An Approach of Designing Conwip Loop For Assembly

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

International Journal of Computer Integrated Manufacturing, 2016

Vol. 29, No. 7, 805–820, http://dx.doi.org/10.1080/0951192X.2015.1130241

An approach of designing CONWIP loop for assembly system in one-of-a-kind production


environment
Guodong Huanga*, Jie Chena, Xiuli Wanga and Yiqun Shib
a
School of Economics & Management, Nanjing University of Science & Technology, Nanjing, Jiangsu Province, P.R. China; bSchool of
Foreign Studies, Suqian College, Jiangsu, P.R. China
(Received 22 August 2014; accepted 14 August 2015)
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

Pursuit of the shortest cycle time does not enhance benefits in one-of-a-kind production (OKP) environment. The quality of
services is affected by the delivery on time for OKP enterprise, and a well-designed constant work in process (CONWIP)
loop structure can effectively improve the performance of assembly system in OKP. The designed CONWIP loop structures
have been tested by simulation under 2 assembly proportion conditions and 14 situations of fluctuation combinations, and
analysed relationship between cycle time and punishment of tardiness. Simulation test shows that some of CONWIP
policies have stronger robustness, which can perform well in several fluctuation situations. The authors propose an idea of
dynamic configuration for CONWIP control. Furthermore, the design procedure of CONWIP can provide a reference and
guidance for the application of CONWIP. This article extends the knowledge of OKP shop floor control and promotes
implementation of CONWIP in pull manufacturing system.
Keywords: OKP; CONWIP; shop floor control; assembly system; due date

1. Introduction changes will increase the lead time of product designing


The studies of one-of-a-kind production (OKP) focus on and process setting, ultimately production lead time will
product development and configuration (Xie and Tu 2006; be extended. In some OKP shop floor with no scientific
Hong, Xue, and Tu 2010; Aleksić et al. 2012; Li and Xie management, work in process (WIP) has to be forced to
2013), and production planning and scheduling (Tu, Chu, rework or scrap because the production data cannot be
and Yang 2000; Choi and You 2006; Li et al. 2011), but updated on a timely basis and is consistently out of date,
there are few studies in OKP shop floor control mechan- which increases the production cycle time and cost.
ism. The shop floor is an important realisation place for Finding a trade-off between enhancing customer invol-
product ideas and production plans. Without the effective vement and improving response capability of manufactur-
control mechanism in the shop floor, the tasks cannot be ing is a challenge for OKP. Applying information
completed on schedule, which will not only influence the technology (IT) should be a fine solution, but the success-
product delivery, but also disturb the established plans and ful OKP case using IT is still little at present, maybe it
strategy level eventually. The effect of OKP to shop floor needs more time to verify. Furthermore, only using com-
is mainly manifested by the fact that OKP pays attention puter-aided technology is hard to bring significant change,
to delivery on time, because customer doesn’t accept early if there is a lack of a control methodology that adjusts
delivery or tardiness. OKP enterprise will increase inven- output according to production status.
tory costs when an order is completed ahead of schedule From the point of shop floor, the actual cycle time of
and risk of tardiness penalty when delivery is delayed. product determines the completion time of order, the fact
Despite the fact that the authors emphasise the impor- that the order completion time is shorter than the due date
tance of delivery on time, in most of the cases, OKP is means the completion in advance and the fact that the
hard to guarantee the promised delivery time in face of order completion time is longer than the due date means
highly personalised products or mass customisation in the delayed delivery. Control of the actual production
practical. The customer involvement of OKP is far greater cycle time is a key factor for OKP shop floor.
than the mass production and the multi-products and small The OKP shop floor shows a characteristic of mixed
batch. Products cannot be converted into profits if the key production. The mixed production needs to consider setup
requirements of customer cannot be satisfied. To some time and bottleneck drift. The setup time is longer, and
extent, the customer has a vote power over any proposal changing frequency is more than multi-products and small
in OKP environment. Sometimes the customer changes batch. Because a product on an OKP flow line is seldom
requirements frequently in product development, these repeated (Wortmann, Muntslag, and Timmermans 1997)

*Corresponding author. Email: hgdtonee@hotmail.com

© 2015 Taylor & Francis


806 G. Huang et al.

and the required processes are very different among types Toyota production system successfully. With Toyota’s glo-
of products, which makes impossible to shorten the setup bal expansion, plenty of experience of Kanban implemen-
time effectively by the learning curve. Setup time often tation is gathered, CONWIP control has a large gap in this
shows an asymmetry, for example the setup time is 10 respect.
units from A product type to B, while 20 from B to A. In The artificial bee colony algorithm is used to minimise
manufacturing, the setup time is an essential factor in OKP the overall makespan by Ajorlou and Shams (2013). Park
shop floor. and Lee (2013) developed an approximation method to
Influenced by product size, complexity, process routes, analyse a multi-product CONWIP assembly system with
enterprise scale and operation strategy, some of the OKP correlated external demands. Harrod and Kanet (2013)
enterprises use flow shop for production, the other use job argue that CONWIP has the lowest mean and variance
shop, especially for larger complex product that is fabri- of tardiness among flow control system in make-to-order
cated in different shop floors and converged into the last (MTO), when finished goods due dates are externally
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

shop floor to assemble, which shows characteristics of determined and early delivery of finished goods is prohib-
assembly flow shop and supply collaboration. In this ited. Renna, Magrino, and Zaffina (2013) proposed a
OKP assembly system, it is hard to control cooperatively dynamic card control methodology based on the observa-
in the last shop that each shop floor has different workload tion of customer demand fluctuations that is evaluated by
and takt time. The convergent position in shop floor needs two moving average.
to be scheduled and controlled scientifically. These researches cover some characteristics of OKP,
Therefore, it is valuable to research control mechanism for instance, they pay more attention to cycle time, con-
of OKP shop floor. At present, two kinds of control sider delivery on time and tardy factors. But these
mechanisms in shop floor have attracted interests of researches are not for CONWIP loop specially. The
researchers. One is proposed by Spearman, Woodruff, scope of CONWIP research is still partial to theory, and
and Hopp (1990), called as constant work in process a few researcher notes that CONWIP is lack of implemen-
(CONWIP) that mainly aimed to control flow shop, the tation guidelines (Pettersen and Segerstedt 2009; Levalle,
other one is named paired-cell overlapping loops of cards Scavarda, and Nof 2013). From perspective of engineer-
with authorisation (POLCA) by Suri (1998). POLCA is ing, a suitable design of CONWIP loop is first considera-
applied more in practice than CONWIP (Riezebos 2010) tion in CONWIP implementation. Huang et al. (2014)
so far. Having a lot of guidelines, design steps or proce- proposed a design pattern of CONWIP loop, eight
dures and practical method, it can direct the implementa- CONWIP policies were designed by this design pattern,
tion of POLCA in the enterprise. The successful cases and then these policies were evaluated in order to obtain a
further perfect POLCA. Reader can refer to Suri and feasible CONWIP scheme. This article is a continuation of
Krishnamurthy (2003), Krishnamurthy and Suri (2009), their work. This article focuses on these policies, which
Riezebos (2010) and reference in their articles. can be considered as a feasible scheme of OKP shop floor
Relatively speaking, CONWIP is short in the aspect of control.
‘how to do’. From the literatures about CONWIP in recent In the following sections, a design pattern of CONWIP
six years, the research focused on the mathematical model loop and procedure, and design CONWIP loop structure for
and algorithm solving (Li et al. 2010; Dalalah and Al- convergent assembly system are introduced in Section 2.
Araidah 2010; Mhada and Malhamé 2011; Ajorlou and Simulation and modelling for loop structures in Section 3.
Shams 2013; Park and Lee 2013; Renna, Magrino, and Performance is evaluated and relationship between cycle
Zaffina 2013), and performance comparison between dif- time and punishment of tardiness is analysed in Section 4.
ferent flow control mechanisms (Jodlbauer and Huber Finally, the conclusion and future work are shown in
2008; Pettersen and Segerstedt 2009; Harrod and Kanet Section 5.
2013). Specifically, Pettersen and Segerstedt (2009)
argued that restricted WIP shortens the lead time and
variation. Although they also mentioned the fact that 2. CONWIP loop structure and the design procedure
CONWIP lacks installation guidelines, they did not intro- 2.1. CONWIP loop structure
duce how to install. Jodlbauer and Huber (2008) measured In this article, CONWIP loop structure is regarded as a
service level performance of MRP (material requirement parameter of CONWIP policy, but it also represents
planning), Kanban, CONWIP and drum-buffer-rope CONWIP policy if the parameter characteristic of
(DBR) due to parameter stability and environmental CONWIP policy is not emphasised. Generally, CONWIP
robustness. Most researchers agree that CONWIP has a loop structure has two ways including single loop and
better performance over Kanban in unstable flow line. multi-loop (or m-CONWIP), and it needs to consider
Kanban is not suitable for OKP environment, its origin factors as follow while choosing the CONWIP loop
is designed to control the stability flow line. Kanban structure.
control, however, has been applied to the well-known
International Journal of Computer Integrated Manufacturing 807

● CONWIP loop identification. Setting begin–end &: means splicing or merging, it can be seen that
position of loop has to consider some hard con- m-loops are merged into a single loop that controls a
straints in practice, such as, in order to prevent route.
bottlenecks drift, it needs to separate bottleneck ^: means several series loop, it can be seen that a
position, business process needs to set push–pull single loop are discrete m-loops that control a route.
interface, geographical space constraints, flow It is the inverse process of merging (&), and often used
shop or job shop, supply chain cooperation agree- in job shop mode.
ment, etc. These factors determine the control span
of CONWIP loop and the number of loop. Production is regarded as a product is processed in the
● Determining appropriate WIP limit. WIP limit upstream, and then enters downstream to process, finally
determines the throughput and cycle time of the finished goods is produced in the production system,
CONWIP loop. This is very important parameter in which the interface point (IP) is a bridge between
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

for pull mechanism. A lot of methods can solve upstream and downstream. IP is also called as the interac-
the WIP limit in literatures, it is worth mentioning tion point (Levalle, Scavarda, and Nof 2013). Workload
that the WIP limit cannot be solved directly by balancing is the core idea in the five basic patterns
using Little’s law (Little 1961), because of the bot- (Figure 1). IP plays an important role of transport hub in
tleneck drift (Hopp and Spearman 2001). convergent and divergent system, and it can be seen as
● Types of CONWIP card. The card has two types bottleneck workstation or hard constraints in a single flow
including the shared card and the independent card line. In other word, a position that IP is put into different
(Khojasteh-Ghamari 2009). CONWIP card is differ- loop determines the scope of the bottleneck drift and the
ent from Kanban card, and it is not directly asso- discrete degree of fluctuation.
ciated with the product class. But in convergent or The five basic patterns can be combined to complex
divergent system, CONWIP card needs to consider patterns. Up&Down is a standard pattern of CONWIP
whether to share a card on many routes to transmit loop as shown in Figure 1(a), the dotted line represents
information. This factor should be considered in the the signal circuit of CONWIP card, and the signal trigger
design of CONWIP loop structure. of CONWIP is located in the entrance of upstream and
exit of downstream. The other patterns are shown in
In order to design a reasonable structure of loop in prac- Figure 1(b)–(e).
tice, the CONWIP designers have to design multiple
schemes to evaluate and select according to the design
objective. The first problem of CONWIP implementation 2.3. Loop structure of assembly flow shop
is how to design the multiple schemes or alternatives. Assembly system is roughly divided into two kinds. One
Because CONWIP lacks design procedure and guidelines kind is that enterprises have their own fabrication lines
for implementation, the designer makes schemes only with where a majority of main parts or components are pro-
experience and feeling, and cannot answer why. cessed and assembly lines where the finished products are
A design pattern of CONWIP loop is proposed to assembled finally. Another is enterprises that have a core
design the CONWIP loop structure. The design pattern technology of product and a few of processing or assem-
gives a feasible guideline to design CONWIP pull bling ability, and most of the parts, components and semi-
mechanism. Under the instruction of design pattern, the finished products are processed by suppliers, such as ori-
designer can get multiple schemes of loop structure pur- ginal equipment manufacturer (OEM). In this article, the
posefully. Design pattern provides a framework for background of assembly system is the former.
designer of CONWIP system. The designer only needs Assembly system (see Figure 2) is composed of fabri-
to pay attention to details and develops CONWIP system cation line-1 (F1), fabrication line-2 (F2) and assembly
according to the actual problems. The understanding level line (AL). AL includes assembly interface (AI) and assem-
of the designer decides the structure of loop and its bly station (AS). AI is IP locating the front end of AL
performance. where upstream lines supplying parts or components to
AL are merger place.
According to the five basic design patterns, the eight
2.2. Loop design pattern in flow shop schemes of CONWIP loop structure are designed respec-
In order to describe the begin–end position of loop clearly, tively, as shown in Figure 3. No. 1 belongs to Up//Down,
the symbols are adopted to represent the design pattern of No. 2 belongs to Up&IP/Down, No. 3 and No. 4 belong to
loop, as follows. Up/IP/Down, No. 5 and No. 6 belong to Up/IP&Down,
and No. 7 and No. 8 belong to Up&Down. These patterns
/: means separation position of loop or push–pull and loop structures are described in details by Huang et al.
interface. (2014).
808 G. Huang et al.
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

Figure 1. (a) Up&Down, (b) Up&IP/Down, (c) Up/IP&Down, (d) Up//Down (or Up/Down), and (e) Up/IP/Down.

Figure 2. The standard assembly system.

2.4. Procedures evaluation) is mainly to determine what method can be


Figure 4 is a general procedure to design CONWIP applied to select a feasible scheme from alternatives. The
scheme. The procedure has four phases, which are implementation tools for CONWIP card trigger and
design, implementation, commissioning and operation transmission are analysed in the initial design and a
and maintenance. Design phase includes initial design, complete implementation scheme will be determined in
detailed design and assessment. The collected production the detailed design. In loop constraints, it needs to be
data (e.g. product parameter, production organisation, considered which factors will interrupt CONWIP loop,
facility layout, production process, etc.) are used to pro- namely, the loop identification will be considered. Loop
vide decision-making for the initial design. Assessment design pattern is to provide a direction for designing
method (e.g. system simulation or comprehensive various loop structure.
International Journal of Computer Integrated Manufacturing 809
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

Figure 3. CONWIP loop structure schemes. (a) No. 1, (b) No. 2, (c) No. 3, (d) No. 4, (e) No. 5, (f) No. 6, (g) No. 7, and (h) No. 8.

The analysis results of loop constraints and design cope with different problem well, the optimal scheme
pattern are used to design specific loop structure in the generated by this comparison may not always be the best
detailed design. There are two ways determining an appro- one under the same theory, and the non-optimal scheme
priate WIP limit. The description of Framinan, González, based on a theory does not mean that others with the same
and Ruiz-Usano (2003) is cited for the two ways. The first theory are still non-optimal.
description refers to given certain manufacturing condi- Furthermore, designer can modify scheme according
tions, finding a procedure to obtain a number of cards that to situation of implementation and commissioning in the
makes the system perform acceptably according to some shop floor. The detailed design can be modified when
pre-defined performance measures, this article called it as some constraints or parameters are changed. If the produc-
static method. The second description refers to the devel- tion style or mode changes, the initial design will be
opment of rules to change or maintain the current number modified and even the design pattern will be rethought.
of cards of a CONWIP system, this article called it as
dynamic control method. In assessment, the comparison
among the CONWIP policies will be recommended. The 3. Simulation modelling
authors hold opinion that comparing with non-CONWIP F1, F2 and AL have two workstations in the assembly
tends to discuss the advantages between theories that can model, respectively. One machine in each workstation,
810 G. Huang et al.
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

Figure 3. (Continued).

capacity of buffer store is unlimited. The sequence of job time is 100, and the standard deviation is 60
that represents product or order is fixed, and machine (CVhigh ¼ 0.60). A job is not processed on a machine
process a job at a time, regardless of the breakdown, when process time is a non-positive in normal distribution.
unit of time is minutes. In addition, α/φ/γ represent a fluctuation combination of
The fluctuation of each line is measured by coefficient multiple lines, that is α field denotes that every machine
of variation. Assuming that the process time is subjected has the same fluctuation level in F1, φ field denotes that
to normal distribution, the fluctuation of process has three every machine has the same fluctuation level in F2 and γ
levels, including: Low level (L) is that the mean process field denotes that every machine has the same fluctuation
time is 100, and the standard deviation is 5 level in AL. For example, L/M/H means that all of coeffi-
(CVlow ¼ 0.05). Medium level is that the mean process cient of variation is 0.05 in F1, all of coefficient of varia-
time is 100, and the standard deviation is 30 tion is 0.30 in F2 and all of coefficient of variation is 0.60
(CVmedium ¼ 0.30). High level is that the mean process in AL.
International Journal of Computer Integrated Manufacturing 811
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

Figure 4. Design procedure.

The assembly model consists of three lines, each line combination of fluctuation, a full factorial experiment is
has three levels, and hence the total combination is 27. But used to get the WIP limit when cycle time is the shortest
13 combinations does not be observed in this article, and throughput is the highest in a designed loop structure.
which are M/M/M, M/L/L, M/H/H, H/M/M, L/M/M, L/ The measurement of cycle time, profits and robustness
M/L, H/M/H, M/L/M, M/H/M, M/M/L, M/M/H, L/L/M under different loop structure is used for performance
and H/H/M. Because the range ability between M level evaluation and analysis
and L level (or H level) is smaller than the range ability
between H level and L level, such as using H/H/L to
observe gains more information than M/M/L does.
An observation repeats 20 times with different random 3.1. Parameter definition and setting
number generation, and a warm-up period of 50 jobs is A concept that each of jobs needs different number of
chosen. Statistics collection then continues for a run length components or parts in OKP was named assembly propor-
of 500 jobs. The experiments are conducted with assembly tion (α).α = 1:n means that a job needs a main body (or
proportion 1:1 and 1:n respectively. The definition of component) and n components (or parts) to assemble. If
assembly proportion and parameter setting are given in components or parts have the same process time repre-
the next section. The experimental purpose is to test the sents the same type, otherwise has different types. The
response capacity of eight loop structures in different concept is often used to draw the assembly flow chart and
combination of fluctuation in F1, F2 and AL. For each make a bill of material (BOM) in practice.
812 G. Huang et al.

Table 1. Batch and assembly proportion.

job1 job2 job3 job4 job5 job6 job7 job8 job9 job10 job11 job12

β 1 1 1 1 1 1 1 1 1 1 1 1
1:1 1 1 1 1 1 1 1 1 1 1 1 1
1:n 3 4 4 3 8 1 3 10 5 1 7 9

Definition 1: J is a set of job type, and j 2 J. N is a set of 3.2. Profit and robustness
parts or components, and n 2 N. The notation of nj The due date is an important factor for OKP. Hence, the
denotes the j-th job needs n parts or components to profit relating due date is a key measurement for perfor-
assemble. The j-th job assembly proportion is defined as mance besides cycle time and throughput.
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

α ¼ 1 :nj , and the set of nj represents {nj |j 2 J, n 2 N},


X   
where nj = 1 denotes that CONWIP card has the shared R¼ λj  ωj max Cj  dj ; 0 (2)
j2J
type, and nj = 1 but not all of jobs denotes the independent
type. Conveniently let 1:n express α ¼ 1 :nj . The para-
In (2), where R is revenue for all of jobs, Cj is a
meters setting are shown in Table 1.
completion time of job j. The average revenue
Definition 2: B is a set of batch, the j-th job assembly R ¼ R=j J j, and j J j denotes the total number of elements
batch is defined as βj , and the set of βj represents {βj |j 2 J, in J set.
β 2 B}, where βj = 1 refers to OKP here exclusively. The X
parameters setting are shown in Table 1. S¼ τ qλ
j2J j j
(3)

Definition 3: Let A and B ∈ J. Two-tuples relation is to In (3), where S is total storage cost of all of jobs that
be defined as (A, B), it represents the setup time from A to early finished in the 3-th party inventory, τ j is a storage
B. (A, B) Þ (B, A) is called as asymmetrical setup. The time of job j, that is τ j ¼  minfCj  dj ; 0g, and q is unit
setup time in this article includes: (job1, job2) = 120, cost of storage per day that less than a day is approxi-
(job2, job1) = 240, (job2, job3) = 100, (job3, mately equal to a day. Therefore the average storage is
job2) = 200, (job3, job4) = 80, (job4, job3) = 160, S ¼ S=j~nj, where ~n ¼ fjjCj  dj < 0; j 2 J g, and the aver-
(job4, job5) = 50, (job5, job4) = 100, (job5, age profit denotes R  S.
job6) = 100, (job6, job5) = 200, (job6, job7) = 120, Robustness is a ratio that a count of the best perfor-
(job7, job6) = 240, (job7, job8) = 100, (job8, mance of a loop structure under all of fluctuation combi-
job7) = 200, (job8, job9) = 80, (job9, job8) = 160, nation is divided by total number of combinations.
(job9, job10) = 50, (job10, job9) = 100, (job10, Robustness approaches to 100%, which indicates that the
job11) = 100, (job11, job10) = 200, (job11, loop structure have more ability to cope with fluctuation.
job12) = 120, (job12, job11) = 240, and the setup time
of others is 0.
4. Results and analysis
The revenue λj is integer value, which were generated
4.1. Observation analysis with 1:1 condition
randomly from a uniform distribution in the interval [1,
100]. The weighed ωj is integer value, which was gener- Table 2 shows the average cycle time of 8 loop structures
ated randomly from a uniform distribution in the interval under the 14 fluctuation combinations. The coefficients of
[1, 10]. Due date dj is set as follows (Harrod and Kanet variation of cycle time shown in Figure 5 are very low,
2013). which means the statistics of cycle time in experiment
have a good representativeness. The throughput changes
between 0.003–0.005 (Figure 6) and their standard devia-
d j ¼ rj þ θ j (1) tion are less than 0.0001.
As is clearly shown from Table 2, No. 1 has the
In (1), where rj is the arrival time of job j and θj is the shortest cycle time under H/L/M and L/L/L, and its
cycle time of job j, θj subjects to uniform distribution in robustness reaches14.29%. No. 2 has the shortest cycle
time under L/M/H, L/H/H, H/L/H and L/L/H, and its
the interval ½bminftjt 2 CTjgc; dmaxftjt 2 CTjge, CTj is
robustness reaches 28.57%. No. 3 has the shortest cycle
a set of planned cycle time of job j that is collected from a
time under L/H/M, M/L/H, H/L/M, H/H/H and L/L/L, its
run of 50 replications in the first come first served system
robustness reaches 28.57%. No. 4 has the shortest cycle
without WIP limit.
time under M/H/L, H/M/L, H/L/L, L/H/L, H/H/L and L/L/
International Journal of Computer Integrated Manufacturing 813

Table 2. Cycle time of 8 loop structures under the 14 fluctuation combinations with 1:1 condition.

Fluctuation No. 1 No. 2 No. 3 No. 4 No. 5 No. 6 No. 7 No. 8

L/M/H 1116.970 1110.769* 1113.221 1116.288 1136.768 1135.102 1123.446 1123.004


L/H/M 1055.952 1074.135 1054.291* 1063.618 1075.801 1080.393 1071.121 1071.402
M/L/H 1123.034 1117.549 1052.014* 1122.847 1136.686 1136.933 1133.869 1132.278
M/H/L 1051.590 1072.715 1052.512 1050.990* 1067.499 1065.852 1071.511 1071.503
H/M/L 1092.855 1113.892 1095.683 1092.368* 1105.497 1105.126 1121.277 1119.660
H/L/M 1105.040* 1120.595 1105.040* 1105.448 1121.926 1121.386 1137.007 1132.429
H/H/H 1151.607 1151.211 1149.610* 1151.067 1161.253 1160.103 1168.831 1170.262
H/L/L 1093.485 1114.404 1094.052 1093.023* 1107.037 1105.535 1120.006 1131.052
L/H/H 1114.416 1108.150* 1151.493 1115.704 1136.901 1135.740 1124.005 1123.250
H/L/H 1154.433 1149.857* 1151.493 1151.157 1163.040 1163.641 1166.852 1170.717
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

L/H/L 1042.007 1057.295 1053.575 1041.228* 1052.522 1052.684 1043.041 1042.661


L/L/H 1112.588 1110.427* 1115.326 1115.013 1136.046 1136.34 1123.721 1123.984
H/H/L 1094.092 1115.208 1093.235 1092.530* 1106.323 1106.122 1121.305 1121.318
L/L/L 1041.193* 1057.239 1041.193* 1041.176* 1052.876 1052.454 1042.635 1042.889
Note: *Denotes the shortest cycle time in 8 loop structures under a fluctuation. In L/L/L No. 1, No. 3 and No. 4 the values are most the same which are
considered the shortest cycle time.

Figure 5. Coefficient of variation of cycle time with 1:1 condition.

Figure 6. Average throughput with 1:1 condition.


814 G. Huang et al.

L, its robustness reaches 42.86%. These four loop struc- of cycle time, which means that the performance of pro-
tures can better cope with all of 14 fluctuation combina- duction system is affected by different loop structures
tions. According to the specific fluctuation, the designer significantly.
may select a scheme of loop structure. For example, No. 1
and No. 3 are acceptable policy for H/L/M, No. 1, No. 3
and No. 4 are acceptable policy for H/L/M, and No. 4 has 4.2. Observation analysis with 1:n condition
the best robustness for cycle time, which represents the Table 4 shows the average cycle time of 8 loop structures
strongest ability to respond to fluctuation. If the fluctua- under the 14 fluctuation combinations. The coefficients of
tions including M/H/L, H/M/L, H/L/L, L/H/L, H/H/L and variation of cycle time shown in Figure 7 are very low,
L/L/L are often encountered in practice, then No. 4 can be which means the statistics of cycle time in experiment
used as a control structure for long term. Considering No. have a good representativeness. The throughput changes
3 copes with 4 fluctuation combinations including L/H/M, between 0.001–0.002 (Figure 8), and their standard devia-
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

M/L/H, H/L/M and H/H/H, No. 3 and No. 4 can cope with tion are less than 0.0001.
10 fluctuation combinations totally, that is 70% fluctua- Table 4 shows that No. 2 has the shorter cycle time
tions can be coped. As is shown in Figure 3, the position than other loop structures, and its robustness reaches
of AI is separated control in No. 3 and No. 4. The only 100%. In the structure of No. 2 (Figure 3(b)), the IP
difference is that No. 4 applies independent card to trans- node (AI position) and an upstream branch are controlled
mit information. Conversion of these two structures only by single CONWIP loop, and which is the Up&IP/Down
involves IP nodes (AI position). Hence, the two structures pattern that uses an independent card to transmit informa-
are relatively easy to convert due to the fact that system tion in different branches of the loop. In the assembly
coupling is low. system with 1:n condition, the production in F1 can be
The above analysis is based on the cycle time. The due taken as OKP, the production in F2 can be taken as multi-
date is important to OKP. The profit is applied to analyse products and small batch, hence AI place needs stronger
in the following, the results are in Table 3. collaboration and response. To improve ability of colla-
As is clearly shown from Table 3, No. 2 has the boration, the control authorisation of AI is transferred to
maximum profit under L/M/H and L/L/L, and its robust- upstream branch of CONWIP loop, different branches
ness reaches 14.29%. No. 3 has the maximum profit under using independent card improves response ability and
M/H/L, H/H/H and H/H/L, and its robustness reaches reduces the waiting time of parts, which is the design
21.43%. No. 4 has the maximum profit under H/M/L, H/ intention for No. 2 policy. In the actual production envir-
L/M and L/H/L, its robustness reaches 21.43%. No. 5 has onment, the condition of assembly proportion 1:n is more
the maximum profit under L/H/M and H/L/L, its robust- common. The authors recommend that the scheme of No.
ness reaches 14.29%. No. 6 has the maximum profit under 2 could be reference structure to design m-CONWIP loop,
L/H/H and H/L/H, its robustness reaches 14.29%. No. 7 when focusing on cycle time.
has the maximum profit under M/L/H and L/L/H, its Table 5 shows that No. 2 has the maximum profit
robustness reaches 14.29%. under H/H/H and H/L/H, and its robustness reaches
The robustness of No. 5, No. 6 and No. 7 are higher in 14.29%. No. 3 has the maximum profit under L/L/L, and
performance of profit than their robustness in performance its robustness reaches 7.14%. No. 4 has the maximum

Table 3. Profit of 8 loop structures under the 14 fluctuation combinations with 1:1 condition.

Fluctuation No. 1 No. 2 No. 3 No. 4 No. 5 No. 6 No. 7 No. 8

L/M/H 49.050 50.558* 49.086 49.530 48.504 46.923 48.974 48.037


L/H/M 49.793 48.254 47.009 48.189 50.506* 46.480 49.145 47.262
M/L/H 49.355 48.620 48.013 47.494 49.745 49.550 50.335* 47.046
M/H/L 48.185 50.040 51.007* 49.996 47.734 48.933 47.256 47.306
H/M/L 48.850 47.945 47.221 49.025* 47.184 48.665 46.642 47.806
H/L/M 47.498 48.959 47.494 49.927* 48.494 49.332 46.258 47.418
H/H/H 48.105 47.794 49.077* 47.369 48.523 48.307 48.272 46.844
H/L/L 47.505 47.456 48.129 49.201 49.551* 46.285 48.900 49.087
L/H/H 49.113 47.677 48.829 45.283 46.819 49.395* 48.272 47.950
H/L/H 46.938 48.055 46.941 46.787 48.565 48.779* 46.718 48.569
L/H/L 47.155 48.079 47.121 50.909* 48.521 48.178 46.948 45.939
L/L/H 46.620 49.004 48.788 48.587 47.747 46.668 49.310* 46.652
H/H/L 46.118 48.187 50.425* 47.925 47.124 48.939 49.738 47.997
L/L/L 49.860 49.969* 49.856 49.855 49.082 49.467 49.737 48.389
Note: *Denotes the maximum profit in 8 loop structures under a fluctuation.
International Journal of Computer Integrated Manufacturing 815

Table 4. Cycle time of 8 loop structures under the 14 fluctuation combinations with 1:n condition.

Fluctuation No. 1 No. 2 No. 3 No. 4 No. 5 No. 6 No. 7 No. 8

L/M/H 1815.689 1322.001* 1914.603 1814.900 1655.961 1656.347 1501.995 1500.783


L/H/M 1903.477 1374.602* 1901.813 1904.103 1751.420 1751.928 1479.569 1479.335
M/L/H 1807.582 1316.594* 1796.067 1805.978 1646.592 1646.592 1504.027 1500.993
M/H/L 1900.370 1375.013* 1900.370 1900.015 1751.238 1751.733 2970.859 1479.882
H/M/L 1790.980 1323.294* 1793.471 1793.914 1646.759 1647.087 1501.483 1501.308
H/L/M 1786.094 1321.326* 1785.704 1819.281 1638.099 1638.005 1500.667 1501.634
H/H/H 1898.221 1381.297* 1898.221 1899.822 1741.063 1741.330 1529.548 1528.106
H/L/L 1785.318 1321.068* 1784.693 1785.123 1639.598 1884.610 1502.793 1502.826
L/H/H 1912.423 1381.216* 1796.116 1910.941 1749.691 1749.784 1503.907 1503.078
H/L/H 1676.221 1337.045* 1795.135 1794.430 1636.818 1636.820 1525.973 1525.289
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

L/H/L 1902.281 1373.009* 1902.530 1900.560 1750.161 1750.415 1480.481 1480.504


L/L/H 1809.573 1317.183* 2372.416 1809.668 1648.971 1649.217 1502.451 1903.926
H/H/L 1893.623 1372.849* 1888.952 1892.670 1743.789 1743.739 1502.139 1501.452
L/L/L 1798.339 1306.031* 1798.365 1797.979 1650.370 1650.287 1480.070 1480.007
Note: *Denotes the shortest cycle time in 8 loop structures under a fluctuation.

Figure 7. Coefficient of variation of cycle time with 1:n condition.

Figure 8. Average throughput with 1:n condition.


816 G. Huang et al.

Table 5. Profit of 8 loop structures under the 14 fluctuation combinations with 1:n condition.

Fluctuation No. 1 No. 2 No. 3 No. 4 No. 5 No. 6 No. 7 No. 8

L/M/H 48.030 44.381 47.980 47.472 45.460 45.460 48.154* 46.035


L/H/M 46.100 46.830 44.892 46.245 48.022* 48.022* 45.890 45.651
M/L/H 48.263 48.532 46.583 49.391* 47.977 47.977 46.333 49.275
M/H/L 47.780 44.557 47.531 44.686 48.163 48.760* 45.394 47.298
H/M/L 46.823 45.124 48.115 49.383* 46.201 46.201 48.148 47.215
H/L/M 45.665 45.176 46.036 47.762* 46.687 46.687 47.155 45.818
H/H/H 47.040 49.334* 46.829 47.883 45.640 45.249 47.829 45.894
H/L/L 46.635 45.951 46.635 49.585* 45.016 44.730 47.561 47.929
L/H/H 46.790 45.456 47.963 48.630* 46.957 47.888 48.113 45.780
H/L/H 47.100 47.865* 47.096 46.264 45.053 45.053 47.195 46.999
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

L/H/L 44.845 46.532 45.271 46.292 49.647* 49.177 47.688 47.249


L/L/H 46.505 46.069 47.573 48.274* 45.554 44.357 45.272 47.891
H/H/L 46.655 45.609 47.100 47.519 46.489 46.489 48.026* 47.759
L/L/L 50.180 49.087 50.184* 48.163 47.567 47.746 46.591 48.921
Note: *Denotes the maximum profit in 8 loop structures under a fluctuation.

profit under M/L/H, H/M/L, H/L/M, H/L/L, L/H/H and L/ assembly systems. Loop structure plays a key role on the
L/H, its robustness reaches 42.86%. No. 5 has the max- performance of the assembly system. There is different
imum profit under L/H/M and L/H/L, and its robustness WIP limit between 1:1 and 1:n conditions, as is shown in
reaches 14.29%. No. 6 has the maximum profit under L/H/ Figures 9 and 10. In addition, different fluctuation combi-
M and M/H/L, and its robustness reaches 14.29%. No. 7 nation makes the WIP limit different, which also means
has the maximum profit under L/M/H and H/H/L, and its that setting an appropriate upper limit is very important for
robustness reaches 14.29%. the loop structure.
The robustness of No. 2 is the strongest in cycle time, The total robustness (Table 6) represents that No. 2
but its robustness is decreased by 85.71% in performance has the strongest robustness and the best ability to cope
of profit. No. 5 and No. 6 are acceptable policy for H/L/M, with fluctuation, the next is No. 4, and then is No. 3. The
and especially No. 5 coping with L/H/M both conditions worst is No. 8 in these schemes of CONWIP loop struc-
1:1 and 1:n well makes maximum profit. ture, which shows that the less amount of loop does not
always get better performance. Observation also shows
that non-optimal performance of cycle time for a loop
4.3. Overall analysis structure is likely to have a better profit, which illustrates
The loop structure not only determinates the control span these eight loop structures have different control
of pull mechanism, but also limits the level of WIP in objective.

Figure 9. WIP limit under 1:1 condition.


International Journal of Computer Integrated Manufacturing 817
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

Figure 10. WIP limit under 1:n condition.

Table 6. Total robustness of 8 loop structures under 1:1 and 1:n conditions.

Robustness No. 1 No. 2 No. 3 No. 4 No. 5 No. 6 No. 7 No. 8

Cycle time 7.14% 64.29% 17.86% 21.43%


Profit 14.29% 14.29% 32.14% 14.29% 14.29% 14.29%
Total 3.57% 39.29% 16.07% 26.79% 7.15% 7.15% 7.15%

Note: Total = 0.5 robustness of cycle time + 0.5 robustness of profit.

P P P
The change of robustness from No. 1 to No. 8 indi- Therefore, j2J λj  j2J ωj maxfCj  dj ; 0g  q j2J
cates that a key factor of profitability is delivery on time τ j λj > 0 and then,
for OKP with tardiness punishment, comparing to pursuit P P
of short cycle time and high throughput purely. A more j2J λj  j2J ωj maxfCj  dj ; 0g
relaxed due date and a shorter cycle time in systems will q< P (4)
j2J τ j λj
lead to lower profits if the cost of inventory is considered.
This article assumes that assembly model has the where maxfCj  dj ; 0g = 0 in (4) if Cj  dj < 0 and τ j
third-party storage, and the storage cost q that is quoted satisfies τ j ¼  minfCj  dj ; 0g ¼ dj  Cj , hence (4) can
by a third-party provider. The storage cost q becomes
be written as:
another important factor to make a profit. Theorem 1
shows if the quoted q is greater than the threshold of q P P
j2J λj j2J λj
that OKP enterprise can afford, it will lead to profits q< P ; let ^q ¼ P (5)
losses. The theorem proving is as follows. j2J ðd j  Cj Þλj j2J ðd j  Cj Þλj

Theorem 1: In the OKP environment, the more relaxed where ^q is a threshold of storage cost. If dj increases, and
due date, the shorter cycle time and lower the threshold of Cj decreases, then dj  Cj increases, that is ^q decreases.
the unit cost of third-party storage per day that OKP can
afford.
4.4. Dynamic configuration of CONWIP control
Proof: According to the formula (2) and (3), the gains WIP limit is an important parameter for CONWIP and is
condition is R  S > 0, that is restricted by the workload of production system. Under
the same assemble system, however, the above results
X    X show that WIP limit setting is restricted by CONWIP
j2J
λj  ωj max Cj  dj ; 0  τ qλ > 0:
j2J j j loop structure. In addition, under the different fluctuation
818 G. Huang et al.

combination, WIP limit (Figures 9 and 10), the shortest Figure 12 is regarded as an expression of distribution
cycle time (Tables 2 and 4) and the maximum profit function, which can be used as dynamic configuration
(Tables 3 and 5) are not identical. An optimal CONWIP rules. On the horizontal axis in Figure 12, the fluctua-
policy is likely to be non-optimal when fluctuation of tion type can be added or deleted according to the
production system has changed. Table 6 shows that some actual situation, and data are also updated regularly.
loop structures have a stronger robustness indeed, such as
No. 2 and No. 4, but CONWIP loop structure with 100% Here, a basic procedure for dynamic configuration apply-
robustness (total = 100%) does not exist. Hence, If ing CONWIP policy set is proposed as follows.
CONWIP policy set is composed of many CONWIP poli- Step 1. Fluctuation type is identified.
cies with stronger robustness, the robustness of CONWIP Step 2. CONWIP loop structure is selected from
policy set will be 100%, which means a feasible CONWIP CONWIP policy set according to Step 1.
policy can be selected based on fluctuation type. That is to Step 3. The optimal WIP of selected loop structure is
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

say, a production system can be controlled by CONWIP set to WIP limit.


policy set dynamically. CONWIP dynamic control and Step 4. Regular monitoring and updating the data, in
dynamic configuration of CONWIP control are different. case of fluctuation changing, Step 1 is to be executed.
The former denotes that the amount of WIP will be There is no doubt that a diversification of policy set is
adjusted according to the system status dynamically, and the premise to realise the dynamic configuration, using the
the latter denotes that CONWIP policy including the loop design pattern to design diversified CONWIP poli-
CONWIP loop structure and the amount of WIP will be cies is a feasible solution.
adjusted according to the system status dynamically. There
are two ways of dynamic configuration in the following.
5. Conclusion and future work
● CONWIP loop structural with a stronger robustness The contribution of this article reveals the fact that only
is selected as CONWIP policy. It adapts to a variety pursuing the shortest cycle time does not increase profit in
of fluctuation dynamically after being configured. OKP environment. Enhancing the accuracy of delivery
But this way is only suitable for a few fluctuations time is much more important to shorten cycle time in
that are just coped with this situation in production OKP. Setting a release date is a method to enhance the
system. accuracy of delivery time. However, due to many factors
● Applying CONWIP policy set to a variety of fluctua- changes in the actual production process, it will result to
tion dynamically. In Section 4.1, for example the tardiness punishment seriously if the release date of orders
analysed results in Table 2 are shown in Figure 11. cannot be complied strictly. Hence, controlling WIP limit
A policy set is composed of No. 2, No. 3 and No. 4, may be more reasonable in practical. Theorem 1 shows
which copes with all of 14 fluctuation combinations. that OKP enterprise cannot blindly adopt the third-party
The shortest cycle time is obtained in each fluctuation storage, an option is to build inventory by itself. Second,
by this way. If the production system shows L/H/M, CONWIP is taken as OKP shop floor control, that is also a
the CONWIP loop structure should select No. 3 and meaningful exploration. The quality of services is affected
WIP limit is set to 15 (see Figure 12). In addition, by the delivery on time for OKP enterprise, and a

Figure 11. Different fluctuations with the same CONWIP control.


International Journal of Computer Integrated Manufacturing 819
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

Figure 12. CONWIP loop structure and WIP limit for different fluctuation.

well-designed CONWIP loop structure can effectively The above problems are overlooked easily in theory
improve the performance of assembly system in OKP. It research, but need to be answered in detail for implemen-
is worthy of mentioning that many studies about assembly tation. In short, these problems will help to develop
systems assume that the assembly proportion is 1:1 con- CONWIP control system that meets the demand of prac-
dition, but this article shows that the performance of the tical enterprise. Especially, the solution of problem (2) is a
assembly system is not always the same under different foundation of realisation for dynamic configuration.
assembly proportion conditions for the same of CONWIP
loop structure. Third the approach of pattern-policy-
scheme is applied to many production environments Acknowledgements
including OKP. Providing a feasible CONWIP scheme is We thank the editors and anonymous referees for their helpful
a prerequisite of successful CONWIP implementation. comments on earlier versions of our article.
CONWIP loop design pattern can generate a great number
of control policies. For the designed CONWIP policies,
Disclosure statement
the response capability test shows that some CONWIP
policies have stronger robustness, and keep up the best No potential conflict of interest was reported by the authors.
performance in many fluctuations. It is realised that
dynamic configuration of CONWIP control may be more
Funding
suitable for OKP environment. Finally, the design proce-
dure of CONWIP in this article provides a reference and This work was supported by the National Natural Science
Foundation of China under [grant number 71571101], [grant
guidance for the application of CONWIP. Meanwhile, the number 71171114], [grant number 71472089] and the
results of this article also help the research based on Fundamental Research Funds for the Central Universities under
virtual supply chain driven by CONWIP. [grant number 30920130111011].
In the future work, a mixed integer programming will
be used to find optimal solution that the completion time
for each order is close to due date. CONWIP loop design References
pattern will be extended into job shop, and needs more Ajorlou, S., and I. Shams. 2013. “Artificial Bee Colony
specification. Furthermore, three main problems need to Algorithm for CONWIP Production Control System in a
be solved in CONWIP implementation, as follows. Multi-Product Multi-Machine Manufacturing Environment.”
Journal of Intelligent Manufacturing 24: 1145–1156.
doi:10.1007/s10845-012-0646-5.
(1) The hardware control technology for CONWIP Aleksić, D. S., D. S. Janković, and L. V. Stoimenov. 2012. “A
Case Study on the Object-Oriented Framework for Modeling
triggering mechanism. Product Families with the Dominant Variation of the
(2) How to construct CONWIP policy based on soft- Topology in the One-of-a-Kind Production.” The
ware method. International Journal of Advanced Manufacturing
(3) Integration supports of problem (1) and (2). Technology 59: 397–412. doi:10.1007/s00170-011-3466-4.
820 G. Huang et al.

Choi, B. K., and N. K. You. 2006. “Dispatching Rules for Dynamic Semiconductor Assembly and Test Factory Based on
Scheduling of One-of-a-Kind Production.” International Performance Evaluation.” Computers & Industrial
Journal of Computer Integrated Manufacturing 19 (4): 383– Engineering 59: 314–322. doi:10.1016/j.cie.2010.05.003.
392. doi:10.1080/09511920500407541. Li, W., B. R. Nault, D. Y. Xue, and Y. L. Tu. 2011. “An Efficient
Dalalah, D., and O. Al-Araidah. 2010. “Dynamic Decentralised Heuristic for Adaptive Production Scheduling and Control in
Balancing of CONWIP Production Systems.” International One-of-a-Kind Production.” Computers & Operations
Journal of Production Research 48 (13): 3925–3941. Research 38 (1): 267–276. doi:10.1016/j.cor.2010.05.002.
doi:10.1080/00207540902998323. Little, J. D. C. 1961. “A Proof of the Queuing Formula: L = λw.”
Framinan, J. M., P. L. González, and R. Ruiz-Usano. 2003. “The Operations Research 9 (3): 383–387. doi:10.1287/
CONWIP Production Control System: Review and Research opre.9.3.383.
Issues.” Production Planning & Control 14 (3): 255–265. Mhada, F., and R. Malhamé. 2011. “Approximate Performance
doi:10.1080/0953728031000102595. Analysis of CONWIP Disciplines in Unreliable Non
Harrod, S., and J. J. Kanet. 2013. “Applying Work Flow Control Homogeneous Transfer Lines.” Annals of Operations
in Make-to-Order Job Shops.” International Journal of Research 182: 213–233. doi:10.1007/s10479-010-0722-1.
Downloaded by [EBSCO Publishing Distribution 2010], [Paige Riordan] at 22:18 15 June 2016

Production Economics 143: 620–626. doi:10.1016/j. Park, C.-W., and H.-S. Lee. 2013. “Performance Evaluation of
ijpe.2012.02.017. a Multi-Product CONWIP Assembly System with
Hong, G., D. Y. Xue, and Y. L. Tu. 2010. “Rapid Identification Correlated External Demands.” International Journal of
of the Optimal Product Configuration and Its Parameters Production Economics 144: 334–344. doi:10.1016/j.
Based on Customer-Centric Product Modeling for One-of- ijpe.2013.02.021.
a-Kind Production.” Computers in Industry 61: 270–279. Pettersen, J.-A., and A. Segerstedt. 2009. “Restricted Work-in-
doi:10.1016/j.compind.2009.09.006. Process: A Study of Differences between Kanban and
Hopp, W. J., and M. L. Spearman. 2001. Factory Physics, CONWIP.” International Journal of Production Economics
Foundations of Manufacturing Management. 2nd ed. Irwin: 118 (1): 199–207. doi:10.1016/j.ijpe.2008.08.043.
McGraw-Hill. Renna, P., L. Magrino, and R. Zaffina. 2013. “Dynamic Card
Huang, G., J. Chen, X. Wang, and Y. Shi. 2014. “A Simulation Control Strategy in Pull Manufacturing Systems.”
Study of CONWIP Assembly with Multi-Loop in Mass International Journal of Computer Integrated
Production, Multi-Products and Low Volume and OKP Manufacturing 26 (9): 881–894. doi:10.1080/
Environments.” International Journal of Production 0951192X.2013.799783.
Research. doi:10.1080/00207543.2014.980458. Riezebos, J. 2010. “Design of POLCA Material Control
Jodlbauer, H., and A. Huber. 2008. “Service-Level Performance Systems.” International Journal of Production Research 48
of MRP, Kanban, CONWIP and DBR Due to Parameter (5): 1455–1477. doi:10.1080/00207540802570677.
Stability and Environmental Robustness.” International Spearman, M. L., D. L. Woodruff, and W. J. Hopp. 1990.
Journal of Production Research 46 (8): 2179–2195. “CONWIP: A Pull Alternative to Kanban.” International
doi:10.1080/00207540600609297. Journal of Production Research 28: 879–894. doi:10.1080/
Khojasteh-Ghamari, Y. 2009. “A Performance Comparison 00207549008942761.
between Kanban and CONWIP Controlled Assembly Suri, R. 1998. Quick Response Manufacturing: A Companywide
Systems.” Journal of Intelligent Manufacturing 20 (6): Approach to Reducing Lead Times. Portland, OR:
751–760. doi:10.1007/s10845-008-0174-5. Productivity Press.
Krishnamurthy, A., and R. Suri. 2009. “Planning and Suri, R., and A. Krishnamurthy. 2003. How to Plan and
Implementing POLCA: A Card-Based Control System for Implement POLCA: A Material Control System for High
High Variety or Custom Engineered Products.” Production Variety or Custom-Engineered Products. Technical Report.
Planning & Control 20 (7): 596–610. doi:10.1080/ Madison: Center for Quick Response Manufacturing,
09537280903034297. University of Wisconsin.
Levalle, R. R., M. Scavarda, and S. Y. Nof. 2013. “Collaborative Tu, Y. L., X. L. Chu, and W. Y. Yang. 2000. “Computer-Aided
Production Line Control: Minimisation of Throughput Process Planning in Virtual One-of-a-Kind Production.”
Variability and WIP.” International Journal of Production Computers in Industry 41: 99–110. doi:10.1016/S0166-
Research 51 (23–24): 7289–7307. doi:10.1080/ 3615(99)00006-8.
00207543.2013.778435. Wortmann, J. C., D. R. Muntslag, and P. J. M. Timmermans.
Li, B. M., and S. Q. Xie. 2013. “Product Similarity Assessment 1997. Customer-Driven Manufacturing. London: Chapman
for Conceptual One-of-a-Kind Product Design: A Weight & Hall.
Distribution Approach.” Computers in Industry 64: 720– Xie, S. Q., and Y. L. Tu. 2006. “Rapid One-of-a-Kind Product
731. doi:10.1016/j.compind.2013.04.001. Development.” The International Journal of Advanced
Li, N., S. Q. Yao, G. Liu, and C. H. Zhuang. 2010. Manufacturing Technology 27: 421–430. doi:10.1007/
“Optimization of a Multi-Constant Work-in-Process s00170-004-2225-1.
Copyright of International Journal of Computer Integrated Manufacturing is the property of
Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted
to a listserv without the copyright holder's express written permission. However, users may
print, download, or email articles for individual use.

You might also like