Professional Documents
Culture Documents
From Loop Structure To Policy Making
From Loop Structure To Policy Making
From loop structure to policy-making: a CONWIP design framework for hybrid flow shop
control in one-of-a-kind production environment
Guodong Huanga*, Jie Chena, Xiuli Wanga, Yiqun Shib and Houping Tiana
a
School of Economics & Management, Nanjing University of Science & Technology, Nanjing, P.R. China; bSchool of Foreign Studies,
Suqian College, Jiangsu, P.R. China
(Received 30 September 2015; accepted 5 September 2016)
A feasible constant work in process (CONWIP) policy can guide developer to better implement CONWIP system. The
feasible policy should be selected from alternatives by evaluation. Therefore, how to generate more than one CONWIP
alternative policy to evaluate is an inevitable problem in CONWIP practice. From the perspective of loop structure, we
propose CONWIP design framework (CDF) which is a systematic design approach to obtain CONWIP alternative poli-
cies. The basic concepts and components for CDF are discussed in this paper. Based on CDF, we make 10 CONWIP
alternative policies for hybrid flow shop in one-of-a-kind production environment, and these alternative policies are eval-
uated by simulation. The simulation result implies that (i) the CONWIP alternative policy with robustness has the poten-
tial to cope with more fluctuations in high-variety production environment; (ii) a better design for CONWIP policy will
be able to enhance the system performance in practice; and (iii) the loop structure can serve as a parameter of CONWIP.
Keywords: CONWIP; OKP; design framework; production control; pull systems
1. Introduction
In the study of constant work-in-process (CONWIP), some researchers have noticed the issues of CONWIP implementa-
tion (Framinan, González, and Ruiz-Usano 2003; Pettersen and Segerstedt 2009). However, the specific approach of
CONWIP implementing is still insufficient. When we develop CONWIP control system for a wire-rope equipment man-
ufacturer located in the east coast of China, the manufacturer emphasises that the feasible policy should be evaluated
from different alternative policies. From the view of project, the request of manufacturer is reasonable and may occur
frequently in CONWIP practice. The parameters setting (e.g. trigger position of CONWIP card, work-in-process (WIP)
limit, etc.), hardware & software installation and even cost accounting are based on the feasible policy, which is viewed
as the blueprint for CONWIP implementing. Due to the fact that the manufacturer is typical one-of-a-kind production
(OKP) enterprise, this paper focuses on how to get a feasible CONWIP policy in OKP environment. We define the shop
floor with OKP characteristics as OKP environment for the remainder of this paper. Some OKP characteristics are sum-
marised as follows.
(1) OKP system and its supply chain are dynamic and complicated. There are great uncertainties for OKP produc-
tion control (Dean, Tu, and Xue 2009).
(2) OKP companies often use an engineer-to-order (ETO) or make-to-order (MTO) marketing strategy. Normally a
pull production system is employed (Tu and Dean 2011).
(3) OKP companies are competing with large manufacturing companies (often conglomerates) by offering the
customised products with shorter lead times and higher quality (Dean, Tu, and Xue 2009).
(4) To gain high production efficiency, OKP companies are suggested to arrange its production in a flow line no
matter that the product is complex or simple (Tu and Dean 2011).
(5) The production line of many OKP companies can be generally considered as a hybrid flow shop (HFS) (Luo
et al. 2010).
The reason why we choose CONWIP as OKP shop floor control is that CONWIP has a simple structure, and the
performance of CONWIP is acceptable and even better in some aspects (e.g. high throughput, short cycle time and low
WIP). Although CONWIP is a pull alternative to Kanban (Spearman, Woodruff, and Hopp 1990), CONWIP system is
more suitable for MTO production than Kanban system (Prakash and Chin 2015). Therefore, CONWIP can satisfy
requirement of characteristics (1) and (2) as mentioned above. Furthermore, CONWIP has the least time delays for deci-
sion-making in MRP (material requirements planning), Kanban and CONWIP (Gong, Yang, and Wang 2014). As a flow
control system, CONWIP has the lowest mean and variance of tardiness than Kanban and POLCA (paired-cell overlap-
ping loops of cards with authorisation) in MTO (Harrod and Kanet 2013). Therefore, CONWIP shows great potential in
satisfying requirement of characteristics (3) and (4) as mentioned above. Obviously, the characteristic (5) mentioned
above is what we are going to consider.
The rest of this paper is organised as follows: Section 2 provides a literature review. A systematic design approach
of CONWIP policy is proposed in Section 3. Based on this approach, CONWIP alternative policies are designed for
HFS in OKP environment in Section 4. These alternative policies are evaluated by simulation in Section 5. In Section 6,
the results are discussed. Section 7 is conclusions and future work.
2. Literature review
At present, one of the production goals is to deliver on time or minimise the missed due-date. The study related to due-
date has been pay attention early in production scheduling. Some researchers discussed job shop scheduling problem
with the due-date (Eilon and Hodgson 1967; Holloway and Nelson 1974; Gee and Smith 1993). Eilon and Chowdhury
(1976) presented that the delivery time can be determined by negotiation with customers. Setting a predictable due-date
can impact the missed due-date performance, and the reasonableness of due-date is also important to meet due-date
(Weeks 1979; Alpay and Yüzügüllü 2009). Combined due-date with dispatching rule is studied to improve the missed
due-date performance (Cheng and Jiang 1998; Tang, Liu, and Liu 2005; Vinod and Sridharan 2011).
Completion time, lead time and set-up time are also considered as influences on due-date in HFS scheduling
problem (Azizoğlu, Çakmak, and Kondakci 2001; Kyparisis and Koulamas 2001; Lee 2009; Mousavi, Zandieh, and
Amiri 2011). Weng, Wei, and Fujimura (2012) presented that the routing strategies can help dispatching rules realise
just-in-time completion for jobs arriving dynamically in HFS. In addition, Ebrahimi, Ghomi, and Karimi (2014) studied
an uncertain problem of due-date. Gupta et al. (2002) considered a generalisation of the permutation flow shop problem
with due-date assignment cost. Branch and bound algorithm, two-phase heuristic algorithm and genetic algorithms are
proposed to minimise the number of tardy jobs (Choi and Lee 2009; Desprez, Chu, and Chu 2009).
Golany, Dar-El, and Zeev (1999) discussed the WIP level and the dispatching rule of order release in a multi-family
and multi-cell manufacturing environment with CONWIP mechanism. Framinan, Ruiz-Usano, and Leisten (2000) stud-
ied the input control and the dispatching rules in a flow shop controlled by CONWIP within a make-to-stock environ-
ment. Combined CONWIP system with learning autonomous pallets, Mehrsai, Karimi, and Scholz-Reiter (2013) built
an exemplary supply network simulation model.
Especially for job shop with CONWIP, Harrod and Kanet (2013) discussed the performance problem of job shop
which is influenced by due-date, queuing rule and pull mechanism. In their opinion, the CONWIP system is recom-
mended, because it is decidedly simpler to implement. Slomp, Bokhorst, and Germs (2009) investigated how lean pro-
duction control principles can be used in make-to-order job shop, and a control system proposed, called as CONWIP/
FIFO/takt time, is applied to a particular manufacturing system. Li (2010) studied the layout change and quality
improvement in job shop with CONWIP control. Ryan, Baynat, and Choobineh (2000) studied the issues related to the
WIP level in a CONWIP controlled job shop. Ryan and Choobineh (2003) described a queuing network model of job
shop operating under single chain CONWIP control and they developed a mathematical model designed to bound
throughput and optimise WIP mix.
In the literature above, the researchers studied many factors that affect delivery on time, such as due-date, dispatch-
ing rule, completion time, lead time, set-up time and so on. In the study of CONWIP, the researchers focused on param-
eter solution and performance comparison with other pull systems. CONWIP is often employed in flow shop and job
shop. However, the study of HFS with CONWIP still lacks, especially for OKP environment.
The original contribution of our study is combining CONWIP theory with OKP practice. CONWIP design frame-
work (CDF) is proposed to obtain CONWIP policy for HFS in OKP environment. We want to extend CONWIP mecha-
nism to OKP shop floor control to improve OKP’s ability to deliver on time, since it is difficult for OKP enterprise to
deliver on time in practice.
implement CONWIP project in different production environments. We hold the opinion that the guidelines of CONWIP
implementation are not only some suggestions, but also a systematic approach. CDF aims to reduce the blindness of
CONWIP policy-making. CDF includes UMD (upstream, midstream and downstream) identification, loop design pattern
(or loop pattern) and pattern refinement (see Figure 1).
In the early stage of CONWIP policy-making, UMD identification is used to classify upstream, midstream and
downstream of production environments or layouts. Based on UMD identification, loop design pattern is used to deter-
mine the macrostructure for CONWIP loop controlling. To get valid loop structure (i.e. the sophisticated loop), the loop
pattern is further refined by loop constraints and generation rules. And then, CONWIP alternative policies are made
according to the generated loop structure and other parameters. Finally, by means of evaluating, one or more feasible
CONWIP policies are selected from these alternatives.
Loop constraint emphasises the actual reason that splits single loop into multiple loops. Generation rule emphasises
the generating principle of valid loop structure (i.e. the generating logic). Essentially, loop constraint and generation rule
are equivalent concepts. Therefore, CDF is also viewed as a principle-parameter approach. To communicate better
between CONWIP policy-makers (or developers) and managers in actual environment, we only need to consider the
loop constraint which is described by natural language (see Section 3.3.1). However, the generation rule, as a logical
description, is also important to develop auto-generating system of CONWIP loop (see Section 3.3.2).
abstract class. UMD can be identified by partitioning or clustering behaviours according to certain attributes. The class
property of UMD can be described by unified modelling language that is an important modelling approach. In addition,
UMD is also viewed as a set. The number of elements is greater than or equal to zero in the subsets U, M and D.
Obviously, factory, shop floor, department, stage and work centre have the property of set.
Some examples are shown in Figure 2 to illustrate UMD identification. In Figure 2(a), the flow assembly system has
higher fluctuation at interface point (IP), since parts and components are gathered to assemble at IP where bottleneck is
formed easily. To avoid bottleneck shifting, IP should be considered as midstream controlled by CONWIP indepen-
dently. Figure 2(b) is a U-shaped line or cellular layout, in which the first workstation and the last workstation are con-
sidered as upstream and downstream, respectively, and the other workstations are considered as midstream. The UMD
identification can also be applied in supply chain, production network or HFS (see Figure 2(c)). The midstream can be
neglected specially for the HFS with small size (see Figure 2(d)).
In some cases, the flow direction of job may be opposite. For example, a work centre in job shop belongs to
upstream for a job, but it also belongs to downstream for another job. To address this issue, the following solutions are
provided for reference.
• Solution 1. The whole job shop belongs to midstream, jobs enter into job shop in upstream and finished goods
leave job shop in downstream (see Figure 3(a)). Upstream and downstream are CONWIP trigger positions (i.e.
check-in/ check-out area).
• Solution 2. Each work centre in job shop is considered as midstream. Jobs enter into work centre in upstream and
leave work centre in downstream (see Figure 3(b)). Solution 2 is an extension of Solution 1.
• Solution 3. Job shop problem can be transformed into flow shop problem by adding machines. Here we reference
to an example from Framinan and Ruiz-Usano (2002) to explain this solution.
A job shop that cyclically processes four types of jobs (J 1 to J 4 ) in three machines M 1 to M 3 is presented according
to the following routing matrix:
0 1
1 2 3
B3 2 1C
M i;j B
¼@ C
3 1 2A
2 3 1
where ði; jÞ values indicate the operation that the i-th job requires in the j-th order. This job shop layout can be
transformed into a flow shop consisting of M 3 -M 1 -M 2 -M 3 -M 1 machines by purchasing new machines of types M 1 and
M 3 . The new matrix is shown in the following.
0 M3 M1 M2 M3 M1 1
x x x
J 1B C
B x C
J 2B x x C
B C
J 3B x x x C
@ A
J4
x x x
where x in the ði; jÞ position indicates that job i is processed in machine j.
International Journal of Production Research 3379
In their example, M 3 -M 1 -M 2 -M 3 -M 1 can be taken as five stages where each stage has a single machine. Inspired by
their work, we suggest using a dynamic dispatching rule (DDR) for each buffer of machine (see Figure 3(c)). If a job
needs to be processed in a machine, this job should be sent to this machine’s buffer first to queue according to the
shortest processing time. Otherwise, this job is directly sent to the next machine’s buffer.
• Solution 4. Purchasing additional machine is heavy burden for small- and medium-sized enterprises. Guinet and
Legrand (1998) proposed a method that the fictitious operation is used to transform job shop scheduling problem
into flow shop problem. Based on their work, we present a solution without adding machine. This solution
requires DDR and circular machine layout which is a special flow shop with closed-loop (see Figure 3(d)).
Solution 4 is an improved version of Solution 3.
d. U/ M/D
Example 1: U/D implies that production system is divided into upstream and downstream.
Example 2: U^M/M^D denotes that upstream and a part of midstream are controlled by one loop, and the down-
stream and the rest of midstream are controlled by another one.
Example 3: U^M/qM/D implies that CONWIP is cancelled for a part of midstream. The composite symbol /q means
a push–pull interface.
Example 4: U/_ni¼1 M ðiÞ /D shows that the midstream is controlled
by n loops. Furthermore, if takt time is known, a
constraint expression can be written as l U = _ni¼1 M ðiÞ =D ¼ takt, where μ denotes interval time between outputs
and i denotes the i-th loop in midstream.
Note that the number of machines, facility layout (e.g. series, parallel and series-parallel) and other parameters in
upstream, midstream and downstream should not be discussed in loop design pattern. These details should be considered
further in patter refinement.
• Production status
- bottleneck drift restricting
- frequency of fluctuation
- machine capacity
- workload balancing
• Production environment
- work study/process analysis/shop-floor activities
- operator skill
- facility layout
- flow shop/job shop
- workshop size
- the number of production resources
- importance and role for storage
- material handling and supply mechanism
- make-to-order/make-to-stock/assemble-to-order/engineering-to-order
• Production requirement
- quality
- process routing
• Administration authority
- span of control and hierarchy of management
- permissions and authorisation for workshop manager
- departmental boundaries/business scope/work content
- setting push–pull interface for business requirement
(2) Outside the enterprise
• Policy
- government policy/legal requirements
- patent barrier
- industry standards
- special and temporary regulatory measures
Note that the loop constraint can also be used as reference for UMD identification. However, the loop constraint
focuses on the control span of CONWIP loop. UMD identification focuses on the allocation of production resources.
For example, if a purchasing department is upstream and a manufacturing department is downstream, CONWIP loop
has to be limited in its own department, respectively. This situation is U/D pattern. When general manager authorises
the head of manufacturing department to manage the purchasing business, the two departments can be completely con-
trolled by single loop, that is U^D pattern. In this example, these two departments are production resources and the
authorisation from general manager is a key constraint for CONWIP loop.
The corresponding relation between set, sequence, loop structure and generation rule is shown in Table 1, in which
the valid sequence is regarded as the loop structure, and a machine is the element (or point) from the sequence (or set).
In Table 1, the fifth row indicates that a valid loop structure with U/qM/D pattern is generated by Rules 1, 3 and Rules
5 to 7 in the following.
Rule 5: there exists hbi, where
m i, mj ∈ b, i < j 8i; j ¼ 1; 2; . . .; j M j.
Rule 6: there exists bp ; bq , where ml 2 bp ; mk 2 bq ; p 6¼ q 8p; q ¼ 1; 2; . . .; jGj,
l\k 8k; l ¼ 1; 2; . . .; j
M j.
Rule 7: If there exists minfjbjg , then q operator is activated in the corresponding loop.
b2G
In CDF, the non-uniqueness of loop design pattern is important to ensure the variety of alternative policy. Due to
the fact that the analysis of loop structure can be converted into the analysis of point-set characteristics, we further
demonstrate that the loop pattern is not unique, at least in theory, based on concepts of the point-set topology (see
below).
Definition 1. Suppose X is a set. A topology on X is a collection τ of subsets of X such that the following are true:
(1) X ; / 2 s.
(2) If V1 ∈ τ and V2 ∈ τ, then V1 ∩ V2 ∈ τ.
(3) If τ1 τ, then [ V 2 s.
V 2s1
We call the pair ðX ; sÞ a topological space, and the elements of τ are called the open subsets of the topological space
ðX ; sÞ.
Definition 2. Let ðX ; sÞ be a topological space. A base for the topology τ is a subcollection B s such that every
element of τ is the union of some subcollection of B. The elements of B are called basic open sets. When the topology
is understood, we say B is a base for ðX ;sÞ
Theorem 1. Let B be a base for a topology on X . The topology τ generated by B is the collection of all subsets V
of X such that if x ∈ V, there exists an element b 2 B with x ∈ b V.
Here the proof of Theorem 1 is omitted (see Davis 2005). Theorem 1 presents that the topology τ is generated if the
base B of ðX ; sÞ is given firstly. Definition 2 presents that the number of open sets in the base B is much less than the
topology τ generally. When the open sets of base satisfy a certain property, the open set in topology also holds for this
property. The base plays an important role in reduction.
Suppose
that a topological space ðX ; =Þ that satisfies Definition 1 is known, where
X ¼ mij ji ¼ 1; 2; . . .; r; j ¼ 1; 2; . . .; k and mij denotes the j-th machine in the i-th region, for i = 1, 2, … , r. It is
clear that X is a non-empty set and = = {all subsets of X }, thus = is a topology of X . According to Theorem 1 and
Definition 2, there exists a base H that generates the topology = in ðX ; =Þ. Let H ¼ fH1 ; H2 ; . . .; Hi ; . . .; Hr g and
mij ∈ Θi, where Θi is defined as follows.
8
< U ; if 1 i\a
Hi , M ; if a i\b ;
:
D; if b i r
where parameters a, b ∈ N+ are the boundary between U (upstream), M (midstream) and D (downstream). The process
of defining Θi is a so-called UMD identification. Every element of = is the union of some subcollection of H. If =0
denotes a collection of some subsets of =, =0 is a so-called loop pattern. Obviously, there is more than one =0 .
policies have intra-group homogeneity and inter-group heterogeneity (see Table 2), which can bring advantage to select
feasible policy. We set other parameters as follows.
• Dispatching rule: CR + OSLK (Critical ratio and operation slack).
• Card type: Independent card, S-closed, S-loop and M-loop.
• WIP level: through a full factorial experiment.
Figure 5. (Continued).
No. 1 ○
No. 2 ○
No. 3 ○
No. 4 ○
No. 5 ○
No. 6 ○
No. 7 ○
No. 8 ○
No. 9 ○
No. 10 ○
3386 G. Huanga et al.
In this formula, di is the due-date assigned to job i, ri is the arrival time of job i, pij is the processing time of the j-th
operation of job i and ni is the total number of operations of job i. μp and μg are the average operation processing time
and the average number of operations per job, respectively. ft denotes the average flow time of a job when the work
load is steady state in the work shop.
dj Due-date of job j.
Lj Lateness of job j, let Lj = Cj – dj.
Tj Tj Tardiness of job j, let Tj ¼ max Lj ; 0 .
τj Storage time of job j, letsj ¼ min Lj ; 0 .
q Unit cost of storage per day.
The average net revenue (ANR), the standard deviation of lateness (SDL) and the semi-quadratic lateness (SQL) are
used to evaluate performance. SDL and SQL were applied to the tardiness problem (Gee and Smith 1993; Cheng and
Jiang 1998; Vinod and Sridharan 2011).
1X n
1 X n
ANR ¼ ðkj xj Tj Þ sj qkj ; (1)
n j¼1 N
~
j¼1
~ ¼ jLj \0; j 2 J , and N
where N ~ denotes the total number of elements in N ~ set.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn
2
j¼1 Lj L
SDL ¼ ; (2)
n1
the average lateness.
where Lis
8P n
>
> Lj =n if Lj \0
<
j¼1
SQL ¼ P (3)
>
>
n
: L2j otherwise
j¼1
In formula (1), we generated kj randomly from the discrete uniform distribution in the interval [1,100], and ωj from
the discrete uniform distribution in the interval [1,10]. The value for q is 0.01.
where Ai,p = 1 if the policy p has the best performance under fluctuation i, and 0 otherwise; and Nf denotes the total
number of fluctuations.
As shown in Figure 7, No.2 has the minimum SDL under L/M/H, M/L/H, H/M/L, H/H/H, H/L/L and H/L/H. No.9
has the minimum SDL under L/H/M, L/H/L and H/H/L. No.5 and No.7 have the minimum SDL under M/H/L. No.6 has
the minimum SDL under H/L/M and L/L/H. No.10 has the minimum SDL under L/H/H. No.2 and No.7 have the
minimum SDL under L/L/L. From the point of view of robustness, No.2 reaches 50%, No.9 reaches 21% and No.5
reaches 7%.
As shown in Figure 8, No.2 has the minimum SQL under L/M/H, M/L/H, H/M/L, H/L/M, H/H/H, H/L/L and L/L/H.
No.9 has the minimum SQL under L/H/M, M/H/L and H/H/L. No.10 has the minimum SQL under L/H/H and L/H/L.
No.7 has the minimum SQL under H/L/H. No.6 has the minimum SQL under L/L/L. From the point of view of
3388 G. Huanga et al.
L/M/H 455.485 ± 9.900 478.238 ± 8.694 472.546 ± 9.482 488.221 ± 9.390 473.556 ± 8.858
L/H/M 472.318 ± 10.459 467.251 ± 9.025 461.973 ± 9.639 459.447 ± 9.762 458.223 ± 9.523
M/L/H 479.872 ± 9.655 468.934 ± 8.088 457.100 ± 8.088 465.740 ± 8.894 453.587 ± 8.466
M/H/L 460.785 ± 9.995 451.361 ± 9.466 466.213 ± 9.580 451.128 ± 10.020 449.998 ± 8.497
H/M/L 482.974 ± 10.130 467.791 ± 7.992 458.847 ± 8.411 462.147 ± 9.408 458.123 ± 8.037
H/L/M 479.249 ± 9.277 468.254 ± 8.061 447.451 ± 8.605 474.998 ± 8.873 457.780 ± 7.942
H/H/H 520.298 ± 11.191 491.211 ± 10.036 519.985 ± 10.632 522.001 ± 11.479 601.111 ± 10.354
H/L/L 459.011 ± 9.995 451.297 ± 7.207 462.111 ± 8.498 479.997 ± 9.303 460.369 ± 8.064
L/H/H 510.110 ± 10.696 523.124 ± 9.883 530.581 ± 9.914 532.255 ± 10.737 500.256 ± 9.500
H/L/H 460.899 ± 10.059 451.132 ± 8.200 456.655 ± 8.971 460.141 ± 9.488 451.998 ± 8.485
L/H/L 459.785 ± 9.96 439.568 ± 9.123 460.774 ± 10.017 458.699 ± 9.764 510.002 ± 9.944
L/L/H 457.698 ± 9.684 429.544 ± 8.456 431.271 ± 8.551 431.258 ± 8.782 427.663 ± 8.223
H/H/L 470.451 ± 10.444 455.178 ± 9.569 460.855 ± 9.890 508.988 ± 10.569 429.561 ± 9.402
L/L/L 432.118 ± 8.784 399.874 ± 7.000 425.011 ± 7.691 428.196 ± 8.279 400.255 ± 7.256
No. 6 No. 7 No. 8 No. 9 No. 10
L/M/H 473.540 ± 8.858 443.604 ± 8.864 472.223 ± 9.482 500.121 ± 9.999 475.446 ± 8.821
L/H/M 475.632 ± 9.361 481.787 ± 9.025 469.772 ± 9.621 472.156 ± 8.518 465.987 ± 8.912
M/L/H 450.688 ± 8.467 462.100 ± 8.340 465.263 ± 9.018 458.669 ± 9.582 466.122 ± 9.068
M/H/L 450.145 ± 9.153 446.236 ± 8.497 478.669 ± 9.560 468.233 ± 8.639 469.147 ± 9.230
H/M/L 449.002 ± 8.279 464.869 ± 8.071 452.030 ± 8.899 472.560 ± 8.919 458.988 ± 9.247
H/L/M 486.101 ± 7.889 446.211 ± 7.895 468.902 ± 8.330 459.980 ± 8.868 462.690 ± 8.980
H/H/H 498.955 ± 10.123 499.788 ± 10.190 516.103 ± 10.804 514.207 ± 10.632 515.000 ± 10.632
H/L/L 472.332 ± 7.591 474.104 ± 7.875 463.100 ± 8.668 458.550 ± 8.498 450.981 ± 8.681
L/H/H 512.124 ± 9.883 532.124 ± 9.883 522.163 ± 9.925 500.098 ± 9.914 589.667 ± 9.361
H/L/H 450.000 ± 8.751 446.889 ± 8.382 452.117 ± 9.218 498.001 ± 10.244 475.588 ± 10.109
L/H/L 493.125 ± 9.578 490.563 ± 9.478 503.450 ± 10.022 425.269 ± 8.487 433.630 ± 8.754
L/L/H 420.021 ± 7.902 422.236 ± 8.068 426.866 ± 8.551 456.882 ± 9.361 422.128 ± 8.551
H/H/L 432.331 ± 9.553 435.111 ± 9.686 439.104 ± 9.890 423.826 ± 8.681 458.980 ± 9.890
L/L/L 398.989 ± 7.091 392.699 ± 7.001 422.300 ± 7.691 422.201 ± 7.691 422.020 ± 7.691
robustness, No.2 reaches 50% and No.9 reaches 21%. However, No.5 has not the strongest robustness under any fluctu-
ations in terms of SQL.
By observing the simulation results, some practical implications are presented below.
• In the production environment with high fluctuation (e.g. OKP shop floor), M-loop policy needs to be optimised
to enhance system performance, since the simulation result shows that there is performance difference between
CONWIP loop policies, and not all of the M-loop policies are superior to the classical CONWIP policy in terms
of a certain indicator (e.g. ANR).
International Journal of Production Research 3389
• In the simulation results, there is no alternative policy that has best performance under 14 fluctuations for all of
indicators. This means that a CONWIP policy is hard to keep the best performance in all situations. Therefore, in
order to make a special CONWIP policy, enterprise characteristics (see Section 3.3.1), shop characteristics (e.g.
production organisation, work station, workload, capacity, utilisation rate, process routing, facility layout, etc.) and
order characteristics (e.g. due-date, process time, operations, bill of materials, etc.) need to be considered in actual
environment.
• The simulation results also reflect that some CONWIP policies have strong robustness. Therefore, selecting the
strongest robustness policy as feasible policy is more adaptable than others.
• Similarly, the policy combination can be considered as another type of the feasible policy. Obviously, the collec-
tion of policies with strong robustness has more adaptable than the single policy with strong robustness.
• Some policies can be replaced by each other, when the performance difference between some alternatives and the
feasible policy is acceptable. In practice, the feasible policy may be changed during and after its implementation.
In this case, the replaceability of CONWIP policy can provide a compromise way.
for HFS and evaluated by simulation in OKP environment. Our study shows that some alternative policies with strong
robustness can cope with more fluctuations, which means that a better design for CONWIP policy will be able to
enhance the system performance in practice, and the loop structure which is regarded as one of the CONWIP parameters
can be optimised.
Kagermann, Wahlster, and Helbig (2013) pointed out that it is possible to manufacture one-off items and have very
low production volumes (batch size of 1) whilst still making a profit, engineering processes can be made more agile
and manufacturing processes can be changed. From this point of view, our study is meaningful. Taking CONWIP as
OKP shop floor control is worth further study. CDF also needs to be validated more in application and improved in the
future work.
Acknowledgements
We thank the editors and anonymous referees for their helpful comments on earlier versions of our article.
Disclosure statement
No potential conflict of interest was reported by the authors.
Funding
This work was supported by the National Natural Science Foundation of China under [grant number 71571101], [grant number
71171114], [grant number 71472089] and the Fundamental Research Funds for the Central Universities under[grant number
30920130111011].
References
Alpay, S., and N. Yüzügüllü. 2009. “Dynamic Job Shop Scheduling for Missed Due Date Performance.” International Journal of
Production Research 47 (15): 4047–4062.
Azizoğlu, M., E. Çakmak, and S. Kondakci. 2001. “A Flexible Flowshop Problem with Total Flow Time Minimization.” European
Journal of Operational Research 132 (3): 528–538.
Cheng, T. C. E., and J. Jiang. 1998. “Job Shop Scheduling for Missed Due-date Performance.” Computers & Industrial Engineering
34 (2): 297–307.
Choi, H., and D. Lee. 2009. “Scheduling Algorithms to Minimize the Number of Tardy Jobs in Two-stage Hybrid Flow Shops.”
Computers & Industrial Engineering 56 (1): 113–120.
Davis, S. 2005. Topology. International ed. New York: McGraw Hill Higher Education.
Dean, P. R., Y. L. Tu, and D. Xue. 2009. “An Information System for One-of-a-kind Production.” International Journal of Production
Research 47: 1071–1087.
Desprez, C., F. Chu, and C. Chu. 2009. “Minimising the Weighted Number of Tardy Jobs in a Hybrid Flow Shop with Genetic
Algorithm.” International Journal of Computer Integrated Manufacturing 22 (8): 745–757.
Ebrahimi, M., S. M. T. F. Ghomi, and B. Karimi. 2014. “Hybrid Flow Shop Scheduling with Sequence Dependent Family Setup Time
and Uncertain Due Dates.” Applied Mathematical Modelling 38 (9–10): 2490–2504.
Eilon, S., and I. G. Chowdhury. 1976. “Due Dates in Job Shop Scheduling.” International Journal of Production Research 14 (2):
223–237.
Eilon, S., and R. M. Hodgson. 1967. “Job Shops Scheduling with Due Dates.” International Journal of Production Research 6 (1):
1–13.
Framinan, J. M., and R. Ruiz-Usano. 2002. “On Transforming Job-shops into Flow-shops.” Production Planning & Control 13 (2):
166–174.
Framinan, J. M., R. Ruiz-Usano, and R. Leisten. 2000. “Input Control and Dispatching Rules in a Dynamic CONWIP Flow-shop.”
International Journal of Production Research 38 (18): 4589–4598.
Framinan, J. M., P. L. González, and R. Ruiz-Usano. 2003. “The CONWIP Production Control System: Review and Research Issues.”
Production Planning & Control 14 (3): 255–265.
Gee, E. S., and C. H. Smith. 1993. “Selecting Allowance Policies for Improved Job Shop Performance.” International Journal of
Production Research 31 (8): 1839–1852.
Golany, B., E. M. Dar-El, and N. Zeev. 1999. “Controlling Shop Floor Operations in a Multi-family, Multi-cell Manufacturing
Environment through Constant Work-in-process.” IIE Transactions 31 (8): 771–781.
Gong, Q., Y. Yang, and S. Wang. 2014. “Information and Decision-making Delays in MRP, KANBAN, and CONWIP.” International
Journal of Production Economics 156: 208–213.
International Journal of Production Research 3391
Guinet, A., and M. Legrand. 1998. “Reduction of Job-shop Problems to Flow-shop Problems with Precedence Constraints.” European
Journal of Operational Research 109 (1): 96–110.
Gupta, J. N. D., K. Krüger, V. Lauff, F. Werner, and Y. N. Sotskov. 2002. “Heuristics for Hybrid Flow Shops with Controllable
Processing times and Assignable Due Dates.” Computers & Operations Research 29 (10): 1417–1439.
Harrod, S., and J. J. Kanet. 2013. “Applying Work Flow Control in Make-to-order Job Shops.” International Journal of Production
Economics 143 (2): 620–626.
Holloway, C. A., and R. T. Nelson. 1974. “Job Shop Scheduling with Due Dates and Variable Processing times.” Management
Science 20 (9): 1264–1275.
Huang, G., J. Chen, X. Wang, and Y. Shi. 2015. “A Simulation Study of CONWIP Assembly with Multi-loop in Mass Production,
Multi-products and Low Volume and OKP Environments.” International Journal of Production Research 53 (14): 4160–4175.
Huang, G., J. Chen, X. Wang, and Y. Shi. 2016. “An Approach of Designing CONWIP Loop for Assembly System in One-of-a-kind
Production Environment.” International Journal of Computer Integrated Manufacturing 29 (7): 805–820.
Kagermann, H., W. Wahlster, and J. Helbig. 2013. Recommendations for Implementing the Strategic Initiative Industrie 4.0. acatech,
Frankfurt, 13–78.
Khojasteh-Ghamari, Y. 2009. “A Performance Comparison between Kanban and CONWIP Controlled Assembly Systems.” Journal of
Intelligent Manufacturing 20 (6): 751–760.
Kyparisis, G. J., and C. Koulamas. 2001. “A Note on Weighted Completion Time Minimization in a Flexible Flow Shop.” Operations
Research Letters 29 (1): 5–11.
Lee, G. 2009. “Estimating Order Lead times in Hybrid Flowshops with Different Scheduling Rules.” Computers & Industrial
Engineering 56 (4): 1668–1674.
Li, J. 2010. “Simulation Study of Coordinating Layout Change and Quality Improvement for Adapting Job Shop Manufacturing to
CONWIP Control.” International Journal of Production Research 48 (3): 879–900.
Luo, X., W. Li, Y. Tu, D. Xue, and J. Tang. 2010. “Optimal Resource Allocation for Hybrid Flow Shop in One-of-a-kind
Production.” International Journal of Computer Integrated Manufacturing 23 (2): 146–154.
Mehrsai, A., H. Karimi, and B. Scholz-Reiter. 2013. “Toward Learning Autonomous Pallets by using Fuzzy Rules, Applied in a
CONWIP System.” The International Journal of Advanced Manufacturing Technology 64 (5–8): 1131–1150. doi: 10.1007/
s00170-012-4057-8.
Mousavi, S. M., M. Zandieh, and M. Amiri. 2011. “An Efficient Bi-objective Heuristic for Scheduling of Hybrid Flow Shops.” The
International Journal of Advanced Manufacturing Technology 54 (1–4): 287–307.
Pettersen, J., and A. Segerstedt. 2009. “Restricted Work-in-process: A Study of Differences between Kanban and CONWIP.” Interna-
tional Journal of Production Economics 118 (1): 199–207.
Prakash, J., and J. F. Chin. 2015. “Modified CONWIP Systems: A Review and Classification.” Production Planning & Control 26
(4): 296–307.
Ryan, S. M., and F. F. Choobineh. 2003. “Total WIP and WIP Mix for a CONWIP Controlled Job Shop.” IIE Transactions 35 (5):
405–418.
Ryan, S. M., B. Baynat, and F. F. Choobineh. 2000. “Determining Inventory Levels in a CONWIP Controlled Job Shop.” IIE
Transactions 32 (2): 105–114.
Slomp, J., J. A. C. Bokhorst, and R. Germs. 2009. “A Lean Production Control System for High-variety/Low-volume Environments:
A Case Study Implementation.” Production Planning & Control 20 (7): 586–595.
Spearman, M. L., D. L. Woodruff, and W. J. Hopp. 1990. “CONWIP: A Pull Alternative to Kanban.” International Journal of
Production Research 28 (5): 879–894.
Tang, L., W. Liu, and J. Liu. 2005. “A Neural Network Model and Algorithm for the Hybrid Flow Shop Scheduling Problem in a
Dynamic Environment.” Journal of Intelligent Manufacturing 16 (3): 361–370.
Tu, Y., and P. Dean. 2011. One-of-a-kind Production. London: Springer.
Vinod, V., and R. Sridharan. 2011. “Simulation Modeling and Analysis of Due-date Assignment Methods and Scheduling Decision
Rules in a Dynamic Job Shop Production System.” International Journal of Production Economics 129 (1): 127–146.
Weeks, J. K. 1979. “A Simulation Study of Predictable Due-dates.” Management Science 25 (4): 363–373.
Weng, W., X. Wei, and S. Fujimura. 2012. “Dynamic Routing Strategies for JIT Production in Hybrid Flow Shops.” Computers &
Operations Research 39 (12): 3316–3324.
Copyright of International Journal of Production Research 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.