You are on page 1of 7

Computers & Industrial Engineering 163 (2022) 107856

Contents lists available at ScienceDirect

Computers & Industrial Engineering


journal homepage: www.elsevier.com/locate/caie

Production planning method for seru production systems under


demand uncertainty
Yuya Fujita a, *, Kazuhiro Izui a, Shinji Nishiwaki a, Zhe Zhang b, Yong Yin c
a
Department of Mechanical Engineering and Science, Kyoto University, Kyoto daigaku-katsura, Nishikyo-ku, Kyoto 615-8530, Japan
b
School of Economics and Management, Nanjing University of Science and Technology, Nanjing 210094, PR China
c
Graduate School of Business, Doshisha University, Karasuma-Imadegawa, Kamigyo-ku, Kyoto 602-8580, Japan

A R T I C L E I N F O A B S T R A C T

Keywords: The seru production system is a new-type manufacturing system arising from Japanese production practices that
Seru loading can simultaneously achieve high efficiency, flexibility, and responsiveness. In this paper, a worker and pro­
Production planning duction quantity allocation optimization method for multi-product seru production systems under demand un­
certainty is proposed. To handle such demand uncertainty, the Monte Carlo technique is employed and the
optimization problem is formulated as a bi-level mixed-integer linear programming problem. At the first step, a
worker allocation problem is solved before demand realization is observed. As the second optimization step, the
production quantity allocation is then conducted based on the observed demand realization. The proposed
method is applied to several numerical problems to illustrate its effectiveness.

1. Introduction addition, this system can reduce the inventory of work-in-process and
finished products, the workforce, lead time, setup time, and shop floor
Recently, since the product life cycle is getting remarkably shorter space, resulting in improved productivity (Sakazume, 2005, 2012;
and customer demand is diversing, a manufacturing system must be Stecke, Yin, Kaku, & Murase, 2012). Consequently, the seru production
constructed adopting high-mix low-volume production (Kaku, 2017; system is suitable for the environment in which the demand changes
Yin, Stecke, & Li, 2018). In such a fluctuating environment, traditional frequently (Yin et al., 2018). Since the first introduction of the seru
line production, which is suitable for low-mix high-volume, can no production system by SONY in 1992, it has been adopted by major
longer be adaptable to such unstable market demand. Therefore, the electronic companies and has brought great benefits (Liu, Stecke, Lian, &
“Seru production system” has attracted attention and been adopted, Yin, 2014; Stecke et al., 2012; Yin et al., 2018; Yin, Stecke, Swink, &
especially among the electronics product manufacturers (Kaku, 2016; Kaku, 2017). Due to its practicality, various studies to obtain optimal
Villa & Taurino, 2013; Yin et al., 2018). configurations of the seru production system have already been con­
The seru production system is a flexible production system which ducted. Aboelfotoh and Md Abdullah (2018) studied the relationship
evolved as an alternative to lean system approaches. The production between the worker skill level of each task and the advisable
system consists of many assembly units called “seru.” In each seru unit, manufacturing strategy, either seru or assembly line. Süer, Ulutas, Kaku,
only one or a few workers are in charge of multiple manufacturing tasks and Yin (2019) proposed to take the product life cycle stage and worker
to fabricate a full product or complete assembly. Since each seru unit is skill level variation into consideration in constructing a seru production
composed to be simple and compact, seru production systems are easy to system. Liu, Li, Lian, and Yin (2012) developed a mathematical model
reconfigure and have high ability to adapt to demand fluctuation. This where the assembly line is reconfigured to a seru-line combined or pure
production system can conduct high-mix production by changing the seru system to optimize the number of serus and worker allocation aiming
products to be manufactured for each seru. Furthermore, since each seru to minimize the makespan. Furthermore, they presented an application
unit is simple and compact, a seru production system can be recon­ of the model in an industrial case. Yılmaz (2020) addressed the workforce
structed very quickly. Owing to the above features, this production sys­ scheduling problem aiming to minimize the makespan taking worker
tem can adjust the type and quantity of products easily and swiftly. In transfer between serus into account using a GA-based algorithm.

* Corresponding author.
E-mail address: fujita.yuya.j20@kyoto-u.jp (Y. Fujita).

https://doi.org/10.1016/j.cie.2021.107856
Received 12 February 2021; Received in revised form 16 October 2021; Accepted 26 November 2021
Available online 2 December 2021
0360-8352/© 2021 Elsevier Ltd. All rights reserved.
Y. Fujita et al. Computers & Industrial Engineering 163 (2022) 107856

Furthermore, since multiple design criteria should be considered in the yield of each seru is almost proportional to the number of assigned
early design phase of the manufacturing system, multiobjective optimi­ workers. Therefore, the number of workers and the production quantity
zation techniques have been also developed for seru configuration allocated to each seru unit are the most important decision variables
problems. Liu et al. (2013) studied a multi-objective optimization tech­ when constructing seru production systems.
nique for worker allocation and training planning problem using a heu­ In this paper, we consider a design problem of a seru production
ristic algorithm aiming to minimize the total training cost and to balance system for multi-generation series products as shown in Fig. 1 where
the processing time of workers in each seru. Yu, Tang, Gong, Yin, and “downward substitution” structure (Hsu & Bassok, 1999) is assumed. In
Kaku (2014) proposed the optimization technique of the number of serus, this assumption, newest generation products are assembled in the seru
worker-to-seru allocation, and batch-to-seru allocation for minimizing labeled with the greatest number. High-skilled workers are allocated to
total labor hours and total throughput time using an NSGA-II based al­ older serus labeled with a smaller number and can assemble products of
gorithm. Yu, Wang, Ma, and Sun (2018) formulated the worker allocation multiple generations. Meanwhile, novice workers in new serus labeled
and loading problem to balance workload among all workers and among with a greater number can only assemble smaller kinds of product
all serus as a bi-objective optimization problem and solved it using an generations. Such implementation of a seru production system can be
improved exact algorithm. Lian, Liu, Li, and Yin (2018) studied a bi- often observed in automobile suppliers.
objective optimization problem to balance workload among all workers All serus are assumed to have existing workers who were allocated to
and among all serus using NSGA-II taking workers’ skill sets and profi­ each seru in a previous production period, and these workers can move
ciency levels into consideration. Yu, Tang, Sun, Yin, and Kaku (2013) to a new seru. Also, new additional workers can be allocated to serus via a
presented a multi-objective line-cell conversion model with two objective certain training process. Consequently, in this design problem, the
functions, reducing the number of workers and minimizing the total number of moved workers and new additional workers must be deter­
throughput time, and provided an improved exact algorithm for small- mined. Furthermore, the production quantity in each seru is also a de­
scale problems to obtain Pareto-optimal solutions. cision variable.
As discussed above, a seru production system is a flexible This paper considers a seru production system for a build-to-order
manufacturing system and effective in the fluctuating environment. In production scheme, and Fig. 2 shows the decision flow for this pro­
addition, since the demand is stochastic in practice (Abolghasemi, Beh, duction. First, only worker allocation is determined taking the variation
Tarr, & Gerlach, 2020; Güler & Bilgiç, 2009), consideration of uncer­ of demand into consideration before the realization is observed, since
tainty is very important to build a production system considering the training of workers must be completed by the start of production. Next,
stochastic characteristic (Wang & Tang, 2018). Many reports which the demand realization is observed, and production quantity allocation
considered uncertainty in various stages of decision making in the design is determined. Then, the production is conducted along the determined
of manufacturing systems have been published (Gabrel, Murat, & Thiele, plan. Therefore, the optimization method proposed in this paper consists
2014), and demand uncertainty is one of the most important topics of of two stages: (1) worker allocation optimization stage and (2) pro­
robust optimization. For examples, recently, Hu and Hu (2018) discussed duction quantity allocation stage. Since the first optimization stage must
a stochastic programming technique for lot-sizing and scheduling prob­ be conducted before demand realization is observed, an optimization
lems where demand is a random variable. Thevenin, Adulyasak, and problem must be formulated considering such demand uncertainty.
Cordeau (2020) proposed optimal material requirement planning which
can handle uncertain demands. Mondal and Roy (2021) and Tolooie, 3. Mathematical model and optimization method
Maity, and Sinha (2020) handled mixed uncertainty including demand in
supply chain design. Chen, Xu, Zou, Peng, & Li, 2021; Xu, Hao, & Zheng, 3.1. Notation
2020 and Sun, Wang, and Xue (2021) formulated multiobjective robust
optimization problems under uncertain environments. However, there Definition of notation used in this paper is as follows.
are few reports for seru production planning under demand uncertainty Indexes:
that discuss worker and production quantity allocation, which are the
i, k index of seru
most important decision variables in seru production.
j index of product generation
Therefore, the aim of this study is to propose the method to obtain
optimal worker and production quantity allocation under the demand
uncertainty environment. In order to consider such uncertainty in the Integer decision variables:
optimization problem, a Monte Carlo approach is utilized and the
pik the number of workers reallocated from seru k to i (integer)
decision-making process is formulated as a bi-level optimization prob­ qi the number of additional workers to seru i (integer)
lem. The remainder of this paper is organized as follows. Section 2 de­
scribes the concept of the seru production system and, in Section 3, a Continuous decision variables:
mathematical model for a worker and production quantity allocation Amij the quantities ofproduct in j − th generation allocated to seru i in scenario m
problem under demand uncertainty is constructed. Then, we formulate
zmij the shortage of product in j − th generation allocated to seru i in scenario m
this design problem of a seru production system as bi-level optimization
problems. The first optimization problem is to determine a worker allo­
Variables, functions and parameters:
cation problem and is formulated as a mixed-integer linear programming
problem, while the second problem is formulated to obtain optimal Dj demand for product in j − th generation
production quantity. In Section 4, the optimization scheme is applied to a dj demand realization for product in j − th generation
numerical example to illustrate its validity through parameter studies. f (d) joint probability distribution function of demand d = (d1, d2, …, dN)
wi the number of workers allocated to seru i after reallocation
2. Production planning in Seru production systems Wi the upper limit of worker number in seru i
w′i the original number of workers allocated to seru i
tri , sali training cost and salary per worker in seru i
Seru production systems can be classified into three types: divisional cj , sj , rj production cost, shortage cost and revenue of generation j per unit
seru, rotating seru, and yatai (Stecke et al., 2012). Among them, rotating ui the yield per worker in seru i
seru can adapt swiftly for the volatile demand by adjusting the number of M the number of Monte Carlo samples
workers in each seru (Iwamuro, 2002), and we focus on a design problem N the number of serus and product generations
in rotating seru in this paper. In rotating seru, workers who are allocated
to a seru unit perform all kinds of tasks in the seru, and the production

2
Y. Fujita et al. Computers & Industrial Engineering 163 (2022) 107856

Fig. 1. Full downward substitution model.

Fig. 2. The optimization flow of the proposed method.

3.2. Assumptions is known, the worker allocation optimization problem can be formulated
as follows,
The following assumptions are considered in this paper to construct { }
∑N ∑ N ∑N ∑N
the mathematical model for the seru production planning problem: Maximize F = − (tri − trk )pik + tri qi − sali wi
(1)
pik ,qi ,Aij ,zij
i=1 k=i i=1 i=1

1. The seru production system meets the features of full downward
+ G(d, w)f (d) dd
substitution, i.e. the workers in seru i can only produce the product
generations j which satisfies i⩽j.
2. The worker allocation is determined before the demand is observed where the objective function F(w) is the expected value of the total
while the product quantity allocation is determined after the demand profit, and w = (w1 ,w2 ,…,wN ). The first term is the total training cost of
is observed. workers and the second term is total wage where (tri − trk ) is training
3. The yield in each seru is proportional to the number of allocated cost required for a worker who moves from seru i to seru k (i > k) to
workers. obtain the skill for assembling product generations k, k + 1, ⋯, i − 1.
4. The demand of products for all generations are random variables, These two terms are not affected by the realization values d, and are
and their joint probability distribution functions are known. decided by only the worker allocation. G in the third term is the
5. Total allocated product quantity for each generation is equivalent to maximum total profit except for the training cost and wage at a certain d,
the observed demand realization. w, and the optimization problem to determine G(d, w) can be formulated
6. There are initially w′i workers allocated in each seru. These workers as follows,
and new additional workers are allocated to each seru after requisite (
j
) (
j
)

N ∑
N ∑ ∑N ∑
minimum training. Since allocation to serus with smaller numbers G(d, w) = maxAij ,zij rj dj − cj Aij − sj zij (2)
requires much training cost, it is assumed that the initially allocated j=1 j=1 i=1 j=1 i=1

workers can be reallocated to seru i′ where i′ ⩽i is satisfied, in order to


restrain total training cost. where “max” is a function to obtain maximum value with respect to Aij
and zij . The first term in Eq. (2) is the total revenue when all the demand
3.3. Formulation is completely satisfied, and the second and third terms represent the
total production cost and shortage cost, respectively. Note that, in this
As mentioned in Section 2, the optimization procedure consists of mathematical model, production yield realization cannot always satisfy
two steps. observed demand and product shortage may occur since the demand
realization is observed only after the worker allocation phase. Consid­
3.3.1. Worker allocation ering such a situation, product shortage zij is introduced as additional
The first step is to solve the worker allocation problem where variables in this formulation.
product demand uncertainty is taken into account in the formulation, To evaluate the integral term in Eq. (1), the Monte Carlo approach
and handle it as random variables. Since the probability distribution Dj (Hsu & Bassok, 1999; Hu & Hu, 2018) can be applied. In the Monte Carlo

3
Y. Fujita et al. Computers & Industrial Engineering 163 (2022) 107856

method, uncertainty is represented using a number of discrete proba­


Subject to :
bilistic scenarios, and the demand realization in each scenario is
generated based on assumed probabilistic distribution. Then, the
j

Amij = dmj for j = 1, …, N; m = 1, …, M
objective function can be transformed into a mixed-integer linear pro­ i=1
gramming problem formulated as follows, where m denotes the scenario

N ∑
N
number and the total sampling scenario number for the Monte Carlo Amij ⩽ui wi + zmij for i = 1, …, N; m = 1, …, M
method is denoted by M. j=i j=i

{ } zmij ⩽Amij for i = 1, …, N; j = i, …, N; m = 1, …, M


∑N ∑N ∑
N ∑
N
Maximize
m m
F = − (tr i − tr )p
k ik + tr q
i i − sali wi ∑
N
pik ,qi ,Aij ,zij
i=1 k=i i=1 i=1 wi = pik + qi for i = 1, …, N
{ ( ) ( )}
j j
1 ∑M ∑
N ∑N ∑ ∑N ∑ k=i
+ rj dmj − cj Amij − sj zmij ∑
k
M m=1 j=1 j=1 i=1 j=1 i=1 w′k = pik for k = 1, …, N
(3) i=1

0⩽pik , qi , Amij , zmij for i = 1, …, N; j, k = i, …, N; m = 1, …, M


In this formulation, the balance between the demand realization and
the production quantity allocated to each seru which corresponds to 0⩽wi ⩽Wi for i = 1, …, N
Assumption 5 is represented as follows.
Note that while production quantity allocation Am
ij and product shortage
zm
ij for each Monte Carlo sampling scenario are handled as variables in
j

Amij = dmj for j = 1, …, N; m = 1, …, M (4)
i=1 the above formulation, Aij and zij are finally fixed in the second opti­
mization step after demand realization is observed, as shown in Fig. 2.
Using this relationship, the term of total production cost in Eq. (3) can be Also note that although the problem includes several integer vari­

represented more simply as Nj=1 cj dm j . ables, since the formulation of the optimization problem is linear as
Furthermore, the following constraints must be satisfied. First, if the shown above, the optimization problem can be solved using a regular
allocated product quantity is higher than the yield at each seru, product optimization tool.
shortage occurs and the amount of product shortage zij can be written as,
3.3.2. Product quantity allocation
∑ ∑ In the second optimization step, since the demand realization values
N N
Amij ⩽ui wi + zmij for i = 1, …, N; ​ m = 1, …, M (5)
j=i j=i for each product generation are already known, the optimization
problem can be formulated as follows,
Next, the shortage of each product allocated to each seru in each scenario (
j
)
∑N ∑
does not exceed the allocated quantity, and the number of workers of ′
Minimize F = sj zij (11)
each seru is the summation of the number of workers moved to the seru Aij ,zij
j=1 i=1
and the number of new no-skill workers added to the seru.
Subject to:
zmij ⩽Amij for i = 1, …, N; ​ j = i, …, N; ​ m = 1, …, M (6)
j


N Aij = dj for j = 1, …, N (12)
wi = pik + qi for i = 1, …, N (7) i=1

k=i

N ∑
N

The following equation represents that the total number of workers Aij ⩽ui wi + zij for i = 1, …, N (13)
j=i j=i
moved from a certain seru to another seru is the number of workers in the
seru before worker assignment. zij ⩽Aij for i = 1, …, N; ​ j = i, …, N (14)

k
w′k = pik for k = 1, …, N (8) 0⩽Aij , zij for i = 1, …, N; ​ j = i, …, N (15)
i=1

where F′ is the second-step objective function, and decision variables are


In addition, the following other numerical constraints must be satisfied. production quantity allocation Aij and the shortage of j-th generation
0⩽pik , qi , Amij , zmij for i = 1, …, N; ​ j, k = i, …, N; ​ m = 1, …, M (9) product allocated to seru i,zij . In this objective function, since the worker
arrangement is already fixed, the training and wage cost terms can be
0⩽wi ⩽Wi for i = 1, …, N (10) eliminated from the first-stage objective function since these two cost
terms are not the function of Aij and zij . Consequently, in this step, the
Summerizing the above, the worker allocation optimization problem
product quantity allocation is determined to minimize the total shortage
is formulated as follows.
cost depending on the observed demand realization.
{ }
∑ N ∑N ∑
N ∑
N
Maximize
m m
F =−
pik ,qi ,Aij ,zij
(tri − trk )pik + tri qi − sali wi 4. Results and discussion
i=1 k=i i=1 i=1
{ ( )}
4.1. Problem settings
j
1 ∑M ∑
N
( ) ∑N ∑
+ rj − cj dmj − sj zmij
M m=1 j=1 j=1 i=1
The proposed method is applied to a numerical example. In this
paper, the worker salary is inclined according to the number of gener­
ations the worker can produce, i.e. the seru number. The basic hourly
wage is set up at 2,500 JPY, and added by 300 JPY for each additional
generation they can process. That is, the hourly wage is described as
follows:

4
Y. Fujita et al. Computers & Industrial Engineering 163 (2022) 107856

sali = 2500 + 300 × (N − i) (16) 4.2. Results

We set that the workers in seru N need training for 32 h before pro­ We conducted the optimization of worker allocation in three
duction, and when they increase the number of generations they can different seru number cases where N = 3, 5, 7 and where the number of
process, additional training for 24 h is required. The training cost per the Monte Carlo sampling points was set to M = 5, 000, where this
person per hour was set at 2500 JPY. Then, number was determined through preliminary numerical experiments.
Fig. 3 and Table 3 show the obtained results. In addition, the optimi­
tri = 2500 × {32 + 24 × (N − i)} (17)
zation result for the deterministic allocation case is shown in Fig. 4 and
In this example, it is assumed that demand follows the normal dis­ Table 4. Note that each value in the profit row in Table 4 is the profit in
tribution (Sun et al., 2021; Thevenin et al., 2020) while any type of the scenario where the demand is fixed at the expected value. Further­
probability distribution can be used in the proposed optimization more, in order to discuss the effect of demand variation parameters,
formulation, since the Monte Carlo method is used to handle demand optimization is conducted for the cases where the standard deviation of
uncertainty in the proposed optimization formulation and the yield is demand is set to μj /6, and the results are shown in Fig. 5 and Table 5.
deterministic. The expected demand and revenue per unit increase
progressively according to the generation of product. Furthermore, since 4.3. Comparison
workers in serus labelled with smaller numbers are more skilled workers,
the value of yield per worker varies according to the seru number. To compare the optimal solutions of the proposed method with
Table 1 shows the income per unit rj , the average demand μj of each deterministic allocation solutions, both solutions are evaluated for the
generation, and the yield per worker ui of each seru. Other parameters scenarios generated using the same random seeds based on the demand
were set as shown in Table 2 regardless of seru or generation number, distribution function used in the optimization. Also, the solutions are
scenarios, and N. compared in the case where the demand is fixed at the expected value. In
In this study, Gurobi Optimizer 8.0.1 is used as the solver. In the this comparison, 10,000 demand scenarios are randomly generated, and
following section, the results obtained by the proposed method are the 2nd-step optimization is conducted for each demand scenario. Then,
compared with the case where the uncertainty is not considered. the profit in the stochastic and deterministic allocation for each scenario
Therefore, the word “stochastic allocation” represents the solutions that is calculated. The average profit and the gap are as shown in Table 6. The
are obtained by the proposed method where the uncertainty is consid­ value in the row indicated by Gap shows the difference of the profits
ered, and the “deterministic allocation” means the results in the case between stochastic and deterministic allocation solutions.Table 7 shows
when all random variables are fixed to their expected values and opti­ the profit for both stochastic and deterministic allocation solutions when
mization is conducted without considering the demand variation. the realization demand is assumed to be equal to the expected values.
Furthermore, Table 8 shows the realization of demand for the scenario
where the profit difference was the largest.

Table 1 4.4. Discussion


rj , μj , ui of each generation.
By comparing Tables 3 and 4, it can be observed that the total
N Seru/Generation number
number of workers in solutions by the proposed method is larger than
1 2 3 4 5 6 7 the deterministic allocation to secure higher production capacity and
3 3000 4400 6200 respond to the uncertainty characteristics of demand. On the other hand,
rj 5 3000 3510 4100 5061 6200 deterministic solutions decrease the number of workers, reducing the
7 3000 3250 3580 3960 4506 5361 6200 training and salary cost, since even the lower production capacity is
3 6300 10654 18000 sufficient to meet the demand in the deterministic case.
μj 5 6300 7330 9130 12210 18000 Fig. 5 and Table 5 shows the optimal solution for a smaller demand
7 6300 6752 7330 9130 10654 13920 18000 variation case, i.e. standard deviation is set to μj /6. By comparing
3 2880 2699 2518 Figs. 3–5 and Tables 3–5, a strong correlation among demand fluctua­
ui 5 2880 2699 2518 2337 2156 tion and the total number of allocated workers is observed.
7 2880 2699 2518 2337 2156 1975 1794 Next, we discuss the average profit for 10,000 scenarios generated
randomly. Table 6 shows that the solutions provided by the proposed

Table 3
Table 2 Total number of workers and expected profit in stochastic optimization case.
Other parameters.
N =3 N =5 N =7
Wi w′i cj sj σj
Total number of workers 17 26 38
10 3 rj /3 2rj μj /3 Expected Profit (10,000 JPY) 9,334 12,851 16,036

Fig. 3. Result of stochastic worker allocation optimization.

5
Y. Fujita et al. Computers & Industrial Engineering 163 (2022) 107856

Fig. 4. Result of deterministic worker allocation.

Table 4 Table 5
Total number of workers and expected profit in deterministic optimization case. Total number of workers and expected profit in stochastic optimization case.
N =3 N =5 N =7 N =3 N =5 N =7

Total number of workers 13 22 33 Total number of workers 15 24 35


Profit (10,000 JPY) 10,022 13,862 17,351 Expected Profit (10,000 JPY) 9,682 13,410 16,708

method offer higher average profit than that of deterministic solutions. Table 6
Such comparisons illustrate the effectiveness of the proposed optimi­ Average profit in 10,000 scenarios (unit: 10,000JPY).
zation method for situations where demand highly fluctuates, while the N =3 N =5 N =7
profit of deterministic allocation is higher than the stochastic one in the
Stochastic Optimization 9,331 12,886 16,092
case where both random variables are fixed to the expected values, as
Deterministric Optimization 8,157 11,339 14,374
shown Table 7. Gap 1,174 1,574 1,718
Furthermore, Table 8 shows the realization values of demand in the
scenario when the profit difference between stochastic and deterministic
solutions is the largest among the randomly generated scenarios. As Table 7
shown in this figure, profit difference becomes significant when the Profit when the demand is fixed to expected value (unit: 10,000JPY).
demand realization of newer product generations is extremely larger N =3 N =5 N =7
than the expected value, since the deterministic allocation solution
Stochastic Optimization 9,510 13,146 16,365
employs fewer workers and the gap between demand and serus capacity Deterministric Optimization 10,022 13,862 17,351
results in large product shortage. In contrast, since the stochastic allo­ Gap -512 -716 -987
cation solution employs many workers to secure higher production ca­
pacity, the shortage and the decrease of profit can be suppressed.
From the above discussions, the importance of variation consider­ Table 8
ation in optimization and the effectiveness of the proposed method is Realization values for the largest profit difference scenario (unit: 10,000JPY).
successfully illustrated. N =3 N =5 N =7
In this example, three cases regarding the number of serus 3, 5 and 7 Stochastic Optimization 8,552 9,552 14,947
are examined. While these numbers are currently enough to describe the Deterministric Optimization − 2,913 754 6,019
practical implementation of seru production systems, if a greater large- Gap 11,464 8,798 8,856
scale production system is assumed, the number of integer variables is dj 1 5,391 5,445 6,033
increased and higher computational cost will be required. In order to 2 10,750 10,811 4,528
handle such cases, a more efficient optimization technique is needed to 3 43,735 11,561 5,508
be developed. This topic will be an important research topic in the 4 20,124 13,915
5 38,085 15,246
future.
6 21,356
7 38,394
5. Conclusion

In this paper, we propose an optimization method for worker allo­ • We constructed a mathematical model for the seru production sys­
cation and production quantity allocation problems in seru production tems to determine worker and production quantity allocation under
systems under uncertain demand. The results of this paper are summa­ uncertain demand.
rized as follows. • An optimization problem was formulated in a build-to-order pro­
duction case by using the Monte Carlo approach. The optimization

Fig. 5. Result of worker allocation optimization when σj = μj /6.

6
Y. Fujita et al. Computers & Industrial Engineering 163 (2022) 107856

process consists of two steps. In the first step, the worker allocation is Liu, C., Yang, N., Li, W., Lian, J., Evans, S., & Yin, Y. (2013). Training and assignment of
multi-skilled workers for implementing seru production systems. The International
determined and the production plan is prepared considering demand
Journal of Advanced Manufacturing Technology, 69, 937–959. https://doi.org/
uncertainty. In the second step, the actual production plan is deter­ 10.1007/s00170-013-5027-5
mined based on the observed demand realization. Mondal, A., & Roy, S. K. (2021). Multi-objective sustainable opened-and closed-loop
• The proposed method was applied to a numerical example. Through supply chain under mixed uncertainty during covid-19 pandemic situation.
Computers & Industrial Engineering, 159, 107453.
the comparison of the obtained solution with deterministic optimal Sakazume, Y. (2005). Is japanese cell manufacturing a new system?: A comparative study
solutions, the effectiveness of the optimization method was between japanese cell manufacturing and cellular manufacturing (<special english
confirmed. issue>); production and logistics. Journal of Japan Industrial Management Association,
55, 341–349. https://doi.org/10.11221/jima.55.341
Sakazume, Y. (2012). The Organizing principles of Assembly Cells (in Japanese) (1st ed.).
CRediT authorship contribution statement Tokyo: Keio University Press.
Stecke, K., Yin, Y., Kaku, I., & Murase, Y. (2012). Seru: The organizational extension of
JIT for a super-talent factory. International Journal of Strategic Decision Sciences, 3,
Yuya Fujita: Methodology, Software. Kazuhiro Izui: Conceptuali­ 105–118. https://doi.org/10.4018/jsds.2012010104
zation. Shinji Nishiwaki: Supervision. Zhe Zhang: Investigation. Yong Süer, G., Ulutas, B., Kaku, I., & Yin, Y. (2019). Considering product life cycle stages and
Yin: Investigation. worker skill level in seru production systems. Procedia Manufacturing, 39,
1097–1103. https://doi.org/10.1016/j.promfg.2020.01.361, 25th International
Conference on Production Research Manufacturing Innovation: Cyber Physical
References Manufacturing August 9-14, 2019 — Chicago, Illinois (USA).
Sun, H., Wang, Y., & Xue, Y. (2021). A bi-objective robust optimization model for disaster
Aboelfotoh, A., & Md Abdullah, G. A. S. (2018). Selection of assembly systems; assembly response planning under uncertainties. Computers & Industrial Engineering, 155,
lines vs seru systems. Procedia Computer Science, 140, 351–358. https://doi.org/ 107213.
10.1016/j.procs.2018.10.304. Cyber Physical Systems and Deep Learning Chicago, Thevenin, S., Adulyasak, Y., & Cordeau, J.-F. (2020). Material requirements planning
Illinois November 5-7, 2018. under demand uncertainty using stochastic optimization. Production and Operations
Abolghasemi, M., Beh, E., Tarr, G., & Gerlach, R. (2020). Demand forecasting in supply Management, 30, 475–493.
chain: The impact of demand volatility in the presence of promotion. Computers & Tolooie, A., Maity, M., & Sinha, A. K. (2020). A two-stage stochastic mixed-integer
Industrial Engineering, 142, 106380. https://doi.org/10.1016/j.cie.2020.106380 program for reliable supply chain network design under uncertain disruptions and
Chen, C., Xu, X., Zou, B., Peng, H., & Li, Z. (2021). Optimal decision of multiobjective and demand. Computers & Industrial Engineering, 148, 106722.
multiperiod anticipatory shipping under uncertain demand: A data-driven Villa, A., & Taurino, T. (2013). From JIT to seru, for a production as lean as possible.
framework. Computers & Industrial Engineering, 159, 107445. Procedia Engineering, 63(956–965), 2013. https://doi.org/10.1016/j.
Gabrel, V., Murat, C., & Thiele, A. (2014). Recent advances in robust optimization: An proeng.2013.08.172. The Manufacturing Engineering Society International
overview. European Journal of Operational Research, 235, 471–483. Conference, MESIC.
Güler, M. G., & Bilgiç, T. (2009). On coordinating an assembly system under random Wang, Y., & Tang, J. (2018). Cost and service-level-based model for a seru production
yield and random demand. European Journal of Operational Research, 196, 342–350. system formation problem with uncertain demand. Journal of Systems Science and
https://doi.org/10.1016/j.ejor.2008.03.002 Systems Engineering, 27, 519–537. https://doi.org/10.1007/s11518-018-5379-3
Hsu, A., & Bassok, Y. (1999). Random yield and random demand in a production system Xu, X., Hao, J., & Zheng, Y. (2020). Multi-objective artificial bee colony algorithm for
with downward substitution. Operations Research, 47, 277–290. https://doi.org/ multi-stage resource leveling problem in sharing logistics network. Computers &
10.1287/opre.47.2.277 Industrial Engineering, 142, 106338.
Hu, Z., & Hu, G. (2018). A multi-stage stochastic programming for lot-sizing and Yılmaz, Ö. F. (2020). Attaining flexibility in seru production system by means of
scheduling under demand uncertainty. Computers & Industrial Engineering, 119, shojinka: An optimization model and solution approaches. Computers & Operations
157–166. Research, 119, 104917. https://doi.org/10.1016/j.cor.2020.104917
Iwamuro, H. (2002). Seru Seisan System (in Japanese). (1st ed.). Tokyo: Nikkan Kogyo Yin, Y., Stecke, K. E., & Li, D. (2018). The evolution of production systems from industry
Shimbun. 2.0 through industry 4.0. International Journal of Production Research, 56, 848–861.
Kaku, I. (2016). A fundamental positive investigation into japanese seru production https://doi.org/10.1080/00207543.2017.1403664
systems. IFAC-PapersOnLine, 49, 337–342. https://doi.org/10.1016/j. Yin, Y., Stecke, K. E., Swink, M., & Kaku, I. (2017). Lessons from seru production on
ifacol.2016.07.627, 8th IFAC Conference on Manufacturing Modelling, Management manufacturing competitively in a high cost environment. Journal of Operations
and Control MIM 2016. Management, 49–51, 67–76. https://doi.org/10.1016/j.jom.2017.01.003
Kaku, I. (2017). Is seru a sustainable manufacturing system? Procedia Manufacturing, 8, Yu, Y., Tang, J., Gong, J., Yin, Y., & Kaku, I. (2014). Mathematical analysis and solutions
723–730. https://doi.org/10.1016/j.promfg.2017.02.093 for multi-objective line-cell conversion problem. European Journal of Operational
Lian, J., Liu, C., Li, W., & Yin, Y. (2018). A multi-skilled worker assignment problem in Research, 236, 774–786. https://doi.org/10.1016/j.ejor.2014.01.029
seru production systems considering the worker heterogeneity. Computers & Yu, Y., Tang, J., Sun, W., Yin, Y., & Kaku, I. (2013). Reducing worker(s) by converting
Industrial Engineering, 118, 366–382. https://doi.org/10.1016/j.cie.2018.02.035 assembly line into a pure cell system. International Journal of Production Economics,
Liu, C., Li, W., Lian, J., & Yin, Y. (2012). Reconfiguration of assembly systems: From 145, 799–806. https://doi.org/10.1016/j.ijpe.2013.06.009
conveyor assembly line to serus. Journal of Manufacturing Systems, 31, 312–325. Yu, Y., Wang, J., Ma, K., & Sun, W. (2018). Seru system balancing: Definition,
https://doi.org/10.1016/j.jmsy.2012.02.003 formulation, and exact solution. Computers & Industrial Engineering, 122, 318–325.
Liu, C., Stecke, K. E., Lian, J., & Yin, Y. (2014). An implementation framework for seru https://doi.org/10.1016/j.cie.2018.05.048
production. International Transactions in Operational Research, 21, 1–19. https://doi.
org/10.1111/itor.12014

You might also like