Content uploaded by Miguel Mujica Mota
Author content
All content in this area was uploaded by Miguel Mujica Mota on Oct 17, 2018
Content may be subject to copyright.
12th International Conference on Industrial Engineering and Industrial Management
XXII Congreso de Ingeniería de Organización
Girona, Spain, July 12-13, 2018
Performance of Unreliable Merging Lines with
Uneven Buffer Capacity Allocation
Abstract The purpose of this study is to investigate the impact that unevenly allo-
cating buffer capacity has on throughput and average buffer level regarding unreli-
able lines to better understand the relevant factors in supply chain design. Results
show that the best patterns for unreliable merging lines in terms of generating higher
throughput rates (TR), as compared to a balanced merging line counterpart, are
those where total available buffer capacity is allocated between workstations in ei-
ther an inverted bowl pattern (i.e. concentrating buffer capacity towards the centre
of the line), or a balanced line pattern. In contrast, when considering the trade-off
between generating revenue resulting from TR and reducing cost created by average
buffer levels (ABL), we found that the balanced pattern was not the best pattern.
The best pattern was dependent on the length of the line and on the total buffer
capacity as shorter lines with very constrained buffers were best served with an
inverted bowl pattern while longer lines had the best results when applying an as-
cending buffer allocation pattern. Longer lines, in contrast, had the best results re-
garding the trade-off between TR and ABL, on average, by allocating buffer capac-
ity evenly in one of the parallel lines while applying any other pattern in the
remaining parallel line.
Keywords: Unreliable merging lines; uneven buffer allocation; average buffer
level; throughput; simulation.
1 Introduction
Parallel merging lines with no mechanical pacing are probabilistic mass production
queueing systems in series. Stocks of partially finished items are usually transferred
to a buffer storage location.
Merging lines that are unbalanced with respect to their buffer capacities are an
important research and practice topic. Often, technical considerations restrict the
amount of space available in the line, thereby making it difficult to allocate total
buffer capacity evenly amongst individual buffers.
CIO2018, 009, v1: ’Performance of Unreliable Merging Lines with Uneven Buffer Capacity . . . 1
2
Taking into account the fact that significant production is taking place in devel-
oping economies, as well as the immense growth in remanufacturing and reverse
logistics, this emphasizes the importance of research work on unreliable assembly
lines with uneven buffer capacity across industries. In addition, queueing networks
with parallel, merging stages are common in a variety of manufacturing systems, as
well as in computer networks and supply chains.
Furthermore, all merging lines are likely to suffer from adverse performance dis-
turbances in the form of downtime resulting from machine failure. This article,
therefore, deals with the twin issues of merging line’s uneven buffer allocation and
unreliability.
How best to allocate buffer space in order to meet the desired performance ob-
jectives contributes value to both research and industry practice, and is a burgeoning
area of investigation. It has long been a general belief that balancing an unpaced
serial production line so that each workstation completes its task at the same rate as
the preceding and subsequent stations and where buffer space is allocated evenly,
gives the best performance (Lambrecht and Segaert, 1990). Some research (see for
instance Conway et al., 1988), however, indicated that that in view of the fact that
real life unpaced assembly lines can never be truly balanced and will certainly suffer
breakdown failures, it is of interest to incorporate the facts of uneven buffer size
allocation and machine breakdowns into the line design to investigate how best to
obtain good performance.
It is the aim of this article to present the results of the impact of unbalancing
buffer capacity on performance, by simulating unreliable merging lines where buff-
ers of unequal sizes are placed between workstations in a variety of patterns, line
lengths and total buffer capacities.
The structure of the remainder of this paper is as follows. Following a brief re-
view of the relevant literature, we present the methodology of the study. Subsequent
sections discuss the results and the conclusions of this investigation.
2 Literature Review
The vast majority of studies on parallel (also known as fork-join, u-shaped, or two-
sided) merging assembly lines have focused on line balancing (see for example,
Akpınar and Bayhan, 2011; Barron, 2015; Purnomo et al., 2013; Sönmez et al.,
2017). For a comprehensive literature review of merging line balancing methods,
see Battaïa and Dolgui (2013) and Sivasankaran and Shahabudeen (2014).
Just a few studies, on the other hand, were done on merging lines with uneven
buffer capacities. They can be divided into two broad categories: reliable and unre-
liable merging lines. Below is a review of pertaining works.
Literature on uneven buffer allocation in reliable merging lines is sparse. Powell
and Pyke (1998) presented general strategies on the efficient placement of buffers
in unbalanced assembly systems with random processing times.
2 CIO2018, 009, v1: ’Performance of Unreliable Merging Lines with Uneven Buffer Capacity . . .
3
Futamura (2000) studied optimal server allocation to tandem queueing networks
with uneven allocation of buffer capacities (BCs), unbalanced mean service times
(MTs) and coefficients of variation (CVs). He showed that optimal configurations
can be predicted and that there is a general interaction between the CV of the service
time distribution and the number of servers at the stations.
Most recently, Shaaban et al. (2017) assessed the performance of unbalanced,
reliable, unpaced merging lines with asymmetric buffer storage sizes. Lines were
simulated with varying line lengths, mean buffer capacities and uneven buffer allo-
cation configurations. They found that higher throughput (TR) and lower average
buffer level (ABL) (as compared to an equivalent balanced merging line) were ob-
tained when total available buffer capacity is allocated as evenly as possible and
with a higher buffer capacity concentration towards the end of the line, respectively.
On the other hand, Gershwin (1991) analyzed a class of assembly/disassembly
network systems in which machines are unreliable, buffers are finite, and machines
perform operations whenever none of their upstream buffers are empty and none of
their downstream buffers are full, and the network structure is a tree. An approxi-
mate decomposition method to estimate TR was presented.
Focusing on three-station assembly systems, Bhatnagar and Chandra (1994) used
simulation to study the effect of variability due to unreliable stations and imperfect
yields on these systems. Greater TR improvements were found from increasing the
production rate of individual stations than from increasing the size of buffers.
Jeong and Kim (2000) investigated buffered production systems with feeder sta-
tions merging into an assembly station. They developed heuristics to determine the
line configuration which would bring about a desired TR at a minimal cost. They
assumed exponential times to failure & repair and processing times with finite
buffer sizes.
Yuan and Liu (2005) studied an unreliable assembly system in which different
types of components are processed by two separate work centers before merging to
an assembly station with random breakdown. They developed formulas for the
probabilities of system state, blocking, starvation, stock-out, and system availability
in the steady state. They also obtained the distributions of blocking and failure
times.
Recently, Jia et al. (2016) studied the transient behavior of assembly systems
with merging serial lines, comprised of Bernoulli machines (subject to failure) with
finite buffers. Formulas were derived to efficiently measure TR, work-in-process
levels, and probability that any one station will be blocked or starved. They devel-
oped an analytical method for dealing with larger and more complex assembly sys-
tems, with multiple feeder lines and merge stations.
We can see from the above review that the focus of merging line research was
on the development of line balancing and mathematical optimization methods. To
our knowledge, there are no studies on the influence of uneven buffer allocation
patterns on TR and ABL in unreliable merging lines.
CIO2018, 009, v1: ’Performance of Unreliable Merging Lines with Uneven Buffer Capacity . . . 3
4
The performance of merging lines with uneven buffer sizes is studied here to
bridge some of the gaps in this research area. To assess if uneven buffer size allo-
cation can generate better results than those obtained from having a constant buffer
capacity level all along the line, this study applies simulation and statistical analysis.
In addition, the effect of various design variables (line length, mean buffer capacity
and buffer imbalance configuration) is evaluated.
3 Methodology
Discrete-Event simulation was viewed as the most appropriate tool for this study
because of the severe limitations of mathematical approaches in dealing with more
realistic and complex merging lines, typically reported with positively skewed op-
eration times. The Simio 9.147 simulation software (Kelton, Smith and Sturrock,
2014) was used to study the behavior of the unreliable, unbalanced merging line.
The merging line’s dependent performance measures used are TR and ABL. Fur-
thermore, we used a representation of the trade-off between TR revenue and ABL
cost. As previous authors have shown that ABL costs could be up to 25% of the
total cost of the product (Azzi et al., 2014), we then considered a combined factor
of 1TR - 0.25ABL.
To generate representative simulation data, a suitable warm-up/transient period
is needed to ensure that observations are very close to normal operating conditions.
Law (2014) suggested running a preliminary system simulation, selecting one out-
put variable for observation. A trial procedure has found that after an initial simu-
lation run of 20,000 minutes, acceptable steady-state behavior was established. So,
all data gathered during the first 20,000 minutes were discarded, and 100 independ-
ent runs of 120,000 minutes each were carried out excluding the first 20,000
minutes of non-steady state data.
Typically, manufacturing equipment on the factory floor is unreliable. In unreli-
able merging lines, the stations are subject to random breakdown and repair events.
In this study, the failure rate used is 0.001 breakdowns per minute and the repair
rate is 0.010 repairs per minute, i.e. MTBF = 1000 minutes and MTTR = 100
minutes. Therefore, line efficiency = 91% [MTBF 1,000 / (MTBF 1,000 + MTTR
100)], which is the same efficiency figure used by Altiok and Stidham (1983) and
Hopp and Simon (1993).
For each station, the processing time was modelled as a Weibull distribution with
mean of 10 time units, whereas the CV was fixed at 0.274, in line with Slack’s
(1982) recommendations.
The independent variables and their levels (for parallel lines 1 and 2) were:
• Line length: N = 5 and N = 8 (i.e. odd and even numbers).
• Mean Buffer Capacity (BC) per station: These values were selected such that
BC ≠ 0, while taking into account that over a certain level of buffer space, the
4 CIO2018, 009, v1: ’Performance of Unreliable Merging Lines with Uneven Buffer Capacity . . .
5
law of diminishing returns sets in, leading to negligible improvement in effi-
ciency as buffer size increases. BC = 2 and BC = 6. Thus, total buffer capacity
of the complete line remained constant and only buffer capacity per individual
station varied.
Six different uneven buffer capacity allocation pattern (Pi) for both lines 1 and 2
were considered (Table 1).
Table 1. Buffer allocation patterns
Line Length (N)
5
8
Mean Buffer Capacity (BC)
2
6
2
6
Ascending (/)
A
1,1,1,5
3,3,3,15
1,1,1,1,1,1,8
3,3,3,3,3,3,24
Descending (\)
B
5,1,1,1
15,3,3,3
8,1,1,1,1,1,1
24,3,3,3,3,3,3
Bowl shape (V)
C1
3,2,1,2
9,6,3,6
4,3,1,1,1,1,3
12,9,3,3,3,3,9
C2
4,1,1,2
12,3,3,6
4,2,1,1,1,1,4
12,6,3,3,3,3,12
inverted bowl
shape (Λ)
D1
1,2,3,2
3,6,9,6
1,1,3,4,3,1,1
3,3,9,12,9,3,3
D2
1,2,4,1
3,6,12,3
1,2,2,4,2,2,1
3,6,6,12,6,6,3
General
E1
2,2,3,1
6,6,9,3
2,2,2,3,3,1,1
6,6,6,9,9,3,3
E2
2,3,2,1
6,9,6,3
2,2,3,3,2,1,1
6,6,9,9,6,3,3
Balanced
2,2,2,2
6,6,6,6
2,2,2,2,2,2,2
6,6,6,6,6,6,6
The patterns used in this study correspond to those used in some previous publi-
cations on reliable unbalanced production lines, such as McNamara et al. (2013)
and Shaaban et al.(2015). Overall, the total number of cells simulated is 324 (2 line
lengths x 2 MB levels x 9 uneven buffer allocation patterns for line 1 x 9 buffer
allocation patterns for line 2).
4 Results
Due to space limitations, Table 2 shows a summary of the simulation experiments
by presenting the average TR, ABL and TR-0.25ABL results for all the combined
experiments. Considering Table 2, “Blcd+Blcd” represents a merging line where
the two parallel lines are balanced, while the results of a “Balanced” pattern for
Parallel line 1 (L1), for example, show the average response value when applying a
“Balanced” pattern only in L1 and every other pattern for Parallel line 2 (L2). Sim-
ilarly, results of an “A” pattern regarding L2 show the average response value when
applying the “A” pattern in L2 and every other pattern in L1.
Grey cells in Table 2 show the best pattern found for the average results per
response. For instance, it can be seen that TR is maximized with a Balanced + Bal-
anced pattern, overall, with the exception of applying an inverted bowl pattern
(“D1” pattern) when BC=2 and N=5. However, caution should be exercised with
this result, as we didn’t find statistically significant differences for this particular
configuration.
CIO2018, 009, v1: ’Performance of Unreliable Merging Lines with Uneven Buffer Capacity . . . 5
6
Furthermore, we found that both ABL and TR-0.25ABL had their best perfor-
mance when N=8 by applying a “Balanced” pattern in combination with any other
rule. In addition, Table 1 shows that when N=5, an ascending order pattern (“A”
pattern) resulted in the best performance. The exception is again found in a line
where BC=2 and N=5, since applying a D1 pattern in either parallel line will pro-
duce the best TR-0.25ABL results.
Table 2. Average response results for different experimental settings depending on buffer patterns
TR
ABL
TR-0.25ABL
N = 5
N = 8
N = 5
N = 8
N = 5
N = 8
BC
Pattern
L1
L2
L1
L2
L1
L2
L1
L2
L1
L2
L1
L2
2
Blcd+Blcd
0,56
0,51
1,31
1,41
0,23
0,16
Balanced
0,56
0,56
0,51
0,51
1,32
1,32
1,15
1,15
0,23
0,23
0,22
0,22
A
0,56
0,56
0,49
0,49
1,24
1,24
1,84
1,84
0,25
0,25
0,03
0,03
B
0,55
0,55
0,51
0,51
1,41
1,40
1,47
1,47
0,20
0,20
0,14
0,14
C1
0,56
0,56
0,50
0,50
1,35
1,35
1,62
1,62
0,22
0,22
0,09
0,10
C2
0,55
0,55
0,50
0,50
1,36
1,36
1,53
1,53
0,21
0,21
0,12
0,12
D1
0,56
0,56
0,51
0,51
1,30
1,30
1,51
1,51
0,24
0,24
0,13
0,13
D2
0,56
0,56
0,51
0,51
1,31
1,31
1,52
1,52
0,23
0,23
0,13
0,13
E1
0,56
0,56
0,50
0,50
1,34
1,34
1,55
1,55
0,22
0,22
0,12
0,12
E2
0,56
0,56
0,50
0,50
1,35
1,35
1,63
1,63
0,22
0,22
0,09
0,09
6
Blcd+Blcd
0,66
0,63
3,80
4,01
-0,29
-0,38
Balanced
0,66
0,66
0,61
0,61
3,83
3,83
3,30
3,31
-0,30
-0,30
-0,21
-0,22
A
0,66
0,66
0,59
0,59
3,54
3,55
5,44
5,43
-0,23
-0,23
-0,77
-0,77
B
0,64
0,64
0,62
0,62
4,13
4,12
4,24
4,24
-0,39
-0,39
-0,44
-0,44
C1
0,65
0,65
0,61
0,61
3,92
3,91
4,73
4,72
-0,33
-0,33
-0,58
-0,57
C2
0,65
0,65
0,61
0,61
3,97
3,96
4,44
4,44
-0,34
-0,34
-0,50
-0,50
D1
0,66
0,66
0,61
0,61
3,73
3,74
4,35
4,36
-0,28
-0,28
-0,47
-0,47
D2
0,66
0,65
0,62
0,62
3,79
3,80
4,37
4,37
-0,29
-0,29
-0,47
-0,48
E1
0,65
0,65
0,61
0,61
3,88
3,89
4,47
4,47
-0,32
-0,32
-0,50
-0,50
E2
0,65
0,65
0,61
0,61
3,93
3,93
4,75
4,76
-0,33
-0,33
-0,58
-0,58
5 Discussion and conclusions
Results from this study show that a balanced buffer allocation pattern in both par-
allel lines, when considering the combined result of TR as a revenue generating
6 CIO2018, 009, v1: ’Performance of Unreliable Merging Lines with Uneven Buffer Capacity . . .
7
variable and ABL as a cost-related variable, is not always the best pattern to increase
performance in merging, unreliable production lines, because an inverted bowl pat-
tern of buffer allocation works best for lines with 5 stations and a mean buffer ca-
pacity per station of 2, while an ascending pattern of buffer allocation is the best
pattern for lines with 5 stations and a mean buffer capacity of 6.
Otherwise, the best pattern, on average, regarding a combined result of TR and
ABL is to assign a “Balanced” pattern to one of the parallel lines while assigning
any other pattern to the other parallel line.
However, our results show that the best pattern regarding Throughput perfor-
mance, by itself, is in fact the “Balanced + Balanced” buffer allocation pattern, even
though the inverted bowl pattern provides very similar throughput performance for
a short line with limited mean buffer capacity.
More research is needed to understand why either an inverted bowl or an ascend-
ing pattern produce the best trade-off between TR and ABL in shorter lines. Fur-
thermore, an analysis of the difference between reliable and unreliable merging
lines is needed.
References
Akpınar, S. and Bayhan, G. M. (2011) ‘A hybrid genetic algorithm for mixed model assembly line
balancing problem with parallel workstations and zoning constraints’, Engineering Applications of
Artificial Intelligence, 24(3), pp. 449–457. doi: https://doi.org/10.1016/j.engappai.2010.08.006.
Altiok, T. and Jr., S. S. (1983) ‘The Allocation of Interstage Buffer Capacities in Production Lines’,
IIE Transactions. Taylor & Francis, 15(4), pp. 292–299. doi: 10.1080/05695558308974650.
Azzi, A. et al. (2014) ‘Inventory holding costs measurement: a multi-case study’, The International
Journal of Logistics Management, 25(1), pp. 109–132. doi: 10.1108/IJLM-01-2012-0004.
Barron, Y. (2015) ‘Mean Sojourn Time in Multi Stage Fork-Join Network: The Effect of
Synchronization and Structure’, International Journal of Operations Research and Information Systems,
6(3), pp. 80–99. doi: 10.4018/IJORIS.2015070104.
Battaïa, O. and Dolgui, A. (2013) ‘A taxonomy of line balancing problems and their
solutionapproaches’, International Journal of Production Economics, 142(2), pp. 259–277. doi:
https://doi.org/10.1016/j.ijpe.2012.10.020.
Bhatnagar, R. and Chandra, P. (1994) ‘Variability in Assembly and Competing Systems: Effect on
Performance and Recovery’, IIE Transactions, 26(5), pp. 18–31. doi: 10.1080/07408179408966625.
Conway, R. et al. (1988) ‘The Role of Work-In-Process Inventory in Serial Production Lines’,
Operations research, 36(2, Operations Research in Manufacturing), pp. 229–241. Available at:
http://www.jstor.org/stable/171278.
Futamura, K. (2000) ‘The multiple server effect: Optimal allocation of servers to stations with
different service‐time distributions in tandem queueing networks’, Annals of Operations Research,
93(1), pp. 71–90. doi: 10.1023/A:1018948512499.
Gershwin, S. B. (1991) ‘Assembly/Disassembly Systems: An Efficient Decomposition Algorithm
for Tree-Structured Networks’, IIE Transactions. Taylor & Francis, 23(4), pp. 302–314. doi:
CIO2018, 009, v1: ’Performance of Unreliable Merging Lines with Uneven Buffer Capacity . . . 7
8
10.1080/07408179108963865.
Hopp, W. J. and Simon, J. T. (1993) ‘Estimating throughput in an unbalanced assembly-like flow
system’, International Journal of Production Research. Taylor & Francis, 31(4), pp. 851–868. doi:
10.1080/00207549308956762.
Jeong, K.-C. and Kim, Y.-D. (2000) ‘Heuristics for selecting machines and determining buffer
capacities in assembly systems’, Computers & Industrial Engineering, 38(3), pp. 341–360. doi:
https://doi.org/10.1016/S0360-8352(00)00045-0.
Jia, Z. et al. (2016) ‘Performance analysis of assembly Systems with Bernoulli machines and finite
buffers’, IEEE Transactions on Automation Science and Engineering, 13(2), pp. 1018–1032.
Kelton, W. D., Smith, J. S. and Sturrock, D. T. (2014) Simio and Simulation: Modeling, Analysis,
Applications. 1st edn. Sewickley, PA: Simio LLC.
Lambrecht, M. and Segaert, A. (1990) ‘Buffer Stock Allocation in Serial and Assembly Type of
Production Lines’, International Journal of Operations & Production Management, 10(2), pp. 47–61.
doi: 10.1108/01443579010000736.
Law, A. (2014) Simulation Modeling and Analysis. 5th edn. New York: McGraw-Hill.
McNamara, T., Shaaban, S. and Hudson, S. (2013) ‘Simulation of unbalanced buffer allocation in
unreliable unpaced production lines’, International Journal of Production Research. Taylor & Francis,
51(6), pp. 1922–1936. doi: 10.1080/00207543.2012.720726.
Powell, S. G. and Pyke, D. F. (1998) ‘Buffering unbalanced assembly systems’, IIE Transactions,
30(1), pp. 55–65. doi: 10.1023/A:1007493512502.
Purnomo, H. D., Wee, H.-M. and Rau, H. (2013) ‘Two-sided assembly lines balancing with
assignment restrictions’, Mathematical and Computer Modelling, 57(1), pp. 189–199. doi:
https://doi.org/10.1016/j.mcm.2011.06.010.
Shaaban, S., McNamara, T. and Dmitriev, V. (2017) ‘Asymmetrical buffer allocation in unpaced
merging assembly lines’, Computers & Industrial Engineering, 109, pp. 211–220. doi:
https://doi.org/10.1016/j.cie.2017.05.008.
Shaaban, S., McNamara, T. and Hudson, S. (2015) ‘The impact of failure, repair and joint imbalance
of processing time means and buffer sizes on the performance of unpaced production lines’, International
Journal of Manufacturing Technology and Management, 29(5–6), pp. 347–365. doi:
10.1504/IJMTM.2015.071228.
Sivasankaran, P. and Shahabudeen, P. (2014) ‘Literature review of assembly line balancing
problems’, The International Journal of Advanced Manufacturing Technology, 73(9), pp. 1665–1694.
doi: 10.1007/s00170-014-5944-y.
Slack, N. (1982) Work time distributions in production system modelling. techreport. doi:
10.1002/9781118785317.weom100178.
Sönmez, E., Scheller-Wolf, A. and Secomandi, N. (2017) ‘An Analytical Throughput Approximation
for Closed Fork/Join Networks’, INFORMS Journal on Computing, 29(2), pp. 251–267. doi:
10.1287/ijoc.2016.0727.
Yuan, X.-M. and Liu, L. (2005) ‘Performance analysis of assembly systems with unreliable machines
and finite buffers’, European Journal of Operational Research, 161(3), pp. 854–871. doi:
http://dx.doi.org/10.1016/j.ejor.2003.09.011.
8 CIO2018, 009, v1: ’Performance of Unreliable Merging Lines with Uneven Buffer Capacity . . .