You are on page 1of 12

CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405

Contents lists available at ScienceDirect

CIRP Journal of Manufacturing Science and Technology


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

Resource allocation methodology based on object-oriented discrete


event simulation: A production logistics system case study
Guangzhen Lia,b,c,* , Shengluo Yangb,c , Zhigang Xub,c , Junyi Wangb,c , Zhaohui Rena ,
Guobing Lia
a
School of Mechanical Engineering and Automation, Northeastern University, Shenyang 110819, China
b
State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Science, Shenyang 110016, China
c
Institutes for Robotics and Intelligent Manufacturing, Chinese Academy of Sciences, Shenyang 110169, China

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

Article history: Resource allocation is always a challenge especially in complicate and uncertain production logistics
systems. Production logistics systems, as integrated production and logistics systems, have attracted
Available online 13 August 2020 increased attention in supply chain and industrial production research. This paper focuses on the
problem of unbalanced and unreasonable resource allocation caused by the lack of an efficient and
Keywords: practical resource allocation methodology in production logistics systems. We propose a resource
Resource allocation
allocation methodology for a production logistics system, and provide the specific modelling and
Production logistics system
simulation procedures. In the case study, we use factor experiment to distinguish bottleneck resources
Object-oriented discrete event simulation
Decision making
from non-bottleneck resources, and compare the performance of 160 different allocation schemes, and
Case study obtain the optimal configuration of each resource including the number of AGV, the speed of AGV and the
Plant Simulation software capacity of each buffer based on the proposed methodology and simulation procedures. The results of the
case study illustrate the effectiveness and practicality of the resource allocation methodology based on
object-oriented discrete event simulation. Both the case study and the allocation methodology can
provide reference for similar resource allocation problems.
© 2020 CIRP.

Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395


Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
Research methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
Case study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
The actual PLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
Determine the simulation input and output . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
Collect the system data and information . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
Construct the simulation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
Validate the simulation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
Factorial experiment and its results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
Bottleneck identification based on the factorial experiment results . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
Multilevel experiment and its results . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
Bottleneck resource allocation based on the factorial experiment results ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
Monitoring experiment and its results . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
Non-bottleneck resource allocation based on the monitoring experiment results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
The time spent on the simulation experiments . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
System performance after resource allocation . . . . . . . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403

* Corresponding author at: School of Mechanical Engineering and Automation, Northeastern University, Shenyang 110819, China.
E-mail address: 2486541829@qq.com (G. Li).

https://doi.org/10.1016/j.cirpj.2020.07.001
1755-5817/© 2020 CIRP.
G. Li et al. / CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405 395

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404

Introduction industrial problems to find the best resource allocation, which has
a direct influence on costs and revenues [22]. Some researchers
Production logistics, as the intermediate link between supply have used this software or technology to solve complex allocation
logistics and sales logistics, refers to all production and logistics problems in various fields. For instance, to configure the number of
activities from rough parts to finished products, including parallel machines for workstations, Al-Turki et al. used Arena
production, transportation and storage [1,2]. A production logistics software to model a flexible shop flow and obtained a considerable
system (PLS) refers to an organic whole system composed of production benefit by taking into account the random arrival of
physical and information subsystems, such as warehouses, products, machine setup time and unexpected failures [23].
machines, materials and production information, which are Seitaridis et al. introduced an agent-based scheme for the
equipped to meet production and logistics needs. Typically, a distribution of electric vehicles across a set of charging stations
PLS involves integrated cyclic processes and can generally be and proposed two offline and online models; then, the operating
divided into three functional areas: storage areas, logistics areas conditions for different numbers of charging stations were
and production areas; moreover, production and logistics areas are considered to determine a reasonable scheme [24]. To promote
often grouped. safe production in strip mines and improve the economic benefit,
Production logistics is a value-added process [1], and the Libing et al. used Flexsim software to simulate the logistics system
acceleration of production logistics can shorten the production of the strip mine and determined the optimal temporary site
cycle time, reduce the backlog of products in the process, and reserves and the optimal number of trucks through orthogonal
speed up the turnover of short-term liquidity [2,3]. Thus, the experimental design [25]. Yang et al. used Plant Simulation
earnings of an enterprise will be immediately influenced by the software to conduct simulations of an assembly workshop and
running status of the corresponding production logistics. In optimized the configuration of the number of workers, the buffer
addition, production and logistics are changing from unrelated capacity and the number of automatically guided vehicles (AGVs)
independent activities to related coordinated activities, and the to obtain the ideal system throughput [26]. Staley and Kim studied
integration of production and logistics is a trend of future the allocation of production buffers and performed simulation
development [4–8], and industrial companies are nowadays acting experiments under various allocation strategies with Petri net
in global production networks [9]. Manufacturing and logistics are simulation technology to obtain the optimal results [27]. Sena et al.
among the industries with the highest energy consumption [10]. applied agent-based simulation technology to study the dynamic
Therefore, PLSs have been increasingly considered in research on allocation problem of additional worker resources and increased
supply chains and industrial production. the average contribution of workers by 117.51% compared with the
Studies of PLSs are complicated because PLSs integrate a large that pre-optimization [28].
number of entities, interactions and external forms of interference Although many researchers have studied resource allocation
[11], and the uncertainty and discrete dynamics of the production problems in related fields, the results of these studies indicated
environment will increase the difficulty of the research [12]. that simulations are an effective way to solve the resource
Resource allocation is a complex discrete optimization problem allocation problem [29,30]. However, these studies have only
that can be solved by establishing the corresponding mathematical individually allocated some significant resources and did not
model and heuristic algorithm. For example, Song et al. established provide global consideration at the system level, nor did they
a mathematical model between the buffer capacity and production realize that the identification of bottleneck resources is the
targets to determine the buffer capacity in a manufacturing system primary task of resource allocation. There is still a gap between
and then obtained the minimum capacity of the buffer using the theoretical research and practical resource allocation in PLSs,
particle swarm algorithm [13]. Nevertheless, Zhang et al. pointed which encompass extremely complex interaction relations.
out that in the case of uncertain production parameters, such Furthermore, many specific domain methods like bottleneck
algorithms may poorly handle actual dynamic production prob- identification are also difficult to integrate into general-purpose
lems, and computer simulations have more advantages in speed simulation software [31,32]. Currently, there is still a lack of object-
and authenticity [14]. oriented discrete event simulation methods to guide the decision-
making procedure of resource allocation. Therefore, the purpose of
Literature review this paper is to propose a general resource allocation method for
PLSs based on object-oriented discrete event simulations and
Object-oriented discrete event simulation is a method for obtain the maximum production capacity with minimum eco-
identifying the significant entities in the real world and is a nomic input.
powerful tool for understanding and explaining how they interact The rest of this paper is organized as follows. In
with one another [15,16], it offers a great potential for modelling Section “Research methodology”, the proposed resource allocation
and analyzing business processes [17], and it is also an important methodology and specific simulation procedures is introduced.
technology to solving material-resource-assignments problem Section “Case study” is a case study of resource allocation problems
[18]. This approach is characterised by inheritance, encapsulation involving an actual PLS. The results of the case study are discussed
and a layered structure, so the construction and maintenance of in Section “Discussion”. Conclusions and future developments are
models is significantly simplified, and the corresponding methods presented in Section “Conclusion”.
are easy to learn and use [19–21].
Many companies have developed a variety of object-oriented Research methodology
simulation software products such as Arena, witness, Flexsim and
Plant Simulation. More importantly, Simulation optimization is a Li et al. noted that bottleneck identification is the primary task
tool commonly used as a decision-making support system on of resource allocation, and a bottleneck generally refers to a
396 G. Li et al. / CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405

Fig. 1. The simulation procedure for resource allocation.

bottleneck resource or bottleneck process that affects the utilization rate and balance rate, are determined as the outputs
workshop output in the current state [33]. According to constraint of the experiment.
theory, bottleneck resources refer to resources with a supply (3) Collect system data and information. Collect information such as
capacity that is less than the demand, and bottleneck resources the processing time and machine properties involved in the
directly affect the production capacity of the whole system [34], actual material flow and fit the information as a precise
which indicates that the allocation strategy should be different for distribution or a certain number as the necessary input set of
bottleneck resources and non-bottleneck resources. For bottleneck the simulation model. To ensure the authenticity of the final
resources, the objective is to obtain the maximum production model, the data and the fitting distribution should be
capacity through appropriate allocation. For non-bottleneck consistent with reality.
resources, the objective is to invest as little as possible without (4) Construct the simulation model. Based on the general modelling
affecting the production capacity. Therefore, the proposed methodology for production logistics [21], establish the
methodology first determines the influence of each resource on simulation model according to the relevant layout, operation
the production capacity through a factorial simulation experiment procedure, characteristics of each object and mutual relation-
and divides all resources into bottleneck resources and non- ships among the objects of the PLS.
bottleneck resources. Then, we design a multilevel experiment for (5) Validate the model. Only when the model is valid can the
bottleneck resources and compare the production information for characteristics of the real system be authentically reflected and
each level of resource allocation to determine the optimal the simulation results be reliable. The validation conditions
allocation of bottleneck resources. Finally, we perform a monitor- include but are not limited to the following conditions: (a) the
ing experiment for non-bottleneck resources to assess the simulation model can run successfully until the termination
maximum utilization of non-bottleneck resources. Thus, resources condition is satisfied; (b) a discussion with management occurs
can be allocated for many work tasks. to determine if the input data and each simulation event reflect
Referring to a methodology to improve the resource consump- the actual situation; and (c) the simulation animation is
tion for the manufacturing of masonry units based on discrete consistent with the actual production process. If the model is
event simulation [9], the simulation steps of resource allocation valid, go to step 6; otherwise, go to step 3 to reconstruct the
methodology in this paper are shown in Fig. 1. The methodology model.
and steps remain abstract enough to include most of the potential (6) Identify the bottleneck resources. According to the main effect of
resource allocation problems. The specific steps are as follows. the factorial experiment and the bottleneck criterion proposed
in 2.1.1, the PLS system resources are divided into bottleneck
(1) Analyse the PLS. Analyse the plant layout, operation procedure, resources and non-bottleneck resources.
logistical strategy and processing time of the PLS; then, (7) Allot bottleneck resources. Analyse the system output of each
communicate with the management personnel and obtain allocation strategy by designing a multilevel experiment, and
feedback to ensure that an accurate understanding of practical then determine the best strategy by comparing these system
problems is achieved. outputs.
(2) Determine the simulation input and output. According to the (8) Allot non-bottleneck resources. Input the bottleneck resource
actual problem, the leading resource types are determined as allocation parameters obtained in step 7 into the simulation
the inputs of the experiment, and the production indicators model, and obtain the minimum allocation quantity of each
used to evaluate the PLS, such as the throughput, equipment non-bottleneck resource by the monitoring experiment.
G. Li et al. / CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405 397

Fig. 2. Material flow diagram.

(9) Present the final simulation results. Input the non-bottleneck the AGV continues along the AGV cycle track and receives the
resource allocation parameters obtained in step 8 into the new distribution task associated with a roughcast warehouse.
model, and obtain the production information for the PLS by
running the model.
Determine the simulation input and output

Case study According to the analysis of the PLS, the primary resources are
the number of AGVs, speed of AGVs, load capacity of AGVs (the
The actual PLS maximum number of parts that can be loaded with AGVs), capacity
of the production buffer and capacity of the logistics buffer. The
In the late stages of the project planning phase, the inputs of allocation of leading resources is taken as the input of the
various resources should be allocated before a factory is built. The simulation experiment, and the throughput in 24 h of continuous
studied PLS involves production logistics activities for seven production is taken as the output. In all the related graphs, tables
system parts. The seven parts are represented by A, B, C, D, E, F and and experiments, ANs represents the number of AGVs, ASs
G in this case. The material flow process is shown in Fig. 2 and represents the speed of AGVs, ALCs represents the load capacity
described as follows: of AGVs, LBC represents the capacity of the logistics buffer, and PBC
represents the capacity of the production buffer.
(1) The control centre publishes the dispatch request according to
the distribution table, which has multiple rows and two Collect the system data and information
columns. Each row represents a dispatch request, and the two
columns record the types of parts and destination of the After extensive communication and discussion with manage-
assignment. The initial distribution table has seven rows, and ment personnel, we determine the following actual production
the initial information is determined by the order of the constraints and relevant data:
production sequence: A–B–C–D–E–F–G.
(2) The available AGV receives the dispatch request and runs to the (1) The processing time, setup time and recovery time of all parts
roughcast warehouse to load the parts according to the of each machine conform to the normal distribution and are
dispatch request. Then, the model deleted the dispatch request shown in Table 1.
of this row in the distribution table after loading, indicating (2) In actual production, since an AGVs needs to unload all the
that this dispatch request has been completed. The AGV follows parts delivered to the logistics buffer, the capacity of the
the circular track to the destination after accepting the mission. logistics buffer is designed to be slightly larger than the load
Each destination equipped with a sensor and the AGV will
make a judgement when it triggers the sensor. The judgement
determines whether the location is the destination of the Table 1
mission and whether the part matches the information related The processing time, setup time and recovery time of each machine.
to the destination.
Machine Process time Setup time Recovery time
(3) If any of the conditions are not satisfied, the AGV continues to (m, s2) (m, s2) (m, s2)
move forward; if both conditions are met, the AGV enters the
PM01 (315, 162) – (15, 42)
unloading track and offloads the parts to the logistics buffer. PM02 (300, 152) – (10, 22)
The parts move from the logistics buffer into the production PM03 (360, 182) (10, 22) –
area through a series of production activities, finally reaching PM04 (240, 122) – –
PM05 (240, 122) – –
the finished-parts warehouse. Additionally, the part type and
PM06 (180, 92) – –
destination of this assignment are added to the last row of the AS01 (330, 172) – –
distribution table as a new dispatch request after the AS02 (270, 142) – –
completion of AGV unloading. Additionally, AGVs are not AS03 (300, 152) – –
allowed to pass while driving and must stop and wait. AS04 (360, 182) – –
AS05 (360, 182) – –
(4) After the unloading task is completed, the AGV can accept a
AS06 (300, 152) – –
new dispatch request as the available vehicle at this time, and
398 G. Li et al. / CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405

Table 2 production area for processing. According to the predetermined


The maximum configurable capacity of the production buffers.
process flow, the part travels from processing machines PM01 to
BF8 BF9 BF10 BF11 BF12 BF13 BF14 BF15 BF16 BF17 BF18 PM06 to assembly machines AS01 to AS06 and finally to the
Cmax 6 6 8 8 2 2 2 3 3 3 3 warehouse2. Therefore, the entire production logistics process
continues running until the termination condition of the model is
satisfied.

capacity of the AGV based on the following constraint Validate the simulation model
relationship: LBC = ALC + 2
(3) When the process layout of the system is determined, the Based on the following points, we confirm that the established
maximum configurable capacity of production buffers under simulation model is valid: (1) the processing time at each
the spatial layout of the current system is shown in Table 2. station, track length and buffer capacity in the simulation model
(4) In actual production, a machine has a certain probability of agree with the actual values; (2) the programmer checks all the
failure. A failed machine needs to be repaired before control procedures and finds no errors; (3) the simulation model
continuing to work. The machine failure rate was set at 5%, runs until the termination conditions are met; (4) the manager
and the time required to repair a machine was set to 1 min. deems that the simulation logic is consistent with the actual
(5) When loading, it takes 20 s to load one part on an on-board production logic by observing the simulation animation step by step
pallet. When unloading, the parts are unloaded with the pallet, (the link to the simulation animation: https://www.bilibili.com/
so the unloading time is 20 s in total. video/BV1ov411z7Fy/); and (5) the actual production line produced
6 h per day, with an average daily output of 52 products. The
deviation with the simulation results was only 6.6%.
Construct the simulation model
Factorial experiment and its results
Establish each object and the external connections among
objects according to the system layout and process flow. Then, Bottleneck identification is a classic problem. Although many
write the production logic sequence for each event and each object studies have proposed corresponding bottleneck identification
in the programming language SimTalk. The 2-dimensional model methods for some typical systems, how to identify the bottleneck
in Plant Simulation software package are shown in Fig. 3. in the actual system is still a realistic challenge, because the current
As shown in Fig. 3, the object “warehouse1” is a roughcast theories are not fully applicable [33,35]. To the best of our
warehouse; the object “warehouse2” is finished-parts warehouse; knowledge, there is no research that combines simulation and
the object “Track_load” is the loading location; the object factor experiments to quantitatively analyze the bottleneck degree.
“Track_unload” is the unloading location; the table file object In the statistical sense, a factorial experiment can determine the
“trans1” is the distribution table; and the object “line” is a main and interaction effects among variables. The main effect
conveyor. represents the variation in the experimental output when each
The running logic of the simulation model is as follows: the input changes independently, and the interaction effect represents
AGVs runs on the track, and when triggering the sensor, it can the variation in the output when both inputs change simulta-
determine whether the loading conditions are satisfied. If the neously. The larger the effect value is, the greater the influence of
conditions are met, the AGVs will enter the track load and load the input on the output.
from warehouse1 according to the distribution table trans1. After From Section “Determine the simulation input and output”, the
loading, the AGVs will move forward towards the destination; if inputs of the factorial experiment are the number of AGVs (ANs),
the conditions are not satisfied, the AGVs continues to move along load capacity of AGVs (ALCs), speed of AGVs (ASs) and capacity of
the track. When the AGVs passes each destination, it will trigger the production buffer (PBC), and the output of the experiment is
the corresponding sensor. If the relevant conditions are met, the the system throughput.
AGVs will enter the corresponding unloading track. When the There are eleven production buffers in the case, and there must
AGVs triggers the unloading sensor, the AGVs stops and unloads be a difference in these buffers. In the factorial experiment, the
the parts to the corresponding logistics buffer. Then, the trans1 capacity of eleven buffers is simplified into one. In other words, the
table rewrites this distribution task to the last row, and the AGVs capacity of all production buffers is assumed to be equal and varies
completes this distribution task; meanwhile, the AGV can accept by an integer variable. The purpose of this approach is to reduce the
the next dispatch request, and the parts begin to enter the experimental complexity. After simplified treatment, the

Fig. 3. The 2D simulation model.


G. Li et al. / CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405 399

Table 3
The results of the factorial experiment.

No. Number of AGVs Load capacity of AGVs Speed of AGVs Capacity of the production buffer Mean throughput
Exp. 1 1 2 0.2 2 28
Exp. 2 1 2 0.2 8 28
Exp. 3 1 2 0.38 2 48
Exp. 4 1 2 0.38 8 48
Exp. 5 1 5 0.2 2 62
Exp. 6 1 5 0.2 8 62
Exp. 7 1 5 0.38 2 107
Exp. 8 1 5 0.38 8 107
Exp. 9 4 2 0.2 27 106.3
Exp. 10 4 2 0.2 8 106.3
Exp. 11 4 2 0.38 2 191.2
Exp. 12 4 2 0.38 8 190.8
Exp. 13 4 5 0.2 2 215.8
Exp. 14 4 5 0.2 8 215.8
Exp. 15 4 5 0.38 2 216.6
Exp. 16 4 5 0.38 8 216.6

experimental complexity was reduced from 214 to 24, and the time 10.4 and 0.4, respectively, which means that the average
spent on the factorial experiment was significantly reduced. throughput is increased by 94.0 when the number of AGVs is
Therefore, the factorial experiment was designed as a four- increased from 1 to 4; the average throughput is increased by 29.9
factor and two-level experiment with a complexity of 16. Each when the load capacity of AGVs is increased from 2 to 5; the
group of experiments was run 100 times with different random average throughput is increased by 10.4 when the AGVs speed is
numbers of seeds. The experimental results are shown in Table 3. increased from 0.22 m/s to 0.38 m/s; and the average throughput is
increased by 0.4 when the AGVs speed is increased from 2 to 8.
Bottleneck identification based on the factorial experiment results The maximum output of the factorial experiment is 216.6, to
quantitatively distinguish the bottleneck and non-bottleneck
The number of resource types is denoted by n, and each resources, we give the following criteria: if the main effect of a
resource is represented by X1, X2, . . . Xn. The simulation function resource is greater than 2% of the maximum output of the factorial
between the target output and the resource parameters is experiment result, the resource is a bottleneck resource; other-
wise, it is a non-bottleneck resource. Therefore, the number of
O ¼ f ðX 1 ; X 2 :::X n Þ
AGVs, load capacity of AGVs and speed of AGVs are bottleneck
The inputs of the factorial experiment are the upper and lower resources, while the capacity of the production buffer is a non-
bounds of each resource. Let Xi-l and Xi-u represent the lower and bottleneck resource in this case.
upper levels of resource inputs, respectively. The main effect and Table 4 also shows the interaction between any two resources.
interaction effect of each resource are calculated as follows, where Nevertheless, the interaction between the AGVs loading capacity
Hi represents the main effect of resource Xi and Wi,j represents the and AGVs speed is 16.3, which is a negative number. The
interaction effect of resource Xi and resource Xj. calculation process of this interaction effect is as follows:

1 1
Hi ¼ ½f ðX 1l ; :::X iu ; :::X nl Þ  f ðX 1l ; :::X il ; :::X nl Þ WALC;AS ¼ f½ð107  62Þ  ð48  28Þ þ ½ð216:6  215:8Þ
2 ð1Þ 4
þf ðX 1u ; :::X iu ; :::X nu Þ  f ðX 1u ; :::X il ; :::X nu Þ
ð190:8  106:3Þg ¼ 14:7
The reason for the negative number is that the third bracket in
1 the above calculation represents the change in throughput when
Wi;j ¼ f½f ðX 1l ; :::X iu :::X ju ; :::X nl Þ  f ðX 1l ; :::X iu :::X jl ; :::X nl Þ
4 the AGVs speed is increased from 0.22 m/s to 0.38 m/s while the
½f ðX 1l ; :::X il :::X ju ; :::X nl Þ  f ðX 1l ; :::X il :::X jl ; :::X nl Þ number of AGVs, load capacity of AGVs and capacity of the
production buffer are all at the upper level. However, when the
þ½f ðX 1u ; :::X iu :::X ju ; :::X nu Þ  f ðX 1u ; :::X iu :::X jl ; :::X nu Þ
AGVs speed is 0.22 m/s or 0.38 m/s, the system has reached the
½f ðX 1u ; :::X il :::X ju ; :::X nu Þ  f ðX 1u ; :::X il :::X jl ; :::X nu Þg maximum throughput, which does not reflect the difference that
should occur, so this term has a negative value. The reason for this
ð2Þ
issue is related to the nature of the PLS; there is a ceiling for system
According to Eqs. (1), (2) and the results of the factorial throughput, and the throughput will not increase further when the
experiment, the upper triangular matrix of the main effect and the resource input is saturated, which is not considered in Eqs. (1) and
interaction effect can be obtained, as shown in Table 4. As shown in (2). Therefore, although the interaction effect is negative, it does
Table 4, the main effects of ANs, ALCs, ASs and PBC are 94.0, 29.9, not mean that the two factors are mutually limiting.

Table 4
Effect analysis for the factorial experiment.

Number of AGVs Load capacity of AGVs Speed of AGVs Capacity of the production buffer
Number of AGVs 94.0 10.6 5.2 0
Load capacity of AGVs 29.9 14.7 0.1
Speed of AGVs 10.4 0
Capacity of the production buffer 0.4
400 G. Li et al. / CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405

Fig. 4. Concept of the multilevel experiment.

Table 5
The results of the multilevel experiment.

ANs ALCs ASs

0.2 0.22 0.24 0.26 0.28 0.3 0.32 0.34 0.36 0.38
1 2 28 30 32 36 38.0 40 42 44 47.5 48
3 41 44 47 50.9 55.2 59.0 62 65 68 71
4 54.0 58 62 66 70 74 78 82 86 90
5 62 67.9 74.0 79.7 84.6 89.2 93.6 97.5 102 107
2 2 54.8 60 64 68.5 74 79.5 84 88 92.0 96.6
3 74.2 85.5 92.2 99.1 106.3 112.9 119.0 125.2 131.9 138.6
4 102 111.0 120.5 129.4 137.6 145.1 153.5 161.9 169.9 177.8
5 122.4 132.6 143.9 155.2 165.5 174.8 184.4 194.3 203.1 210.9
3 2 80.8 87.9 95.3 102.3 109.6 116.5 123.8 130.7 137.5 144.3
3 114.2 123.4 135.5 146.8 157.9 168.1 178.6 188.4 197.5 206.2
4 152.0 165.5 178.8 191.7 203.1 212.7 216.6 216.6 216.6 216.6
5 183.7 199.5 212.7 216.3 216.6 216.6 216.6 216.6 216.6 216.6
4 2 106.3 116.2 126.1 135.8 145.0 154.4 163.9 172.9 181.9 190.8
3 145.7 164.1 180.9 194.8 207.1 216.0 216.6 216.6 216.6 216.6
4 199.4 214.0 216.4 216.5 216.6 216.6 216.6 216.6 216.6 216.6
5 215.8 216.1 216.2 216.3 216.4 216.5 216.5 216.6 216.6 216.6

Fig. 5. System throughput when the number of AGVs is 1.

Multilevel experiment and its results allocation schemes for each level of each parameter. Finally, the
optimal allocation strategy can be obtained based on the
For a specific PLS, we first determine the configurable range of experimental results and manual decisions.
each bottleneck resource configuration according to the actual Through the bottleneck analysis in Section “Bottleneck identi-
situation and divide each bottleneck resource parameter into many fication based on the factorial experiment results”, the bottleneck
levels within the actual range. Then, we formulate a host of resources are determined as ANs, ASs and ALCs. According to the
G. Li et al. / CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405 401

actual situation, the configuration of ANs is divided into four levels: divided into four levels: 2, 3, 4, and 5. The concept of the multilevel
1, 2, 3, and 4. The configuration of ASs is divided into ten levels: experiment is shown in Fig. 4. An experiment for each level is run
0.20 m/s, 0.22 m/s, 0.24 m/s, 0.26 m/s, 0.28 m/s, 0.30 m/s, 0.32 m/s, 100 times with different random number seeds. The results are
0.34 m/s, 0.36 m/s, and 0.38 m/s. The configuration of ALCs is shown in Table 5.

Fig. 6. System throughput when the number of AGVs is 2.

Fig. 7. System throughput when the number of AGVs is 3.

Fig. 8. System throughput when the number of AGVs is 4.


402 G. Li et al. / CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405

Table 6
Alternative allocation strategy for bottleneck resources.

No Number of AGVs Load capacity of AGVs Speed of AGVs (m/s) Mean throughput
1 3 4 0.32 216.6
2 3 5 0.26 216.3
3 4 3 0.30 216.0
4 4 4 0.24 216.4
5 4 5 0.20 215.8

Bottleneck resource allocation based on the factorial experiment the capacity of the production buffer BF8 to BF18. The monitoring
results experiment code is embedded at the entrance and exit of each
buffer. Each trigger will read the number of parts in the buffer and
For the PLS, the objective is to obtain the maximum throughput retain the historical maximum Smax. The number of AGVs = 3, load
and the best economic benefit with the minimum resource input. capacity of AGVs = 4, speed of AGVs = 0.32 m/s (obtained by the
Therefore, this paper chooses the most reasonable allocation analysis in Section “Bottleneck resource allocation based on the
scheme among 160 configurations based on this approach. Four factorial experiment results”) and the maximum configurable
graphs are used to represent the throughput when the number of capacity of the buffers BF8-BF18 (6, 6, 8, 8, 2, 2, 2, 3, 3, 3, and 3) were
AGVs is 1, 2, 3, and 4; the abscissa is the AGVs speed; and the input into the simulation model, and the monitoring experiment
ordinate is the throughput. Different curves represent the different was run with different random number seeds 1000 times. The
load capacities of the AGVs. The results of 160 groups of multilevel results of the experiment are shown in Table 7.
experiments are shown in Figs. 5–8.
According to Figs. 5 and 8, the minimum throughput of the Non-bottleneck resource allocation based on the monitoring
system is 28, and the maximum throughput is 216.6. The upper and experiment results
lower bounds of throughput are approximately 7.7 times different,
indicating that reasonable and efficient resource allocation is The monitoring experiment was carried out after the allocation
crucial to the production capacity of the PLS. According to Figs. 5 scheme of bottleneck resources was determined. The experiment
and 6, the system throughput is not saturated when the number of was repeated 1000 times to obtain the maximum instantaneous
AGVs is 1 or 2. According to Figs. 7 and 8, the system throughput is capacity of each buffer under various conditions.
saturated when the number of AGVs is 3 and 4. Therefore, we A monitoring experiment is designed for the non-bottleneck
consider the situation when the number of AGVs is 3 and 4. resources. The purpose of the experiment is to monitor the
As shown in Fig. 7, if the number of AGVs is set to 3, the system maximum utilization of non-bottleneck resources during a model
throughput does not reach the maximum when the load capacity run. Let Cmax represent the maximum number of non-bottleneck
of the AGVs is 2 or 3. When the load capacity of the AGVs is 4 and resources that can be allocated in a real PLS, Smax represent the
the speed of the AGVs is 0.32 m/s, the system throughput reaches maximum utilization level in the monitoring experiment, and R
saturation. When the load capacity of the AGVs is 5 and the speed is represents the final configuration parameter for the non-bottle-
0.26 m/s, the system throughput is saturated. neck resource. The calculation equation for R is
As shown in Fig. 8, if the number of AGVs is configured as 4, the
R ¼ minðC max ; Smax Þ ð3Þ
system throughput cannot reach the maximum when the load
capacity of the AGVs is 2. When the load capacity of the AGVs is 3 The maximum instantaneous capacity and the actual maximum
and the speed of the AGVs is 0.30 m/s, the system reaches configurable capacity in the 1000 experiments involving BF8–BF18
saturation. When the load capacity of the AGVs is 4 and the speed is are shown in Table 8, and it can be seen that most of the
0.24 m/s, the system throughput reaches saturation. When the load buffers have been fully utilized. The actual maximum instanta-
capacity of the AGVs is 5 and the speed is 0.20 m/s, the system neous capacities of BF10 and BF18 are only 3 and 2, respectively,
throughput reaches saturation. which are less than the maximum configurable capacities of 8 and
From the above analysis, it can be seen that the five 3. This means that buffers BF10 and BF18 with capacities of 3 and 2
configuration groups in Table 6 are all ideal results that can can meet the production requirements without affecting the
achieve the maximum throughput of the system with small throughput.
resource inputs.
It can be seen from Table 6 that the differences among the five
configurations are not significant. However, from an economic point
Table 7
of view, it is very cost effective to increase the load capacity of AGVs The results of the monitoring experiment.
and the speed of AGVs and reduce the use of one AGVs, so the number
BF8 BF9 BF10 BF11 BF12 BF13 BF14 BF15 BF16 BF17 BF18
of AGVs used is 3. Similarly, comparing the first and second scenarios,
it is much more cost effective to increase AGVs speed by 0.06 m/s Smax 6 6 3 8 2 2 2 3 3 3 2
than to increase AGVs load and capacity of the logistics buffer by 1
(the capacity of the logistics buffer is related to load capacity of
AGVs). Therefore, the first group of configuration plans was selected,
and the final bottleneck resource allocation strategy was as follows:
Table 8
the number of AGVs is 3, the load capacity of AGVs is 4, the speed of Allocation strategy of non-bottleneck resources.
AGVs is 0.32 m/s and the capacity of the logistics buffer is 6.
BF8 BF9 BF10 BF11 BF12 BF13 BF14 BF15 BF16 BF17 BF18
Monitoring experiment and its results Smax 6 6 3 8 2 2 2 3 3 3 2
Cmax 6 6 8 8 2 2 2 3 3 3 3
R 6 6 3 8 2 2 2 3 3 3 2
Through the analysis in Section “Bottleneck identification based
on the factorial experiment results”, the non-bottleneck resource is Bold indicates variance exist in buffer utilization.
G. Li et al. / CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405 403

Table 9
The time spent on the simulation experiment.

Factorial experiment Multilevel experiment Monitoring experiment Total


Experimental frequency 1600 16,000 1000 18,600
Total run time (s) 200.532 2214.674 161.446 2576.652
Average run time (s) 0.1253 0.1384 0.1614 0.1385

Discussion finding indicates that the system throughput under this allocation
strategy has reached a maximum. Moreover, it can be seen from
The time spent on the simulation experiments Fig. 9(b) that the production capacity is very stable.
The utilization of each work station and AGVs are shown in Fig.10.
All experiments are performed in Plant Simulation 14.0 The horizontal coordinate represents each work station or AGVs, and
software and run on a CPU i3-7100 computer with 4.00G memory. the vertical coordinate represents the proportion of each state of the
The time spent on the experiment in the application case is shown work station to the total time. As shown in Fig. 10, it can be seen that
in Table 9. As shown in Table 9, the three types of experiments had the utilization rate of each station is kept at a high level, and the
a total of 18,600 repetitions, with a total time cost of 2576.652 s. balance rate of the system is also in line with the reality.
Each experiment simulated the production situation within 24 h For the logistics system, the AGVs, as a bridge between the
and output the corresponding statistics; the average cost of each warehouse and production areas, play an important role;
experiment was 0.1385 s, which shows that the simulation is a very therefore, they are objects of great concern. The activities
high-speed process and that the experimental results can be performed by the studied AGVs can be divided into four cyclic
obtained in approximately 0.15 s. However, if the production steps: loading the parts from the warehouse, transporting the parts
experiment is carried out in a real environment, the real system to the destination, unloading the parts at the logistics buffer, and
needs to run continuously for 24 h. Not only is the time difference running empty to the warehouse for the next distribution task.
approximately 576,000 times different, but the resources con- These four activities correspond to the four operating states of
sumed in the production experiment can be reduced. In addition, it AGVs: occupied-runing, empty-running, loading and unloading.
is impractical to use actual production experiments to study Fig. 10 also shows the utilization of AGVs. The ratios of the occupied
throughput under all allocation schemes. These analyses indicate running and empty running statuses for the AGVs are approxi-
that the use of discrete event simulation software can significantly mately 0.3 and 0.54, respectively. According to the simulation
reduce the planning time and unnecessary production experi- model, this phenomenon is influenced by the distance between
ments while still achieving desirable allocation results. warehouse 1 and each logistics buffer. The proportions of the AGV
loading time and unloading time are approximately 0.12 and 0.03,
System performance after resource allocation respectively, with a difference of a factor of 4, which is due to the
loading and unloading strategies. Notably, the loading time is 20 s
The final allocation strategy is as follows: the number of AGVs is to load of each part, and the unloading time is 20 s in total, which is
3, load capacity of AGVs is 4, speed of AGVs is 0.32 m/s, capacity of consistent with the actual specifications.
the logistics buffer is 6, and capacities of the production buffers With all the analyses above taken into account, the production
BF8-BF18 are 6, 6, 3, 8, 2, 2, 2, 3, 3, 3, and 2. To test the final system capacity is highly efficient and stable after allocation, the
throughput, we ran the model 100 times and obtained the box plot equipment utilization rate is consistent with the actual situation,
in Fig. 9(a); the completion time of each part is shown in Fig. 9(b). which is also satisfactory. The AGV distribution is reasonable.
According to Fig. 9(a), the system throughput fluctuates between These results suggest that the final allocation strategy is highly
215 and 218 due to the uncertainty in the production process. The satisfactory, which indicates that the resource allocation strategies
median system throughput is 217, with an average of 216.6. This proposed in this paper is useful and practical.

Fig. 9. (a) Box plot of system throughput; (b) the completion time of each part.
404 G. Li et al. / CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405

Fig. 10. The utilization of each work station and AGVs.

Conclusion References

[1] Zhang, Z., David, J., 2019, An entropy-based approach for assessing the opera-
This paper proposes a methodology and simulation procedure tion of production logistics. Expert Syst Appl, 119:118–127. http://dx.doi.org/
for resource allocation in PLSs based on object-oriented discrete 10.1016/j.eswa.2018.10.044.
event simulation. Then, based on the proposed methodology, a [2] Huang, S., Guo, Y., Zha, S., Wang, Y., 2019, An internet-of-things-based produc-
tion logistics optimization methodology for discrete manufacturing. Int J
specific problem was solved through a case study to obtain the Computer Integr Manuf, 32/1: 13–26. http://dx.doi.org/10.1080/
optimal number of AGVs, speed of AGVs, load capacity of AGVs, and 0951192X.2018.1550671.
capacities of the logistics buffers and production buffers among [3] Agnetis, A., Aloulou, M.A., Kovalyov, M.Y., 2017, Integrated production sched-
uling and batch delivery with fixed departure times and inventory holding
160 schemes, and the maximum system throughput was achieved costs. Int J Prod Res, 55/20: 6193–6206. http://dx.doi.org/10.1080/
with the minimum economic input. Finally, the effectiveness of the 00207543.2017.1346323.
proposed resource allocation methodology is illustrated. Notably, [4] Zhang, Y., Liu, S., Liu, Y., Li, R., 2016, Smart box-enabled product-service system
for cloud logistics. Int J Prod Res, 54/22: 6693–6706. http://dx.doi.org/10.1080/
this paper has solved the following problems:
00207543.2015.1134840.
[5] Luo, H., Wang, K., Kong, X., Lu, S., Qu, T., 2016, Synchronized production and
(1) We used the methodology effectively solved the problem of logistics via ubiquitous computing technology. Rob Comput Integr Manuf, 45/
unbalanced and unreasonable resource allocation in a real PLSs and C: 99–115. http://dx.doi.org/10.1016/j.rcim.2016.01.008.
[6] Fu, L., Aloulou, M., Triki, C., 2017, Integrated production scheduling and vehicle
maximized the system output with minimized economic inputs. routing problem with job splitting and delivery time windows. Int J Prod Res,
(2) The bottleneck resources of the PLSs are identified through a 55/20: 5942–5957. http://dx.doi.org/10.1080/00207543.2017.1308572.
simulation factorial experiment, and the influence of various [7] Müller-Boyaci, P., Wenzel, S., 2016, Simulation toolkit for autonomous control
in serial production networks of automotive suppliers. J Simul, 10/2: 123–136.
resources on the production capacity is quantified: the higher the http://dx.doi.org/10.1057/jos.2016.5.
main effect is, the greater the influence degree, which also means [8] Hwang, H., Ahn, H., Kaminsky, P., 2016, Algorithms for the two-stage produc-
that when the available economic capacity is limited, the tion-capacitated lot-sizing problem. J Global Optim, 65/4: 1–23. http://dx.doi.
org/10.1007/s10898-015-0392-2.
factorial experiments can provide the priority for each resource. [9] Lanza, J., Ferdows, K., Kara, S., Mourtzis, D., Schuh, G., Váncza, J., Wang, L.H.,
Wiendahl, H.P., 2019, Global production networks: design and operation. CIRP
The novelty of the paper is that the methodology aims at PLSs, Ann, 68/2: 823–841. http://dx.doi.org/10.1016/j.cirp.2019.05.008.
[10] Schmidt, M., Schafers, P., 2017, The hanoverian supply chain model: modeling
which integrate many interactions, perform bottleneck analysis by
the impact of production planning and control on a supply chain’s logistic
a new approach, and propose allocation strategies according to objectives. Prod Eng, 11/1: 1–7.
different types of resources. The novelty of the study also lies in the [11] Trentesaux, D., Giret, A., 2015, Go-green manufacturing holons: a step towards
sustainable manufacturing operations control. Manuf Lett, 5:29–33. http://dx.
fact that it is not only applicable to factories in use but also
doi.org/10.1016/j.mfglet.2015.07.003.
applicable to factories in the late stage of planning when not yet [12] Nouiri, M., Bekrar, A., Trentesaux, D., 2019, An energy-efficient scheduling and
under construction, which is difficult to achieve for general rescheduling methodology for production and logistics systems. Int J Prod Res,
industrial advanced planning and scheduling (APS) systems. 1–21. http://dx.doi.org/10.1080/00207543.2019.1660826.
[13] Song, S.G., Li, A.P., Xu, L.Y., 2008, Buffer capacity optimization in reconfigurable
In the comparison and selection of allocation strategies in the manufacturing system. Comput Integr Manuf Syst, 10.
multilevel experiment, the evaluation criterion of “maximize the [14] Zhang, Z., Wang, X., Wang, X., Cui, F., Cheng, H., 2019, A simulation-based
production capacity while minimizing economic inputs” is still approach for plant layout design and production planning. J Ambient
Intell146?Hum Comput, 10/3: 1217–1230. http://dx.doi.org/10.1007/
vague. To overcome this deficiency, we will establish a mathemat- s12652-018-0687-5.
ical evaluation model of PLSs in future research. [15] Sawhney, A., Abudayyeh, O., Monga, A., 1999, Modelling and analysis of a mail
processing plant using Petri nets. Adv Eng Software, 30/8: 543–549. http://dx.
doi.org/10.1016/S 0965-9978(99)00011-00013.
Acknowledgments [16] Dimitris, M., 2019, Simulation in the design and operation of manufacturing
systems: state of the art and new trends. Int J Prod Res, 1–23. http://dx.doi.org/
10.1080/00207543.2019.1636321.
The authors are grateful for the support provided by the [17] Coelho, P., Silva, C., Ferreira, L.M., Franca, B., 2018, Operational improvement of
Shenyang Two-hundred Engineering Science and Technology an industrial equipment rental system using discrete event simulation. IFAC
Papers Online, 51/11: 478–483. http://dx.doi.org/10.1016/j.ifacol.2018.08.364.
Program (grant number: Z17-7-002) and thank the managers of
[18] Donhauser, T., Rackow, T., Hirschbrunn, J., Schuderer, P., Franke, J., 2016, Valid
the production logistics system investigated in the case study for methodology for using discrete event simulation to improve the resource
their timely communication and feedback. We also thank the consumption for the manufacturing of masonry units. Procedia CIRP, 41:57–
anonymous reviewers and editors for their valuable comments. 62. http://dx.doi.org/10.1016/j.procir.2015.12.016.
G. Li et al. / CIRP Journal of Manufacturing Science and Technology 31 (2020) 394–405 405

[19] McNally, P., Heavey, C., 2004, Developing simulation as a desktop resource. Int J [27] Staley, D.R., Kim, D.S., 2012, Experimental results for the allocation of buffers in
Computer Integr Manuf, 17/5: 435–450. http://dx.doi.org/10.1080/ closed serial production lines. Int J Prod Econ, 137/2: 284–291. http://dx.doi.
09511920310001654283. org/10.1016/j.ijpe.2012.02.011.
[20] Wang, H., Li, L., Jiao, Y.Y., Ge, X.R., Li, S.C., 2014, A relationship-based and object- [28] Sena, D.C., Silva, E.M.M., Costa, A.P.R., Montevechi, J.A.B., Pinho, A.F., Miranda,
oriented software for monitoring management during geotechnical excava- R.C., 2017, Dynamic allocation of additional human resources using hybrid
tion. Adv Eng Software, 71:34–45. http://dx.doi.org/10.1016/j.adven- simulation. Int J Simul Modell, 16/1: 84–95. http://dx.doi.org/10.2507/
gsoft.2014.02.001. IJSIMM16(1)7.371.
[21] Gowri, A., Venkatesan, K., Sivanandan, R., 2009, Object-oriented methodolo- [29] Lafortune, S., Lin, F., Hadjicostis, C.N., 2018, On the history of diagnosability and
gyology for intersection simulation model under heterogeneous traffic con- opacity in discrete event systems. Annu Rev Control, 45:257–266. http://dx.
ditions. Adv Eng Software, 40/10: 1000–1010. http://dx.doi.org/10.1016/j. doi.org/10.1016/j.arcontrol.2018.04.002.
advengsoft.2009.03.015. [30] de Sousa Junior, W.T., Montevechi, J.A.B., de Carvalho Miranda, R., Campos, A.T.,
[22] Junior, W.T.D.S., Montevechi, J.A.B., Miranda, R.D.C., Oliveira, M.L.M.D., Cam- 2019, Discrete simulation-based optimization methodologys for industrial
pos, A.T., 2020, Shop floor simulation optimization using machine learning to engineering problems: a systematic literature review. Comput Ind Eng,
improve parallel metaheuristics. Expert Syst Appl, 150113272. http://dx.doi. 128:526–540. http://dx.doi.org/10.1016/j.cie.2018.12.073.
org/10.1016/j.eswa.2020.113272. [31] Fortmann-Roe, S., 2014, Insight Maker: a general-purpose tool for web-based
[23] Al-Turki, U.M., Saleh, H., Deyab, T., Almoghathawi, Y., 2012, Resource allocation modelling & simulation. Simul Modell Pract Theory, 47:28–45. http://dx.doi.
and job dispatching for unreliable flexible flow shop manufacturing system. org/10.1016/j.simpat.2014.03.013.
Advanced Materials Research, vol. 445. Trans. Tech. Publications. p. 947–952. [32] Shang, S., Yun, G.J., 2013, Stochastic finite element with material uncertainties:
http://dx.doi.org/10.4028/www.Scific.net/AMR.445.947. Implementation in a general purpose simulation program. Finite Elem Anal
[24] Seitaridis, A., Rigas, E.S., Bassiliades, N., Ramchurn, S.D., 2019, An agent-based Des, 64:65–78. http://dx.doi.org/10.1016/j.finel.2012.10.001.
negotiation scheme for the distribution of electric vehicles across a set of [33] Li, X., Yuan, Y., Sun, W., Feng, H., 2016, Bottleneck identification in job-shop
charging stations. Simul Modell Pract Theory, 102040. http://dx.doi.org/ based on network structure characteristic. Comput Integr Manuf Syst, 4:023.
10.1016/j.simpat.2019.102040. [34] Watson, K.J., Blackstone, J.H., Gardiner, S.C., 2007, The evolution of a manage-
[25] Libing, Y., Hanhong, C., Yuncai, C., Haiyang, Y., 2008, Simulating and optimizing ment philosophy: The theory of constraints. J Oper Manage, 25/2: 387–402.
of logistics system in strip mines. Proceedings 3rd international symposium on http://dx.doi.org/10.1016/j.jom.2006.04.004.
modern mining & safety technology proceedings, 138–143. [35] Shi, W., Yan, H., 2006, Method of shifting bottleneck analysis in knowledge-
[26] Yang, S.L., Xu, Z.G., Wang, J.Y., 2019, Modelling and production configuration oriented manufacturing system. Comput Integr Manuf Syst, 12/2: 271–279.
optimization for an assembly shop. Int J Simul Modell (IJSIMM), 18/2. http://dx. http://dx.doi.org/10.13196/j.cims.2006.02.113.shiW.020.
doi.org/10.2507/IJSIMM18(2)CO10.

You might also like