Professional Documents
Culture Documents
2, 2012 197
Jun Duanmu*
Old Dominion University,
Virginia Modeling Analysis and Simulation Center,
1030 University BLVD,
Suffolk, VA 23435, USA
Fax: +1 757 686 6214
E-mail: jduanmu@odu.edu
*Corresponding author
Kevin Taaffe
Clemson University,
Department of Industrial Engineering,
130-A Freeman Hall,
Clemson, SC 29634, USA
Fax: +1 864 656 0795
E-mail: taaffe@clemson.edu
Abstract: This paper is motivated by a firm that manufactures two main types
of products. In an effort to increase their throughput, the company created a
production scheme using takt times. To achieve a smooth flow of production,
they desired low work-in-process inventory in order to make all components
move simultaneously. However, the process includes parallel assembly lines
that converge to or diverge from common resources. A simple takt time
calculation cannot provide enough information to achieve the desired
throughput. The authors identify solutions that improve throughput. One
solution, based on takt times or each station’s processing times (PTs) for a zero
waste schedule, can be used in many other scheduling problems. With the aid
of Little’s Law and simulation, the authors also show how the system flow can
be influenced via managerial settings. Across all production configurations, a
10–15% increase in throughput over the original design can be achieved.
1 Introduction
Many firms today have initiatives to reduce waste and improve product flow, and a
common solution or implementation approach is to apply lean principles within the firm’s
production system. This revolution in the production and manufacturing environment
began many years ago when Womack et al. (1990) created the lean production system, a
method of measuring a company’s manufacturing effectiveness. It is most commonly
known as a systematic approach to reducing (or eliminating) seven critical wastes, which
will ultimately lead to reducing variability in supply, processing time (PT) and demand
(Shah and Ward, 2007). For many manufacturers, lean is the set of ‘tools’ that assist in
the identification and steady elimination of waste. Examples of such tools are value
stream mapping (VSM), Five S, Kanban (pull systems), takt time and poka-yoke (error-
proofing). These tools have been so well adopted across the manufacturing industry that
Womack and Jones (2003) believe the lean production system would be the predominant
method for conducting any manufacturing operation in the 21st century. However, it is
important to note that the individual application of any of the above tools will not always
yield the improvement in performance that the company seeks.
In the motivating example provided in this research paper, one such company wanted
to arrange a new production layout to support a significant increase in the throughput of
their two main products. The company’s two product lines could have extensive
customisation, which further complicated the ability to maintain the philosophy of lean
thinking. With subgoals of having synchronised production, one-piece-flow and low
WIP, a production design based on takt times was adopted, however desired throughput
levels could not be confirmed. In this paper, the authors describe some key aspects of
lean, takt time analysis, Little’s Law and how these concepts can be combined with
simulation to address complex manufacturing issues. Implementation of a single lean
principle tool can be insufficient without considering a more rigorous mathematical
analysis and, in this research, a process flow simulation.
The remainder of this paper is structured as follows. Section 2 provides a brief
overview of the literature related to the research. Section 3 describes the production
scheme. Section 4 presents the base simulation model and the initial issues that were
faced. A throughput analysis follows in Section 5, which then provided motivation for the
sensitivity analyses performed in Section 6. The research conclusions are shown
in Section 7.
Production capabilities using takt times, requirements analysis and simulation 199
2 Related literature
Takt time only defines the synchronised production pace in a company under a complex
manufacturing environment. Such an environment can include assembly lines that merge
or diverge and multiple products that use the same assembly lines yet they require unique
PTs. Often a detailed production schedule is necessary to fulfil the goal of lean
production, and research has been carried out in the modelling of such problems
(Balakrishnan and Vanderbeck, 1999; Lee and Vairaktarakis, 1997; Miltenburg, 2007).
In fact, improving the productivity of manufacturing has been well studied, without a
consensus on the best method for attaining the desired improvements (Gershwin, 2000).
Muthiah and Huang (2006) provide a review of the various measurement systems in
place, and how metrics-based diagnosis and productivity measurement can achieve
improvements in manufacturing. Using a survey and analysis of four unique industry
partners, Guinery et al. (2010) present a toolkit for improving or redesigning the
production planning, scheduling and control practices.
There can be different goals in the modelling of a manufacturing process. Heike et al.
(2001) summarise the general types and approaches to solve these problems. They also
put forward a model for a low volume, mixed one-piece-flow model for an aerospace
company. This model has similarities with the model in this paper.
Researchers have used many methods in identifying the interrelationship between
resources, components and the underlying stochastic process. In queuing theory, the most
fundamental relationship is Little’s Law, which is expressed as
WIP = TH × CT (1)
where WIP is the work-in-process inventory (WIP), TH represents the throughput of the
system, and CT represents the cycle time per unit processed in the system. This
relationship is the basis for Hopp and Spearman’s (1999) Factory Physics approach to
managing a manufacturing process. It is clear that even though one system may have
higher WIP, if each unit requires a longer cycle time, then the system loses any benefit of
the increased WIP. If the system is very complex and Equation (1) is difficult to use
directly, simulation is often used. AlDurgham and Barghash (2008) highlight the nature
of relationships between different decision areas and how simulation can be used in a
more systematic way to support decision-making. They present a framework for applying
simulation in areas ranging from layout to scheduling and how to use a feedback-
adjustment procedure with progressive improvement. Ali et al. (2005) analyse a lean
production environment, which combines product mix and production volume with the
aid of simulation. The performance under different production scenarios is developed to
find the optimal combination of product mix to meet future customer demands. This
research provides a re-configurable assembly system modelling by adding flexibility and
evaluates alternative designs. In the authors’ research, a similar methodology is used in
creating different production schemes as well as incorporating an optimisation function to
identify the best parameters.
Czarnecki and Loyd (2001) present the application of simulation in a lean assembly
line for high-volume manufacturing. Using concepts such as one-piece-flow and takt
times, they provide a simple example about two different models based on both
traditional and lean systems. However, it is difficult to determine exactly how beneficial
any new policies or changes to process flows would be for more complex problems.
Schroer (2004) also used simulation to understand lean manufacturing. Using a software
200 J. Duanmu and K. Taaffe
product called Modular Manufacturing Simulator, they illustrate the relationship between
cycle time, WIP and resource allocation. Although they didn’t directly analyse the
problem with the theory of Little’s Law, their work indicates the relationship between
these variables and why lean is so effective.
There are several examples of researchers using simulation in various production
environments. Lian and Van Landeghem (2002) compare pull and MRP systems based
on two simulation models. Park et al. (1998) presents a simulation model to assess
maximum throughput at the Mercedes Benz AAV production facility. While buffer sizes
were calculated and bottleneck analysis was performed, knowledge of Little’s Law and
lean on their model was not directly applied. Hung and Liker (2007) also address how
various production conditions (e.g. batching, reliability and sequencing) can have a
significant impact on the ability of a pull system to maintain the expected production
responsiveness and throughput capability.
Adams et al. (1999) introduce how simulation can be used as a tool for continuous
process improvement. They presented several model structures to represent different
improvement plans and compared their effects. While simulation is often chosen as a tool
to evaluate process improvement, firms understand that time and modelling expertise are
essential in conducting sound projects. Through a parametric approach, Farr et al. (2010)
also show how simulation can be effectively implemented without the typical investment
in time and modelling resources.
A new trend in simulation modelling research is to combine design of experiments
(DOE) with simulation to identify the significant system factors and develop a
mathematic regression model representation. In particular, Sandanayake and Oduoza
(2009) estimate throughput, using takt times and process times obtained from a
regression model. Our research differs from this approach in that an exact mathematical
sequencing analysis is conducted to provide an understanding of the limits on achievable
throughput. In addition, the simulation-optimisation approach within this paper indicates
how process times and takt times can be manipulated to reach maximum throughput
while adhering to constraints in the production process.
All in all, there is a great deal of literature in several areas within production
planning. However, combining these issues in greater detail has been less explored.
Moreover, little documented research exists in combining these topics with an
optimisation approach. In fact, simulation is a powerful tool for understanding how these
various tools and concepts work together.
In addition, there are two types of component 1c, depending on the level of specialisation
equipped, namely 1ch (high-spec) and 1cl (low-spec), which represent 25% and 75% of
the component c, respectively. Figure 1 depicts this process flow for each component and
product type. There are roughly eight stages of production as identified in Figure 1 by the
bottom row. The main stages and component process flows are described next.
1 Components a, b and c enter the production system from their own pre-treat lines,
however 1b and 2b share the same resource of pre-treatment.
2 After the pre-treat, 1a, 2a and 1c enter the metal work stage, where they are
processed separately.
3 After the metal work stage, all components will converge and undergo processing
one-by-one during the Common stage (a case in which the components will be
batched as a group of two will be discussed also in later parts).
4 After the Common stage, 1a and 1c enter the Type 1 Paint stage, 2a enters the Type 2
Paint stage and 1b and 2b proceed directly to the Custom 1 stage. Note that all 1a’s
might flow through Type 1 Paint twice, i.e. in some cases this station has a re-entrant
flow of 1a. After Paint, 1c, 1a and 2a enter their respective custom stages.
5 After the Custom 1 stage, components a and b are assembled together for each
product and form component ab. Component 1c awaits the matched 1ab coming
from the Check stage.
6 The assembled ab components share the resources at the Check Stage, and then they
continue to Custom 2 for their own product types. After Custom 2 stage, Product P2
completes a Final check, while Product P1 will be created in Assembly 2 by
combining component 1ab with 1c. The assembled Product 1abc then proceeds to the
Final check.
There are no queues between each stage. However, the stage can have multiple stations
for processing by allowing multiple parts to be processed within the same stage. It is a
single-flow process, where the next task cannot be performed until the component in
front finishes its processing and moves forward. Using this as motivation, the company
calculated appropriate station PTs based on takt time calculations.
where Nbatch is the number of components that need to be batched, while Nnot batch is the
number that are not batched. This results in a takt time of 450/24 = 18.75 min.
After step 2 in the Common stage, component 2a will immediately move to its Paint
stage. It is assumed that each batch has a 2a and all batches which have 2a split
into individual components after step 2. The new takt time from station 3 is 450/(3 + 3 +
3 + 12) = 21.42 min. The real time can be longer since more components might be
batched together throughout the whole stage.
4 Base models
stable output as it was expected. Given that the production facility did not exist yet, a
review by expert team members was conducted. Using a walk-through of modelling
assumptions and preliminary results, the team was able to provide a validation of the
modelling approach.
Table 2 LimWIP base throughput (12/3 throughput)
Base limited WIP model Product 1 Product 2 Total throughput Planned throughput Utilisation
Throughput 2.65 10.83 13.48 15 0.9
5 Throughput analysis
Permutation 1 Permutation 2
ac1 utilisation 0.909 0.769
a2 utilisation 0.909 0.769
b1 utilisation 0.909 0.769
b2 utilisation 0.909 0.769
Common stage utilisation 0.909 0.769
208 J. Duanmu and K. Taaffe
Alternatively, for the second permutation, two A’s are separated by one B, again with
each B going to a unique downstream station. Now the cycle loss can be quantified as:
d = max ( 0, t2 A − 2t1, t2B − 5t1 )
Based on the above equations, it can be found that wait times equal to 11.25 min for
Sequence 1 and 3.75 min for Sequence 2. Quantifying utilisation as CT/(CT + d), each
wait time (or loss) then translates into a utilisation of 0.769 and 0.909, respectively,
which is exactly the same result achieved in Table 5. Such analysis can be performed
when the total number of components is not large. Otherwise, the sheer number of
permutations becomes overwhelming, and heuristic approaches to identify a preferred
sequence are necessary.
{ } { }
S1 = {a1 , !, aα } , S2 = b1 , !, bβ , S3 = c1 , !, cγ , !, S n = {m1 , ! , mω }
For example, there are α components in set S1, β components in set S2, etc. Each
component in set Si has a throughput of Ti. In order to reach a ‘0’ waste time at the stage
interfaces, the following mathematic relationship should be obeyed:
1 For the throughput relationship of sets S1 and S2:
T2 = S1 × T1 = α T1
2 For any set beyond S2, the following throughput rule should be applied:
Ti = ( Si −1 + Ti −2 ) × Ti −1 , i = 3,…, n
Using the firm’s 3/12 production plan, the throughput of each component group after
Common stage is:
1ac = 6, 1b = 3, 2a = 12, 2b = 12
For simpler assembly lines, an analytical calculation might help engineers directly reach
the desired throughput result. However, more complex systems involving batching,
splitting and re-entrant flow may be better suited for analysis, using a simulation model
along with production system knowledge based on Little’s Law. In this section a brief
summary about factors that might affect the throughput and utilisation is shown.
6.1 Overview
According to Little’s Law (see Equation (1)), if we want to increase throughput, we can
either increase WIP or shorten the cycle time. However, the relationship between WIP
and cycle time (CT) is non-linear since they have a complex interaction with throughput;
the change is contingent upon how WIP and cycle time interact, and this change and
interaction can be checked by simulation.
In setting up the alternatives, WIP and CT are manipulated by either adding buffer
areas or reducing station PTs. The motivation here was to identify how much PT
reduction is associated with a particular throughput level. Then, the company can
determine whether the reduction is possible.
6.2 Scenarios
In this section, several scenarios are defined based on a relatively wide range of settings
for the buffer and PT parameters. Then OptQuest is used to perform simulation-
optimisation to find the maximum throughput. The following is a description of the five
scenarios that were examined.
S1: Add buffers: By adding buffers, components are allowed to move downstream to a
buffer when otherwise they would have been held at an upstream station. This would also
allow one line to accumulate components even when the other line does not have a
matching component that is ready. This may result in smooth flow or possible higher
throughput, but it is opposite to the rules proposed in lean production. In Arena, the range
of buffer is defined from 0 to 8 units.
S2: Reduce PTs through Common stage: From Little’s Law, it is expected that shortening
the cycle time would improve throughput. Two possible ways to reduce cycle time are to
1 decrease the PT for the workstation and
2 lower the WIP.
The authors first consider only PT reductions before and including the shared resource
(i.e. the Common stage). The downstream station PTs are untouched and are still the
same as shown in the takt times. Arena (and OptQuest) is programmed to select between
50% and 100% of the current takt time values for a cycle time reduction. Specifically, a
lower bound PT of 50% or 90% of the original PT is enforced two separate simulation
optimisation tests.
S3: Reduce PTs after Common stage: This is an effort to compare the effect of obtaining
PT reductions before and after Common stage. The allowable processing reduction is the
same as in S2.
210 J. Duanmu and K. Taaffe
S4: Combine WIP and PT factors before and including Common stage: In this scenario,
OptQuest can determine the best combination of buffer size and PT before and including
the Common stage to maximise throughput. Buffer size is also allowed to increase (but
not PT reductions) after the Common stage.
S5: Simulation-optimisation of all parameters: Here, all input parameter options are
allowed to be set within the simulation-optimisation model. The authors want to see
whether a 100% throughput (based on the amount of orders being sent through the
system) could be obtained. The ranges for buffer and PTs are the same as in earlier
scenarios.
merging and Common stage are accelerated a little, they can easily act off the idle time
on the diverging stages. However, if we keep any two stages including the Common stage
unchanged, the interface delay will still exist no matter how fast the other stations
function. Given that the production loss occurs in the stations prior to and including
Common stage, this scenario provides little improvement to overall throughput.
Table 6 Scenario results for 3/12 throughput
S4: Combine WIP and PT factors before and including Common stage: There is no clear
difference when compared against Scenario 2. The buffer helps smooth the production.
However, the buffer actually causes a longer cycle time, which works against throughput.
In takt time design, the buffer is not as important a factor in throughput control. This
scenario also shows the same recommendation as in Scenario 2 – a slightly reduced PT
works just as well as a much smaller PT.
Production capabilities using takt times, requirements analysis and simulation 213
Figure 2 Throughput comparison for all configurations based on 3/12 desired production
(see online version for colours)
214 J. Duanmu and K. Taaffe
Figure 3 Throughput comparison for all configurations based on 12/12 desired production
(see online version for colours)
7 Conclusions
Through the significant testing and analysis conducted on this problem, the contributions
of both a mathematical and simulation approach is presented to improve production
throughput in an assembly line system. In adhering to a design that is driven by takt
times, engineers must be aware of the throughput loss between stages when each stage
cannot provide a seamless handoff to the subsequent stage. While this may seem to be a
small issue on the surface (simply reduce all PTs), it is critical to be able to measure the
required PT reductions, as well as to understand the extent by which throughput is
reduced due to the delay-or-wait nature of the stages.
Using simulation, engineers can find that a theoretical takt time plan cannot be
fulfilled under a complex assembly and manufacturing process. Even if the utilisation can
be calculated by hand, such systems ultimately require a combination of simulation and
takt time analysis to achieve maximum throughput and utilisation. Moreover, for systems
with many components and many stages, using a mathematical approach to component
sequencing becomes extremely difficult.
Across all production configurations tested in the case study, a 10–15% increase in
throughput is achieved over the firm’s original design. The batching and re-entrant loop
configurations (HBNC, HBHC) provide additional throughput complications, and that a
batching configuration seems to work well when both products are produced in more
balanced quantities (i.e. the 3/12 production is not as efficient as the 12/12 production for
the batching configurations). Additional areas for future work include the consideration
Production capabilities using takt times, requirements analysis and simulation 215
of additional product types being produced on the same line, and the comparison of
throughput (and machine/labour requirements) when the component production is more
(less) dedicated.
References
Adams, M., Componation, P., Czarnecki, H. and Schroer, B.J. (1999) ‘Simulation as a tool for
continuous process improvement’, in P.A. Farrington, H.B. Nembhard, D.T. Sturrock and
G.W. Evans (Eds). In Proceedings of Winter Simulation Conference, pp.766–773, The
Institute of Electrical and Electronics Engineers, Piscataway, NJ.
AlDurgham, M.M. and Barghash, M.A. (2008) ‘A generalized framework for simulation-based
decision support for manufacturing’, Production Planning and Control: The Management of
Operations, 1366-5876, Vol. 19, pp.518–534.
Ali, S.A., Seifoddini, H. and Sun, H. (2005) ‘Intelligent modeling and simulation of flexible
assembly system’, in M.E. Kuhl, N.M. Steiger, F.B. Armstrong and J.A. Joines (Eds).
Proceedings of Winter Simulation Conference, 2005, pp.1350–1358, Orlando, Florida.
Balakrishnan, A. and Vanderbeck, F. (1999) ‘A tactical planning model for mixed-model
electronics assembly operations’, Operations Research, Vol. 47, No. 3, pp.395–409.
Czarnecki, H. and Loyd, N. (2001) ‘Simulation of lean assembly line for high volume
manufacturing’, In Proceedings of the Huntsville Simulation Conference, 2001, Available at:
http://www.scs.org/confernc/hsc/hsc02/hsc/papers/hsc037.pdf, Accessed on 2 Feburary 2009.
Farr, R., Tannock, J. and Almgren, T. (2010) ‘A parametric approach to logistic control within
manufacturing simulation’, Int. J. Industrial and Systems Engineering, Vol. 5, No. 3,
pp.313–326.
Gershwin, S.B. (2000) ‘Design and operation of manufacturing systems: the control-point policy’,
IIE Transactions, Vol. 32, No. 100, pp.891–906.
Guinery, J., Wang, K. and Maccarthy, B. (2010) ‘Development and evaluation of a production
planning, scheduling and control re-design Toolkit’, Int. J. Industrial and Systems
Engineering, Vol. 5, No. 3, pp.251–267.
Heike, G., Ramulu, M., Sorenson, E., Shanahan, P. and Moinzadeh, K. (2001) ‘Mixed model
assembly alternatives for low-volume manufacturing: the case of the aerospace industry’,
Int. J. Production Economics, Vol. 72, No. 2, pp.103–120.
Hopp, W.J. and Spearman, M.L. (1999) Factory Physics (2nd ed). Boston, MA: McGraw-Hill.
Hung, K.T. and Liker, J.K. (2007) ‘A simulation study of pull system responsiveness considering
production condition influences’, Int. J. Industrial and Systems Engineering, Vol. 2, No. 2,
pp.123–136.
Lee, C. and Vairaktarakis, G. (1997) ‘Workforce planning in mixed model assembly systems’,
Operations Research, Vol. 45, No. 4, pp.553–567.
Lian, Y. and Van Landeghem, H. (2002) ‘An application of simulation and value stream mapping
in lean manufacturing’, In Proceedings of 14th European Simulation Symposium, 2002,
Dresden, Germany, Available at: http://www.scs-europe.net/services/ess2002/PDF/log-11.pdf,
Accessed 20 March 2007.
Miltenburg, J. (2007) ‘Level schedules for mixed-model JIT production lines: characteristics of the
largest instances that can be solved optimally’, Int. J. Production Research, Vol. 45, No. 16,
pp.3555–3577.
Muthiah, K.M.N. and Huang, S.H. (2006) ‘A review of literature on manufacturing systems
productivity measurement and improvement’, Int. J. Industrial and Systems Engineering,
Vol. 1, No. 4, pp.461–484.
216 J. Duanmu and K. Taaffe
Park, Y.H., Matson, J.E. and Miller, D.M. (1998) ‘Simulation and analysis of the Mercedes-Benz
all activity vehicle (AAV) production facility’, in D.J. Medeiros, E.F. Watson, J.S. Carson and
M.S. Manivannan (Eds). In proceedings of Winter Simulation Conference, 1998, Vol. 2,
pp.915–920, SCS, San Diego.
Sandanayake, Y.G. and Oduoza, C.F. (2009) ‘Dynamic simulation for performance optimization in
just-in-time-enabled manufacturing processes’, Int. J. Advanced Manufacturing Technology,
Vol. 42, Nos. 3–4, pp.372–380.
Schroer, B.J. (2004) ‘Simulation as a tool in understanding the concepts of lean manufacturing’,
Simulation, Vol. 80, No. 3, pp.171–175.
Shah, R. and Ward, P.T. (2007) ‘Defining and developing measures of lean production’, Journal of
Operations Management, Vol. 25, pp.785–805.
Womack, J.P. and Jones, D.T. (2003) Lean Thinking: Banish Waste and Create Wealth in Your
Corporation (2nd ed.). New York: Simon and Schuster.
Womack, J.P., Jones, D.T. and Roos, D. (1990) The Machine That Changed the World, New York:
Harper Perennial.