1 if machine k is assigned to operation j of job i; 0,
otherwise
In this section, a mathematical model is developed for uiji′j′k 1 if operation j′ of job i′ is scheduled before
the FJSP-APP. This model will be used to evaluate the operation j of job i where both operations are
quality of small- and medium-sized solutions obtained by processed on machine k; 0, otherwise
the proposed hybrid GA. In the development of the vijj′ 1 if operation j′ of job i is scheduled before
mathematical model, the following generally accepted operation j of this job; 0, otherwise
assumptions are made:
The FJSP-APP is formulated into a MILP. The mathe-
& There is no priority among the jobs and all of them are matical model, which is based on Manne’s model for the
available at time 0. JSP [31], is as follows:
& No job preemption or cancellation is allowed. min Cmac
& All machines are available at time 0 and machine failure subject to:
is not considered. P
Cmax  ct ijk 8i 2 N
& The processing time of each operation on each k2Mij ð1Þ
alternative machine is known and fixed. 8j 2 Oi
& Each machine can process one job at a time.
& The different operations of a job cannot be processed P
xijk ¼ 1 8i 2 N
simultaneously. k2Mij
8j 2 Oi ð2Þ
& Setup times on each machine are either negligible or
independent of job sequence and can be included in the
processing times.
& The transportation time among machines is negligible. ct ijk  H:xijk 8i 2 N
& Each operation can be started when all of its prede- 8j 2 Oi ð3Þ
cessors have been completed. Nonlinear precedence 8k 2 Mij
relations among the operations of each job are known
and can be depicted as directed graphs. ct ijk  st ijk ¼ pt ijk :xijk 8i 2 N
The following notations are used in the model. 8j 2 Oi ð4Þ
Indices, sets, and parameters: 8k 2 Mij

i, i′ Indices for jobs

st i0 j0 k  ct ijk þ H:uiji0 j0 k  0 8i; i0 2 N ; i < i 0
j, j′ Indices for operations
8j 2 Oi
k, k' Indices for machines ð5Þ
8j 0 2 Oi 0
N Set of jobs 8k 2 Mij \ Mi0 j 0
M Set of machines
Oi Set of operations in job i  
Mij Set of machines on which operation j of job i can be st ijk  ct i0 j0 k þ H  1  uiji0 j0 k  0 8i; i0 2 N ; i < i0
processed; Mij
M 8j 2 Oi
ð6Þ
Pi Set consisting of ordered pairs which represent the 8j0 2 Oi0
8k 2 Mij \ Mi0 j0
precedence relations among operations in job i, i.e.,
hj; j0 i 2 Pi iff operation j should precede operation j′
P P
Qi Set consisting of ordered pairs of operations in job i st ij0 k 0  ct ijk þ H  vijj0  0 8i 2 N
with no precedence relation, i.e., hj; j0 i 2 Qi iff k 0 2Mij0 k2Mij ð7Þ
operation j has no precedence relation with j′ 8h j; j0 i 2 Qi
278 Int J Adv Manuf Technol (2012) 59:273–287

P P  
st ijk  ctij0 k 0 þ H  1  vijj0  0 8i 2 N Constraint sets 10 through 14 are the non-negativity and
k2Mij k 0 2Mij 0 ð8Þ integrality constraints on the corresponding decision
8h j; j0 i 2 Qi
variables.

P P
st ij0 k 0  ct ijk  0 8i 2 N
k 0 2Mij0 k2Mij ð9Þ 4 Proposed hybrid genetic algorithm
8h j; j0 i 2 Pi
GA is a stochastic optimization approach that has been
widely applied to many hard optimization tasks, including
ct ijk  0 8i 2 N scheduling problems [32]. GAs, in their basic form, are
8j 2 Oi ð10Þ excellent in global searching and can find promising
8k 2 Mij solution areas efficiently. But they are poor at exploiting
these areas without the risk of being trapped in local
optima. The incorporation of appropriate local search
st ijk  0 8i 2 N heuristics into conventional GA is an effective method to
8j 2 Oi ð11Þ overcome this weakness and enhance the optimization
8k 2 Mij performance of the algorithm [33].
There is much empirical and theoretical evidence in
support of enhancing the performance of GA via the
xijk 2 f0; 1g 8i 2 N incorporation of neighborhood search. According to these
8j 2 Oi ð12Þ research works, the competence of GAs in any given domain
8k 2 Mij depends on the amount of problem-specific knowledge
incorporated within them. But the integration of problem-
specific knowledge with GA is not restricted to utilizing
uiji0 j0 k 2 f0; 1g 8i; i0 2 N ; i < i0 specially designed local search operators. Initialization of GA
8j 2 Oi
ð13Þ with solutions generated by efficient heuristics and developing
8j0 2 Oi0
intelligent genetic operators are two other ways in which a GA
8k 2 Mij \ Mi0 j0
can be combined with domain-specific information [34].
In the following, the proposed hybrid GA is described in
detail. In Section 4.1, we explain the adopted representation
vijj0 2 f0; 1g 8i 2 N
ð14Þ scheme which can handle the characteristics of the FJSP-
8h j; j0 i 2 Qi
APP completely. Generation of the initial population is
Constraint 1 enforces the makespan to be equal to the described in Section 4.2. This subsection describes how
maximum completion time of all jobs. Constraint 2 states effective heuristics have been adapted to the special structure
that each operation must be assigned to exactly one of the problem and combined with each other to initialize the
machine. Constraint 3 requires the completion time of an hybrid GA intelligently. Calculations related to fitness
operation on machine k to be zero if this operation is not function are discussed in Section 4.3. Genetic operators,
assigned to the machine k. Constraint 4 states that the mainly crossover and mutation, are described in Section 4.4,
interval between the start and finish times of an operation is and finally in Section 4.5, we present a novel neighborhood
equal to the processing time of the operation. Furthermore, function based on which a local search procedure is provided
this equation satisfies the assumption that operation and embedded into the GA. The overall flow of the proposed
preemption is not allowed. Constraints 5 through 8 are algorithm is shown in Fig. 1.
two pairs of disjunctive constraints which are used to
prohibit two operations assigned to the same machine or 4.1 Representation scheme
belonging to the same job to be processed simultaneously.
Constraints 5 and 6 prevent two operations of different jobs The choice of a proper representation scheme is crucial
to overlap each other on the same machine, while to the performance of GA. In the proposed hybrid GA,
constraints 7 and 8 prevent the concurrent processing of the task sequencing list [35] is used to encode each
two operations of the same job regardless of the assigned candidate solution into a chromosome. A chromosome is
machine. Constraint 9 requires that the starting time of an composed of NO triples, where NO is equal to the total
operation cannot be less than the completion time of its number of operations. Each triple is represented as (i, j, k),
predecessors. This constraint guarantees the respect of where i, j, and k denote job, operation, and machine,
precedence relations among the operations of each job. respectively. The operation indicated at locus r has the rth
Int J Adv Manuf Technol (2012) 59:273–287 279

Start 4.2 Initialization

Some heuristics from scheduling literature are utilized in the

Generate initial population
using a combination of initialization phase of the proposed hybrid GA to provide a
random and heuristic methods certain quality of the initial population. To prevent premature
convergence of the algorithm and to ensure diversity of the
Gen_count := 1 initial population, these heuristic-based individuals are com-
bined with randomly generated solutions to create the initial
Apply tournament population. Constructing a solution for the FJSP-APP problem
selection operator involves three decisions: determining a route for each job
considering the precedence constraints, assigning each opera-
Apply modified tion to a feasible machine, and, finally, sequencing the assigned
order-based crossover operations on each machine. To select a feasible machine for
each operation, a procedure called approach by localization
Apply double (AL), which was proposed by Kacem et al. [36], is adopted.
insertion mutation This heuristic employs an iterative algorithm to assign each
operation to a feasible machine while both the processing
Apply machine times and workload of the machines are taken into account.
mutation In each iteration, the machine assignment for one operation is
done and, depending on how this operation is selected, two
Apply adaptive local variants of the AL heuristic are developed. In variant 1,
search operator which we call global min assignment (GMA), the operation
in which one of its feasible machines yields the minimum
Gen_count := Gen_count+1 workload among all machines is selected. In variant 2, called
random permutation method (RPM), the operation selection
is done randomly. The detail of these heuristics can be found
Is stop
Create new No
criterion in [35]. The machine assignment of PGMA percent of
population
satisfied? individuals in the initial population is done by the GMA
Yes rule, PRPM percent by the RPM rule, and machine assignment
for the remaining individuals is done randomly.
STOP To create a process plan for each job and define the
operation sequence on each machine, it is enough to specify
Fig. 1 Flowchart of the proposed hybrid GA priority of each operation through a proper procedure which
respects the precedence constraints. Given the priority of
priority. An example of the encoding scheme is shown in and the selected machine for each operation, an algorithm is
Fig. 2. The leftmost gene, (2, 1, 4), indicates that operation then applied to decode each individual in the initial
1 of job 2 holds the highest priority and should be population into a feasible schedule. To specify the priority
scheduled first on machine 4. Then, regarding (1, 2, 2), of each operation, two known dispatching rules, namely
operation 2 of job 1 should be scheduled on machine 2, shortest processing time (SPT) and most work remaining
and so on. (MWR), have been adopted and modified to suit the FJSP-
As it will be clear in subsequent sections, based on this APP problem. The selection of these two dispatching rules
encoding method, the initialization procedure and mutation is due to their better performance among common
and crossover operators are designed so that no infeasible dispatching rules in makespan problems [37].
individuals can be produced. This representation is capable The SPT rule used here is slightly different from the
of encoding all possible solutions and can embody all of the conventional SPT. Based on the conventional SPT rule, the
constraints of the problem. operation with the minimum processing time from the set of
schedulable operations is selected in each step. A schedu-
(2,1,4) (1,2,2) (3,1,1) (1,1,3) (3,2,3) (1,3,4) (2,2,3) (3,3,1) (1,4,2)
lable operation is an operation whose preceding operations
are scheduled and at most one operation from each job can
Operation: O21 O12 O31 O11 O32 O13 O22 O33 O14
appear in the schedulable operation set. In the SPT rule
Priority: 1 2 3 4 5 6 7 8 9
used in the proposed algorithm, due to the nonlinear
Machine Assignment: M4 M2 M1 M3 M2 M4 M3 M1 M2
precedence relations, more than one operation belonging
Fig. 2 Chromosome structure to each job may appear in the schedulable operation set.
280 Int J Adv Manuf Technol (2012) 59:273–287

»
However, as standard SPT, the operation with the shortest Step 4. Select the operation oij from CO(k) with the highest
processing time is selected from this set. priority, i.e., the operation which occurs leftmost in
»
Based on conventional MWR rule, the schedulable the chromosome. Append oij to the schedule so as
»
operation associated with the job having the most work to be started at its earliest starting time sij .
»
remaining is selected for scheduling in each iteration. When Step 5. Form SO(k+1) by removing oij from SO(k) and
»
the precedence relations are not linear, the number of inserting direct successors of oij into it. Increment
schedulable operations belonging to this job is not k by 1.
necessarily 1. In this case, based on the modified MWR Step 6. If SOðkÞ 6¼ ;, go to step 2; otherwise, terminate.
rule used in this paper, an operation is selected at random.
Assigning priority to individuals in the initial population The order in which operations appear in the corresponding
is accomplished partly by applying these two dispatching chromosome is not necessarily as their starting time order in
rules and partly by random assignment. PSPT is the percent the decoded schedule. After the decoding process, the
of initial chromosomes whose priority assignment is operation sequence in a chromosome is reordered according
performed through applying the SPT rule, PMWR is the to the starting time of the operations in the decoded schedule.
percent corresponding to MWR rule, and the priority This chromosome modification facilitates the inheritance of
assignment of the remaining chromosomes is done ran- sequencing information from parents to offspring [39].
domly. Note that random priority assignment should be
consistent with precedence relations, i.e., no operation can 4.4 Genetic operators
appear in a chromosome before its predecessors.
Selection, crossover, and mutation are the main genetic
4.3 Fitness evaluation operators which are applied to each successive generation to
produce improved population. Selection operator is responsi-
The fitness function is the means by which the quality of each ble for choosing chromosomes from the current population
individual in a population is evaluated. Since the objective of that will compose the next generation, either by being directly
the algorithm is to minimize the makespan, each individual is copied or by being evolved through crossover and mutation.
directly evaluated based on the makespan of the schedule it In the proposed hybrid GA, binary tournament selection
represents. So a procedure is needed to decode each individual method is used due to its ability to balance between
into a schedule. Considering that in a scheduling problem with exploration and exploitation of the mating pool [18].
makespan criterion the optimum solution is within the set of Common crossover operators which have been devel-
active schedules [38], a decoding procedure is developed to oped for permutation representation such as partially
convert each chromosome into a unique active schedule. mapped crossover, cycle crossover, and order-based cross-
This procedure is an extension of the Giffler and Thompson over can be applied to the priority-based representation
algorithm which was originally proposed to generate active adopted in this research. But these operators are likely to
schedules for a JSP [38]. The notations and steps of this produce infeasible individuals and the incorporation of
procedure are as follows: repairing procedures is time-consuming. We thus use a
special crossover operator which maintains the feasibility of
SO(k) The set of schedulable operations at stage k
individuals. This crossover operator, which is a combina-
CO(k) The set of candidate operations from which one
tion of two-point and order-based crossover, works as
should be scheduled at stage k
follows: First, two cut points are chosen randomly in the
sij The earliest time at which operation oij 2 SOðkÞ
two selected parents. The elements of left-side and right-
could be started
side subsections of parent 1 (P1) are passed on to the same
fij The earliest time at which operation oij 2 SOðkÞ
positions as in offspring 1 (O1). The remaining operations
could be finished
are brought in from parent 2 (P2) by a left-to-right scan and
mij The machine to which operation oij is assigned
placed into the positions between two cutting points of O1.
Offspring 2 (O2) can be obtained by reversing the roles of
Step 1. Let k:=0 and initialize SO(k) to contain all P1 and P2. Since the adopted crossover operator preserves
operations with no predecessors.
   precedence constraints, given two feasible parents, the
Step 2. Determinef þ :¼ min fij oij 2 SOðkÞ . If oþ
ij is the offspring produced through this operator are feasible too
operation by which f+ is defined, denote the [14]. The process of crossover is illustrated in Fig. 3.
machine assigned to and the job corresponding In the proposed algorithm, two mutation operators are
to oþ + +
ij by m and j , respectively. applied to each generation. The first mutation operator
Step 3. Build the set of candidate
  operations, CO :¼  alters the routing of jobs and the sequence of operations
oij 2 SOðkÞ sij < f þ ^ mij ¼ mþ _ j ¼ jþ assigned to each machine, and the second one alters the
Int J Adv Manuf Technol (2012) 59:273–287 281

Cut point 1 Cut point 2 Selected

element

Parent 1: (2,1,4) (1,2,2) (3,1,1) (1,1,3) (3,2,3) (1,3,4) (2,2,3) (3,3,1) (1,4,2)
Original
chromosome: (3,1,1) (2,1,4) (1,2,1) (1,1,3) (3,2,3) (1,3,4) (2,2,3) (3,3,1) (1,4,2)

Offspring 1: (2,1,4) (1,2,2) (1,1,1) (3,1,3) (1,3,1) (3,2,2) (2,2,3) (3,3,1) (1,4,2)
Mutated
(3,1,1) (2,1,4) (1,2,1) (1,1,3) (3,2,4) (1,3,4) (2,2,3) (3,3,1) (1,4,2)
chromosome:
Parent 2: (1,1,1) (3,1,3) (2,1,4) (1,2,3) (3,3,2) (1,3,1) (1,4,3) (2,2,4) (3,2,2) Fig. 5 Machine mutation operator

G=(N, A, E) is formed by nodes (N), conjunctive arcs (A),

Offspring 2: (1,1,1) (3,1,3) (2,1,4) (1,2,2) (1,3,4) (3,3,1) (1,4,3) (2,2,4) (3,2,2)
and disjunctive arcs (E). The node set is composed of
operations plus two dummy nodes, 0 and *, which are the
starting and ending nodes, respectively. The weight of
Parent 1: (2,1,4) (1,2,2) (3,1,1) (1,1,3) (3,2,3) (1,3,4) (2,2,3) (3,3,1) (1,4,2)
each node is given by the processing time of the
corresponding operation. Conjunctive arcs connect two
Fig. 3 Crossover operator
consecutive operations of the same job based on the
processing route of the job, and disjunctive arcs define the
machine assignment of an operation. Since known mutation sequence of operations on each machine. The longest path
operators such as insertion, inversion, and swap, have the from the starting to the ending node of the graph is the
potential to yield infeasible individuals, a new mutation critical path which is equal to the makespan of the
operator, called double insertion mutation, has been devised corresponding schedule. The operations on this path are
as the first mutation operator. In this mutation procedure, called critical operations. The disjunctive graph
two elements e1 and e2, which are placed at positions l1 and corresponding to the solution represented by the chromo-
l2, respectively (l1 <l2), are picked up randomly from the some in Fig. 2 is shown in Fig. 6. The critical path is
selected chromosome. First, e2 is inserted at the leftmost highlighted with bold arcs.
feasible position within the interval l1 to l2 (A position is The neighborhood function used in the local search is
feasible for an operation if it is located after the operation based on moving an operation. For a given solution with
predecessors and before the operation successors). Then, in disjunctive graph G, each neighbor is obtained by deleting
a similar manner, e1 is inserted at the rightmost feasible a node corresponding to an arbitrary operation from its
position within the interval l1 to l2. An example of mutation current position and inserting it at another feasible position
operation is shown in Fig. 4. in the graph. The resulting graph is feasible if it (1) has no
In the second mutation operator, called machine mutation, directed cycle, (2) the chosen operation is assigned to one
an operation is randomly chosen from the selected individual of its alternative machines, and (3) the precedence relations
and the machine assigned to this operation is changed to within the jobs are taken into account. Moving an
another alternative machine at random. An example of operation, say v, may result in the modification in some or
machine mutation operation is shown in Fig. 5. all of the following: the machine on which v is processed,
the sequence of operations on that machine, and the
4.5 Local search processing route of the job to which v belongs. In the local
search process, only neighbors which are obtained by
The disjunctive graph model [39] can be used to represent moving critical operations are considered. This is because
the feasible solutions of a FJSP-APP. A disjunctive graph

8 8
First Second
element 6 3
element O11 O13

O12 O14

Original 0 6 5 0
(3,1,1) (2,1,4) (1,2,1) (1,1,3) (3,2,3) (1,3,4) (2,2,3) (3,3,1) (1,4,2)
chromosome: 0 O21 O22
*
10 8
Mutated O31 12 O33
(3,1,1) (2,1,4) (3,3,1) (1,1,3) (3,2,3) (1,2,1) (1,3,4) (2,2,3) (1,4,2) O32
chromosome:

Fig. 4 Double insertion mutation operator Fig. 6 A disjunctive graph

282 Int J Adv Manuf Technol (2012) 59:273–287

moving non-critical operations could not reduce the make- Cmax (G) The length of the longest path from node
span of a given schedule [40]. 0 to node * in graph G
The detailed flowchart of the local search procedure is CP(G) The set of critical operations in graph G
shown in Fig. 7. The notations used in the flowchart are as G− (v) The disjunctive graph obtained by
follows: removing operation v from graph G. To
derive G− (v) from G, the node
M(v) The set of alternative machines for corresponding to v and the conjunctive
operation v and disjunctive arcs coming into
pvk The processing time of operation v on and going out of v are deleted, the job
machine k predecessor of v is connected to the job
J(v) The job to which operation v belongs successor of v by a conjunctive arc, and

Fig. 7 Detailed flowchart of

Begin
local search procedure

Set v to be the first operation in CP(G)

Set w to be the first operation in RT v
Set k to be the first machine in M(v)
Set u to be the first operation in SQ v (k )

'
Cmax := Cmax (G )

Set u to be the
(
E v (w) + Lv JS v ( w) <) No
1 first operation
'
C max − p vk in SQ v (k )

Yes Are all

the machines No Set k to be the
in M(v) next machine
checked? in M(v)
No E v (w) + Lv (MS v (u ) ) <
Set k to be the
'
C max − p vk Yes first machine
in M(v)
Are all Are all
Yes Set w to be the
Yes the operations the operations No
1 in SQ v (k ) in RT v next operation
v
checked? checked? in RT
No E v (u ) + Lv (JS v (w)) <
1 Set w to be the
No '
C max − p vk Yes first operation
Set u to be the in RT v
next operation Yes Are all
No Set v to be the
in SQ v (k ) the operations
next operation
in CP(G)
checked? in CP(G)
(
E v (u ) + Lv MS v (u ) <) Yes
'
C max − p vk Yes

No Stop

No C max (G ′(v, k , w, u ))
Yes
< C max
'

Are all
1
Yes the operations
in SQ v (k )
'
Cmax (
:= Cmax G ' (v, k , w, u ) )
checked?

No
Are all
Set u to be the Yes the operations
next operation 1 in SQ v (k )
in SQ v (k ) checked?

No
Set u to be the
next operation
in SQ v (k )
Int J Adv Manuf Technol (2012) 59:273–287 283

the machine predecessor of v is connected conjunctive arc connecting w and its job
to the machine successor of v by a successor is deleted, the disjunctive arc
disjunctive arc. connecting u and its machine successor is
SQv (k) The ordered set of operations which are deleted, a node corresponding to v with
processed on machine k in graph G− (v), weight pvk is added to the graph, w is
including dummy operation 0. The connected to v and v connected to the job
operations are ordered based on their successor of w with two conjunctive arcs,
starting time. and u is connected to v and v connected to
RTv The set of operations belonging to the job the machine successor of u with two
J(v) in graph G− (v) after which v can be disjunctive arcs.
placed without violating the precedence
relations. The operations are ordered based The local search improves a given solution with
on their starting time. This set includes corresponding graph G by examining the critical operations
node 0 if v has no predecessors. of G one by one. Each critical operation v is first deleted
JSv (w) The job successor of operation w in graph from G, which yields the graph G− (v). Then, reinserting v
G− (v), i.e., the operation which belongs to in all the feasible and promising positions is examined. The
the same job as w and directly follows w. process of finding a feasible position for v involves three
For the last operation in the processing decisions: the processing route of the job to which v
route of a job, operation * is considered as belongs, the machine on which v is processed, and the
the job successor. sequence of operations on that machine. For each critical
MSv (u) The machine successor of operation u in operation, all the possible choices of these three decisions
graph G− (v), i.e., the operation which is are considered in a hierarchical manner, i.e., first the job
processed on the same machine as u just processing route, then the alternative machine, and finally
after it. For the last operation in a machine the machine sequence. The feasible processing routes for J
sequence, operation * is considered as the (v) generated by inserting v into G− (v) are represented by
machine successor. the set of operations after which v can be placed (RTv).
Ev (w) The length of the longest path from node 0 Similarly, after the selection of a feasible machine from M
to node w in graph G− (v), including w (v), the possible sequences for this machine generated by
Lv (w) The length of the longest path from node w inserting v into G− (v) are represented by the set of
to node * in graph G− (v), including w operations after which v can be placed (SQv (k)).
G′ (v, k, w, u) The disjunctive graph obtained by reinserting Since combining all the alternative choices for these
v in graph G− (v), where k is the machine parameters leads to combinatorial explosion, a neighborhood
to which v is assigned, w is the operation reduction scheme based on Gao et al. [39] which reduces the
in processing route of J(v) after which v is search space significantly has been incorporated into the
placed, and u is the operation in machine k local search procedure. Based on this reduction technique,
sequence after which operation v is inserted. only promising positions are considered. A promising
To derive G′ (v, k, w, u) from G− (v), the position is a position in G− (v) at which inserting v results

Table 2 Parameters of the proposed hybrid GA

Parameter Description Candidate values Best value

Pop_size Population size – 200

Max_gen Maximum number of generations – 50
PGMA Percent of initial chromosomes whose machine assignment is done by the global min assignment rule 0, 10, 20, 30, 40, 50 30
PRPM Percent of initial chromosomes whose machine assignment is done by the random permutation method 0, 10, 20, 30, 40, 50 10
PSPT Percent of initial chromosomes whose priority assignment is done by the shortest processing time rule 0, 10, 20, 30, 40, 50 10
PMWR Percent of initial chromosomes whose priority assignment is done by the most work remaining rule 0, 10, 20, 30, 40, 50 30
Pc Crossover probability (%) 50, 60, 70, 80, 90 90
Pm Double insertion mutation probability (%) 0, 5, 10, 15, 20 10
Pmm Machine mutation probability (%) 0, 1, 2, 3, 4, 5 0.5
PDEV Initial value of the permitted relative deviation for a solution to undergo local search (%) 0, 5, 10, 15, 20 5
PLS Local search probability in final generation (%) 0, 10, 30, 50, 70 50
284 Int J Adv Manuf Technol (2012) 59:273–287

Table 3 Numerical results for small size problems

Problem No. of jobs No. of machines No. of operations Solution Adopted from

Exact Hybrid GA Reported

P1 5 5 20 33 33 38 Sundaram and Fu [43]

P2 6 5 18 27 27 27 Li et al. [9]
P3 5 6 21 27 27 28 Shao et al. [17]
P4 5 5 13 14 14 16 Moon et al. [12]
P5 8 6 20 23 23 – Lee et al. [22]

in a solution whose makespan is no longer than Cmax (G). values, which are usually spaced over a range evenly, are
More specifically, for a promising position, inequality 15 considered for each parameter. Then, one of the parameters
holds. The left-hand side of inequality 15 represents the is varied over its range while keeping other parameter
longest path from 0 to * which passes through node v in G′, values fixed at an initial value. The varying parameter is
where G′ is the graph generated by inserting v into G− (v), fixed at the best value obtained based on the performance
and the right-hand side of inequality 15 is equal to the criterion, i.e., the average of solutions found for some
makespan of G. randomly selected problems. The process is repeated for
other parameters, and an iteration is completed when all the
maxf E v ðwÞ ; Ev ðuÞg þ maxf Lv ðJS v ðwÞÞ ; Lv ð MS v ðuÞÞg parameters are fixed. The best setting of parameters found
ð15Þ
þ pvk < Cmax ðGÞ in an iteration is used as the initial value for the next
iteration. The iterations are repeated until no change in the
This inequality has been used in the local search procedure value of parameters occurs.
with two modifications. First, it was turned into four inequal- The set of candidate values and the best value found for
ities which are evaluated in a ranked order to identify and each parameter are presented in Table 2. The parameter
discard the infeasible and unpromising neighbor solutions as tuning process described above was applied to all the
soon as possible. Second, if during the search a better solution parameters, except two—pop_size and max_gen—which
than the current best one is found, then the right-hand side of were fixed at 200 and 50, respectively, because increasing
15 is replaced by its makespan. them gives a negligible improvement in solution quality.
A local search operator is applied to every individual in a
standard genetic local search. This approach not only
5 Experimental results increases computational effort but also does not necessarily
improve the performance of the hybrid algorithm [42]. In
5.1 Adjustment of hybrid GA parameters the proposed algorithm, adaptive strategies are adopted to
limit the local search application to more promising
The proposed hybrid GA includes several parameters individuals. For this purpose, a local search probability
whose values influence its performance. To adjust these has been introduced with which local search is applied to an
parameters, which are listed in Table 2, the iterative individual. Starting from zero, this probability linearly
parameter tuning method proposed by Franca et al. [41]
has been adopted. In this method, first a set of candidate
M1 O8,1 O3,1
M1 O2,3 O1,2 O4,5
M2 O1,1 O6,1 O3,2 O5,1 O1,2
M2 O5,2 O4,2 O3,3 O2,2
M3 O7,1 O2,1 O8,2
M3 O4,1 O4,3 O3,1 O1,3
M4 O5,3 O5,2 O7,2 O2,2
M4 O1,1 O5,1 O2,1 O5,4
M5 O8,3 O4,1 O8,4 O5,4
M5 O5,3 O3,5

M6 O3,2 O3,4 O4,4 O1,4 M6 O6,2 O3,3

2 4 6 8 10 12 14 16 18 20 22 24 26 time 2 4 6 8 10 12 14 16 18 20 22 time

Fig. 8 Gantt chart for the solution of P3 Fig. 9 Gantt chart for the solution of P5
Int J Adv Manuf Technol (2012) 59:273–287 285

Table 4 Numerical results for extended problems

Extended problem Base problem Demand size No. of jobs No. of machines No. of operations Makespan CPU time (s)

Best Average

EP1 P4 5 25 5 65 62.0 62.0 5

EP2 P4 10 50 5 130 124.0 124.0 32
EP3 P4 20 100 5 260 247.0 247.3 267
EP4 P5 5 40 6 100 105.0 105.0 27
EP5 P5 10 80 6 200 208.0 208.1 196
EP6 P5 20 160 6 400 416.0 416.4 635

increases with progressive generations up to PLS for the For each problem, the output of the proposed algorithm
final generation. Through this mechanism, in early gen- is compared with the exact solution and the solution
erations, the GA operators play the dominant role to reported in the paper from which the problem is adopted.
explore the promising solution areas, and in later gener- To obtain the exact solution, the mathematical model
ations, these areas are exploited more intensely by developed in Section 3.2 was solved by LINGO 9 solver.
increasing the rate of local search. The results show that the proposed algorithm can obtain
Fitness-based adaptation is another method to assign optimum solution in all cases. Also, in most cases, the
computational resources to local search more effectively proposed hybrid GA gives better results in comparison to
[42]. The idea is to bias the selection of individuals for local the results reported in the related works. The solution
search to those with better fitness function. To incorporate generated by the proposed algorithm for P3 and P5 are
this idea into the proposed hybrid GA, the relative deviation presented in Fig. 8 and Fig. 9, respectively.
of an individual’s fitness value from the best value found so
far is used as the criterion to select candidates for local 5.3 Numerical analysis for large-sized problems
search. Specifically, the local search operator is applied to an
individual if its relative deviation satisfies inequality 16. To test the proposed algorithm for large-sized problems,
» two of the benchmark problems (P4 and P5) are selected,
f ði; t Þ  f ðt  1Þ
< PDEV  idleðt  1Þ ð16Þ and by changing the total demand for each job, six
f » ð t  1Þ extended problems have been developed. The character-
In 16, f(i, t) is the fitness value of individual i in istics and numerical results for the extended problems are
generation t, f* (t − 1) is the best fitness value found until presented in Table 4.
generation t − 1, idle(t − 1) is the number of consecutive The proposed algorithm was run 10 times for each
generations with no improvement in the best solution until problem to reduce the stochastic effects, and the average
generation t − 1, and PDEV is the initial value for the
permitted relative deviation.
The proposed hybrid GA was coded in Visual Basic
Application and implemented on a personal computer with
a Pentium IV 2.40-GHz CPU.

5.2 Numerical analysis for small-sized problems

To evaluate the performance of the proposed hybrid GA, five

problems have been adopted from the IPPS literature. The
specifications of each problem and the solution results are
summarized in Table 3. The benchmark problems are selected
so that their objective function and assumptions are the same
as the ones considered in this paper, and thus the results are
comparable. The only exception is P5 which is adopted from
Lee et al. [22]. Since the problem they have considered is
slightly different, only the data (disregarding due dates and
possibility of outsourcing) are taken from Lee et al. [22]. Fig. 10 Convergence curves for EP5
286 Int J Adv Manuf Technol (2012) 59:273–287

and the best makespan obtained over 10 runs as well as the The numerical results showed that the proposed hybrid
average CPU time are reported. To illustrate the conver- GA obtains optimal or near-optimal solutions and has a
gence performance of the algorithm, its convergence rates is good convergence performance. Considering the efficiency
compared with two other cases in Fig. 10: when no local of the proposed algorithm, it can be used in planning and
search is applied and when local search is applied with a scheduling systems of industrial and service firms. Because
certain probability in all generations. Figure 10, which of the stochastic behavior of parameters and dynamic
shows the average fitness function for EP5 over 10 runs, changes of characteristics in most manufacturing environ-
demonstrates the advantages of the proposed algorithm: It ments, as the directions for future studies, the proposed
yields better solutions and converges more quickly while algorithm can be extended to include stochastic processing
fighting premature convergence efficiently. times and dynamic conditions. Developing efficient algo-
rithms for the IPPS problem in which other performance
criteria such as total production cost, number of tardy jobs,
6 Conclusion machine workload, tardiness, and lateness are considered
separately or together within a multi-objective framework is
In this paper, the problem of integrating process planning another interesting direction for future studies.
and scheduling, which has extensive applications in modern
manufacturing systems, was considered. Given a set of
jobs, each of which consists of a number of operations, the
References
aim of the problem is to find the optimum schedule. There
are alternative processing routes based on precedence
relations for each job and alternative machines for each 1. Kim KH, Egbelu PJ (1999) Scheduling in a production environ-
ment with multiple process plans per job. Int J Prod Res 37:2725–
operation. 2753. doi:10.1080/002075499190491
The problem was first formulated as a mixed integer 2. Kang M, Han J, Moon JG (2003) An approach for interlinking
linear model. Although the numerical results on some test design and process planning. J Mater Process Technol 139:589–
problems demonstrated the ability of the mathematical 595. doi:10.1016/S0924-0136(03)00516-8
3. Guo YW, Li WD, Mileham AR, Owen GW (2009) Applications
model to obtain exact solutions, as the size of problem of particle swarm optimisation in integrated process planning and
increases due to the NP-hardness nature of the problem, scheduling. Robot Comput-Integr Manuf 25:280–288.
finding exact solutions becomes intractable. Therefore, a doi:10.1016/j.rcim.2007.12.002
hybrid GA which integrates the advantages of evolutionary 4. Zhou X, Qiu Y, Hua G, Wang H, Ruan X (2007) A feasible
approach to the integration of CAD and CAPP. Comput-Aided
algorithms and local search was developed to generate Des 39:324–338. doi:10.1016/j.cad.2007.01.005
high-quality solutions in a reasonable time. The proposed 5. Tan W, Khoshnevis B (2000) Integration of process planning and
algorithm can determine the optimum schedule for an IPPS scheduling—a review. J Intell Manuf 11:51–63. doi:10.1023/
problem by simultaneously generating a feasible processing A:1008952024606
6. Saygin C, Kilic SE (1999) Integrating flexible process plans with
route for each job which respects nonlinear precedence scheduling in flexible manufacturing systems. Int J Adv Manuf
relations, assigning each operation to an appropriate Technol 15:268–280. doi:10.1007/s001700050066
machine, and sequencing the operations assigned to each 7. Jain A, Jain PK, Singh IP (2006) An integrated scheme for
machine. The unique features of the proposed hybrid GA process planning and scheduling in FMS. Int J Adv Manuf
Technol 30:1111–1118. doi:10.1007/s00170-005-0142-6
can be briefed as follows: 8. Li X, Gao L, Zhang C, Shao X (2010) A review on integrated
process planning and scheduling. Int J Manuf Res 5:161–180.
& Employing specialized genetic operators which enhance doi:10.1504/IJMR.2010.031630
the global search ability of the algorithm and always 9. Li X, Gao L, Shao X, Zhang C, Wang C (2010) Mathematical
modeling and evolutionary algorithm-based approach for integrated
yield feasible offspring process planning and scheduling. Comput Oper Res 37:656–667.
& Initializing the GA population with quality individu- doi:10.1016/j.cor.2009.06.008
als which are generated based on problem-specific 10. Morad N, Zalzala A (1999) Genetic algorithms in integrated
heuristics process planning and scheduling. J Intell Manuf 10:169–179.
doi:10.1023/A:1008976720878
& Incorporating an ad hoc local search procedure into the 11. Zhao F, Hong Y, Yu D, Yang Y, Zhang Q (2010) A hybrid particle
evolutionary algorithm whose efficiency is improved swarm optimisation algorithm and fuzzy logic for process
through applying an appropriate neighborhood reduc- planning and production scheduling integration in holonic
tion scheme manufacturing systems. Int J Comput Integr Manuf 23:20–39.
doi:10.1080/09511920903207472
& Selective application of local search through adopting 12. Moon C, Lee YH, Jeong CS, Yun YS (2008) Integrated process
adaptive methods during the selection of individuals for planning and scheduling in a supply chain. Comput Ind Eng
local search 54:1048–1061. doi:10.1016/j.cie.2007.06.018
Int J Adv Manuf Technol (2012) 59:273–287 287

13. Chryssolouris G, Chan S, Cobb W (1985) Decision making on the International Conference on Systems, Man, and Cybernetics
factory floor: an integrated approach to process planning and 5:3212–3217. doi:10.1109/ICSMC.2000.886495
scheduling. Robot Comput-Integr Manuf 1:315–319. doi:10.1016/ 28. Tan W, Khoshnevis B (2004) A linearized polynomial mixed
0736-5845(84)90020-6 integer programming model for the integration of process
14. Kim YK, Park K, Ko J (2003) A symbiotic evolutionary algorithm planning and scheduling. J Intell Manuf 15:593–605.
for the integration of process planning and job shop scheduling. doi:10.1023/B:JIMS.0000037710.80847.b6
Comput Oper Res 30:1151–1171. doi:10.1016/S0305-0548(02) 29. Sherali HD, Sarin SC, Desai R (1990) Models and algorithms for
00063-1 job selection, routing, and scheduling in a flexible manufacturing
15. Baykasoglu A, Özbakır L (2009) A grammatical optimization system. Ann Oper Res 26:433–453
approach for integrated process planning and scheduling. J Intell 30. Özgüven C, Özbakır L, Yavuz Y (2010) Mathematical models for job-
Manuf 20:211–221. doi:10.1007/s10845-008-0223-0 shop scheduling problems with routing and process plan flexibility.
16. Li WD, McMahon CA (2007) A simulated annealing-based Appl Math Model 34:1539–1548. doi:10.1016/j.apm.2009.09.002
optimization approach for integrated process planning and 31. Manne AS (1960) On the job-shop scheduling problem. Oper Res
scheduling. Int J Comput Integr Manuf 20:80–95. doi:10.1080/ 8:219–223
09511920600667366 32. Choy KL, Leung YK, Chow HKH, Poon TC, Kwong CK, Ho
17. Shao X, Li X, Gao L, Zhang C (2009) Integration of process GTS, Kwok SK (2011) A hybrid scheduling decision support
planning and scheduling—a modified genetic algorithm-based model for minimizing job tardiness in a make-to-order based
approach. Comput Oper Res 36:2082–2096. doi:10.1016/j. mould manufacturing environment. Expert Syst Appl 38:1931–
cor.2008.07.006 1941. doi:10.1016/j.eswa.2010.07.125
18. Li X, Shao X, Gao L, Qian W (2010) An effective hybrid 33. Wang L, Zhang L, Zheng DZ (2006) An effective hybrid genetic
algorithm for integrated process planning and scheduling. Int J algorithm for flow shop scheduling with limited buffers. Comput
Prod Econ 126:289–298. doi:10.1016/j.ijpe.2010.04.001 Oper Res 33:2960–2971. doi:10.1016/j.cor.2005.02.028
19. Candido MAB, Khator SK, Barcia RM (1998) A genetic 34. Hart WE, Krasnogor N, Smith JE (2005) Memetic evolutionary
algorithm based procedure for more realistic job shop scheduling algorithms. In: Hart WE, Krasnogor N, Smith JE (eds) Recent
problems. Int J Prod Res 36:3437–3457. doi:10.1080/ advances in memetic algorithms. Springer, Berlin, pp 3–27
002075498192148 35. Pezzella F, Morganti G, Ciaschetti G (2008) A genetic algorithm
20. Kis T (2003) Job-shop scheduling with processing alternatives. Eur J for the flexible job-shop scheduling problem. Comput Oper Res
Oper Res 151:307–332. doi:10.1016/S0377-2217(02)00828-7 35:3202–3212. doi:10.1016/j.cor.2007.02.014
21. Mati Y, Xie X (2003) A polynomial algorithm for a two-job shop 36. Kacem I, Hammadi S, Borne P (2002) Approach by localization
scheduling problem with routing flexibility. Proceedings of IEEE and multiobjective evolutionary optimization for flexible job-shop
International Conference on Robotics and Automation 1:157–162. scheduling problems. IEEE Trans Syst Man Cybern C Appl Rev
doi:10.1109/ROBOT.2003.1241589 32:1–13. doi:10.1109/TSMCC.2002.1009117
22. Lee YH, Jeong CS, Moon C (2002) Advanced planning and 37. Zhou H, Feng Y, Han L (2001) The hybrid heuristic genetic
scheduling with outsourcing in manufacturing supply chain. Comput algorithm for job shop scheduling. Comput Ind Eng 40:191–200.
Ind Eng 43:351–374. doi:10.1016/S0360-8352(02)00079-7 doi:10.1016/S0360-8352(01)00017-1
23. Moon C, Seo Y, Yun Y, Gen M (2006) Adaptive genetic algorithm 38. Baker KR, Trietsch D (2009) Principles of sequencing and
for advanced planning in manufacturing supply chain. J Intell scheduling. Wiley, New York
Manuf 17:509–522. doi:10.1007/s10845-005-0010-0 39. Gao J, Sun L, Gen M (2008) A hybrid genetic and variable
24. Shen W, Wang L, Hao Q (2006) Agent-based distributed neighborhood descent algorithm for flexible job shop scheduling
manufacturing process planning and scheduling: a state-of-the- problems. Comput Oper Res 35:2892–2907. doi:10.1016/j.
art survey. IEEE Trans Syst Man Cybern C Appl Rev 36:563–577. cor.2007.01.001
doi:10.1109/TSMCC.2006.874022 40. Mastrolilli M, Gambardella LM (2000) Effective neighbourhood
25. Wong TN, Leung CW, Mak KL, Fung RYK (2006) Integrated functions for the flexible job shop problem. J Sched 3:1–20.
process planning and scheduling/rescheduling—an agent-based doi:1 0.10 02/(SICI)10 99-1 425 (2 000 01/0 2)3:1 <3 ::AID-
approach. Int J Prod Res 44:3627–3655. doi:10.1080/ JOS32>3.0.CO;2-Y
00207540600675801 41. Franca PM, Mendes A, Moscato P (2001) A memetic algorithm
26. Leung CW, Wong TN, Maka KL, Fung RYK (2010) Integrated for the total tardiness single machine scheduling problem. Eur J
process planning and scheduling by an agent-based ant colony Oper Res 132:224–242. doi:10.1016/S0377-2217(00)00140-5
optimization. Comput Ind Eng 59:166–180. doi:10.1016/j. 42. Hart WE (1994) Adaptive global optimization with local search.
cie.2009.09.003 Doctoral dissertation, University of California, San Diego
27. Ey H, Sackmann D, Mutz M, Sauer J (2000) Adaptive job-shop 43. Sundaram RM, Fu SS (1988) Process planning and scheduling—a
scheduling with routing and sequencing flexibility using expert method of integration for productivity improvement. Comput Ind
knowledge and colored Petri nets. Proceeding of the IEEE Eng 15:296–301. doi:10.1016/0360-8352(88)90102-7