International Journal of Production Research

ISSN: 0020-7543 (Print) 1366-588X (Online) Journal homepage: http://www.tandfonline.com/loi/tprs20

Value creation through design for scalability of

reconfigurable manufacturing systems

Yoram Koren, Wencai Wang & Xi Gu

To cite this article: Yoram Koren, Wencai Wang & Xi Gu (2016): Value creation through design
for scalability of reconfigurable manufacturing systems, International Journal of Production
Research, DOI: 10.1080/00207543.2016.1145821

To link to this article: http://dx.doi.org/10.1080/00207543.2016.1145821

Published online: 15 Feb 2016.

Submit your article to this journal

Article views: 10

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=tprs20

Download by: [University of California, San Diego] Date: 19 February 2016, At: 08:17
 International Journal of Production Research, 2016
http://dx.doi.org/10.1080/00207543.2016.1145821

Value creation through design for scalability of reconfigurable manufacturing systems

Yoram Koren*, Wencai Wang and Xi Gu

Department of Mechanical Engineering, University of Michigan, Ann Arbor, MI, USA

(Received 11 November 2015; accepted 18 January 2016)

Rapid and cost-effective scalability of the throughput of manufacturing systems is an invaluable feature for the
management of manufacturing enterprises. System design for scalability allows the enterprise to build a manufacturing
system to supply the current demand, and upgrade its throughput in the future, in a cost-effective manner, to meet possi-
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

ble higher market demand in a timely manner. To possess this capability, the manufacturing system must be designed at
the outset for future expansions in its throughput to enable growths in supply exactly when needed by the market. A
mathematical method that maximises the system throughput after reconfiguration is proposed, and an industrial case is
presented to validate the method. The paper offers a set of principles for system design for scalability to guide designers
of modern manufacturing systems.
Keywords: capacity scalability; reconfigurable manufacturing systems; managerial insights

1. Introduction
The surge of globalisation in the late 1990s created a fierce competition that is causing abrupt variations in product
demand, which makes it harder for manufacturing enterprises to predict the future demand for new products (Koren
2010). Prior to the mid-1990s, high-volume manufacturers, such as automakers, enjoyed stable markets with long pro-
duct lifetimes, in which products were manufactured using fixed transfer lines (Dolgui et al. 2005). By contrast, manu-
facturing companies in the twenty-first century are facing increasingly frequent and unpredictable market changes,
including rapid introduction of new products, and frequently varying product demand.
Usually, manufacturing systems are designed with a specific capacity to fulfil a forecasted demand (Tang et al.
2004b). The designers of manufacturing systems face a tough dilemma regarding the capacity of new manufacturing
systems: If the new system is designed to produce a smaller throughput than the market will require in the future, a
tremendous financial loss in losing market share will take place. And if the new manufacturing system is designed to
produce a larger throughput than the actual market will need in the future, then the system will be partially idle, which
means a considerable loss in capital investment – purchasing, installing and maintaining machines that are not operating.
In some cases, even if there are periods in which the system is operated at full capacity, these periods are short com-
pared to the system’s entire life cycle (DeGarmo, Black, and Kohser 1998). Therefore, if the investment on the excess
capacity could be delayed until the capacity is actually needed to supply a steady real demand, the system lifetime cost
can be significantly reduced.
The solution to this dilemma is designing modern manufacturing systems that possess a new characteristic that
enables future upgrading of their capacity (i.e. maximum annual production, which is proportional to production rate).
This is the characteristic of scalability, which enables the option of future incremental throughput upgrading in a cost-ef-
fective and timely manner (Koren and Ulsoy 1997; Koren, Hu, and Weber 1998). The capacity of manufacturing sys-
tems that possess the characteristic of scalability could be increased (1) rapidly, exactly when the market needs more
products, and (2) incrementally, exactly by the additional capacity required by the market (Koren and Ulsoy 2002a).
The option of incremental-capacity additions enables cost-effective expansion of the entire system capacity, which allows
production of more products without increasing the product cost. Strategic incremental-capacity planning improves life-
time system economics (Bard, Srinivasan, and Tirupati 1999; Koren 2005; Zhang, Fu, and Zhu 2008; Geng and Jiang
2009). Thus, system scalability is an imperative characteristic to enable cost-effective competition, thereby creating a
vital business value for manufacturing enterprises.

*Corresponding author. Email: [email protected]

© 2016 Taylor & Francis

 2 Y. Koren et al.

If the enterprise management is convinced at the outset of the design process of a new production system that its
throughput upgrading will be possible in a cost-effective and timely manner in the future, it facilitates the managerial
decision of building new production plants and new manufacturing systems. Designing new manufacturing systems with
a potential of volume scalability is imperative in contributing to the profitability of manufacturing enterprises, and satis-
fying in a timely manner the society needs for new products. Scalability is therefore a significant system characteristic
for both the enterprise management and society, and consequently, it derives intensified attention in recent years (Putnik
et al. 2013).
Reconfigurable manufacturing systems (RMS) possess six characteristics (Koren et al. 1999), of which scalability is
perhaps the most important characteristic. One method for designing manufacturing systems for future scalability is opti-
mising the system configurations. Numerous models and methods have been developed to optimise the system configu-
ration. For example, Zhang, Wang, and Luo (2000a) developed a probabilistic model to represent an RMS, and solved
the optimal configuration problem (Zhang, Wang, and Luo 2000b). Abdi and Labib (2003) utilised an Analytical Hierar-
chical Process to select a manufacturing system among feasible alternatives, and to select products and group them into
families (Abdi and Labib 2004a). Meng, Heragu, and Zijm (2004) investigated the optimal reconfigurable layout prob-
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

lem based on open queuing network model. Youssef and ElMaraghy (2006) utilised a Genetic Algorithm (GA) to opti-
mise the arrangement of machines, the selection of equipment and assignment of operations. Dou, Dai, and Meng
(2010) also used a GA method to optimise the multi-part flow-line configuration of an RMS. Goyal, Jain, and Jain
(2012) used Shannon entropy to rank different configurations with the assignment of reconfigurable machine tools.
Design for future capacity expansions requires mathematical methods. Deif and ElMaraghy (2006) addressed capacity
scalability in RMS based on a control approach. Our approach is based on minimising the total number of machines and
maximising system throughput by concurrently reconfiguring and rebalancing the system to match new market demand.
Wang and Koren (2012) offered such mathematical analysis for systems that did not have buffers. In the current
paper, we extend the analysis to systems with buffers. In addition, we propose a set of principles for system design-for-
scalability that maximises the economic value of manufacturing systems.
Section 2 of this paper describes the system architecture that enables easy design for incremental scalability. Sec-
tion 3 defines the system scalability and formulates several principles for achieving incremental scalability. Section 4
presents the mathematical formulations needed to minimise the total number of machines and maximise system through-
put by concurrently reconfiguring and rebalancing the system. Our method is based on GA. To validate our proposed
approach, Section 5 presents a case study of system design for scalability, which was performed for the automotive
industry. Section 6 concludes the paper.

2. System architecture for incremental capacity scalability

The cost and practicality of capacity expansion depends on the architecture of the production system. Dedicated lines do
not have scalable capacity and cannot cope with large fluctuations in product demand (Dolgui, Guschinsky, and Levin
2009). Therefore, the production systems of high-volume mechanical parts (such as engine blocks) have undergone a
huge transformation over the past two decades (Brooke 1998; Krygier 2005). Serial transfer lines with dedicated machi-
nes have been intensively replaced with systems composed of flexible CNC machines. One of the world’s first algo-
rithms that addressed capacity scalability was developed by Son, Olsen, and Yip-Hoi (2001) for upgrading the capacity
of serial lines composed of CNC machines. However, in recent years, industry is utilising complex production systems
that are designed with reconfigurable manufacturing system architecture.
RMS, originally introduced by Koren and Ulsoy (1997), have a configuration architecture that can be easily scaled
up in order to rapidly react to changing market demand (Koren and Ulsoy 2002b). An early approach to scalability was
presented by Spicer, Koren, and Shpitalni (2002), where scalability was analysed as one of the critical issues in design-
ing large, complex machining systems. More recently, it was shown that the RMS architecture enables system design
for scalability in incremental capacity steps that precisely follow market changes, which is the most economical way to
satisfy demand (Wang and Koren 2013).
In this paper, we consider the design optimisation of the system layout such that adding machines will be done
quickly and cost effectively. Koren and Shpitalni (2010) introduced a dynamic model for capacity scalability, which
helps to minimise the delay in scaling the system capacity, thereby improving the system response to sudden demand
changes. In practice, however, the delay in scaling up the system capacity depends on the type of the material handling
devices. Each time that machines are added, the material handling system must be adapted to serve the new system lay-
out. There are systems whose material transport system is formed by automated guided vehicles (AGVs), which facili-
tate the part transfer to and from the new machines. However, AGVs are expensive and slow, and consequently do not
fit high-capacity manufacturing systems. Instead, gantries and conveyors are usually utilised.
 International Journal of Production Research 3

A typical, modern RMS architecture is illustrated in Figure 1. This architecture is composed of many stages (in the
auto industry – 6 to 25 stages) of multiple parallel CNC machines at each stage, with all machines at a stage performing
exactly the same sequence of machining tasks. Each stage has a gantry that loads parts and unloads the machined parts,
and moves them to the system part-transfer mechanism (may be a gantry or a conveyor, as shown in Figure 1). Small
buffers (5–10 parts) exist between stages. A part that is processed on a machine in one stage may be transferred to any
machine in the next stage. Some researchers refer to this feature as a system with crossovers (Freiheit, Shpitalni, and
Hu 2004a).
As can be seen in Figure 1, the original system has empty spaces that are reserved for future added machines
(marked in Figure 1 ‘space for scalability’). This enables adding machines rapidly and very easily to enhance the system
throughput, if and when the market will demand more products. This is the scalability feature of RMS. An AI-based
algorithm may assist the designer at the reconfiguration planning stage to allocate optimal spaces for future increase in
the system capacity (Tang et al. 2004a).
One of the most critical issues in operating efficiently any manufacturing system is the balancing of its operations
(Battaïa and Dolgui 2013). System balancing means that each station (e.g. machine, assembly post, etc.) along a produc-
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

tion line has almost the same cycle time, so that the idle time when some machines complete their tasks quicker than
other machines in the system, is minimised. System balancing enables synchronisation of the operations and is essential
in obtaining high throughput in both machining and assembly systems (Sotskov, Dolgui, and Portmann 2006). Sophisti-
cated mathematical methods have been developed for balancing transfer lines (Dolgui, Guschinsky, and Levin 2006,
2009). System balancing is usually accomplished by shifting operations among machines in the system, but it may be
also achieved by scaling the capacity of individual pieces of equipment (Spicer, Yip-Hoi, and Koren 2005; Lorenzer
et al. 2007; Youssef and ElMaraghy 2008). The most practical approach to scale up the capacity of reconfigurable sys-
tem is adding CNC machines to existing manufacturing systems. However, achieving cost-effective scalability depends
on the original system design layout (Freiheit, Wang, and Spicer 2007).
Note that system configurations and system flexibility are two distinct characteristics. The system configuration
depends on (1) the arrangement of the machines in the system, and (2) the way that they are connected. The system
flexibility depends on (1) the type of machines (e.g. CNC, dedicated) and (2) the method that they are connected (e.g. a
rigid synchronised mechanism, a gantry). For example, assume that to supply a market demand, there is a need for 18
machines. These machines can be arranged as a serial line of 18 machines, two parallel lines of 9 machines, or a system
composed of three lines, each of 6 machines. These are three distinct system configurations (Lorenzer et al. 2007). If
these machines are CNC machines and they are connected via gantries, then the system is flexible (Maier-Speredelozzi,
Koren, and Hu 2003). If the machines are dedicated to perform a unique set of operations, the system is not flexible
and is usually built as a transfer line. A modern RMS is usually designed to produce a family of products, so that both
system flexibility and system configuration are being considered at the system design stage (Krygier 2005).

Figure 1. The layout of a reconfigurable manufacturing system.

 4 Y. Koren et al.

3. Definition and principles of system scalability

System scalability can be easily accommodated in multi-stage RMS (Figure 1), which are composed of CNC machines
or reconfigurable machines (Koren and Kota 1999; Landers, Min, and Koren 2001). Note that after reconfigurations, the
number of stages in the system always remains unchanged, and scalability is obtained by adding machines to the
system.

3.1 Definition
System scalability, in percentage, is defined as (Wang and Koren 2012):
Scalability = 100 × (1 − Smallest Incremental Capacity/Existing Capacity)%
By this definition, a system that can be adjusted to meet a new market demand by adding a small incremental capac-
ity is highly scalable.
The scalability of a serial line is 0% – if a serial line production capacity needs to be increased to satisfy a larger
market demand, an entire new line must be added, which doubles the production capacity of the system. Therefore, the
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

minimum increment of adding production capacity in a serial line is adding a whole new line (i.e. 100% of the system),
making the scalability of a serial line 0%. Doubling the line capacity (when double capacity is not really needed) will
be enormously expensive because there is no guarantee that the extra capacity will ever be fully utilised. A more cost-
effective solution for scaling-up capacity of serial lines may be designing them with small reserved capacity (Freiheit,
Shpitalni, and Hu 2004b), but this design has not been acceptable by industry. To conclude, when markets are volatile,
designing a manufacturing system with zero scalability is a bad economic decision, and the enterprise management
should not allow it.
Figure 2 shows five 8-machine manufacturing systems with different configurations, and their scalabilities are sum-
marised in Table 1 based on our scalability definition.
For the five configurations in Figure 2, both configurations (d) and (e) have scalability of 87.5% − the highest possi-
ble for 8-machine configurations. A minimum increment of 12.5% − in these cases, one machine − can be added to
increase the system capacity in these two configurations; for example, a machine can be added to stage 2 of configura-
tion (d), as shown in Figure 2.
It is worthwhile to point out that a trade-off between system scalability and investment cost is considered in
selecting the optimal configuration. To demonstrate this point, we compare the configurations depicted in Figure 2(a)
and (e). The system in Figure 2(a) has four identical two-stage serial lines that are arranged in parallel. In this case, if a
product requires, for example, machining on both the upper and side surfaces, the four machines in the first stage might
be three-axis vertical milling machines to cut the upper surface, and the four machines in the second stage might be
three-axis horizontal milling machines to operate on the side surface. Conversely, in a parallel system, all eight machines
in Figure 2(e) must be five-axis milling machines – making the system far more expensive, and not economical. In the
system in Figure 2(a), capacity scalability must be performed in steps of 25% by adding one vertical machine and one
horizontal machine, rather than in steps of 12.5% as with the parallel configuration. Adding a step of 12.5% in Fig-
ure 2(e) means in practice adding one five-axis machine with a large tool magazine that contains every tool needed for
the whole part processing – an expensive addition.

(d)
(a)

(b)

(c) (e)

Figure 2. Five 8-machine configurations: only configurations (d) and (e) enable incremental scalability of adding one machine.
 International Journal of Production Research 5

Table 1. Scalability values of the configurations in Figure 2.

Configuration (a) (b) (c) (d) (e)

Scalability 75% 50% 0% 87.5% 87.5%

In general, the smallest scalability adjustment steps can be accomplished when the original system is purely parallel
(e.g. Figure 2(e)). However, the initial cost of a parallel system is the highest of all system configurations. In parallel
configurations, each machine must perform all the manufacturing tasks needed to complete the part. Therefore, each
machine must have the entire set of tools needed to produce the whole part, and should also be able to perform more
functions, for which more axes of motion are needed. As a result, the capital cost per additional volume increment
added to a parallel configuration is the highest of all configurations. Therefore, purely parallel systems are rarely found
in industrial plants.
In conclusion, comparing these five configurations from economic scalability viewpoint, the configuration in
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

Figure 2(d) is most likely to be the configuration of choice in this case – it has a reasonable economic compromise
between investment cost and future scalability cost. The following example clarifies the method of adding a small incre-
mental capacity.

3.2 Example
On a system composed of eight machines, as shown in Figure 3, we have to process a part that requires 36 machining
tasks of 50 s each, totalling 1800 s (or 30 min) that are needed to machine each part.
(a) Design a scalable system configuration when the required demand is 128 parts per 8-h shift.
(b) After 1 year, the demand has grown, and 144 parts per shift are needed. How many machines should be added,
and what is the recommended new configuration?
Solution:
(a) In 8 h, there are 480 min. Therefore, the required cycle time is 480/128 = 3.75 min/part. The cost-effective scal-
able system configuration is depicted in Figure 2(d), and is shown in detail in Figure 3(a). Here, each machine
does nine tasks of 50 s, totalling 450s per machine. The cycle time is 225s/part = 3.75 min/part.
(b) When the demand grows from 128 parts/shift to 144 parts/shift, the capacity should be increased by 144/128–
1 = 0.125. One additional machine is needed to add 1/8 capacity. For all the five configurations in Figure 2,
only configurations (d) and (e) can achieve the throughput requirement by adding one machine, but only config-
uration (d) yields a cost-effective solution. Assume that the new machine is added to stage 2 of configuration
(d), as shown in Figure 3(b). Line balancing is performed as follows: The last task that was performed on stage
1 is shifted forward from stage 1 to stage 2, the first two tasks that were performed on stage 3 are shifted back-
wards from stage 3 to stage 2, and the first task that was performed on stage 4 is shifted backwards from stage
4 to stage 3. In total, three tasks are added to stage 2, and one task is removed from each of the other three
stages. Therefore, each machine in stage 2 will now operate for 600 s on each part, and each machine in the
other three stages operates for 400 s on each part. The system cycle time becomes 200 s per part, and the sys-
tem throughput becomes 144 parts/shift.

600

450 450 450 450 400 600 400 400

450 450 450 450 400 600 400 400

(a) 9 tasks per machine; (b) 8 tasks per machine in stages 1, 3, and 4;
system cycle time: 225 s/ part 12 tasks per machine in stage 2;
system cycle time:200 s/ part

Figure 3. When demand grows, the initial system (a) is cost-effectively scaled-up to configuration (b) to meet the new demand.
 6 Y. Koren et al.

The initial capital investment in the system configuration in Figure 3 is a bit higher than that in two serial lines of
4-machine each, because the material handling system is more complex, and we reserve a spot in stage 2 for the
option of adding a ninth machine if needed. However, one may consider the extra capital investment similar to buying
an insurance premium for a future event that is likely to occur. If the demand does rise, the system can be easily
scaled up and the new demand can be supplied in a short time, at a minimum additional investment. If the demand is
unchanged during the lifetime of the system, a small capital investment on the more sophisticated material handling
system is lost.
Ideal scalability results are obtained by adding r machines to a configuration that has L stages and N machines per
stage, if the number of equal tasks needed to complete the part is k(LN + r)L (k = 1,2,….). In the above example, 36
equal tasks were needed to complete the part, and a system with four stages and two machines per stage was precisely
balanced and perfectly scalable. Note that in this example r = 1 and k = 1, and the specific task time (50 s) does not
affect the solution. This example may be generalised by articulating a set of design-for-scalability principles.

3.3 Design-for-scalability principles

Downloaded by [University of California, San Diego] at 08:17 19 February 2016

We formulate the following five principles for the optimal design of scalable manufacturing systems:
(1) The architecture of manufacturing systems must be reconfigurable to enable the integration of additional produc-
tion resources, when needed.
(2) The RMS capacity must be designed at the outset to be scalable in optimal increments.
(3) To be rapidly scalable, the RMS requires additional investment in its initial infrastructure (e.g. space for adding
machines, extended gantries.)
(4) To be economically scalable, the RMS contains a mix of flexible and reconfigurable equipment.
(5) To minimise the number of additional machines needed to scale-up the system capacity, reconfiguration planning
and system rebalancing must be performed simultaneously.
Principles 1 and 2 guarantee that the production system will supply any future market demand, exactly when
needed, resulting thereby a significant business value for the manufacturing enterprises. The initial capital investment is
smaller, although small extra investment is required according to Principle 3. Principles 4 and 5 guide the system design
engineers.
The above example presented an ideal case for demonstrating the concept of scalability and its principles. To imple-
ment this concept in practice, however, various constraints must be taken into consideration. The major practical con-
straints are task precedence, unequal task times and machine characteristics to perform designated tasks. These issues
are discussed in the following sections.

4. Formulation of scalability planning

We propose below a practical method to determine the most cost-effective scalability planning of RMS to meet a new
market demand. In the planning, many factors need to be taken into consideration, including a detailed process plan,
set-up plan, machine capability and the number of spots in the original system that are reserved for adding machines in
the future. Note that simultaneous reconfiguration planning and system rebalancing are required to maximise the system
throughput and minimise the number of machines to be added. In this section, we introduce an optimisation model for
scalability planning.

4.1 Assumptions
The following assumptions are made based on the current manufacturing practice in the powertrain industry.
(1) A multi-stage system with RMS configuration (i.e. the configuration similar to that shown in Figure 1 or
Figure 3) is considered. Parts are moved from one stage to another through conveyors (or gantries), and deliv-
ered to different machines within a stage using stage-gantries.
(2) The number of stages must remain unchanged during any reconfiguration process. This keeps the system set-up
plan unchanged in order to avoid major adjustments of process plans, and thus it minimises the impact of sys-
tem reconfiguration on product quality.
(3) All the machines within the same stage are identical, and perform exactly the same sequence of tasks.
 International Journal of Production Research 7

(4) There are reserved spots for adding new machines in the stages, and material handlers can be extended to deli-
ver parts to the newly added machines.

4.2 Inputs
Scalability planning requires four types of inputs:

4.2.1 Configuration information and machine reliability information

We use a set M to store all the system configuration information and machine reliability information. The system config-
uration can be characterised by the following inputs: the number of stages: L; the number of machines in stage i: Ni
(i = 1, 2, ..., L); the maximum number of machines allowed in stage i: Mi (i = 1, 2, ..., L); and the capacity of the buffer
between stages i and i + 1, Bi (i = 1, 2, ..., L – 1). The machine reliability information consists of two parameters: MTBF
(Mean Time Between Failures), and MTTR (Mean Time To Repair).
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

4.2.2 Stage characteristics

Each manufacturing stage has capabilities that are defined by a set of characteristics of the stage. These include machine
tool characteristics such as functionality, power, accuracy and machining ranges, as well as fixture features such as face
accessibility, which defines the faces that are accessible by the cutting tool. When a set of tasks is assigned to a stage,
the necessary capabilities must fall into the set of characteristics of the stage. Assuming that the number of all-possible
total characteristics is C, then a capability matrix S 2 RLC can be constructed to store all possible characteristics of
each stage. The elements in matrix S are shown in Equation (1).


1; if stage i has characteristic j
Sij ¼ ð1  i  L; 1  j  C Þ (1)
0, otherwise

4.2.3 Manufacturing tasks

There are three types of inputs that are related to the manufacturing tasks.
Task characteristics: These include task type, access direction, dimension, accuracy and power needed to perform
the task. For a task to be assigned to a stage, its characteristics must be included in the set of characteristics of that
stage. A task characteristic matrix K 2 RN C is used to store the characteristics of all tasks.


1; if task i has characteristic j
Kij ¼ ð1  i  N ; 1  j  C Þ (2)
0, otherwise
Task precedence: manufacturing tasks must be performed at a certain order. Each task can only be performed after all
its parent tasks have been completed. A two-dimension binary matrix P 2 RN N , where N is the number of tasks to be
processed, is used to represent the precedence tree, as shown in Equation (3):


1; if task i must be performed before task j
Pij ¼ ð1  i; j  N Þ (3)
0, otherwise
Task time: the time for operating task i is defined by ti ði ¼ 1; :::; N Þ. We use an array t 2 RN 1 to store all the operation
time for all tasks.

4.2.4 Throughput requirement

The system must be reconfigured so its new capacity will fulfil the new throughput requirement, THnew, which can be
calculated from the new system demand.

4.3 Decision variables

Two decision variables need to be determined:
 8 Y. Koren et al.

(1) The number of machines being added to stage i: Δi (i = 1, ..., L). The machine allocation array determines how
many machines are to be added to the system and where to add them. Δi > 0 and Δi < 0 represent adding machi-
nes to stage i and removing machines from stage i, respectively. Note that the machine allocation array will
change the configuration matrix M.
(2) The binary task allocation variable Tij, as

1; if task i is assigned to stage j
Tij ¼ ð1  i  N ; 1  j  LÞ (4)
0, otherwise
The task allocation matrix describes how the tasks should be reallocated when the new machines are added to or
removed from the system.
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

4.4 Constraints
The following constraints are considered in the optimisation problem:

4.4.1 Task assignment constraints

Each task must be assigned to one and only one stage
X
L
Tnl ¼ 1 ð8n ¼ 1; . . .; N Þ (5)
l¼1

4.4.2 Precedence constraints

PL
PL where task n is assigned (i.e. l¼1 l  Tnl ) cannot be
If task n must be performed before m (i.e. Pnm = 1), then the stage
on the downstream of the stage where task m is assigned (i.e. l¼1 l  Tml ).
!
XL X L
Pnm  l  Tnl  l  Tml ð8n ¼ 1; . . .; N ; 8m ¼ 1; . . .; N Þ (6)
l¼1 l¼1

4.4.3 Characteristic constraints

If task n has characteristic j (i.e. Knj = 1) and is assigned to stage l (i.e. Tnl = 1), then stage l must have characteristic j
(i.e. Slj = 1).

Knj  Tnl  Slj ð8n ¼ 1; :::; N ; 8j ¼ 1; . . .; C; 8l ¼ 1; . . .; LÞ (7)

4.4.4 Space constraints

The number of machines in each stage is constrained by the maximum number of machines allowed in that stage.
N l þ D l  Ml ð8l ¼ 1; . . .; LÞ (8)

4.4.5 Throughput constraints

The throughput requirement should be satisfied.
THðM; T; tÞ  TH new (9)
 International Journal of Production Research 9

In Constraint (9), the throughput of the system can be calculated based on the system configuration (after machines are
added/removed), the task allocated in every stage, the task time for each task, and the MTBF and MTTR of all machi-
nes in the system. In this paper, the system throughput THðM; T; tÞ will be calculated using an ERC/RMS developed
software PAMS, based on the analytical algorithm in Yang, Wu, and Hu (2000).

4.5 Optimisation problems

The first optimisation problem that we present deals with minimising the number of additional machines needed to meet
a specific market demand. If two solutions have the same number of additional machines, then the one with a larger sys-
tem throughput is better. This can be formulated as follows:

X
L
min Dl þ TH new =THðM; T; tÞ (10)
l¼1
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

subject to: Constraints (5) to (9) P

In the objective function, the first term Ll¼1 Dl is the total number of machines in the system. Based on constraint
Psecond term TH new =THðM; T; tÞ  1, and it is decreasing in THðM; T; tÞ. Therefore, this objective is to min-
(9), the
imise Ll¼1 Dl first; and then to maximise THðM; T; tÞ.
Another optimisation problem is: given the total number of machines that can be added to the system or removed
from the system (denoted as Δtotal ), calculate the maximal system throughput. It also involves the simultaneous analysis
on where the machine should be added or removed, and on how to rebalance the system by task reallocation. Mathemat-
ically, the problem can be formulated as:
max THðM; T; tÞ (11)
PL
subject to: l¼1 Dl  Dtotal and Constraints (5) to (8)

4.6 Genetic Algorithm

An optimisation tool that is based on GA was developed here to solve these optimisation problems so as to determine
the optimal system reconfiguration when market demand changes. John Holland, who was a professor of Electrical
Engineering at the University of Michigan, introduced GA in 1973 (Holland 1973). GA has been extensively utilised to
solve optimisation problems in production engineering (Rekiek et al. 2002; Su and Shiue 2003; Youssef and ElMaraghy
2008; Dou, Dai, and Meng 2010).
The steps of GA are summarised as follows. First, populations of solutions are randomly initiated. These populations
are represented in a string format, and are evaluated based on a fitness function. Then, using some genetic operators,
including mutation, crossover and selection, these populations can evolve into future generations. After a number of
generations, the algorithm converges to the near-optimal solution (Gen and Cheng 2000).
A typical GA requires a genetic representation (usually in a string format) of the solution domain, and a fitness func-
tion that evaluates the performance of candidate solutions. Next, we introduce the string representation and the fitness
function of the optimisation problems.

4.6.1 String representation of task allocation

First, we introduce the string representation of manufacturing tasks designed for task allocation. Each digit of the string
contains two kinds of information: its position is gene, and its value is allele. The length of each string is equal to the
number of manufacturing tasks (genes) that need to be assigned to the system (i.e. N). The digits of the strings represent
the stage index number (allele). For task n, the total number of stages that it can be assigned to can be calculated as
P Q
An ¼ Ll¼1 Slj , so the digit at the nth position of the string can take values among 1, 2, …, An. Moreover, we con-
j:Knj ¼1
struct a task-to-stage (T2S) index table to map a task to the actual stage that it is assigned to: if the digit at the nth posi-
tion of the string is m (i.e. gene is n and allele is m), then without considering the precedence constraint, task n should
be assigned to stage T2S(n, m).
Moreover, the following logic is used to ensure precedence constraints are not violated:
 10 Y. Koren et al.


T 2Sðn; mÞ if T 2Sðn; mÞ  sðkÞ 8k : Pkn ¼ 1
sðnÞ ¼ (12)
T 2Sðn; m þ lmin Þ if T 2Sðn; m þ lÞ  sðkÞ 8k : Pkn ¼ 1; 8l  lmin
where s(n) is the stage number where task n is assigned to. Therefore, if task k is a predecessor of task n, then s(n) ≥ s
(k).

4.6.2 String representation of system reconfiguration

We use another string to represent where the machines should be added to. The length of the string equals the maximum
number of machines to be added. The digits in the string represent the stages where the corresponding machine is added
to. A digit ‘0’ is used to represent the case where the machine is not needed. For example, a string [1 2 2 3 0] indicates
that the first machine is added to stage 1, the second and third machines are added to stage 2, the fourth machine is
added to stage 3 and the fifth machine is not needed.
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

4.6.3 Fitness function

Based on objective function (10), the fitness function can be defined as:
8 L
> P
>
< Dl þ TH new =THðM; T; tÞ THðM; T; tÞ  TH new
f ¼ l¼1
(13)
>
> PL
: Dl þ TH new =THðM; T; tÞ þ Y THðM; T; tÞ\TH new
l¼1

where Y is a penalty cost when the new system demand is not met. In our example, we set Y to be a large number so
that the fitness function is only considered when TH ðM; T; tÞ  TH new .

5. Case study
This section presents a case study that we conducted with our industry partners from automotive powertrain companies
to examine and validate the proposed approach. The case selected is the rough machining process of an automotive V6
cylinder head provided by an industrial partner of the ERC-RMS. There are 141 features on the part, which can be pro-
cessed by 43 machining tasks that are performed on six faces, including milling, drilling, boring, spot-facing and tap-
ping. The total time needed for the rough machining is 1019 s. The precedence relationships of these 43 tasks are
shown in the graph in Figure 4. Because of its complexity, this part is ideal for the study as it permits many process
design solutions for different system configurations. The machines used for all stages are four-axis CNC machining cen-
tres (three perpendicular axes and a rotary table). Each machine has three accessible faces, and is capable of completing
machining tasks on these faces. Figure 5(a) shows the part and Figure 5(b) shows the machine features. MTBF and
MTTR of all the CNC machines are 193 and 16.7 min, respectively.

5.1 Baseline configurations

In this case study, three system configurations, 3 × 6, 6 × 3 and 9 × 2, are used as baselines to study the scalability plan-
ning. Figure 6 gives three system configurations and their throughputs that are associated with the line balancing results.

5.2 Scalability results

Assume that the 3 × 6 configuration that is depicted in Figure 6(a) is currently being used to fulfil a production demand
of 50 JPH (jobs per hour). When the new production demand is increased by 15% (namely to 57.5 JPH), the proposed
scalability planning algorithm calculates that two new machines have to be added to the system, as shown in Table 2.
The rebalancing results per-machine and per-stage, as was calculated by our algorithm, are also shown in Table 2. After
adding the two machines, the system capacity is increased to 59.6 JPH. Note that compared to duplicating a full three-
machine serial line, the new configuration needs only two new machines to entirely fulfil the new production demand.
Table 2 also shows that, contrary to the common sense that the two machines should be added to two different
stages, the two machines are added to the same stage (i.e. stage 3). The reason is that by adding the two machines to
stage 3, task reallocation and line rebalancing enable higher throughput with this configuration. In this case study, the
machining tasks are not evenly distributed at each accessible face in the selected set-up. Machines tend to be added to
 International Journal of Production Research 11

5 8 10 11 12 15

6 9 13 16
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

7 14 17

18 19 20 21 22 23 24 27 31 42 43

25 26 32 33 34 39

28 29 30 35 36 37 38 40

41

Figure 4. Precedence graph of all the tasks.

Figure 5. Part and machine configuration.

the stages with a set-up that allows access to more tasks. Note that in the reconfigured system, although the total task
time on stage 3 is larger than that of stages 1 and 2, the cycle time for all the three stages is almost equal (50.1, 50.9,
51.6 s.) Therefore, the system throughput is maximised, while the number of additional machines is minimised.
Even more remarkable results are obtained when even a larger throughput is required, and to achieve it, three, four
and five machines are added to the configuration depicted in Figure 6(a). Our algorithm has a constraint (dictated by
our industry sponsor) that maximum three machines may be added to any stage while the set-up plan remains
unchanged. Our line-balancing software yields the results shown in Figure 7.
 12 Y. Koren et al.

machine C buffer with capacity C

20 20

8 8 8 8 8

(a) 3×6 configuration, (b) 6×3 configuration,

throughput: 53.58 JPH throughput: 47.27 JPH

5 5 5 5 5 5 5 5

(c) 9×2 configuration,

throughput: 40.42 JPH
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

Figure 6. Three configurations of 18 machines and their throughput in jobs-per-hour.

Table 2. Comparison of the original system and the new system.

Original system New system

Required throughput (JPH) 50 57.5

Stage # 1 2 3 1 2 3
Number of machines 6 6 6 6 6 8
Number of tasks 14 16 13 12 14 17
Total task time (s) 330.42 340.34 348.38 300.64 305.71 412.79
Stage cycle time (s) 55.07 56.72 58.06 50.11 50.95 51.60
Actual throughput (JPH) 53.58 59.61

The locations of the additional machines in the system are optimised by our software, which is based on the method
that was explained above. Note that the machines are not evenly added to each stage. If the manufacturer requires sys-
tem scalability of adding a maximum of five machines, the original system architecture must reserve only one spot for a
new machine in stages 1 and 2, and three reserved spots in stage 3. Such a detailed planning of production lines for
future capacity expansions has a significant economic value to the manufacturing cooperation.
In general, we suggest that scalability planning be performed concurrently with the design of the new manufacturing
system. This way, optimal locations for future additional machines can be identified at the outset. Thus, material
handling systems and buffer capacities can be optimised for future expansion, which, in turn, reduces the lifetime invest-
ment cost in the production system.
Another important parameter in system design is the buffer capacity. Table 3 summarises the system productivity of
the three configurations in Figure 6 (i.e. 3 × 6, 6 × 3 and 9 × 2) with different buffer capacities, and the new productiv-
ity when 1–5 machines are added to existing systems.
Moreover, the comparison of the throughputs of systems with different configurations and different buffer capacities
is shown in Figure 8 (a) and (b), respectively. It can be seen from Figure 8(a) that the 3 × 6 configuration (depicted in
Figure 6(a)) gives both the highest original system throughput and the largest throughput gain per new added machine.
This is because when the number of stages in the system increases, both system reliability and system balance tend to
decline. Moreover, Figure 8(b) shows that larger buffer capacities will increase both the original system throughput and
the throughput gain per added machine. Note that the allocation of the buffer capacities in the system can also be opti-
mised (Dolgui et al, 2013; Shi and Gershwin 2014). When the production demand increases for a given existing system,
Table 3 is very conveniently helping to decide how many new machines are needed and at which locations they should
be added.
 International Journal of Production Research 13

20 20 20 20 20 20

(a) baseline, (b) with one machine added, (c) with two machines added,
throughput: 53.58 JPH throughput: 56.48 JPH throughput: 59.61 JPH

20 20 20 20 20 20
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

(d) with three machines added, (e) with four machines added, (e) with five machines added,
throughput: 62.47 JPH throughput: 65.33 JPH throughput: 68.29 JPH

Figure 7. Reconfiguration for scalability planning for the 3 × 6 system.

Table 3. Throughput (in JPH) of systems with different configurations and buffer capacities, when new machines are added.

3×6 6×3 9×2

Configuration
Capacity per Buffer 0 20 40 0 8 16 0 5 10

No. of machines added +0 49.5 53.6 55.0 41.3 47.3 50.1 31.9 37.7 40.4
+1 52.8 56.5 58.0 43.8 49.8 52.7 34.0 40.6 43.6
+2 55.7 59.6 61.2 46.4 52.2 55.3 36.2 42.8 46.1
+3 58.7 62.5 64.1 48.2 54.8 58.0 38.2 44.8 48.1
+4 61.5 65.3 67.0 50.1 57.3 60.6 39.5 46.7 50.1
+5 64.2 68.3 70.1 52.6 59.8 63.2 42.3 49.2 52.6
Avg. throughput gain per 2.94 2.94 3.02 2.26 2.50 2.62 2.08 2.30 2.44
machine added

6. Conclusion and discussion

This paper elaborates on the economic value of scalability in RMS, and develops design-for-scalability principles. A sys-
tematic approach for scalability planning was developed, so that the exact capacity needed to satisfy surging market
demand could be added to the system at a minimum investment cost. The scalability planning process requires simultane-
ous changing of the system configuration and rebalancing the reconfigured system. An optimal scalability-planning
problem, which is subject to realistic constraints, was formulated. The optimisation problem was solved by utilising GA.
The proposed approach was validated through a real industrial case.
This paper urges and demonstrates that scalability planning should be performed concurrently with the design of a
new manufacturing system. With this approach, the material handling system can be optimised for future scalability
actions, reducing thereby the capital investment cost. This design methodology contributes to the rapid responsiveness
of RMS to abrupt market events (Koren 2013). Experimental results have demonstrated that the proposed approach
addresses the scalability-planning problem cost-effectively and efficiently.
For the purpose of simplicity, this paper only considers minimising the total number of machines, or maximising the
system throughput as the optimisation objective. However, in real practice, other cost factors should be taken in consid-
 14 Y. Koren et al.

throughput of different systems throughput of different systems throughput of different systems

with total buffer capacity 0 with total buffer capacity 40 with total buffer capacity 80
80 80 80

60 60 60
throughput (JPH)

throughput (JPH)

throughput (JPH)
40 40 40

20 3 6 20 3 6 20 3 6
6 3 6 3 6 3
9 2 9 2 9 2
0 0 0
0 2 4 0 2 4 0 2 4
no. of machines added no. of machines added no. of machines added

(a) Comparison of systems throughputs under different configurations

Downloaded by [University of California, San Diego] at 08:17 19 February 2016

throughput of 3 6 system throughput of 6 3 system throughput of 9 2 system

with total buffer capacity C with total buffer capacity C with total buffer capacity C
80 80 80

60 60 60
throughput (JPH)

throughput (JPH)

throughput (JPH)
40 40 40

20 C=0 20 C=0 20 C=0

C = 40 C = 40 C = 40
C = 80 C = 80 C = 80
0 0 0
0 2 4 0 2 4 0 2 4
no. of machines added no. of machines added no. of machines added
(b) Comparison of systems throughputs with under buffer capacities

Figure 8. Comparison of system throughputs under different configurations and different buffer capacities.

eration as well. These include labour, tooling, utility, floor space, operating cost and the efficient response to machine
failures that disrupt production. Gu et al. (2015) proved that by utilising the built-in reconfigurability feature, the RMS
could respond rapidly to disruptive events (e.g. machine failures), which makes the RMS the preferred system choice.
The selection of a manufacturing system may also depend on the product life cycle and the frequency of new product
introduction (Niroomand, Kuzgunkaya, and Bulgak 2012), as well as on a variety of practical challenges (Bi et al.
2008; Rösiö and Säfsten 2013), such as the machine functionality (Abdi and Labib 2004b), and the control design (Tang
and Qiu 2004). A suitable index that accounts for the scalability and other system characteristics (Gumasta et al. 2011)
should be developed at the design stage of new manufacturing systems.

Disclosure statement
No potential conflict of interest was reported by the authors.

References

Abdi, M. R., and A. W. Labib. 2003. “A Design Strategy for Reconfigurable Manufacturing Systems (RMSs) Using Analytical Hier-
archical Process (AHP): A Case Study.” International Journal of Production Research 41 (10): 2273–2299.
Abdi, M. R., and A. W. Labib. 2004a. “Grouping and Selecting Products: The Design Key of Reconfigurable Manufacturing Systems
(RMSs).” International Journal of Production Research 42 (3): 521–546.
Abdi, M. R., and A. W. Labib. 2004b. “Feasibility Study of the Tactical Design Justification for Reconfigurable Manufacturing Sys-
tems Using the Fuzzy Analytical Hierarchical Process.” International Journal of Production Research 42 (15): 3055–3076.
 International Journal of Production Research 15

Bard, J. F., K. Srinivasan, and D. Tirupati. 1999. “An Optimization Approach to Capacity Expansion in Semiconductor Manufacturing
Facilities.” International Journal of Production Research 37 (15): 3359–3382.
Battaïa, O., and A. Dolgui. 2013. “A Taxonomy of Line Balancing Problems and their Solution Approaches.” International Journal
of Production Economics 142 (2): 259–277.
Bi, Z. M., S. T. Y. Lang, W. Shen, and L. Wang. 2008. “Reconfigurable Manufacturing Systems: The State of the Art.” International
Journal of Production Research 46 (4): 967–992.
Brooke, L. 1998. “GM Launches First CNC-based Engine Plant.” Automotive Industries 178 (11): 23.
DeGarmo, E., J. T. Black, and R. A. Kohser. 1998. Materials and Processes in Manufacturing. 7th ed. London: Macmillan.
Deif, A. M., and W. H. ElMaraghy. 2006. “A Control Approach to Explore the Dynamics of Capacity Scalability in Reconfigurable
Manufacturing Systems.” Journal of Manufacturing Systems 25 (1): 12–24.
Dolgui, A., B. Finel, N. Guschinsky, G. Levin, and F. Vernadat. 2005. “A Heuristic Approach for Transfer Lines Balancing.” Journal
of Intelligent Manufacturing 16 (2): 159–172.
Dolgui, A., N. Guschinsky, and G. Levin. 2006. “A Special Case of Transfer Lines Balancing by Graph Approach.” European Jour-
nal of Operational Research 168 (3): 732–746.
Dolgui, A., N. Guschinsky, and G. Levin. 2009. “Graph Approach for Optimal Design of Transfer Machine with Rotary Table.” Inter-
national Journal of Production Research 47 (2): 321–341.
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

Dolgui, A., A. Eremeev, M. Y. Kovalyov, and V. Sigaev. 2013. “Complexity of Buffer Capacity Allocation Problems for Production
Lines with Unreliable Machines.” Journal of Mathematical Modelling and Algorithms in Operations Research 12 (2): 155–
165.
Dou, J., X. Dai, and Z. Meng. 2010. “Optimisation for Multi-part Flow-line Configuration of Reconfigurable Manufacturing System
Using GA.” International Journal of Production Research 48 (14): 4071–4100.
Freiheit, T., M. Shpitalni, and S. J. Hu. 2004a. “Productivity of Paced Parallel-serial Manufacturing Lines with and without Cross-
over.” Journal of Manufacturing Science and Engineering 126 (2): 361–367.
Freiheit, T., M. Shpitalni, S. J. Hu, and Y. Koren. 2004b. “Productivity of Synchronized Serial Production Lines with Flexible
Reserve Capacity.” International Journal of Production Research 42 (10): 2009–2027.
Freiheit, T., W. Wang, and P. Spicer. 2007. “A Case Study in Productivity-cost trade-off in the Design of Paced Parallel Production
Systems.” International Journal of Production Research 45 (14): 3263–3288.
Gen, M., and R. Cheng. 2000. Genetic Algorithms and Engineering Optimization. New York: Wiley.
Geng, N., and Z. Jiang. 2009. “A Review on Strategic Capacity Planning for the Semiconductor Manufacturing Industry.” Interna-
tional Journal of Production Research 47 (13): 3639–3655.
Goyal, K. K., P. K. Jain, and M. Jain. 2012. “Optimal Configuration Selection for Reconfigurable Manufacturing System Using
NSGA II and TOPSIS.” International Journal of Production Research 50 (15): 4175–4191.
Gu, X., X. Jin, J. Ni, and Y. Koren. 2015. “Manufacturing System Design for Resilience.” Procedia CIRP 36: 135–140.
Gumasta, K., S. K. Gupta, L. Benyoucef, and M. K. Tiwari. 2011. “Developing a Reconfigurability Index Using Multi-attribute Util-
ity Theory.” International Journal of Production Research 49 (6): 1669–1683.
Holland, J. H. 1973. “Genetic Algorithms and the Optimal Allocation of Trials.” SIAM Journal on Computing 2 (2): 88–105.
Koren, Y. 2005. “Reconfigurable Manufacturing and beyond.” Keynote Speech at 3rd CIRP International Conference on Reconfig-
urable Manufacturing, Ann Arbor, MI, USA.
Koren, Y. 2010. The Global Manufacturing Revolution. Hoboken, NJ: John Wiley & Sons.
Koren, Y. 2013. “The Rapid Responsiveness of RMS.” International Journal of Production Research 51 (23–24): 36817–36827.
Koren, Y., U. Heisel, F. Jovane, T. Moriwaki, G. Pritschow, A. G. Ulsoy, and H. Van Brussel. 1999. “Reconfigurable Manufacturing
Systems.” CIRP Annals – Manufacturing Technology 48(2): 6–12.
Koren, Y., S.J. Hu, and T.W. Weber. 1998. “Impact of Manufacturing System Configuration on Performance.” CIRP Annals – Manu-
facturing Technology 47(1), 689-698.
Koren, Y. and S. Kota. 1999. Reconfigurable Machine Tool. US patent # 5,943,750.
Koren, Y., and M. Shpitalni. 2010. “Design of Reconfigurable Manufacturing Systems.” Journal of Manufacturing Systems 29 (4):
130–141.
Koren, Y., and A.G. Ulsoy. 1997. “Reconfigurable Manufacturing Systems.” Engineering Research Center for Reconfigurable Manu-
facturing Systems, ERC/RMS Report #1. Ann Arbor, MI: University of Michigan.
Koren, Y., and A. G. Ulsoy. 2002a. “Vision, Principles and Impact of Reconfigurable Manufacturing Systems.” Powertrain Interna-
tional 5 (3): 14–21.
Koren, Y., and A. G. Ulsoy. 2002b. Reconfigurable Manufacturing System Having a Production Capacity, Method for Designing
Same, and Method for Changing its Production Capacity. US patent # 6,349,237, Patent filling date: May 1998.
Krygier R. 2005. “The Integration of Flexible, Reconfigurable Manufacturing with Quality.” 3rd CIRP International Conference on
Reconfigurable Manufacturing, Ann Arbor, MI.
Landers, R. G., B. K. Min, and Y. Koren. 2001. “Reconfigurable Machine Tools.” CIRP Annals – Manufacturing Technology 50(1):
269–274.
Lorenzer, T., S. Weikert, S. Bossoni, and K. Wegener. 2007. “Modeling and Evaluation Tool for Supporting Decisions on the Design
of Reconfigurable Machine Tools.” Journal of Manufacturing Systems 26 (3–4): 167–177.
 16 Y. Koren et al.

Maier-Speredelozzi, V., Y. Koren, and S. J. Hu. 2003. “Convertibility Measures for Manufacturing Systems.” CIRP Annals – Manu-
facturing Technology 52(1): 367–370.
Meng, G., S. S. Heragu, and H. Zijm. 2004. “Reconfigurable Layout Problem.” International Journal of Production Research 42
(22): 4709–4729.
Niroomand, I., O. Kuzgunkaya, and A. Bulgak. 2012. “Impact of Reconfiguration Characteristics for Capacity Investment Strategies
in Manufacturing Systems.” International Journal of Production Economics 139 (1): 288–301.
Putnik, G., A. Sluga, H. ElMaraghy, R. Teti, Y. Koren, T. Tolio, and B. Hon. 2013. “Scalability in Manufacturing Systems Design
and Operation: State-of-the-art and Future Developments Roadmap.” CIRP Annals – Manufacturing Technology 62(2): 751–
774.
Rekiek, B., A. Dolgui, A. Delchambre, and A. Bratcu. 2002. “State of Art of Optimization Methods for Assembly Line Design.”
Annual Reviews in Control 26 (2): 163–174.
Rösiö, C., and K. Säfsten. 2013. “Reconfigurable Production System Design – Theoretical and Practical Challenges.” Journal of Man-
ufacturing Technology Management 24 (7): 998–1018.
Shi, C., and S. B. Gershwin. 2014. “A Segmentation Approach for Solving Buffer Allocation Problems in Large Production Sys-
tems.” International Journal of Production Research. doi:10.1080/00207543.2014.991842.
Son, S. Y., T. L. Olsen, and D. Yip-Hoi. 2001. “An Approach to Scalability and Line Balancing for Reconfigurable Manufacturing
Downloaded by [University of California, San Diego] at 08:17 19 February 2016

Systems.” Integrated Manufacturing Systems 12 (7): 500–511.

Sotskov, Y. N., A. Dolgui, and M. C. Portmann. 2006. “Stability Analysis of an Optimal Balance for an Assembly Line with Fixed
Cycle Time.” European Journal of Operational Research 168 (3): 783–797.
Spicer, P., Y. Koren, and M. Shpitalni 2002. “Design Principles for Machining System Configurations.” CIRP Annals – Manufacturing
Technology, 51(1),276–280.
Spicer, P., D. Yip-Hoi, and Y. Koren. 2005. “Scalable Reconfigurable Equipment Design Principles.” International Journal of Produc-
tion Research 43 (22): 4839–4852.
Su, C.-T., and Y.-R. Shiue. 2003. “Intelligent Scheduling Controller for Shop Floor Control Systems: A Hybrid Genetic Algorithm/
Decision Tree Learning Approach.” International Journal of Production Research 41 (12): 2619–2641.
Tang, Y., and R. G. Qiu. 2004. “Integrated Design Approach for Virtual Production Line-based Reconfigurable Manufacturing Sys-
tems.” International Journal of Production Research 42 (18): 3803–3822.
Tang, L., D. Yip-Hoi, Y. Koren, and W. Wang. 2004a. “An AI-based Computer-aided Reconfiguration Planning Framework for
Reconfigurable Manufacturing Systems.” ASME Proceedings on Manufacturing Engineering, Anaheim, CA, Paper No.
IMECE2004-60744, 411–423.
Tang, L., D. Yip-Hoi, W. Wang, and Y. Koren. 2004b. “Concurrent Line-balancing, Equipment Selection and Throughput Analysis
for Optimal Line Design.” International Journal for Manufacturing Science & Production 6 (1): 71–81.
Wang, W., and Y. Koren. 2012. “Scalability Planning for Reconfigurable Manufacturing Systems.” Journal of Manufacturing Systems
31 (2): 83–91.
Wang, W., and Y. Koren. 2013. “Design Principles of Scalable Reconfigurable Manufacturing Systems.” Manufacturing Modelling,
Management, and Control 7 (1): 1411–1416.
Yang, S., C. Wu, and S. J. Hu. 2000. “Modeling and Analysis of Multi-stage Transfer Lines with Unreliable Machines and Finite
Buffers.” Annals of Operations Research 93 (1–4): 405–421.
Youssef, A., and H.A. ElMaraghy. 2006. “Modelling and Optimization of Multiple-aspect RMS Configurations.” International Journal
of Production Research 44 (22): 4929–4958.
Youssef, A., and H. A. ElMaraghy. 2008. “Availability Consideration in the Optimal Selection of Multiple-aspect RMS Configura-
tions.” International Journal of Production Research 46 (21): 5849–5882.
Zhang, M. T., J. Fu, and E. Zhu. 2008. “Dynamic Capacity Modeling in Semiconductor Assembly Manufacturing.” International
Journal of Production Research 46 (3): 739–752.
Zhang, X., J. Wang, and Z. Luo. 2000a. “A Stochastic Model of a Reconfigurable Manufacturing System Part 1: A Framework.”
International Journal of Production Research 38 (10): 2273–2285.
Zhang, X., J. Wang, and Z. Luo. 2000b. “A Stochastic Model of a Reconfigurable Manufacturing System Part 2: Optimal Configura-
tions.” International Journal of Production Research 38 (12): 2829–2842.