Effect of Maintenance Resource Constraints on Flow-Shop Environment in a Joint

Production and Maintenance Context

Sandeep Kumar*, Bhupesh Kumar Lad
Intelligent Manufacturing Planning Lab, Indian Institute of Technology Indore, Indore, India
Email*: [email protected]; Email: [email protected]

Abstract - Production and maintenance planning decisions such as spare parts; maintenance technicians, etc. [3].
for any manufacturing industry are very crucial to make Sometimes situation may arise where a number of
shop floor operations effective. Despite the dependencies, machines require maintenance and concern department
these policies are often considered and optimized may have limited technicians and spares to tackle the
independently in literature and practice. Recently,
researchers have started focusing more on integrating these
requirement. Due to this maintenance of some machines
policy options. However, the effect of maintenance resource may be delayed, which may affect the production
constraints during such consideration is missing in the schedule. Furthermore, in manufacturing systems, some
literature. The present paper investigates the effect of spare machines are critical and require maintenance in priority.
parts and maintenance technician’s unavailability on joint As a consequence decision on the quantity of maintenance
production and maintenance planning decisions for flow- resources; their provision policy and their allocation are
shop environment. The performance measures considered crucial for maintenance planning.
here are makespan, total production cost and system In the context of spare parts and maintenance
utilization. Simulation-based optimization techniques are planning, [4] insisted on how pivotal spare parts
used to obtain the optimal production and maintenance
plan. Different cases of spare parts lead times and
management was within the scope of a maintenance
maintenance technician’s unavailability are considered for strategy. In [5], author proposed continuous and periodic
investigation. For each case production and maintenance review policies to supply the necessary spare parts for
planning decisions are analyzed. It is found that spare parts multiple identical machines subjected to failures. These
and maintenance technician’s unavailability have a strategies are based on a joint optimization of
significant impact on production performance. maintenance and spare parts ordering policies. Jin et al.
[6] proposed a policy which jointly optimizes the
Keywords - Sequencing, maintenance resources, non- inventory of spares, the capacity of repair and the
identical machine system, modeling and simulation maintenance under the game-theoretical framework.
I. INTRODUCTION In recent works, human resources are exploited
separately in production [7], [8] and maintenance
Nowadays manufacturing systems are facing more planning [9], [10] and not in integrated production and
competitive environment due to globalization. To stay maintenance scheduling. In [7], authors investigated the
competitive, a manufacturing system must satisfy integration of the employee timetabling and production
demands timely, at minimum cost with limited use of scheduling problems. In [11], authors studied a practice
resources. These goals can be achieved by efficient and maintenance labor-constrained scheduling problem which
effective production scheduling and maintenance is formulated as a bi-objective mixed-integer
planning. Production scheduling is an important activity programming model with the aim of simultaneously
in a manufacturing system and is concerned with minimizing the labor requirements and maximizing the
allocating limited machines to a set of jobs efficiently system availability.
over time and utilize production capacity in meeting To address the above mentioned issues, where a set of
demands for finished products. In real manufacturing jobs are to be scheduled on deteriorating machines under
systems, machines may be unavailable due to unplanned limited maintenance resources; requires to solve NP-hard
or planned maintenance activities which may affect the combinatorial problem [3]. Simulation based optimization
production schedule. On the other hand, maintenance plan tool is widely used to solve such problems [12].
may be affected due to the tight production schedule. In From above discussion it is clear that, production and
the past decade, researchers have been worked maintenance planning are interdependent. Also,
exhaustively on joint production and maintenance maintenance resources affect both of these planning
planning problems for different configuration of the variables. Despite the interdependencies and effect of
workshops like a single machine, multiple machines, maintenance resource constraints, production and
flow-shop, job-shop, etc. The developed models give maintenance planning considering spares provision policy
insights of interdependency of production and and workforce sizing for complex production
maintenance. However, most of the works consider environment is not addressed in literature.
unlimited availability of maintenance resources [1], [2]. The current paper accommodates above research gap.
Maintenance is performed on machines to minimize First, production scheduling and maintenance planning
downtime and to maintain in good working condition. It is decisions are evaluated jointly. Then the effects of
often impossible to perform all the desirable maintenance maintenance resource constraints on these decisions are
actions due to the limitation on maintenance resources analyzed.

978-1-5090-3665-3/16/$31.00
Authorized licensed use©2016
limited IEEE
to: MULTIMEDIA UNIVERSITY. Downloaded641
on December 19,2023 at 15:01:54 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 2016 IEEE IEEM

The paper is organized as follows. Section II presents before its due date an earliness cost is imposed and if it
the problem description and formulation. Section III gives processed after its due date a tardiness cost is imposed.
details of mathematical model. Solution method and Earliness and tardiness cost for a job per hour are 15%
obtained results are discussed in Section IV. Comparison and 20% of job manufacturing cost respectively.
of computational time using different algorithms is The methodology for above production and maintenance
presented in Section V. Effects of maintenance resource planning problem with limited resources is as follows:
constraints is analyzed in Section VI. Section VII First, optimal job schedule and PM frequency is obtained
concludes the paper and identifies the future scope of by minimizing makespan; minimizing total production
work. cost and maximizing system utilization separately,
considering unlimited maintenance resources. Then
II. PROBLEM DESCRIPTION AND FORMULATION various cases are made considering spare parts lead times
and maintenance technician’s availability and for each
Let us consider a flow-shop kind of production case job scheduling decision and PM frequency are
environment. The system consists of six non-identical analyzed again. The obtained results are compared with
machines where = 1, … . ,6 having multiple binary different performance measures.
components where = 1, … . ,4 . Binary means the Based on above description, the problem is formulated as
component has two states either working or failed. The follows:
machines ( , , … . ) have 3, 2, 2, 4, 3 and 1 Performance measure (I): Minimize,
different binary components respectively and are shown = ( , , , , , ) (1)
in Table I. For example machine has three Performance measure (II): Minimize,
components i.e. , and . The failure pattern of Total production cost = + + + (2)
components follows two-parameter Weibull distribution Performance measure (III): Maximize,
with a shape parameter and scale parameter Ƞ . ∑
Whenever a machine component fails, the machine stops Utilization( )=∑ ∑ ∑ (3)
production and a Corrective Maintenance (CM) is where, is completion time of job and is
performed on the machine to restore it back to working operation time of job sequenced on machine at
condition by replacing failed component. Machines also
position for operation.
receive Preventive Maintenance (PM) at intervals .
Makespan is total time elapsed when all the jobs have
In PM operations like cleaning; lubrication; change of
completed their processing. While utilization defined here
filters; etc. are performed. Time to carryout PM is 8 hours
and time to replace the failed components for CM is is a portion of available time that the system is operating.
hours. Fixed costs to carryout per PM and per CM are The total production cost ( ), considered here is the
1000 and 5000 Monetary Units (MU) respectively. The sum of earliness cost ( ), Tardiness cost ( ), PM cost
maintenance activity is dependent on availability of ( ) and CM cost ( ).
resources like spare parts and maintenance technicians. These performance measures are a function of sequencing
Unlimited maintenance resources provide instant variables ( ), and other model parameters.
availability but cost more. Thus, in current work limited
spare parts i.e. machine’s components are kept with Lead The problem is subjected to a constraint which ensures the
Time (LT). Similarly, limited maintenance technicians are sequencing of job on scheduled machine at
considered here. The cost per technician per hour is 500 position for an operation.
MU. At shop floor, some machines are occupied more
compare to others and delay in maintenance on these
machines may affect production schedule. Therefore on =1 ℎ , = 1, 2 … .5 (4)
these machines, maintenance is needed in priority. Thus, a
priority is given to occupied machines based on their load. The next provides the details of the mathematical model
Properties of machine’s components are shown in Table I.
The above mentioned system processes six jobs III. CALCULATION OF MAKESPAN AND TOTAL
where = 1, … . ,6. Let these jobs are to be scheduled non- PRODUCTION COST
preemptively on above non-identical machines. Let us
consider a fix process flow for each job with a number of The following assumptions are made in development of
operations performed in each flow as . The process flow mathematical model:
is shown in Table I. For example, the first operation of job • Each job is available at the beginning of production
1 (O1J1) is on machine and second operation of same period.
job (O2J1) is on machine . These jobs have fixed • Each machine can handle only one job at a time.
Processing Times ( ), Set-Up Times ( ), Demand ( ) • Failure of machine’s components is independent.
and Batch Sizes ( ). The job properties are shown in • Failure of machine’s component is independent.
Table I. The monthly demand is 3000 for each job. The • Machines are available at the start of production.
due date for all jobs is 500 hours. If a job is produced

Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded642

on December 19,2023 at 15:01:54 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 2016 IEEE IEEM

• Quality rejection cost is not considered in the The down time of machine due to
problem.
Table I Machine’s and Job’s Properties preventive maintenance before processing of job
sequenced at position occur if PM is performed before
PT ST
Ƞ Cost the same job and is:
Jobs BS
(hrs.) In (MU)
Minutes = ( )+ + (8)
C11 20 2 1800 O1J1 9.6 90 600 540
where, is time to repair for PM of machine.
M1 C12 16 1.8 2000 O1J2 4.2 60 500 800
C13 15 2.5 1500 O1J3 2.6 42 300 620 is waiting time for maintenance technicians and is
C21 10 1.5 1000 O1J4 8.1 30 400 360 waiting time for spare parts for machine.
M2
C22 30 2.1 1200 O1J5 1.5 60 400 600
C31 25 3 1800 O1J6 4 30 300 450 where, =
M3
C32 20 2.7 2400 O2J4 6.8 120 200 500 1,
O2J1 3.2 20 500 700
C41 15 2 3000 ℎ
O2J2 4.8 80 500 900
M4 C42 18 1.6 1600 O2J3 2.8 60 500 800 0, ℎ
C43 14 1.8 1800 O2J5 3 90 500 850
C44 16 2.1 2000 O2J6 2.1 60 500 600 The total number of PM performed on machine is
O3J3 6.2 30 600 1000 evaluated by optimizing .
C51 24 1.8 1800
O4J4 4.1 360 300 900
M5
C52 16 2.5 1600 O3J5 1.4 60 200 1000
= (9)
C53 20 3 1200 O4J6 3.2 30 400 1000
O3J4 5.6 120 300 650
M6 C61 25 2 3000
O3J6 1.9 90 400 800 is the down time of machine due to
As discusses in the previous section, = corrective maintenance occur during the processing of
( ).
job sequenced at position and is:
where, is the sum of operation times of the job
scheduled on various machines according to the process = ( )+ + (10)
flow. It can be expressed as:
where,
1,
= ℎ , = 1, 2 … . (5)
= ℎ
0, ℎ
where, is the sum of operations times of job
and preceding jobs sequenced on machine. It can be Generally in manufacturing system maintenance
expressed as: technicians are limited and may be engaged in CM when
they are needed for PM and vice versa. Thus, PM and CM
= (6) may be delayed. Also, at shop floor some machines are
critical and may need maintenance on priority. So, a
Mathematically, operation time of job processing priority is given based on their load.
on machine for operation is sum of waiting time ∑
( ) of job sequenced at position that is = (11)
additional to processing time of previous sequenced jobs where, T is planning horizon and its value is 500 hours.
on same machine, set-up time ( ) of batch, processing Spare parts may not be available at the time of PM and
CM due to variation in lead time and may be delay the
time ( ) of batch and machine downtime due to
maintenance activity. Thus, an appropriate spare parts
preventive maintenance ( ) and corrective provision policy is necessary to minimize such delay.
maintenance ( ) . Then, The ingredient costs of total production cost (as
= + + + mentioned in section II) are calculated using equations
below:
+ (7)
= 0, ( ) (12)
1,
= ℎ
= 0, ( ) (13)
0, ℎ

Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded643

on December 19,2023 at 15:01:54 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 2016 IEEE IEEM

The optimal production sequence and PM frequency

= ( ) + (14) decisions are obtained by minimizing makespan, total
production cost and system utilization by using eq. (1), (2)
and eq. (3) respectively. The results are shown in Table II
= ( ) + (15) where I, II and III represents the results obtained by
minimizing makespan, minimizing total production cost
where, D is monthly demand. and are earliness and system utilization respectively. The corresponding
and tardiness cost of job per hour. C is labor cost. makespan is 19.2 days; total production cost is 489,370
and are fixed costs to carryout per PM and MU and system utilization is 84%. The number of
maintenance technicians required corresponding to each
per CM. Next section presents solution method used to
performance measure is 15.
solve the problem and obtained results.
Table II Obtained Production Sequence and
IV. SOLUTION METHOD AND RESULTS Sequence Spare
Jobs ( ) (Weeks) Quantity
Job sequencing and PM decisions are to be I II III I II III I,II, III
determined simultaneously in the problem presented in O1J1 3 2 1 C11 5
section II. This problem is of combinatorial type and also M1 O1J2 1 1 2 2 1 2 C12 7
strongly NP-hard [3]. To solve this, combined simulation O1J3 2 3 3 C13 3
and optimization is performed using Witness 14 O1J4 1 2 1 C21 8
M2 1 3 2
manufacturing performance edition tool, which is a O1J5 2 1 2 C22 4
product of Lanner Group. In current work, Adaptive O1J6 2 1 1 C31 6
M3 2 3 2
Thermo-statistical Simulated Annealing (ATSA) O2J4 1 2 2 C32 7
algorithm is used for optimization. It gives rapid O2J1 2 2 3
C41 10
convergence to high quality solutions using a very modest O2J2 1 1 4
number of evaluations. The algorithm parameters; initial M4 O2J3 3 4 1 1 1 1 C42 6
temperature (T), cooling rate, cooling steps and number of O2J5 5 3 5 C43 8
O2J6 4 5 2 C44 3
without improvement scenarios for termination are 100,
O3J3 1 2 2
0.91, 25 and 1000 respectively. The simulation and C51 5
O4J4 3 4 3
optimization process is described by pseudo code M5
O3J5 4 3 4
3 2 2
C52 11
presented below. O4J6 2 1 1 C53 9
Pseudo Code O3J4 2 1 1
X0 = GenerateSolution(); M6 2 1 2 C61 7
O3J6 1 2 2
T = INITIAL_TEMP;
Z = 0 // iteration count; V. COMPARISON OF COMPUTATIONAL TIME
while(Termination_condition(X0, T, Z))
X1 = Next(X0) // find a solution near of X0 To check the value of Adaptive Thermo-statistical
if(Fitness(X1) < Fitness(X0)) Simulated Annealing (ATSA), the problem of section II is
X0 = X1 also solved by using Hill Climb and Random Solution
else if(rand() < tempFunc(X0, X1, T, Z)) optimization algorithms. Experiments were run on PC
X0=X1//if X1 is poor than X0//conditionally accept with Intel Core i7-373000 CPU @ 3.40 GHz.
Simulate(X1) // for uncertain model parameters Optimization results is presented in Table III. Simulated
end if annealing achieved the best result providing 5% and 3%
AnnealingSchedule(X0, T, Z) // anneal the temp reduction in makespan; 12% and 4% reduction in TPC
Z = Z + 1 // increment and 0.25% and 0.5% reduction in system utilization
end while compared to Hill Climb and Random Solution
Figure I shows the progress of the simulation and respectively. On the other hand, simulated annealing
optimization process for the case of minimizing makespan performed poorly regarding computational time requiring
in days. almost 2% to 6% and 1% to 8% more time compared to
Hill Climb and Random Solution respectively. From the
results it can be concluded that a production manager
should go for ATSA algorithm for better results.
However, it may take little more time.
Table III Optimization computation time and best results
Computation
Best result
S. Optimization time (minutes)
N. algorithm I II III
I II III
(Days) (MU) (%)
Figure I Results from successive trials Random
1 59 65 63 20.7 510,061 80
solutions
2 Hill Climb 56 60 64 20.1 555,573 82
Simulated
3 60 62 68 19.2 489,370 84
Annealing

Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded644

on December 19,2023 at 15:01:54 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 2016 IEEE IEEM

VI. INVESTIGATION AND OBSREVATIONS flow-shop system and can be extend to different
To investigate the effect of the unavailability of spare production environments considering various operation
parts and maintenance technicians on joint production and planning.
maintenance planning decisions; total nine cases are ACKNOWLEDGMENT
investigated. The spare parts lead times variation The authors would like to thank the management and staff
considered is 0 to 2 weeks while technician’s availability of AVTEC Limited, Pithampur (M.P.), India, for
is varied in the range of 50% to 100%. For each case the generously providing the support for this study.
problem of section II is solved using thermo-statistical
Simulated Annealing algorithm for various performance REFERENCES
[1] C.R. Cassady and E. Kutanoglu, "Minimizing job tardiness
measures separately. The value of performance measures using integrated preventive maintenance planning and
for each case and the changes in optimal production production scheduling," IIE Transactions, vol. 35, no. 6, pp.
sequences and PM frequency decisions comparing to 503–513, 2003.
results of section IV are shown in Table IV. From Table [2] S. Kumar, B.S. Purohit and B.K. Lad, “Integrated approach for
IV following observations have been made: job scheduling and multi-component maintenance planning in
a production system”, 5th International and 26th All India
• The spare parts and technicians unavailability affects Manufacturing Technology Design and Research Conference
production and maintenance planning decisions (AIMTDR), Guwahati, India, pp. 482(1–6), 2014.
irrespective of performance measures. [3] P. Do, H.C. Vu, A. Barros, and C. Bérenguer, "Maintenance
• The maintenance resources have different effect on grouping for multi-component systems with availability
operations planning decisions for different constraints and limited maintenance teams," Reliability
performance measures and the affect is more for Engineering & System Safety, vol. 142, pp. 56–67, 2015.
[4] W. Van Jaarsveld and R. Dekker, "Spare parts stock control for
minimizing makespan case. redundant systems using reliability centered maintenance
• For performance measures I and III the optimal data," Reliability Engineering & System Safety, vol. 96, no. 11,
values are obtained for unlimited maintenance pp. 1576–1586, Nov. 2011.
resources case while for performance measure II, the [5] S. Panagiotidou, "Joint optimization of spare parts ordering
optimal value is obtained for limited maintenance and maintenance policies for multiple identical items subject to
resources case. silent failures," European Journal of Operational Research,
vol. 235, no. 1, pp. 300–314, 2014.
• The effect of spare parts unavailability is more [6] T. Jin, Z. Tian, and M. Xie, "A game-theoretical approach for
compared to technician’s unavailability on joint optimizing maintenance, spares and service capacity in
production and maintenance planning decision. performance contracting," International Journal of Production
• For the production scenarios with restricted due date Economics, vol. 161, pp. 31–43, 2015.
case, performance measure I may be useful. Where [7] O. Guyon, P. Lemaire, É. Pinson, and D. Rivreau, "Cut
system idleness cost is high, more focus should be on generation for an integrated employee timetabling and
production scheduling problem," European Journal of
performance measure III. If production cost is a Operational Research, vol. 201, no. 2, pp. 557–567, 2010.
criterion, then production manager should go for [8] M. Sabar, B. Montreuil, and J. Frayret, "A multi-agent-based
performance measure II. approach for personnel scheduling in assembly centers,"
Engineering Applications of Artificial Intelligence, vol. 22, no.
VII. CONCLUSION 7, pp. 1080–1088, 2009.
This paper investigates the effect of maintenance [9] C. de Snoo, W. van Wezel, J.C. Wortmann, and G.J.C.
resource constraints on joint production and maintenance Gaalman, "Coordination activities of human planners during
rescheduling: Case analysis and event handling procedure,"
planning decisions. A heuristics based Adaptive Thermo-
International Journal of Production Research, vol. 49, no. 7,
statistical Simulated Annealing algorithm is used to solve pp. 2101–2122, 2011.
the joint planning problem. For various unavailability [10] F. Marmier, C. Varnier, and N. Zerhouni, "Proactive, dynamic
cases of maintenance resources, optimal production and multi-criteria scheduling of maintenance activities,"
sequencing and PM frequency decisions changes. Also, International Journal of Production Research, vol. 47, no. 8,
variations in the optimal values of different performance pp. 2185–2201, 2009.
measures are found in the range of 14 to 30 percent. [11] N. Safaei, D. Banjevic, and A.K.S. Jardine, "Bi-objective
workforce-constrained maintenance scheduling: A case study,"
These variations and changes in the optimal planning Journal of the Operational Research Society, vol. 62, no. 6, pp.
decisions may vary, depending upon the type of the 1005–1018, 2010.
production environment. But still maintenance resource [12] S.M. Asadzadeh and A. Azadeh, "An integrated systemic
constraints have a significant impact on production model for optimization of condition-based maintenance with
performance. The presented investigation is limited for human error," Reliability Engineering & System Safety, vol.
124, pp. 117–131, 2014.

Table IV Results of various cases of spare parts lead time variation and technicians unavailability
Lead I (Minimize Makespan) II (Minimize TPC) III (Maximize System Utilization)
Technician
Case Time in Makespan TPC
Availability Utilization
Weeks change (Days) change (MU) change
1 0 100% No No 19.2 No No 489,370 No No 84%
2 0 75% Yes No 20.7 Yes Yes 496,785 Yes No 82%
3 0 50% Yes Yes 22 Yes Yes 504,398 Yes Yes 81%
4 1 100% Yes No 21.2 Yes Yes 512,215 No No 80%
5 1 75% Yes Yes 22.8 Yes Yes 456,862 Yes Yes 79%
6 1 50% Yes Yes 23.4 Yes Yes 492,652 Yes Yes 78%
7 2 100% Yes Yes 22.5 Yes No 508,269 Yes No 76%
8 2 75% Yes Yes 24.1 Yes Yes 515,196 Yes Yes 75%
9 2 50% Yes Yes 25 Yes Yes 546,304 Yes Yes 72%
Authorized licensed use limited to: MULTIMEDIA UNIVERSITY. Downloaded645
on December 19,2023 at 15:01:54 UTC from IEEE Xplore. Restrictions apply.