Professional Documents
Culture Documents
To cite this article: K. M. N. Muthiah & S. H. Huang (2007) Overall throughput effectiveness (OTE)
metric for factory-level performance monitoring and bottleneck detection, International Journal of
Production Research, 45:20, 4753-4769, DOI: 10.1080/00207540600786731
Taylor & Francis makes every effort to ensure the accuracy of all the information (the
Content) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
International Journal of Production Research,
Vol. 45, No. 20, 15 October 2007, 47534769
1. Introduction
attempts to reveal the hidden costs associated with a piece of equipment. When it is
applied by autonomous small groups on the shop-floor together with quality control
tools, OEE is an important complement to the traditional top-down-oriented
performance measurement systems (Jonsson and Lesshammar 1999). Hansen (2002)
mentioned that using OEE metrics and establishing a disciplined equipment
performance reporting system will help a manufacturing system to focus on the
parameters critical to its success. Analysing OEE categories can reveal the
greatest limits to success. If a plant is in the envious position of selling everything
it can make, then hidden factory is the difference between what good products were
transferred out in the past 12 months and what could be made in 8760 h of perfect
production. World class companies are known for an attribute. They are built
around the concept that an effective factory producing good goods as needed
to meet market demands is a valuable asset for any company to have. Companies
Downloaded by [Dicle University] at 11:14 12 November 2014
with the most effective factories will have the staying power to be the long-term
survivors, assuming that the need for the product is continuous. The size of
the opportunities for business improvement varies proportionately with the level
of information-sharing throughout the company.
Scott and Pisa (1998) pointed out that the gains made in OEE, while important
and ongoing, are insufficient. It is necessary to focus ones attention beyond the
performance of individual tools towards the performance of the whole factory.
The ultimate objective is a highly efficient integrated system, not brilliant individual
tools. They coined the term overall factory effectiveness (OFE), which is about
combining activities and relationships between different machines and processes, and
integrating information, decisions, and actions across many independent systems and
subsystems.
There is considerable amount of literature on manufacturing system productivity
measurement and improvement. Muthiah and Huang (2006) reviewed and
categorized various productivity improvement methods. Operations research-based
methods (e.g. optimization, network computation) and control theoretical methods
are based on rigorous mathematical modelling. System analysis based methods
(e.g. IDEF, Petri net) for productivity improvement are driven by information
systems. Continuous productivity improvement methods (e.g. lean manufacturing,
six Sigma) are primarily empirical methods established by practice. The focus of
this paper is performance metrics-based methods. The review of performance
metrics-based methods indicates that quantitative metrics for measuring
factory-level productivity and for performing factory-level diagnostics (bottleneck
detection, hidden capacity identification) are lacking.
This paper addresses this gap and presents the OTE metrics that could be
used for factory-level performance monitoring and diagnostics. The paper is
organized as follows. Following the Introduction section, section 2 presents related
work in factory-level performance diagnostics. Section 3 presents the concept
of theoretical OEE, followed by the development and validation of OTE metrics.
In section 4, OFE computation and the bottleneck detection ability of OTE
are illustrated. To illustrate the application of the developed metrics in a real-
world scenario, two case studies are discussed in section 5. Finally, section 6 draws
conclusions regarding the utility and applicability of OTE and it discusses future
research scope.
OTE metric for factory-level performance monitoring and bottleneck detection 4755
2. Related work
where Aeff is the availability efficiency that captures the deleterious effects due to
breakdowns, set-ups, and adjustments; Peff is the performance efficiency that
captures productivity loss due to reduced speed, idling, and minor stoppages; and
Qeff is the quality efficiency that captures loss due to defects, rework, and yield.
These three components of OEE are calculated as follows:
Downloaded by [Dicle University] at 11:14 12 November 2014
TU
Aeff 2
TT
TP Ract
Peff NOR SR 3
TU Rth
Pg
Qeff , 4
Pa
where
TU uptime of the equipment,
TT total observation time of the equipment,
NOR net operating rate,
SR speed ratio,
TP production time of the equipment,
Ract actual processing rate of the equipment,
Rth theoretical processing rate of the equipment,
Pg good product output from the equipment,
Pa actual products processed by the equipment during the observation time.
OEE has been used extensively for equipment productivity improvement,
especially in the semiconductor industry (Ames et al. 1995, Bonal et al. 1996,
Giegling et al. 1997, Leachman 1997, SEMI E79-0200 2000). Semiconductor
Equipment and Materials International (SEMI) is now pioneering the development
of OFE metrics (SEMI E124-0703 2003). However, the metrics are geared towards
semiconductor manufacturing. SEMI metrics are intended for evaluating the overall
efficiency of semiconductor manufacturing, not for diagnosing problems (identifying
bottleneck/opportunities for improvement) in the factory. Therefore, their utility is
limited.
An attempt was made by Huang et al. (2002, 2003) to come up with factory-level
diagnostic metrics. However, there were major unaddressed issues. (1) The OTE
metrics defined needed good product output from the factory for computation.
Hence, it cannot be used for factory-level diagnostics such as bottleneck detection.
(2) The OTE metrics developed then were for series and parallel subsystems alone.
This was not sufficient to compute OFE as it lacked constructs to model a factory by
capturing the equipment connectivity information. (3) The OTE metrics developed
4756 K. M. N. Muthiah and S. H. Huang
were not statistically validated. This paper, apart from addressing all these issues,
explains the theoretical OEE concept, OTE development methodology, validates the
developed OTE metrics, and demonstrates its diagnostic ability.
when equipment is ready for production but there are no parts available to be
processed. This productivity loss is due to poor factory operations (because of
material-handling problems or factory design flaws such as unbalanced line, etc.)
rather than caused by the equipment. Therefore, we argue that it should be captured
by factory-level metrics. This is the motivation for proposing the theoretical OEE
concept, which captures productivity loss caused solely by the equipment alone.
Availability Efficiency
Plant
Plant operating time (Aeff ) = Tu /Tt
Shutdown
Setup &
Scheduled Failure Equipment
adjustment
downtime Time operation time (Tu )
time
Performance Efficiency
(Peff ) = Tth /Tact
Minor
stoppages & Theoretical processing time (Tth)
speed losses
Quality Efficiency
(Qeff ) = Pg /Pa
1. Series 2. Parallel
1
2
1 2 n
3. Assembly 4. Expansion
1
1
2 a
e 2
n n
Downloaded by [Dicle University] at 11:14 12 November 2014
Subsystem OTE
n n Q o o
min mini1,2,...,n1 OEEi Rthi nji1 Qeff j ,OEEn Rthn
Series
mini1,2,...,n fRthi g
Pn
i1 OEEi Rthi
Parallel Pn
i1 Rthi
min mini1,2,...,n fOEEi Rthi =kAi Qeffa g,Rtha OEEa
Assembly
min mini1,2,...,n fRthi =kAi g,Rtha
Pn
i1 minfRthe OEEe kEi Qeffi ,Rthi OEEi g
Expansion Pn
i1 minfRthe kEi ,Rthi g
X
n
Qeffi Xij Qeffij : 8
j1
OTE metric for factory-level performance monitoring and bottleneck detection 4759
This OEE(i) is then used in the appropriate OTE metric shown in table 1.
follows. For each subsystem, a simulation model is built capturing the architecture as
proposed in figure 2. The theoretical subsystem throughput is obtained by setting
the theoretical OEE of individual pieces of equipment to 100%. The actual
subsystem throughput is obtained by simulating the subsystem under various loss
scenarios. It should be noted that, there are multiple equipment in each subsystem
simulation model considered for validation. Detailed subsystem configurations
used for validation including efficiency levels can be found in Muthiah (2003).
Specifically, each of the three components of equipment theoretical OEE (Aeff, Peff,
and Qeff) is set at two levels, high and low, to generate eight cases for validation. For
example, when Aeff is low, it is set at low for all the equipment in the subsystem. The
subsystem throughput efficiency is the ratio of the actual throughput to theoretical
throughput equation (6) and let us call it the simulation OTE. If the OTE metric is
valid, it should be the same as simulation OTE. Validation is conducted by testing
the following hypothesis:
Hypothesis testing requires the determination of significance level and sample size.
A significance level of 5% ( 0.05) is used as it is typical in any hypothesis testing
procedure. The sample size plays a critical role in hypothesis testing and there are
different methods for determining an appropriate sample size (Banks et al. 2001).
The percentage change in the first four moments (mean, variance, skewness, and
kurtosis) of subsystem throughput is recorded as sample size increases. Maximum of
four moments converging point is taken as the appropriate sample size for better
precision. Arena 5.0 was used to build subsystem simulation models and to
determine appropriate sample size.
The simulation results for all the subsystems are shown in tables 25, respectively.
Multiple product and rework scenarios are also validated which is not shown here
for brevity. However, one can refer to Muthiah (2003) for details. The
validation tables showed that the null hypotheses (H0: OTE simulation OTE)
cannot be rejected in any case for all the subsystems. This could be observed from the
p-values in tables 25. Thus, the developed OTE metrics are validated.
4760 K. M. N. Muthiah and S. H. Huang
Simulation
Simulation OTE (standard
Case number Qeff, Aeff, Peff OTE OTE (mean) deviation) p
Simulation
Simulation OTE (standard
Case number Qeff, Aeff, Peff OTE OTE (mean) deviation) p
Simulation
Simulation OTE (standard
Case number Qeff, Aeff, Peff OTE OTE (mean) deviation) p
4. Bottleneck detection
Simulation
Simulation OTE (standard
Case number Qeff, Aeff, Peff OTE OTE (mean) deviation) p
Actual processing time (min) TRI(6.1, 6.2, 6.3) TRI(7.4, 7.68, 7.8) TRI(9.0, 9.4, 9.7)
Theoretical processing time (min) 5 7 9
Total time (min) 1440 1440 1440
Uptime (min) 1393.86 1393.54 1398.05
Quality level (%) 100 100 100
1 5.4309 2.6481
2 5.3467 2.0538
3 5.3424 2.1806
Average queue
Case number lengths compared p Mean queue length
Now, we will show how the OTE metrics can be used for bottleneck detection.
Note that the KANBAN system has three machines in series. From the OTE metric
for a series subsystem shown in table 1, one can see that the numerator (and hence
the OTE value) Q is limited by the piece of equipment whose
OEEi Rthi nji1 Qeff j (or if it is the last piece of equipment then Qeff is
dropped) is minimum, which indicates the bottleneck. In our example, this term,
called bottleneck indicator, for equipment 1, 2, and 3 is 0.1561, 0.1260, and 0.1033,
respectively. Therefore, equipment 3 is the bottleneck.
The fact that equipment 3 is the bottleneck can also be shown through sensitivity
analysis. The system is simulated for 1 day with 40 replications for all cases in the
sensitivity analysis. The current system throughput is normally distributed with a
mean of 126 and a standard deviation of 29.048. We then double the capacity
of equipment 1 (by adding an identical piece of equipment in parallel) and obtained
129.75 and 27.609 as the system throughput mean and standard deviation,
respectively. If the capacity of equipment 2 is doubled, then the system
throughput mean and standard deviation are changed to 139.00 and 34.0509,
respectively. In both cases, no significant throughput improvements are observed
at 0.05. On the other hand, when the capacity of equipment 3 is doubled, the
system throughput increased to a mean of 247.38 with a standard deviation
of 75.387, which is significantly different from the original throughput at
0.05. Hence, the sensitivity analysis validates the bottleneck detection method
using OTE.
Likewise if a factory is a combination of more than one subsystem the OTE for a
series subsystem is used to arrive at the OFE and to find the bottleneck. Let us
consider the system in figure 3 for illustration.
The production system in figure 3 has nine machines (M19). The arrows show
the process flow. According to the subsystem definitions (specified in figure 2)
OTE metric for factory-level performance monitoring and bottleneck detection 4763
Parallel Subsystem
(M4, M5, M6) Assembly Subsystem
(M7, M8, M9)
M4
M1 M2 M3 M5 M7 M9
M6
Series Subsystem M8
(M1, M2, M3)
Downloaded by [Dicle University] at 11:14 12 November 2014
Theoretical Theoretical
processing rate of the quality efficiency of
Subsystem subsystem (Rth) the subsystem (Qeff)
machines M1, M2 and M3 form a series subsystem, machines M4, M5 and M6 form
a parallel subsystem, and machines M7, M8 and M9 (assembly station) form an
assembly subsystem. Let us call the three subsystems Subsystems1, 2 and 3,
respectively. To compute the OFE of this production system, OTE for each
of Subsystems1, 2 and 3 are computed from the appropriate OTE formula in table 1.
It should be noted that Subsystems1, 2 and 3 themselves are in series with respect
to each other. Hence, from table 1, OFE is computed from OTE formula for a
series subsystem by replacing OEE with the computed OTE values. Then the
numeratorQ value would be limited by the subsystem whose OTEi
Rthi nji1 Qeffj (or OTE(i) Rth(i) for the final subsystem in the series) value
is the least. This term as mentioned above is called the bottleneck indicator. The
subsystem with the least value of this bottleneck indicator is determined to be the
bottleneck subsystem. The bottleneck equipment in the bottleneck subsystem is
identified by using the corresponding subsystem bottleneck indicator from the OTE
formulae listed in table 1. Rth and Qeff for subsystems are computed as shown
in table 9.
4764 K. M. N. Muthiah and S. H. Huang
5. Case study
a time. The process data and equipment data are shown in tables 10 and 11,
respectively. During the total time of observation, 985 500 wafers were processed
with 46 929 scrapped units. This yielded an overall quality efficiency rate of 0.9524.
The following subsystems are recognized in the wafer fab:
. Two pieces of diffusion equipment form a parallel subsystem.
. Two pieces of dry etch equipment form a parallel subsystem.
. Diffusion subsystem, dry etch subsystem, lithography equipment and
transport are in series.
A single part is processed multiple times (six times, to be exact) in the wafer fab.
Assuming the quality efficiency of each piece of equipment is the same. Since the
overall
p quality
efficiency is 0.9524, individual equipment quality efficiency is thus
6
0:9524 0:9916. A part processed the second time in the same equipment is treated
Diffusion
Dry Etch
Lithography
Figure 4. Case study: process flow of a wafer fab model (SEMI E124-0703 2003).
OTE metric for factory-level performance monitoring and bottleneck detection 4765
as a different part type. Based on the quality efficiency at each piece of equipment,
the product mix for each tool set can be computed as follows. Take the diffusion tool
set for an example. The number of wafers coming for the second time would be
(0.99194) times the original, since the wafers are processed four times previously
before coming again to the diffusion tool set. The proportion of wafers processed for
the first time and the second time are then 0.5081(1/10.99194) and 0.4919(0.99194/
(10.99194)), respectively. Similarly, the product mix at dry etch is 0.5040 first time
processing and 0.4959 second time processing; while the product mix at lithography
is 0.5061 first time processing and 0.4939 second time processing.
Assuming actual processing rate equals theoretical processing rate, the theoretical
OEEs of all pieces of equipment are calculated as shown in table 12. Now we are
ready to calculate the OTE metrics for the subsystems. Take the diffusion tool set
as an example. From table 10, we found that it takes 225 min to process a batch
Downloaded by [Dicle University] at 11:14 12 November 2014
of wafer the first time and 255 min the second time. Since the batch size is 75 (from
table 11), first time processing rate is 0.3333 wafers/min (75/225), second time
processing rate is 0.2941 wafers/min (75/255). From equation (7), we can calculate
the theoretical processing rate of a diffusion machine as [(0.5081 0.3333)
(0.4919 0.2941)] 0.3140 wafers/min. (Assuming both diffusion machines performs
equal number of first and second time processing.) According to the OTE metric for
parallel subsystems shown in table 1 and the theoretical OEE values for the two
diffusion machines shown in table 12, we can calculate the diffusion subsystem OTE
as [(0.9225 0.3140) (0.9621 0.3140)]/[0.3140 0.3140] 0.9423. Since the two
Time to Theoretical
Time to travel to Cycle production
Process step Equipment process Time to the next time time/unit
number set name Time to load batch unload step (min) (min/wafer)
diffusion machines are in parallel (from table 9), the theoretical processing rate of
the subsystem is thus 0.3140 0.3140 0.6280 wafers/min and its quality efficiency is
(0.9919 0.9919)/2 0.9919. Table 13 shows the calculated OTE metrics for all
subsystems. Note that material-handling vehicle (transport) is treated as a piece of
processing equipment with a quality efficiency of 100%.
Since the three subsystems are in series, we are ready to calculate the OFE using
the OTE metric for series subsystems. Note that this time a subsystem is treated as a
virtual piece of equipment, whose OEE is the OTE of the subsystem. First, we
calculate the bottleneck indicator for each subsystem. Take the diffusion subsystem
for an example. The bottleneck indicator is 0.9423 0.6280 0.99194 0.5728.
Similarly for dry etch, lithography and transport, the bottleneck indicators are
1.0733, 0.4251 and 2.7679, respectively. The smallest bottleneck indicator is that of
lithography (0.4251), indicating that it is a bottleneck. From table 13, we find that
the smallest subsystem processing rate is also that of lithography (0.4545). Therefore,
OFE is 0.4251/0.4545 0.9353, indicating that the factory effectiveness is at 93.53%
of the ideal throughput.
be done to increase the OFE. Three scenarios result in increased theoretical OEE of
the bottleneck equipment and hence OFE of the manufacturing system:
. Decreasing the actual processing time of the bottleneck equipment (should
aim to eliminate minor stoppages and speed losses).
. Decreasing the failure time (decreasing plant shutdown, decreasing scheduled
downtime, decreasing failure time, decreasing set-up and adjustment time,
i.e. downtime losses).
. Increasing good product production at the bottleneck equipment (decreasing
scrap, i.e. defect losses).
The bottleneck equipment is Inspection1. Let us observe what happens if the
bottleneck equipments actual processing time is made three-quarters of its current
processing time. Decreasing the actual processing time at Inspection1 (bottleneck
equipment) resulted in an increase of OFE to 0.4261. The new bottlenecks of the
system are Molder-a and Molder-b. Now to improve the OFE further if the actual
processing time of Molder-a and Molder-b are decreased by three-fourths, OFE
increased to 0.4767. Computations showed that Inspection2 is the bottleneck now.
Now finally if we reduce the actual processing time of Inspection2 to three-fourths
of its current time, OFE increased to 0.5372. Hence, OTE, apart from measuring
performance at the factory level also indicates the bottleneck equipment. Using OTE,
factory professionals can quantitatively find out areas that constrain the factory
productivity.
This paper presents the developed OTE metrics for factory-level performance
monitoring and diagnostics. The concept behind OTE development and the OTE
validation methodology is explained. By using examples, the OFE computation and
OTE metric for factory-level performance monitoring and bottleneck detection 4769
References
Downloaded by [Dicle University] at 11:14 12 November 2014
Ames, V.A., Gililland, J., Konopka, J., Schnabl, R. and Barber, K., Semiconductor
Manufacturing Productivity Overall Equipment Effectiveness (OEE) Guidebook.
International SEMATECH, Report Technology Transfer 95032745A-GEN, 1995.
Banks, J., Carson, J.S. and Nelson, B.L., Discrete Event System Simulation, 2001 (Prentice-
Hall: Upper Saddle River, NJ).
Bonal, J., Ortega, C., Rios, L., Aparicio, S., Fernandez, M., Rosendo, M., Sanchez, A. and
Malvar, S., Overall fab efficiency, in Proceedings of the 7th Annual IEEE/SEMI
Advanced Semiconductor Manufacturing Conference, Cambridge, MA, USA, 1996.
Giegling, S., Verdini, W.A., Haymon, T. and Konopka, J.M., Implementation of overall
equipment effectiveness (OEE) system at a semiconductor manufacturer, in Proceedings
of the 1997 IEMT Symposium, Austin, TX, USA, 1997.
Hansen, R.C., Overall Equipment Effectiveness A Powerful Production/Maintenance Tool
for Increased Profits, 2002 (Industrial Press: New York, NY).
Huang, S.H., Dismukes, J.P., Shi, J., Su, Q., Razzak, M.A., Bodhale, R. and Robinson, D.E.,
Manufacturing productivity improvement using effectiveness metrics and simulation
analysis. Int. J. Prod. Res., 2003, 41(3), 513527.
Huang, S.H., Dismukes, J.P., Shi, J., Su, Q., Wang, G., Razzak, M.A. and Robinson, D.E.,
Manufacturing system modeling for productivity improvement. J. Manuf. Sys., 2002,
21(4), 249259.
Jonsson, P. and Lesshammar, M., Evaluation and improvement of manufacturing
performance measurement systems: the role of OEE. Int. J. Oper. Prod. Manag.,
1999, 19(1), 5578.
Korgaonker, M.G., Just in Time Manufacturing, 1992 (Macmillan India: Bangalore).
Leachman, R.C., Closed-loop measurement of equipment efficiency and equipment capacity.
IEEE Trans. Semiconduct. Manuf., 1997, 10(1), 8497.
Muthiah, K.M.N., Diagnostic factory productivity metrics. MS thesis, University of
Cincinnati, 2003.
Muthiah, K.M.N. and Huang, S.H., A review of literature on manufacturing systems
productivity measurement and improvement. Int. J. Ind. Sys. Eng., 2006, 1(4).
Nakajima, S., Introduction to Total Productive Maintenance (TPM), 1988 (Productivity Press:
Cambridge, MA).
Scott, D. and Pisa, R., Can overall factory effectiveness prolong Moores law? Solid State
Tech., 1998, March, 7582.
SEMI E124-0703, Provisional guide for definition and calculation of the overall factory
efficiency (OFE) and other associated factory-level productivity metrics, in
Semiconductor Equipment and Materials International, 2003.
SEMI E79-0200, Standard for definition and measurement of equipment productivity, in
Semiconductor Equipment and Materials International, 2000.