Professional Documents
Culture Documents
com
ScienceDirect
Procedia CIRP 37 (2015) 140 – 145
1
Daimler AG, Hanns-Martin-Schleyer-Str. 21-57, 68305 Mannheim, Germany
2
Daimler AG, Hedelfinger Str. 4 – 11, 73734 Esslingen,Germany
3
TU Darmstadt Technical University, Otto-Berndt-Straße 2, 64287 Darmstadt,Germany
Abstract
To identify the momentary bottleneck of serial production lines with finite buffers is still a challenge in the automobile industry. In literature
there are already some detection methods. However, these methods lack the applicability for a real-time fault repair prioritization. A new
method is being developed because of this disadvantage to detect the bottlenecks by analyzing the increase and decrease of buffer levels. This
method has the benefit of detecting the real-time bottlenecks as well as giving a forecast bottleneck emerging in the near future by indicating
the remaining time until a machine becomes one. The new fault repair prioritization method was validated in a material flow simulation of a
production line. This simulation indicates a significant output increase. In the following steps, the new method is going to be implemented in a
case study.
© 2015
© 2015 The
The Authors.
Authors. Published
Published by
by Elsevier
Elsevier B.V.
B.V. This is an open access article under the CC BY-NC-ND license
Selection and peer-review under responsibility of the International Scientific Committee of the “4th CIRP Global Web Conference” in the
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
person of theunder
Peer-review Conference Chair of
responsibility Dr.the
John Ahmet Erkoyuncu.
organizing committee of CIRPe 2015 - Understanding the life cycle implications of manufacturing
2212-8271 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the organizing committee of CIRPe 2015 - Understanding the life cycle implications of manufacturing
doi:10.1016/j.procir.2015.08.071
Michael Wedel et al. / Procedia CIRP 37 (2015) 140 – 145 141
repair (as is shown in Fig. 1) at the faulted real-time downstream of the buffer with most of the parts [17, 18]. This
bottleneck which also will reduce the key performance method works well by analyzing job flow systems. If the
indicator “mean time to repair” (MTTR; often used in lean production line has finite buffers [19] or no buffers at all [16],
production). Therefore, an effective fault repair prioritization this method cannot be used and it will not be applicable if
focusing on bottlenecks is needed which is already a known several machines get their parts from one buffer [20].
method for maintenance prioritization [4]. Most long-term or The second method is the active period method by Roser
short-term bottleneck detection methods are used for et al.. Roser’s method analyzes how long a machine is in
maintenance prioritization, which determine the bottleneck for active mode (produces parts, is faulted, has a tool change etc).
a selected longer time period (long-term) [5] or for a day If the machine becomes idle, the active mode will change into
inactive. The momentary bottleneck at a point of time is the
(short-term) [6]. It is also known that maintenance
machine with the longest uninterrupted active time [16]. Yang
prioritization based on short-term bottlenecks results in a
et al. modified the active period method by defining the active
higher output increase than focusing on long-term bottlenecks
time as the time only when the machine is in production [6].
[7, 8]. The reason behind this fact is that in balanced Eskandari et al. declared the “criticality indicators based
production lines with nearly identical cycle times [9] the method” as a further method for detecting momentary
bottleneck shifts randomly [10] with different probabilities. bottlenecks [21]. This method was derived from Králová et al.
When taking into account this behavior a fault repair and has the aim to detect machines with critical capacities [22,
prioritization according to real-time bottlenecks and in 23]. The criticality indicators based method relies entirely on
addition to near future bottlenecks are a challenge but it averages, and this aspect emphasizes the methods inability to
should even be more effective, too. fully capture the bottleneck shiftiness of real-time bottlenecks
[14].
2. Previous Work on bottleneck definitions and detection
methods Another possible time focus for the bottleneck detection is
the bottleneck prediction [24, 25], which is also an important
A sizeable amount of literature about bottleneck piece of information for the fault repair prioritization: Ni and
definitions exists but none of these definitions are for general Jin see a high potential in prioritizing maintenance tasks by
usage [11]. All definitions agree to the fact that the bottleneck analyzing future bottlenecks [26]. In literature, there are
has a negative impact on the output of the production system. already some bottleneck prediction methods. These are based,
The negative impact is commonly described as: “constraining for example, on stochastic models like the ANFIS [27] or the
the system” [12]. ARMA model [24]. In comparison to the long-term, to the
short-term and to the momentary bottleneck detection
As already described in the introduction, bottlenecks are methods, the bottleneck prediction received less focus in
not stable in serial production lines, as they shift randomly due literature the few last years. Only Lee and Ni used a
to machine downtimes (faults, tool changes, preventive bottleneck prediction for a fault repair prioritization (fault
maintenance…) [13]. The more a line is balanced, the more repair prioritization according on the urgency) [28], by
the bottleneck shifts [14]. For the evaluation of the balanced analyzing the increase and decrease of the buffer level. While
rate of the production line the utilization formula of Lawrence the first step of the newly derived method in this paper is also
et al. can be used [10]: based on this idea, it can also be applied for more complex
production lines with parallel machines as well as to detect
the real-time bottleneck in a serial production line.
ߩ ൌ (1)
ఓ
3. Conclusion of the literature review
In this formula ߩ is the utilization rate, ɉ the arriving rate
of parts and ߤ the distribution of the service time. If the ߩ of In literature there are a lot of bottleneck detection methods
every machine of the line has the same value, the bottleneck which can be sectioned into long-term, short-term, momentary
shiftiness will occur most frequently [10]. The shifting bottleneck detection methods, or bottleneck prediction
bottleneck phenomenon is the reason why long-term and methods. However, there is no method for a fault repair
short-term bottleneck detection methods oversee temporary prioritization that focuses on the real-time bottlenecks as well
bottlenecks [15] which implies the need to prioritize fault as on the near future bottlenecks in a complex production line
repair based on real-time bottlenecks. Generally bottleneck with parallel machines and finite buffers. This information is
detection methods can be separated into simulation based and
crucial for a fault repair prioritization in the daily business of
analytical methods [7], the latter of which is the focus of this
the automotive industry. Therefore, the lack of applicability
paper.
was the motivation to derive a new and faster method. When
The real-time bottleneck is a special form of a momentary
implementing this method, a fault repair prioritization for the
bottleneck. Momentary bottlenecks are ones that can occur at
any point in time [16], and any point in time can even be operator or for maintenance personnel can be issued at the
defined as the current moment (real-time). In literature there serial production line on the plant floor. This change will help
are two main methods for detecting momentary bottlenecks. to reduce the reaction time on bottlenecks and increases the
The first method analyzes the waiting time of the parts in output by repairing the machines in the order they are actually
front of a machine, which is comparable to analyzing the limiting the production line output. The benefit of the new
queue length in front of this machine [10]. The momentary method will be shown by a simulation at the end of this paper.
bottleneck of a production line is the machine, which is
142 Michael Wedel et al. / Procedia CIRP 37 (2015) 140 – 145
4. A method for detecting real-time and near future and the buffer Bi downstream of machine Mi will be empty
bottlenecks after the time duration
In this paper a bottleneck is defined as a machine that has a ݐ ൌ ܤ ሺݐ ሻ ݐܿ כାଵ . (3)
negative impact on the output of other machines by a
disturbance of the throughput (as previously mentioned). This In both formulas ܿ ݐis the cycle time (once for machine
adverse impact results in idle time at nearby machines. The Mi-1 and once for machine Mi+1). ܤିଵ௫ is the maximal
reason for the idle time at upstream machines are full buffers number of parts that can be stored in buffer ܤିଵ and ܤ ሺݐ ሻ is
and at downstream machines empty buffers. This phenomenon the number of parts in buffer ܤ at the time ݐ (respectively
will result if the bottleneck machine has, for example, a fault, ܤିଵ ሺݐ ሻ ). Analogous to formula (2) and (3) the time
a tool change, a cleaning or if its cycle time is higher than the durations ݐିଵ and ݐ can be calculated for the buffers next
cycle time of the nearby machine. The new method focuses on to the faulted machine Mn. After this, the shortest time ܵܶ
the unplanned standstills (faults) because this is the point of until a nearby machine will be negatively impacted can be
interest for the operator and maintenance. The idea of the determined by:
method is to concentrate on the decrease and increase of the
buffer levels between the machines (step 1, according to the ܵܶ ൌ ሺݐିଵ Ǣ ݐ Ǣ ݐିଵ Ǣ ݐ ሻ (4)
idea of Lee and Ni [28]) as well as analyzing the prioritization
of the fault repairs if the capacity of the buffers reaches its The faulted machine with the shortest time ܵܶ until a
limit (step 2). This approach allows for detecting the real-time nearby buffer first reaches its capacity limits (full or empty),
must be prioritized for the fault repair (according to [28]).
bottleneck as well as giving a forecast when a machine will
Formula (4) also gives the information of how long the repair
become a bottleneck by analyzing the remaining time until a
of the prioritized machine may continue until this machine
nearby machine gets idle. Finally, the case will be analyzed
will become a bottleneck (bottleneck prediction).
only if parallel machines are faulted (step 3).
The serial production lines in the automobile industry are
4.1. Step 1
mostly more complex than the line in Fig. 2. A large portion
of serial production lines have parallel machines like in Fig. 3:
The remaining time till the buffer is empty (ܤ௫ ሺݐሻ ൌ Ͳሻ can be 3. The downstream buffer is empty or there is no
analyzed analogously: buffer (high priority).
4. The upstream buffer is full or there is no buffer
ሺ௧బ ሻכ௧ כ௧శభ
ݐ ൌ ቚ ቚ (7) (high priority).
௧ ି௧శభ
5. The cycle time of still running parallel machines has
For formula (6) we suppose, that the buffer level will the same value as the downstream or upstream
increase and for formula (7) that the buffer level will nearby machine(s) (no priority).
decrease. In the special case, where a buffer level after a block
of parallel machines can increase and upstream buffer level If there is no upstream or downstream buffer or the buffer
can decrease, the buffers must first be analyzed in more detail capacities are exhausted (category 3 and 4), the faulted
before using the above mentioned formulas. Hence, the first machine will have a negative impact on other machines
step is to investigate every buffer to see if the buffer level is
increasing or decreasing by: (bottleneck). If more than one machine belongs to category 4
and 5, it must be analyzed which of these machines must be
ܤሶ௫ ሺݐሻ ൌ
ଵ
െ
ଵ
(8) repaired first. To identify this machine, it must be understood
௧ೣ ௧ೣశభ which machine causes the biggest loss of value added. Cycle
time can be seen as the representative of value, which a
If ܤሶ௫ ሺݐሻ ൏ Ͳ, the buffer level will decrease; if ܤሶ௫ ሺݐሻ Ͳ,
machine adds to the part. According to this viewpoint, the
the buffer level will increase and if ܤሶ௫ ሺݐሻ ൌ Ͳ, the buffer level
will stay constantly. In the case that the buffer level in front of faulted machine that generates the highest loss of value added
the block of parallel machines in Fig. 3 is decreasing then to idle machines has the highest priority:
instead of formula (6) the formula (9) has to be used:
ܲ ݅ݎൌ ݄݉ܽܿ݅݊݁ ݔ݄ܽ݉ݐ݅ݓσ௫ୀଵ ܿݐ (13)
൫షభ ሺ௧బ ሻ൯כ௧ כ௧షభ
ݐିଵ ൌ ቚ ቚ (9) In formula (14) x is the number of machines that are idle
௧ ି௧షభ
because of the faulted machine and ܿݐ is the cycle time that
If the downstream buffer is increasing the following formula every particular machine needs to process parts.
has to be taken instead of formula (7):
4.3. Step 3
ሺషభೌೣ ିషభ ሺ௧బ ሻሻכ௧ כ௧శభ
ݐ ൌ ቚ ቚ (10)
௧ ି௧శభ
For the special case that only machines are faulted parallel
to each other, downstream and upstream buffers do not need
In formula (6), (7), (9) and (10) ܿݐ is the resulting cycle time
to be analyzed. The parallel machines can be seen as one
of the still processing parallel machines:
machine on the line that has a reduced resulting cycle time if
ିଵ one or more of the parallel machines are faulted. For the fault
ଵ
ܿݐ ൌ σ௭௬ୀଵ ൨ (11) repair prioritization it is more interesting to know which
௧
machine can be repaired in the shortest time. Therefore, the
analysis of the fault specific mean time to repair (MTTRF) is
In formula (11) ܿݐ௬ is the current cycle time of every important.
parallel machine of the parallel machine block. If the machine The analysis of only the MTTRF would provide the right
is faulted, the cycle time of this machine will be λ . By priority if all parallel machines had the same cycle time. In
analyzing the limiting value, the cycle time of a faulted practice the cycle time of parallel machines slightly differs.
machine is not taken into consideration in formula (11). Hence, the minimal value of the multiplication of the fault
specific mean time to repair and the cycle time characterizes
The faulted machine with overall minimum value of tfi-1, tei, the machine that has priority:
tfn-1 or ten has to be prioritized:
ܲ ݅ݎൌ ݄݄݉ܽܿ݅݊݁݅݊݅݉ݐ݅ݓሾܴܶܶܯி ݐܿ כ ሿ (14)
ܲ ݅ݎൌ ൣݐିଵ Ǣ ݐ Ǣ ݐିଵ Ǣ ݐ ൧ (12)
When examining parallel machines, there is another
4.2. Step 2 special scenario to consider if the loader L in Fig. 3 (e.g. a
gantry or a robot) of the parallel machines is faulted. In this
For the special case that a faulted machine has no buffer or case, every single parallel machine will stop working after
the buffer has reached its capacity limit, this machine has an finishing the current cycle. If the loader is faulted, it will be
even higher priority because the faulted machine causes a reflected in the algorithm as if all parallel machines, which get
negative impact on nearby machines (bottleneck). Hence, the supplied by the loader, are faulted.
case ܤሶ௫ ሺݐሻ ൌ Ͳ has to be analyzed. The reasons for ܤሶ௫ ሺݐሻ ൌ The derived method can also be used for a priority
analysis of more than two faulted machines. In order to do so,
Ͳcan be:
the buffers next to all faulted machines have to be analyzed.
With the help of Step 1, 2 and 3 the machine with the highest
1. The downstream buffer is full (no priority).
priority, as well as priority ranking of other faulted machines
2. The upstream buffer is empty (no priority). can be detected.
144 Michael Wedel et al. / Procedia CIRP 37 (2015) 140 – 145
5. Simulation of the fault repair prioritization method Table 2. Buffer capacities (maximal number of parts)
Cycle
Machine MTTR MTBF time
M2 0:04:55 0:50:59 17.6
Loader L 0:02:41 0:30:01
M3-1 0:07:22 0:41:21 24.4
M3-2 0:06:55 0:25:27 24.4
M4 0:00:45 0:51:58 19.0
M5 0:04:28 1:20:13 18.6
M6 0:03:08 1:17:18 17.5
M7 0:07:58 1:20:12 17.6
M8 0:10:44 1:34:22 17.9 Fig. 4. Simulation results
M9 0:07:01 0:42:21 18.4
M10 0:10:21 0:24:19 17,6 Fig. 4 shows that the newly developed prioritization
M11 0:04:56 1:09:53 19.2 method increases the throughput in all case compared to the
M12 0:07:41 2:27:36 17.6 “first come first serve” prioritization. In the special case only
M13 0:03:22 1:50:11 17.6 one worker is assigned to the production line, the throughput
M14 0:05:58 1:29:00 17.6 is low because one worker cannot handle the fault repairs
adequately. Consequentially, in many cases the buffers reach
Table 1. Provided data for the simulation
their capacity limits and the throughput of the line collapses.
If two workers are available, the impact of the
Table 1 shows the mean time to repair (MTTR) and mean prioritization method will show its highest effect (gaining
time between failure (MTBF) data from a real production line throughput of 4.5%). If more than two workers are available
which was used for the simulation. For the loader L no cycle in the production line, a positive but a smaller impact can be
time is specified because in the simulation it receives a observed. This conclusion is due to the fact that both methods
driving command from M3-1 or from M3-2 as soon as they result in a saturated condition. Using the “first come first
have finished their production cycle. Table 2 shows the buffer serve” prioritization method, the saturated condition (nearly
capacities that got used for the simulation. maximum throughput) is achieved with five workers. On the
If a fault occurs the worker will be sent to the faulted other hand, the new prioritization method already delivers the
machine. As long as more workers are available than faults maximum output with three workers. Hence, the new method
occur, a prioritization method is not needed. If more faults are allows the manager the possibility of running the production
in the production line than workers are available, the new line more effectively (throughput increase), more efficiently
method will determine which machine will have to be (decrease the number of worker), or both.
repaired first. The assumption must be made that all workers
are equally qualified for repair.
Michael Wedel et al. / Procedia CIRP 37 (2015) 140 – 145 145
6. Conclusion and future work [7] Chang Q, Ni J, Bandyopadhyay P, Biller S, Xiao G. Supervisory
factory control based on real-time production feedback. Journal of
Manufacturing Science and Engineering 2007; Vol. 129, p. 653-660.
In this paper a new fault repair prioritization method for
[8] Li L, Ni J. Short-term decision support system for maintenance task
serial interlinked production lines with finite buffers was prioritization. International Journal of Production Economics 2009;
developed that aims to decrease the reaction time on 121, p. 195-202.
throughput critical machines. The prioritization method is [9] Conway R, Maxwell W, McClain JO, Thomas LJ. The role of work-in-
focusing on real-time and near future bottlenecks because process inventory in serial production lines. Operation Research 1988;
Vol. 36, No. 2, p. 229-241.
these bottlenecks are decreasing the throughput of the
[10] Lawrence SR, Buss AH. Shifting production bottlenecks: causes, sures,
production line currently or will decrease it in near future. and conundrums. Production and Operations Management 1994; Vol.
Therefore, two new bottleneck detection methods were 3, No 1; p. 21-37.
developed in this paper. The newly introduced real-time [11] Wang Y, Zhao Q, Zheng D. Bottleneck in production networks: An
bottleneck detection method analyzes the loss of value added overview. Journal of Systems Science and System Engineering 2005;
Vol. 14, No. 3, p. 347-363.
of idle machines. The second method, which detects the near
[12] Kuo CT, Lim JT, Meerkov SM. Bottlenecks in serial production lines:
future bottleneck, is focusing on the buffer levels and is a a system-theorethic approach. MPE 1996;Vol. 2, p. 233-276.
further development of the ideas of Lee and Ni [28]. The [13] Sengupta S, Das K, VanTil RP. A new method for bottleneck
simulation of the new prioritization method compared to the detection. Proceedings of the 2008 Winter Simulation Conference
“first come first serve” method shows a throughput increase 2008; p. 1741-1745.
[14] Roser C, Lorentzen K, Deuse J. Reliable shop floor bottleneck
up to 4.5%. In exchange for output, it is also possible to
detection for flow lines through process and inventory observation.
economize repair personnel. Procedia CIRP 19 2014; p. 63-68 .
The introduced method is a link between machines and [15] Wang Z, Chen J, Wu Q. A new method of dynamic bottleneck
operators, and specifically helps the worker to work more detection for semiconductor manufacturing line. The International
effectively. Future research work should quantify the benefit Federation of Automatic Control 2008; p. 14840-14845.
[16] Roser C, Nakano M, Tanaka M. Shifting bottleneck detection.
of the new method for different production lines in the
Proceedings of the 2002 Winter Simulation Conference. 2002; p. 1079-
automotive industry, especially production lines with a high 1086.
product variety. Another application possibility can be found [17] Li L, Chang Q, Ni J, Xiao G, Biller S. Bottleneck detection of
in the field of logistic processes, but further work must be manufacturing systems using data driven method. International
conducted. Symposium on Assembly and Manufacturing 2007; p. 76-81.
[18] Betterton CE, Silver SJ. Decting bottlenecks in serial production lines
In the following steps, the new prioritization method is
– a focus on interdeparture time variance. International Journal of
going to be implemented in a case study to verify the output Production Research 2012; Vol. 50, No. 15, p. 4158-4174.
increase. Therefore, the data of the manufacturing data [19] Roser C, Nakano M, Tanaka M. A practical bottleneck detection
collecting system is going to be obtained and the fault repair method. Proceedings of the 2001 Winter Simulation Conference 2001;
prioritization is going to be displayed on the Andon Boards as p. 949-953.
[20] Roser C, Nakano M. Constraint management in manufacturing
well as on computer tablets. The new method should also
systems. JSME International Journal 2003; Vol. 46, No. 1, p. 73-80.
contribute to the science project “Industry 4.0”. [21] Eskandari H, Babolmorad N, Farrokhnia N. Bottleneck analysis in a
pharmaceutical production line using simulation approach. Proceedings
7. Acknowledgement of the 2013 Summer Computer Simulation Conference 2013.
[22] Leporis M, Králová Z. A simulation approach to production line
The authors thank Ulrich Burges (SimPlan AG) for bottleneck analysis. International Conference Cybernetics and
Informatics 2010.
implementing the new fault repair prioritization method into [23] Králová Z, Bielak M. Synchronizácia výrobného proceso s využitím
an existing production line simulation model and Heiko simulácie vo witnesse. 7. Konference Witness 2004.
Noessler (Daimler AG) for providing the data for the [24] Li L, Chang Q, Xiao G, Ambani S. Throughput bottleneck prediction
simulation. We would also like to thank Philipp Noessler of manufacturing systems using time series analysis. Journal of
(Daimler AG) for his assistance in the review process. Manufacturing Science and Engineering 2011; Vol. 133.
[25] Arab A, Ismail N, Lee LS. Maintenance scheduling incorporating
dynamics of production system and real-time information from
8. References workstations. J Intell Manuf 2013; 24, p. 695-705.
[26] Ni J, Jin X. Decision support systems for effective maintenance
[1] Lemessi M, Rehbein S, Rehn G. Semi-automatic simulation-based operations. CIRP Annals – Manufacturing Technology 61 (2012); p.
bottleneck detection approach. Proceedings of the 2012 Winter 411-414.
Simulation Conference 2012. [27] Cao Z, Deng J, Liu M, Wang Y. Bottleneck prediction method based
[2] Goldratt EM, Cox J. The Goal: A process of ongoing improvement. on improved adaptive network-based fuzzy interference system
20th anniversary edition, Edition: 3, Productivity & Quality Publishing (ANFIS) in semiconductor manufacturing system. Chinese Journal of
Private Limited 2004. Chemical Engineering 20 (6), 2012; p. 1081-1088.
[3] Nakajima S. Introduction to TPM: Total productive maintenance. [28] Lee S, Ni J. Joint decision making for maintenance and production
Productivity Press Cambridge 1988. scheduling of production system. International Journal of Advanced
[4] Li L, Ambani S, Ni J. Plant-level maintenance decision support system Manufacturing Technology 2013; p. 977-990.
for throughput improvement. International Journal of Production
Research 2009; Vol. 47, No. 24, p. 7041-7061.
[5] Roser C, Nakano M, Tanaka M. Detecting shifting bottlenecks.
International Symposium on Scheduling 2002; p. 59-62.
[6] Yang K, Chung Y, Park SC. Short-term bottleneck detection for
process planning in a FAB. International Journal of Computer and
Communication Engineering 2014; Vol. 3, No. 6, p. 442-445.