Sequencing and scheduling of part and tool in a flexible manufacturing system

with non-identical parallel machines

Naveen Kumar1, Pankaj Chandna2
1
Department of Mechanical Engineering
Sir Padampat Singhania University, Udaipur-313601, INDIA
2
Department of Mechanical Engineering
National Institute of Technology Kurukshetra-136118, INDIA

ABSTRACT
Scheduling problem in a flexible manufacturing system (FMS) has been considered to be NP-
hard and it is quite cumbersome to solve such problems by conventional optimization techniques.
Recently, researchers have focused on metaheuristic techniques to solve FMS scheduling
problems, however, a significant part of literature dealing with FMS scheduling problems,
consider parallel identical machine environment. FMS scheduling problems with non- identical
machines environment have received less attention from the researchers though quite prevalent
in the industry. This paper attempts to solve sequencing and scheduling of part and tool with an
aim to minimize the makespan, for a flexible manufacturing system which has non-identical
machines and a common tool magazine. The tools in the FMS are shared among different
machines from common tool magazine through a material handling system And also each tool
type is single in number. The authors have proposed two heuristic techniques, priority
dispatching rule algorithm (PDRA) and modified genetic algorithm (MGA). PDRA, are the
simple priority rules and are most frequently applied to scheduling problems because of their
ease of implementation as compared to excel algorithms. The proposed MGA uses three parent
crossover and a diversity operator to find the optimal solution to the sequencing and scheduling
problem. Three parents crossover produces three offspring by linear combination of three parents
to improve the fitness and diversity of the offspring. The diversity operator further diversifies the
generated offspring. Both the techniques use Giffler and Thompson procedure to generate active
feasible schedule. The proposed algorithms are tested on a set of 3 test problems, each for two
machines, three machines and four machines environment. The result for the makespan obtained
by the two proposed algorithms is compared by calculating a parameter called percentage
difference from the best (PDB). The analysis of result reveals that MGA gives better results for
all the test problems.

Keywords: Scheduling, Flexible manufacturing system, Modified genetic algorithm

NOMENCLATURE

estik Earliest start time of kth operation of ith job

th
eft ik Earliest finishing time k operation of i th job
DT Datum time
n Number of jobs

1
IP Initial population size calculated by (n 2 n) / 2
s Population size of the selection pool A . s 0.5IP
Ji Job number ( i 1 to n )
J ik kth operation of ith job
C.O.J Conflict of jobs
Mj Machine number (j=1 to m )
m Number of machines
MT Makespan time
p Diversity probability
MGA Modified Genetic Algorithm
MAXGEN Maximum number of generations for MGA
P.O.J Processed operation of job
Pl l th Chromosome ( l =1 to 3IP )
Pl i i th element in l th chromosome ( i =1to n )
th
t ijk Processing time of k operation of i th job on j th machine

1. INTRODUCTION

The fast changing demand of customers necessitates the manufacturer to produce verities of
product types with minimum lead time. The advent of computer numerical controlled machines
has resulted in the development of a flexible manufacturing system (FMS). In Flexible
manufacturing systems verity of part types in smaller batch size can be produced with minimum
setup time. An FMS consists of number of general purpose numerically controlled machines,
interconnected by automated material handling system and the control is being done by a central
computer. With an FMS flexibility of job shop and efficiency of mass production can be
attained.

2. LITERATURE REVIEW

There are four stages of decision problems for successful installation and implementation of an
FMS: designing, planning, scheduling and control (Stecke, 1983). Major part of available
research in FMS is related to scheduling problem as FMS scheduling is much more complicated
than job shop scheduling due to its inherent flexibility. A few researchers in their surveys
presented that lack of tooling consideration leads to poor performance of an FMS (Gray et al.
1993, Veeramani et al. 1992). Most of the research reported in the field of scheduling considers
part and tool flow as separate issues and often, the effect of one of the pair on the other is
neglected (Praharan et al., 2006). In this work part and tool flow have been considered
simultaneously. There are two approaches to solve scheduling problem in an FMS; part
movement approach or tool movement approach. In part movement approach the parts are routed
through different machines whereas in the tool movement approach the cutting tools are shared
among different machines by using a tool transport system. The second approach is useful when
the tool cost becomes significant proportion of the part cost, sometimes as high as 25-30%
(Selim and Siraceddin, 1999). Scheduling problems under the part movement policy have been
studied extensively in many papers but there are relatively few research articles on the problems

2
considering the tool movement policy (Roh and Kim, 1997). Solving the loading and scheduling
sequentially does not produce optimal results as the solution of loading problem becomes a
limitation for the scheduling problem (Roh and Kim, 1997). Considering these aspects, in this
paper the part scheduling and tool allocation problem is solved simultaneously.

FMSs with non-identical machine environment are not uncommon in the industry, but there are
few research articles (Caker at al. 2008, Raja et al. 2008, Balin, 2011) available in the literature.
In this work we considered non-identical machining environment.

Scheduling of parts in an FMS even without additional resources is a NP-hard problem and
requires heuristic approach for solution (Agnetis et al., 1997). An FMS scheduling problem is
generally solved by using optimization based methods and heuristic based methods.
Optimization methods, e.g. branch and bound, integer programming, dynamic programming etc.,
become cumbersome to solve with increase in problem size. Therefore, efficient heuristic
methods should be developed for large sized problems. Priority dispatching rule algorithm
(PDRA) are the popular simple heuristics which can be applied to any FMS scheduling problem
with ease. Giffler and Thompson algorithm (Giffler and Thompson, 1960) was developed for the
job shop manufacturing environment can also provide quality results for an FMS scheduling
problem (Nasciment, 1993). Prabhaharan et al. (2006) used combined PDRA and simulated
annealing algorithm for the scheduling and sequencing of parts and tools in a flexible
manufacturing cell (FMC). Udhayakumar and Kumanan (2010) used Ant colony algorithm to
schedule tools and parts in an FMS environment having identical machines with an objective to
minimize makespan. Recently, researchers have been increasingly using artificial intelligence
techniques for solving scheduling problem in FMSs. One of such intelligent probabilistic search
technique is genetic algorithm (GA) (Holand, 1975). According to Goldberg (1989), GAs can be
adapted to treat the complexity levels required to provide adaptive search at the required
robustness. Genetic algorithm has been successfully used to solve verities of optimization
problems including problems related to manufacturing, i.e scheduling, process planning and
system design. Keung et al. (2003) in their paper solved sequencing problem in FMS
environment with one material handling device, by using GA. The performance measure was to
minimize the penalty cost. Shanker et al. (2005) in their research used GA to solve scheduling
considering robots and AGVs, in order to minimize the makespan and delay time. Chan et al.
(2008) used concept of dominant genes in GA to solve the scheduling problem in an FMS. They
compared the results for the minimization of makespan with other techniques like Ant Colony
Optimization (ACO) and Petri Nets (PNs) and found that the proposed algorithm gives better
results. Kim et al. (2007b) used Symboitic Evolutionary Assymmetric Multileveled Algorithm
(ASMEA) to solve the scheduling problem in an FMS environment with four types of
flexibilities i.e. machine, tools, processing and sequencing. They reported that the results
obtained were of high quality and the speed of convergence of the algorithm was also fast. Gang
and Wu (2004) solved sequencing problem in FMS by using hybrid approach (GA and PN). In
GA evolutionary process; crossover and mutation are the important search operators. Most of the
research which use GAs in solving FMS scheduling problem use two parent crossover in order to
produce offspring. Elsayed et al. (2014) proposed a new GA with multi-parent crossover with a
diversity operator. They tested this algorithm on number of optimization problems and the results
show its better performance. In this paper we propose a modified GA which uses three-parent
crossover with a diversity operator.

3
 3. PROBLEM FORMULATION

In this paper problem of simultaneous sequencing and scheduling of tools and part has been
attempted for a flexible manufacturing system with non-identical work centres (WCs) and a
CTM. Each work center has different velocity ( V j ) due to its maximum available power. The
processing time of k th operation of i th job on different machines can be determined by using
following expression:
ti1 j V1 ti 2 j V2 (1)
None of WCs has dedicated tool magazine. All the tools are stored in the CTM and shared
among all the work centres through single tool grip. Each tool type in CTM is one in quantity.
The processing requirements of the jobs are:
All the machines are capable of doing all the operations of any of the jobs.
Each job has a number of operations and individual operation requires a specific tool.
Sequence of operations and respective tools vary from job to job.
The operations sequence and type of tool required for each operation with processing
time of operations are pre-specified.
One job can be processed on one work centre/tool at a time.
All the operations of a job are processed during a single work centre visit.
A job does not visit the same work centre twice.
Availability of tools in each verity is considered as one, which can be used for more than
once for any job.
The operation time of a job includes the loading, unloading, tool changeover and setup
times (both tool and job) along with the processing time.
A work centre is capable of doing many operations.
Once the operation is over the tool returns to the common tool magazine with negligible
transfer time immediately and is available for the next operation.

Considering the high initial cost of investment in FMSs, it is important for the manufacturer to
maximize the machine utilization. In this context, maximum makespan that increases the
utilization of machines is considered as the appropriate measure of system performance.

4. PROPOSED HEURISTICS

PDRA and Modified Genetic Algorithm (MGA) heuristics are used to find the optimal or near
optimal solution for the proposed problem. Both the algorithm use a new algorithm (PCN_SCH)
which is an extension of logic of GT algorithm, for generating active feasible schedule of joint
operation-tool schedule. The new algorithm (PCN_SCH) is described as below:

4.1 PCN_SCH algorithm

The proposed new algorithm (PCN_SCH) is developed by extending the concept of GT

algorithm. Whenever two or more jobs require the same tool at the same time (conflict of jobs),
the tool is assigned arbitrarily to a job and hence an active feasible schedule is obtained. The
optimal solution can be obtained by evaluating the feasible schedules with the desired objective
criterion.

4
Step 1: Construct a table having columns one for each work centre, job assigned, datum time and
processed operation of a job. The column for each work centre is further subdivided for
each tool type.
Step 2: Select randomly the number of jobs equal to the number of machines and assign one job
to one machine arbitrarily. Enter in the first line of the table estik + processing time in
appropriate block of work centre under appropriate tool. The total of the two values give
earliest finish time ( eft ik ) of the immediately waiting operation of the assigned job.
Step 3: Enter the datum time ( DT ) equal to the smallest of the eft ik entries, in the appropriate
column.
Step 4: Select the jobs whose eft in the table matches with the current datum time. In case there
are more than one jobs whose eft matches with current datum time; select one job
arbitrarily.
Step 5: Check for the tool conflict? If yes, go to step 6 else go to step 7.
Step 6: List the operations of all the jobs contending for the same tool. Select a job arbitrarily
and earmark its operation. Assign the tool to the earmarked operation of the job.
Step 7: Process the earmarked operation along with its work centre and tool. Enter the job
number and operation completed in the format of i k in the column P.O.J .
Step 8: Now, write down the earliest finishing time of next waiting operation as earliest finishing
time of immediate processed operation plus processing time of the waiting operation.
Also, update the eft of the contending operation of other jobs (who are not assigned the
required tool during conflict resolution) as sum of earliest finish time of the operation of
the job assigned with the tool plus processing time.
Step 9: Update the datum time as minimum of earliest finishing time. Check, if all the operations
of all the jobs are scheduled. If yes, go to step 10, else go to step 4.
Step 10: Update the DT as largest entry of earliest finishing time. The table gives the active
feasible schedule and DT represents the makespan time.

4.1.1 Numerical example

A numerical example of scheduling problem of three jobs and three tools on two work centres is
taken to explain the detailed procedure of the proposed algorithm (PND_SCH). This algorithm
generates active feasible schedule based on input sequence of the jobs generated by proposed
algorithms. In this paper, we have used two PDRs namely shortest processing time (SPT) and
longest processing time (LPT) which use processing time for the sequencing of jobs. In this
illustration we describe SPT rule to generate the sequence of jobs. According to this rule, the
jobs are sequenced in non-decreasing order of their processing time. The problem data are
summarized in Table 1.
Table 1- Job-operation-tool matrix
Machine No
Job No Operation No Tool Type
1 2
I 2 4 2
1 Ii 1 10 5
Iii 3 8 4
I 3 6 3
2 Ii 2 5 2.5
Iii 1 6 3
3 I 1 10 5

5
 Ii 3 6 3
Iii 2 8 4

The velocities of machines is assumed to be ( V1 , V2 ) (1,2) . From the given problem data, job
number 2 has the least total processing time and the job number 3 has the highest total
processing time. Hence the sequence generated by SPT is 2-1-3. The procedure for generation of
active feasible schedule is shown in the Table 2.

In row number 5 in Table 2, the operation number 3 of job number 2 is waiting on machine
number 1 and on machine 2, operation number 1 of job number 3 is waiting to be processed and
both require same tool type 1. This is a condition for job conflict and is resolved by randomly
selecting job 3 to be supplied with the tool. Now, the operation 3 of job 2 has to wait till the tool
1 is free from the assigned job.

Table 2- Active schedule generation procedure

Machine No 1 Machine No 2 Job Assigned
Tool Type Tool Type
DT COJ POJ
Machine 1 Machine 2
1 2 3 1 2 3
0+6 0+2 2 1 2 - 11
6 2*
0+6 2+5 2 1 6 - 21
6* 7
6+5 2+5 2 1 7 - 12
11 7*
6+5 7+4 2 1 11 - 13&22
11* 11*
11+6 11+5 2 3 16 2,3 31
17 16*
16+6 16+3 2 3 19 - 32
22 19*
16+6 19+4 2 3 22 - 23
22* 23
19+4 - 3 23 - 33
23*

4.3 Modified genetic algorithm

It is important to control the distribution of offspring in comparison to their parents in order to

improve the performance of GAs. The narrow distribution of offspring lead to premature
converge of the algorithm and wider distribution of offspring result in the algorithm taking a
long time to converge to the optimal solution. In this context, in this research paper we use a
modified GA with three-parent crossover and a diversity operator. Three parents crossover
produces three offspring by their linear combination to improve the fitness and diversity of the
offspring. The diversity operator further diversifies the generated offspring.

Initial population of chromosomes of size IP equal to (n2-n)/2 is generated. Each chromosome in

this population contains numbers from 1 to j randomly generated by permutation. The structure
of the generated chromosome for five jobs is shown in the Figure 2.

Figure 2- Structure of a chromosome

5 3 1 2 4
6
 job number
sequence of jobs
The numbers in the chromosome represent the job numbers and from left to right is their
sequence of assignment on machines. If there are 2 machines then the job number 5 is
assigned to machine 1 and job number 3 is assigned to machine 2.
From the generated initial population s best solutions are selected and stored in an archive pool
A. Now using tournament selection ( 2), the chromosomes are selected and stored in the
selection pool B. The size of pool is 3IP. Three consecutive parents from the selection pool are
selected and are arranged in increasing value of fitness that is f ( P1 ) f ( P2 ) f ( P3 ) . Now the
crossover of these three chromosomes P1 , P2 and P3 is performed by their linear combinations.
The crossover is performed in two steps. First, the intermediate offspring IM 1 , IM 2 and IM 3
are generated by single point crossover of P2 with P3 , P3 with P1 and P1 with P2 . Final
offspring O1 , O2 , and O3 are generated by two point crossovers of P1 with IM 1 , P2 with IM 2
and P3 with IM 3 . The crossover of P1 , P2 and P3 to generate O1 is illustrated in Figure 3(a), (b).

Figure 3(a) - Single point crossover of parent chromosomes and generation of offspring IM1
crossover point x1
P2 4 1 3 5 2

P3 1 3 2 4 5

IM 1 4 1 2 4 5

Figure 3(b) - Final offspring obtained by crossover of IM1 with P1 at point x1and x2

IM 1 4 1 2 4 5

x1 x2
P1
5 3 1 2 4

O1
4 1 1 2 5

The offspring obtained after crossover and diversity operator may provide an infeasible solution.
As shown in Figure 3(b) that the generated offspring O1 has job number 1 at two places,
whereas, job number 3 is missing. A repairing algorithm is used to repair each of the infeasible
offspring by replacing the repeating job number by that of the missing job number in the
chromosome. The offspring and archive pool individuals are then merged and the best s
individuals are selected to as new population for the next generation.

7
Various parameters for the modified GA, like s 0.5 * IP , p 0.05 and MAXGEN 100 are
decided based on trial runs of modified genetic algorithm on five different scheduling problems
from the literature(Prabaharan et al. 2006] for m=4 case.

4.3.1 Steps of MGA

Step 1: Set p 0.05 , IP (n 2 n) / 2 , MAXGEN 100 and s 0.5 * IP .

Step 2: An initial population of size is generated by random permutation of number of jobs j.
Step 3: An archive A is filled with s best solutions.
Step 4: Using the tournament selection process with size TC=2, a selection pool B is filled with
selected chromosomes with size 3* .
Step 5: Select three consecutive chromosomes from the selection pool and arrange them so
that f ( P1 ) f ( P2 ) f ( P3 ) . Replace any duplicate chromosome with randomly selected
chromosome from the selection pool.
Step 6: Generate two random numbers x1 and x2 as crossover points for generation of each
offspring. The three offspring are generated through the mating of selected chromosomes
as mentioned below.
[ x 1....x2 ] [ x 1......j ] [1....x1 & x2 1..... j ]
O1 P1 1 [ P2 P3 1
[1, 2.....x1 ]
] (2)
[ x1 1....x2 ] [ x1 1......j ] [1....x1 & x2 1..... j ]
O2 P2 [ P3 P1
[1, 2.....x1 ]
] (3)
[ x 1....x ] [ x 1......j ]
O3 P3 1 [ P1 P2 1 ][1....x1 & x2 1..... j ]
[1, 2.....x ]
2 1
(4)
Step 7: Generate a number u; u [0,1]. If u<p, apply the diversity operator as per following
pseudocode (Elsayed et al. 2014):
for i {1,2....3IP}
for j {1,2....n}
generate u [0,1]
if u<p
replace
End
End
End

Step 8: Check the feasibility of all the generated offspring. If not feasible go to step 9, else go to
step 10.
Step 9: Repair the infeasible offspring by replacing the repeating job number in the offspring by
a missing job number.
Step 10: using PND_SCH algorithm and calculate the makespan time.
Step 11: Combine chromosomes from archive pool and offspring. Select the best individuals to
make a new population of size IP.
Step 12: Check, if termination criterion is met? If yes go to step 12 Otherwise go to step 2.
Step 13: Stop and note down the optimal sequence

5. RESULTS AND DISCUSSIONS

8
To obtain the test results for makespan time for the proposed sequencing and scheduling
problem, a set of 3 problems, each for m=2, m=3 and m=4, are taken from the literature
(Prabaharan et al. 2006). The velocity of machines are given as ( V1 , V2 , V3 , V4 ) (1,2,4,5) . The
results obtained for makespan time using proposed modified GA are compared with the results
obtained by using SPT and LPT dispatching rules in Table 3, Table 4 and Table 5.

The best value of makespan time is marked with boldface, whereas the second best result is
marked with #. The computational time taken by the proposed MGA is shown in the Figure 5.
Figure 5. Computational time taken by MGA

Table 3- Comparison of results obtained for m=2

Problem No No of Jobs No of SPT algorithm LPT Algorithm Proposed
1 6 Operations
4 124.5 94# MGA
82.5
2 8 7 222.5# 239 196.5
3 12 4 222 188# 167.5

Table 4- Comparison of results obtained for m=3

Problem No No of Jobs No of SPT algorithm LPT Algorithm Proposed
1 6 Operations
4 47.5# 87 MGA
35.5
2 8 7 119.25# 125.5 90.25
3 12 4 118 104.75# 79.25

Table 5- Comparison of results obtained for m=4

Problem No No of Jobs No of SPT algorithm LPT Algorithm Proposed
1 6 Operations
4 54.45# 59.65 MGA
35.7
2 8 7 85.85# 101.35 62.95
3 12 4 64.7# 85.25 52.4
The results show that the MGA algorithm gives better results, for all the problems than SPT and
LPT techniques. To measure the performance of proposed MGA in giving better results over
SPT and LPT dispatching rules, percentage difference from best (PDB) is calculated as per
following expression:
PDB (MTbest MTMGA ) *100 / MTbest (5)

Where, MTbest is the best makespan time value by the one of proposed PDRs and MTMGA is the
value obtained by MGA.
The average percentage difference calculated are 11.60% for m=2, 24.63 % for m=3 and 26.70%
for m=4. The computational time taken by MGA is reasonable. The computational time increases
exponentially with higher number of jobs. But the higher time does not delay the implementation
of solution as the scheduling problems can be solved offline prior to the actual loading of the
batch of jobs.

9
 6. CONCLUSIONS

The results obtained for the makespan by the PDRs and MGA are compared and it was found
that the proposed MGA algorithm gives better results in almost all the test problems. The
computation time taken by the MGA problem is comparable to PDRs for m=2 and m=3, but it is
higher for m=4. Since the loading and scheduling problems can be solved offline before the
actual production starts, hence the computational time is justifiable.

REFERENCES
Stecke, K E (1983) Formulation and solution of non-linear integer production planning problem for flexible
manufacturing system Management Science, 29(3), 273-288.
Gray, A E, Seidmann A, Stecke K E (1993) A synthesis of decision models for tool management in automated
manufacturing Management Science, 39, 549-567.
Veeramani D,Upton D M, Barash M M (1992) Cutting tool management in computer integrated manufacturing
International Journal of Flexible Manufacturing Systems, 4, 237-265.
Praharan T, Nakkeeran P R, Jawahar N (2006) Sequencing and scheduling of job and tool in a flexible
manufacturing cell Int Journal of Advanced Manufacturing Technology, 29(7-8), 729-745.
Selim Akturk M, Siraceddin O (1999) Joint lot sizing and tool management in a CNC environment Compt Ind
Eng, 40, 61-74.
Roh H K, Kim Y D (1997) Due date based loading and scheduling methods for a flexible manufacturing system
with an automatic tool transporter Intl J Prod Res, 35(11), 2989-3003.
Caker T, Koker R, Demir H.I (2008) Parallel robot scheduling to minimize mean tardiness with precedence
constraints using a genetic algorithm Adv in Eng Soft, 39, 47-54.
Raja K, Arumugam C, Selladurai V (2008) Non-identical parallel machine scheduling using genetic algorithm and
fuzzy logic approach Intl J of Services and Operation Management 4(1), 72-101
Balin S (2011) Non-identical parallel machine scheduling using genetic algorithm. Expert System with Applications
38: 6814-6821
Agnetis A, Alfieri A, Brandimarte P, Prinsecchi P (1997) Joint job/tool scheduling in FMC with no on-board tool
magazine Comput Integr Manuf Syst 10(1), 61-68.
Giffler B, Thompson G L (1960) Algorithms for solving production scheduling problems Int J of Oper Res 8, 487-
503.
Nasciment M A (1993) Giffler and Thompson algorithm for job shop scheduling is still good for flexible
manufacturing systems J Oper Res Soc 44(5), 521-524.
Prabaharan T, Nakkeeran P R, Jawahar N (2006). Sequencing and scheduling of job and tool in a flexible
manufacturing cell Int J Adv Manuf Technol 29(7-8), 729-745.
Udhayakumar P, Kumanan S, (2010) Sequencing and scheduling of job and tool in a flexible manufacturing system
using ant colony optimization algorithm Int J Adv Manuf Technol, 50, 1075-1084.
Holland H H (1975) Adaption in Natural and Artificial Systems University of Michigan Press, Detroit MI
Goldberg D E (1989) Genetic Algorithm in Search, Optimization, and Machine Learning Addition-Wesley,
California.
Keung K W, Ip W H, Yuan D (2003) An intelligent hierarchical workstation control model for FMS Journal of
Material Processing Technology, 139, 134-139.
Shankar S, Ponnamabalam S G, Gurumarinuthu M (2005) Scheduling flexible manufacturing systems using
parallelization of multi-objective evolutionary algorithms Int J Adv Manuf Technol, 30:279-285.
Chan F.T.S., Chung S H, Chan L Y (2008) An introduction of dominant genes in genetic algorithm for FMS Int J
Prod Res, 46(16), 4369-4389.
Kim Y K, Kim J Y, Shin K S (2007b) An asymmetric multileveled symbiotic evolutionary algorithm for integrated
FMS scheduling J Intell Manuf, 18(6), 631-645.
Gang X Z, Wu (2004) Deadlock-free scheduling strategy for automated production cell IEEE trans Syst Man
Cybern Part A Syst Hum, 34(1), 113-122.
Elsayed S M, Sarkar S A, Essam D L (2014) A new genetic algorithm for solving optimization problems Engg
Applications of Artificial Intelligence, 27, 57-69.

10