République Tunisienne

Ministère de l’Enseignement Supérieur

et de la Recherche Scientifique

Université de Tunis El Manar

École nationale d’ingénieurs de Tunis

Industrial engineering department

second year final project

Towards Digital Twin of a Flexible manufacturing

system with AGV

Prepared by :
Yasmine BEL HADJ SALAH

Wassim SOUISSI

President : Atidel HAJ ALOUANE

Supervised by : Amel JAOUA

Reporter : Moez KALEL

Class : 2AGI1

Academic year 2020/2021

Abstract

In today’s competitive environment, the Flexible Manufacturing System (FMS)

has evolved as a solution to the requirements of modern manufacturing industries.
The Automated Guided Vehicle (AGV) system is the most widely used material
handling system in FMS that improves the flexibility in material transfer between
workstations (WS). However, managing a processing system based on automatic
controlled vehicles (AGV) is a complex entry. In fact, the AGV configuration,
has become the biggest problem in the AGV system. The implementation
and management of such AGV networks, which are characterized by the
interdependence of several elements, has become too complex. So, we present two
approaches for modeling these systems: analytic approach and simulation. Then
we stick to the second approach to come up with a simulator that will drive these
types of AGV-based systems. In our work, we determined the optimal number of
AGVs by comparing the analytical result with the simulation taking into account
the traffic management and dispatching strategies. We then determined the
robustness of this solution to the variability in operating times.

ii
Dedication

We begin by thanking Ms. Amel JAOUA who gave us the honor to be our
supervisor. We deeply thank her for her continuous encouragement and for being
always there to listen to us, to help us and to guide us to find the right path by her
wisdom and her precious advice. As well as her moral support and her proof of
understanding, which gave us the strength and courage to accomplish this project.
We would also like to thank the respectable members of the jury for giving us their
precious time to comment, discuss and judge our work.
With great respect, we would like to express our great gratitude to all the teachers
and administrators of the National School of Engineering of Tunis.

iii
Contents

General introduction 1

1 State of the art 2

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Digital twin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 the relationships between digital twins and simulation models . . . . . . . . . . . . . . 3
1.3 Flexible Manufacturing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.2 Basic components of FMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3.3 Objectives of FMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Automated Guided Vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.2 Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.3 Industrial applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4.4 Example of an automated guided vehicle . . . . . . . . . . . . . . . . . . . . . 8
conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Simulation Modeling on ARENA 10

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1 Definition and ways of simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 The role of simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Presentation of the ARENA tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Setting windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4.1 AGVs’ parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4.2 AGVs’ management algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4.3 The production line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Implementation for a typical application 16

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1 Presentation of the model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.1 Path layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

iv
 3.1.2 Conflict-Free Checking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Presentation of the simulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Estimation of AGV’s number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3.1 Deterministic model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3.1.1 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3.1.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.2 The effect of process time variability on AGVs’ number . . . . . . . . . . . . . . 28
3.3.2.1 Number of replications . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.2.2 Process time variability . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

General conclusion 33

Bibliography 34

v
List of Figures

1.1 Digital model,shadow,and twin[7] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2 types of flexibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 characteristics of FMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 The AGV 2ACB2.5 [8] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 The AGV 2ACB2.5 caracteristics [9] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.6 The AGV 2ACB2.5 specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1 ARENA library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.2 The setting window . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 The setting window . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 The setting window . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.1 General 2D model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.2 Path layout of bi-directional model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Detailed 2D model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4 Model simulator screen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.5 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.6 The flow chart of the estimate and simulation approach [10] . . . . . . . . . . . . . . . 21
3.7 Machines’ time operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.8 The nearest AGV dispatching rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.9 Total Makespan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.10 Workstation Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.11 Cyclic AGV dispatching rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.12 Total Makespan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.13 Workstation 5 Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.14 Random choice of free AGV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.15 Total Makespan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.16 Workstation 5 Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.17 Process time using uniform distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.18 Low variability’s results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.19 Overlapping using output analyzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.20 Medium variability’s results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

vi
3.21 Overlapping using output analyzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.22 High variability’s results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.23 Overlapping using output analyzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

vii
General introduction

Manufacturing systems have become more and more efficient but also more and more complex. Controlling
and manipulating them successfully has become a very delicate and difficult task to implement.
In this context, several research works have been carried out to improve the control of these manufacturing
systems. These works are mainly oriented towards the control of machines, scheduling, operations
planning, and supervision. Indeed, the arrival of AGVs in manufacturing systems has made a radical
change in the production environment thanks to their great flexibility, possibility of adaptation to
other production equipment and reliability. They are nowadays considered as the most appropriate
handling equipment for a flexible production environment.
In addition, optimizing machines’ time operation is not enough to improve productivity. In fact,
parts generally only spend a small part of their manufacturing time on processing stations. Therefore,
reducing handling times becomes an important criterion in assessing the production function since
the input of the raw material until the finished or semi-finished product is released. We then became
interested in the handling function in the manufacturing systems.
So, in this report, we will begin with explaining the basic notions of our project such as AGVs, the
simulation software, and simulation models and their relationships with digital twin, then, we will use
analytical approches to determinate the optimal AGVs’ number in manufactering systems, and we will
use a case of study to explain it better.

1
 Chapter 1

State of the art

Plan
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1 Digital twin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 the relationships between digital twins and simulation models . . . . . . . 3

3 Flexible Manufacturing System . . . . . . . . . . . . . . . . . . . . . . . . . . 4

4 Automated Guided Vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Chapter 1. State of the art

Introduction

In this chapter, after first elaborating the digital twin and his relationship with simulation models,
we will define the flexible manufacturing systems FMS and discuss the role of AGVs in such system.
Moreover, we will clarify the technical side of AGVs, then we will discuss the technologies used on
AGVs, and we will cite an example of AGV used in our case. Finally, we conclude with a summary of
the different parts presented in this chapter.

1.1 Digital twin

The Digital Twins, themselves are characterized by a seamless integration of the cyber and physical
spaces. The importance of Digital Twin is increasingly recognized by universities and industry. It’s
been almost 15 years since the concept was first proposed.
Many Digital Twin applications are implemented with success in several industries including product
design, production, forecasting and health management, and a few other areas. However, at this point
no article has focused on reviewing Digital Twin’s claims in the industry.
The first appearance of the Digital Twin dates back to 2003, Grieves introduced the concept, for the
first time, in his course on " product lifecycle management".[1]
Although the concept is not precise enough at this point, a preliminary kind of Digital Twin was
proposed to incorporate three parts : the physical product, the virtual product and their connections.
Enabling technologies for Digital Twins have grown since then. In 2012, Digital Twin’s concept was
revisited by the National Aeronautics and Space Administration, which defined Digital Twin as a
multiphysical, multiscale, probabilistic, ultrafidelity simulation reflects, in a timely manner, the state
of a corresponding twin supported historical data, data from real-time sensor physical model.[2]
Digital Twins are becoming a popular research subject. Conforming to Gabor et al [3], the Digital
Twin can be a special simulation, built based on expert knowledge and the actual data from the
current system, to understand a more precise at multiple scales of your time and space. According
to Maurer [4], the Digital Twin can be a representation that will illustrate the assembly process and
the performance of the commodity. The meaning of Digital Twin becomes more and more since then,
giving particular notions like cell digital twin and experimental digital twin.[5]

1.2 the relationships between digital twins and simulation models

Many people might imagine that simulation models are digital twins. the very fact is that a digital
twin are often a simulation model, but a simulation model might not necessarily be a digital twin.

3
Chapter 1. State of the art

Digital models utilized in simulations often have an equivalent sort of sensor information and controls
of a digital twin, but the information might be generated and manipulated within the simulation. The
simulation may replicate what could happen within the world , but not necessarily what’s currently
happening (Wong 2018). Kritzinger[6] et al. (2018) propose a classification of digital models into
three subcategories supported their level of knowledge integration between the physical and the digital
counterparts: (1) Digital model: a digital representation of an existing or planned object with none
sort of automated data exchange between the physical and digital objects. Most of the present offline
simulation models are this type of digital model; (2) Digital shadow: a digital model with an automatic
one-way data flow between the physical and digital objects, e.g., a simulation model using real-time
sensor data as inputs (Yang et al. 2017); (3) Digital twin: a digital model with bi-directional data
flow between the physical and digital objects, e.g., a simulation model that uses real-time sensor data
as inputs and updates a number of the parameters of a producing process or equipment.

Figure 1.1: Digital model,shadow,and twin[7]

1.3 Flexible Manufacturing System

Face to the complexity, the competitivity, and the speed-change of the market, the development and
implementation of a flexible manufacturing system became a necessity for all companies.

1.3.1 Definition

“FMS consists of a group of processing workstations interconnected by means of an automated material

handling and storage system and controlled by integrated computer control system.” (flexible manufacturing
system, book).
FMS is based on automatization and flexibility of the process. It can process various part styles
simultaneously at the workstation, handling flexible routing of parts through machines in a manufacturing
system, and quantities of production are often adjusted in response to changing demand patterns.

4
Chapter 1. State of the art

Figure 1.2: types of flexibility

An FMS can have the efficiency of a transfer line and the flexibility of a job shop. The flexibility
enables an FMS to rapidly adjust to the parts varieties according to the changing market conditions.

Figure 1.3: characteristics of FMS

1.3.2 Basic components of FMS

The basic components of FMS are:

• Workstations:
It is a set of machines or other equipment capable to operate one or many tasks at the same time.
The various workstations are machining centers; load and unload station; Assembly workstation;
inspection stations. . .

5
Chapter 1. State of the art

• Automated Material Handling and Storage system:

Their principal function is transporting the parts between the workstations. There are many
types of automated material handling but the most known are automated guided vehicle (AGV).

• Computer Control System:

It is used to monitor all the activities. Its main goal is to optimize the traffic and the production
process performance.

1.3.3 Objectives of FMS

The principal objectives of FMS are:

• To improve operational control through:

* Reduction in the number of uncontrollable variables
* Providing tools to recognize and react quickly to deviations in the manufacturing plan
* Reducing the dependence of human communication

• To reduce direct labor:

* Removing operators from the machining site (their responsibilities activities can be broadened)
* Eliminating dependence on highly skilled machines (their manufacturing skills can be better
utilized in manufacturing engineering functions)
* Providing a catalyst to introduce and support unattended or lightly attended machining
operation

• To improve short run responsiveness consisting of:

* Engineering changes
* Processing changes
* Machining downtime or unavailability
* Cutting tool failure
* Late material delivery

• To improve long-run accommodations through quicker and easier assimilation of:

* Changing product volumes
* New product additions and introductions

6
Chapter 1. State of the art

* Differentiation part mixes

* Increase Machine Utilization by:
- Eliminating machine setup
- Utilizing automated features to replace manual intervention. - Providing quick transfer devices
to keep machines in the cutting cycle
* Reduce inventors by:
- Reducing lot sizes
- Improving inventors turn-over
- Providing the planning tools for JIT manufacturing

1.4 Automated Guided Vehicle

1.4.1 Definition

First of all, an automated guided vehicle or automatic guided vehicle (AGV) is a mobile robot that
follows the markers or the wires in the soil or uses a vision or laser. They are most often used
in industrial applications to move materials around a production facility or stock. AGV increases
efficiency and reduces the cost of helping to automate a production facility or stock.
Feature-wise, any installation requiring speed and regularity is potentially automated using AGVs.
Such an achievement must essentially respect constraints following:
* The structure of the AGV traffic network will have to take particular account of medium and
instantaneous speed constraints of production machines.
* The space reserved for the circulation of AGVs shall take account of the flows insured and the
concentration of access points. The more positions are grouped and the more access shall be planned.
This can be seen by analogy with road traffic where the transport capacity depends on the size of the
circuit.

1.4.2 Function

The main function of AGVs is transport for production. So they’re connected directly with production
equipment. Requirements information is centralized by the AGV system which then supplies the
requested products and evacuates the output products. The fact that AGVs are not integrated into
the production function can allow the transport of the products by conventional means..
AGVs have two modes of work:

• Carrier:

7
Chapter 1. State of the art

they carry the load, and they can be equipped with a lifting device, allowing them to adapt to
different levels of intake and removal. The loading/unloading system may be a chain or roller
element, in a technology similar to that of the ground conveyors; or a push-chain device, a
shooter; etc. The maximum loads that can be reasonably transported by carrier vehicles are in
the range of 1 500 to 2 000 Kg;

• Tractor:
they pull one or more loaded trailers. The coupling of the trailers shall then be manual; however,
the detachment can be automatic. A vehicle can thus tow up to 10 tons and more.

1.4.3 Industrial applications

Automated Guided Vehicle (AGV) is a robot that moves autonomously without human intervention.
Traditional automation technologies are wire-guided, laser-guided and self-guided. However, other
equally powerful technologies exist such as geo guidance and ultrasonic guidance. AGVs are most often
used in industrial applications to move goods autonomously in a factory, a warehouse or a workshop.
Thus, the term automatic truck for an AGV originates from the first AGVs on the market: industrial
trucks for transporting pallets. Indeed, the first AGV saw the light of day in the the 1950s, initially
developed for the automotive industry. Nowadays, their applications have been extended and they are
no longer limited to the industrial sector. They are also used in hospitals, museums, airports...

1.4.4 Example of an automated guided vehicle

We have chosen the "2ACB2.5" ( Fig 1.4) as an example of an Automated Guided Vehicle.

Figure 1.4: The AGV 2ACB2.5 [8]

8
Chapter 1. State of the art

This vehicle is a 3-wheel type truck. The carrier comes into the roller table (2-serial, 3-serial,
and 2-stage type), lifter, and push-pull type useful in a variety of applications. Also, the automatic
charging and autonomous radio standby functions are available as options. This AGV is 250kg loading.
The technical characteristics of the AGV "2ACB2.5" are presented in Fig.5.

Figure 1.5: The AGV 2ACB2.5 caracteristics [9]

We present the specifications of the 2ACB2.5 in this figure below.

Figure 1.6: The AGV 2ACB2.5 specifications

Conclusion

As a conclusion, we notify that digital twin will be the future of basic simulation. The integration
of such tool in a flexible manufacturing system requires real time information and automatization of
operations. Finally, we see the importance of AGV as a material handling system.

9
 Chapter 2

Simulation Modeling on ARENA

Plan
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1 Definition and ways of simulation . . . . . . . . . . . . . . . . . . . . . . . . . 11

2 The role of simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Presentation of the ARENA tool . . . . . . . . . . . . . . . . . . . . . . . . 12

4 Setting windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Chapter 2. Simulation Modeling on ARENA

Introduction

In this chapter, we will explain the role of simulation and we will present the Arena software which is
a mixture of the two others, called Hybrid, and how to use it, and its setting windows.

2.1 Definition and ways of simulation

Simulation is one of the ways to model complex systems such as production systems. It is a technique
which consists, often with the help of a computer, in the construction of a model of an existing system,
or that one wants to conceive; with the aim of analyzing its behavior under certain conditions.One
of its main advantages is its ability to model dynamic behaviors over time. The main advantage for
simulation is to have benefited from the advances in computer software. And even if it does not bring
an "analytical" solution insofar as the results cannot be exact, it makes it possible to work on realistic
and complex models, which make it possible to exploit its power at best.
If we start from the following principle: "It is better to obtain approximate answers to an exact
problem, than to obtain exact answers to an approximate problem", we understand why the simulation
is regarded in certain cases as much more useful and powerful than the mathematical approaches.
There are several ways to develop simulation models:

• Programming language: this is a rather slow but extremely versatile approach. We can
compare it to languages like C++ or Pascal although these two softwares are harder to learn.
We quote for example SLAM, ECSL, GPSS, SIMSCRIP.

• Simulators: an easy and fast approach thanks to very complete graphical interfaces nowadays.
However, they are less versatile, as they are often restricted to certain fields of application
(production, communication). Example: Witness, ProModel, Taylorll.

2.2 The role of simulation

Simulation is one of the ways to study the temporal or dynamic behavior of a system. For that there
are well defined methods to build the model, but most importantly to ensure its validity. We work in
three steps:

• Modeling: Using a program, we build a computer model of the system to be studied

• Simulation: We run the model in order to generate results

• Analysis of the results: From these results, we analyze the functioning of the system through
performance criteria and we take decisions at the conceptual or operational level.

11
Chapter 2. Simulation Modeling on ARENA

If the results are not as we expected then we can revise the model.

The construction of the model always involves a form of software. The two main parameters for discrete
time simulation are the entities and the logical relations.
Entities represent tangible elements of the real world, for example product in a production system,
or information in a communication system. They can be temporary; like parts that pass through the
system; or permanent; that remain in the model.
Logical connections interconnect the different entities and resources, for example a machine that will
work on a part. They are one of the key elements of the simulation since they allow to define the global
behavior of the model.
Another key element of all simulation systems is the executor because it controls the time advance. It
checks the logical relationships between the entities and advances the clock to the next moment. It is
the central element that gives its temporal and dynamic behavior to the system. The last two elements
are the random number generator that is used to obtain stochastic behaviors similar to those of the
real world and the results collection and analysis module that allows the user to use the simulation
tool to analyze the system he is working on.

2.3 Presentation of the ARENA tool

The "SIMAN-ARENA" simulation software was created in 1982 by C.D Pedgen of System Modeling
Corporation. It is a simulation language of process interaction type.
This simulation software is especially adapted to the simulation of manufacturing systems. We will
first look at the basic notions of its use.
The basic idea when modeling a system with Arena is that of flow, which is the movement of entities
along a given path, and entities are the basic object that represents a physical or informational entity,
means that any object moving in the system.
The system will be modeled by a set of interconnected blocks that will describe the flows of the system.
The path followed by the entities will be modeled by a succession of interconnected blocks that each
have one or more inputs and outputs. When an entity crosses a block, the function associated with it
is performed.
Beyond the model construction with the blocks, the Arena logic part has two other important aspects
such as animation, we can perfectly content ourselves with the logic part, however it is much more
interesting to be able to visualize and study the dynamics of flows in the model.
In addition, several statistical data can be extracted through a simulation report that shows many
important performance indicators related to resources, stocks or entities.

12
Chapter 2. Simulation Modeling on ARENA

Moreover, ARENA can be integrated with Microsoft technologies such as Visual Basic, Microsoft Visio,
Excel or Access. This advantage allows to automate several models and therefore to support models
with a large amount of data, which become closer to the real system.
So, some of the existing blocks in the ARENA library can be listed in the following table below:

Figure 2.1: ARENA library

2.4 Setting windows

Before runing the model in Arena, we have to configure it using an input windows that allows the user
to enter the parameters of the model.
in our model, we have three setting windows to configure; AGVs’ parameters, AGVs’ managment
algorithm, and the production line.

13
Chapter 2. Simulation Modeling on ARENA

2.4.1 AGVs’ parameters

In this setting window, the user have to enter the AGVs’ parameters such as the AGVs’s number,
AGVs’ speed, the loading and the unloading time.

Figure 2.2: The setting window

2.4.2 AGVs’ management algorithm

In this setting window, we have to choose the AGVs’ management algorithm that we want to use in
our model. In fact, there are three types of AGVs’ scheduling; the selection of the nearest free AGV,
the cyclic selection of AGV, and the random choice of free AGV.

Figure 2.3: The setting window

2.4.3 The production line

In this window, we have to choose the line production of our three products; P1, P2, P3, and we have
the number of products that we want to create. Also, we choose the "Job Shop" scheduling to manage
our production line.

14
Chapter 2. Simulation Modeling on ARENA

Figure 2.4: The setting window

Conclusion

As a conclusion, this simulation software called Arena is too powerful because it facilates modeling
complex systems without using any physical equipment, added to that, the speed of Arena’s calculation
is one of its major assets.

15
 Chapter 3

Implementation for a typical

application

Plan
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1 Presentation of the model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2 Presentation of the simulator . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3 Estimation of AGV’s number . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Chapter 3. Implementation for a typical application

Introduction

In this chapter, we will present our 2D model made with Visual Studio, and our simulator made
with SIMAN/ARENA, then we will explain how to search the optimal number of AGVs with fixed
machines time operation using two steps: First step is analytical and the second using simulation.
After that, we will add variability to the process time and study the effect of stochasticity on the
system.

3.1 Presentation of the model

In our case, we have decided to work on the example of a flexible workshop made up of five workstations
that can handle three product lines ; P1, P2, P3.
The transports are performed with a fleet of automated guided vehicles (AGV) following unidirectional
circuits in a conventional configuration. The products arrive at the entrance of the system in a main
stock. In addition, each workstation has two unrestricted stocks at the input and at the output in
order to simplify the study. The upstream stock of each machine allows the waiting for the availability
of the station. The downstream stock allows the waiting of the AGV. In the study of this system, we
have taken into consideration the following conditions:

• Each product has a configurable routing.

• The number of vehicles as well as their travel speeds are controllable parameters in our system.

• The scheduling of the vehicles can follow three different algorithms:

a)Selection of the nearest free AGV:
This technique consists in selecting all free vehicles and then calculating the distance, separating
each vehicle from the location of the request. The closest vehicle will then be called.
b)Cyclic selection of AGV:
The idea of this technique is that none of the vehicles is overused; even if the vehicle i is the
closest to the request, its next mission will only be possible once all the other vehicles have also
completed a number of missions greater than or equal to that completed by this vehicle.
c)random choice of free AGV:
The idea of this technique is that we choose randomly an AGV i to do the mission j.

• The operation of the vehicles can be seen as a succession of periods of good operation which
means that all operations are achieved without breaking down.

• The operating times for the five stations and particularly the loading and unloading times are

17
Chapter 3. Implementation for a typical application

deterministic variables.

• The simulations are terminating, means that we start at t=0 with an empty system and we end
when all the parts to be processed are in the depots.

3.1.1 Path layout

To study this system we have established the surface plan of the process (Fig.3.1). This consists in
describe the essential structure of the system in a graphical formalism.

Figure 3.1: General 2D model

In fact, this plan corresponds to the representation of the surfaces used by the studied process with
the indication of the distances covered by the flow. The path layout is as follow:
The Park is the initial position of AGVs. After each trip, they return to the Park and stay on standby.
The first line “N” where the 5 workstations are located, and the second line “M” is parallel to “N”.
Between the two lines we have the bridges which are short ways to move from “N” to “M” or the
inverse.All bridges have the same length. Each station has two buffers; One for the pickup and another
for delivery.In fact, buffer is a bridge that transports the part from the AGV to the station (pickup
action) and the inverse (delivery action).
B+ (the red lines): one direction bridges from line “M” to line “N”.
B- (the blue lines): one direction bridges from line “N” to line “M”.

3.1.2 Conflict-Free Checking

For the calculation, we have been inspired by “QIU Ling, HSU Wen-Jing” article: conflict free AGV
routing in a Bi-directional path layout. As a conclusion, to avoid collision between AGVs we must
have:
D ≥ 2(1 + r)l + (1 + r)la (3.1)

18
Chapter 3. Implementation for a typical application

1+r 1 1 1+r l
l + la ≤ lb ≤ D − l − la (3.2)
2r 2r 2r 2r 2

with :
D: the distance between two adjacent stations.
r: the velocity of AGVs running on bridges, where r 1.
Lb: the length of a bridge.
la: the length of an AGV, including safety allowance which protects vehicles from collisions.
l: the half length of the edges of the junction.

Figure 3.2: Path layout of bi-directional model

In our case, our model is unidirectional, so we made some modification to get the following results.First,
let us introduce the following notations:
S: minimal distance between a delivery buffers of workstation i and the pickup buffers of workstation
i+1.
e: minimal distance between two buffers.
Db: minimal distance between two bridges.
Lb: minimal distance between the two lines ‘M’ and ‘N’.
sd: security distance (10% of the calculated distance).

Figure 3.3: Detailed 2D model

19
Chapter 3. Implementation for a typical application

S ≥ w/2 + w + w + w/2 + sd = 3 ∗ 1, 6 + 0, 48 = 5, 28m (3.3)

e ≥ w/2 + w/2 + sd = 1, 6 + 0, 16 = 1.76m (3.4)

Db ≥ w/2 + w/2 + sd = 1, 6 + 0, 16 = 1.76m. (3.5)

Lb ≥ w + w/2 + sd = 1, 6 + 0, 8 + 0, 24 = 2.64m. (3.6)

Because arena does not work with float measures, we had to take the following distances: S=6m;
e=2m; Db=2m ; lb=3m

3.2 Presentation of the simulator

This simulator was developed with SIMAN/Arena using the approach and design results of the previous
results of the previous paragraph.
The graphical interface of this simulator allows us to visualize the operation of the system in an
animated way (Fig.3.3). We can also observe the transfer of parts, the machines in production, the
AGVs in movement in the network.

Figure 3.4: Model simulator screen

This graph can be described as follow:

20
Chapter 3. Implementation for a typical application

Figure 3.5: Nomenclature

3.3 Estimation of AGV’s number

The number of AGVs influences the performance of a flexible manufacturing system. These vehicles
are usually expensive and the utilization of smaller or higher vehicles number then it is needed will
affect the performance of the system as we will see further.
Practically, the precise number of required vehicles cannot be given by any models. The exact number
can be smaller or higher than the estimated on depending on characteristics of the estimation model.
To determine the best number of AGV we will estimate the number by calculation as a first step then
we will validate with simulation.

Figure 3.6: The flow chart of the estimate and simulation approach [10]

21
Chapter 3. Implementation for a typical application

The optimal number of AGVs cannot be researched by any mathematical model but we do the
calculation at first in order to decrease the simulation time.
In the second step, the simulation of the material handling system has operated to optimal the number
of AGVs. We will take the estimate value in first step as the initial number in the simulation then
we will increase or decrease this number and evaluate the system performance in order to validate the
results. Without simulation it will be impossible to determine the optimal number of AGVs, due to
the complexity of this type of systems.
Added to the system complexity, the AGVs has some most key related issues including guide
path design, vehicle dispatching, idle vehicle positioning, battery management, vehicles speed, vehicle
capacity and deadlock resolution. To decrease the difficulty and the complexity of the system we did
the following assumption:

• AGVs are always available, and we do not considerate the battery management as a problem.

• We have three dispatching rules.

• All AGVs have a constant speed equal to 1m/s.

• Each vehicle can transport one product at each trip.

• We have one-direction path.

• Idle vehicles remain at their last position until they get a new assignment.

• The loading time is 10s and the unloading time is 10s.

• The number of products that we have fixed is equal to 50.

• the production line of the 3 products are as follows:

P1:M4=>M1=>M5
P2:M2=>M5
P3:M3=>M5
The speed of AGV, Loading/Unloading time are the closes to real situation, and the processing
time is fixed. We choose 50 products because they are quite enough to show us the performance
of the system because the time between arrivals is fixed : one product each 30s with equal
distribution for the three products.

3.3.1 Deterministic model

In this type of model, we choosed that the machines’ time operation is determinist. In fact, we have
fixed these processing time as follow:

22
Chapter 3. Implementation for a typical application

Figure 3.7: Machines’ time operation

The estimation model:

Many models have been proposed in the literature to estimate the number of AGVs. We have the
method proposed by P.J Egbelu :[11]

nv = T T /60T − Lt (3.7)

with:
- Nv: number of required vehicles.
- TT: estimated total operational time for vehicles(min).
- T: available time (min).
- Lt: expected lost time by each vehicle during a time period or shift due to battery change (min).
This model takes into consideration the battery management.
In our case we will use a simple equation used to determine the number of required resources:

nv = T W/f (3.8)

with:
- TW: total work time from and to the pickup station.
- f: the frequency of products arrival.

T W = T T + LU T + AM T (3.9)

With:
-TT: total trip time from the pickup station and back to it
-LUT: average total load and unload time
-AMT: average total machine processing time
This model gives us the required number of vehicles in order to manage the arrival frequency. In our
model, we have three different jobs, so we will take the average machine working time, the average

23
Chapter 3. Implementation for a typical application

total load/unload time and add to it the time needed to complete a trip.
For P1 MT=90s, for P2 MT= 45s, for P3 MT= 50s:

AM T = 62s (3.10)

T T = 104s (3.11)

LU T = 70s (3.12)

T W = T T + LU T + AM T = 104 + 70 + 62 = 236s (3.13)

f = 30s (3.14)

N v = 7, 86 = 8vehicles (3.15)

So, we will simulate only for numbers close to 8. It is a waste of time to simulate a smaller or a higher
number then the estimated value and this is the importance of this first step.

3.3.1.1 Simulation results

We did the simulation by increasing the number of AGV from 3 to 12 and see the effect on the makespan
and the performance of the system. As a key performance indicator, we took the makespan and the
bottleneck machine number 5 utilization rate. The number of AGVs which has the smallest makespan
and the maximum resources’ utilization is the optimal value.

• The nearest AGV dispatching rule:

Figure 3.8: The nearest AGV dispatching rule

24
Chapter 3. Implementation for a typical application

Figure 3.9: Total Makespan

Figure 3.10: Workstation Utilization

For this dispatching rule the optimal number is 9 AGVs with a total makespan equal to 1765s
and rate of utilization equal to 57%.

• Cyclic AGV dispatching rule:

Figure 3.11: Cyclic AGV dispatching rule

25
Chapter 3. Implementation for a typical application

Figure 3.12: Total Makespan

Figure 3.13: Workstation 5 Utilization

For this dispatching rule the optimal number is 9 AGVs with a total makespan equal to 1836s
and rate of utilization equal to 54%.

26
Chapter 3. Implementation for a typical application

• Random AGV dispatching rule:

Figure 3.14: Random choice of free AGV

Figure 3.15: Total Makespan

Figure 3.16: Workstation 5 Utilization

For this dispatching rule the optimal number is 9 AGVs with a total makespan equal to 1909s and rate
of utilization equal to 52%.

27
Chapter 3. Implementation for a typical application

3.3.1.2 Analysis

We see that in this condition of simulation the nearest AGV dispatching rule give us the best makespan,
and the maximum rate of machine utilization.
We conclude that there is no big difference between the various dispatching rules, but we think that in
real conditions like AGV breakdowns, AGV battery live, and stochastic process time those dispatching
strategies will affect the system performance.
In our case 9 AGVs is the optimal number.It is close to the estimated number by calculation but the
simulation gives us more accurate results because it takes into consideration the dependence between
resources and entities.
We clearly see that adding more AGV’s (nv >9) does not improve the system performance.
Practically, it increases congestion in the system and add more deadlocks.
We conclude that this study shows us the importance of the research of the optimal AGV number in
the system and how it affects the system performance.

3.3.2 The effect of process time variability on AGVs’ number

In the previous part, this process time was fixed. In this part we want to study the effect of process
time variability on AGVs’ number needed. We have chosen three types of variability: low, medium,
and high.

• For the low variability; the coefficient of variability is equal to 0,2.

• For the medium variability;the coefficient of variability is equal to 0,4.

• For high variability;the coefficient of variability is equal to 0,5.

First of all, we will determine the right number of replications for each variability model, then
we will analyze the effect of this different type of stochasticity on our system.

3.3.2.1 Number of replications

The number of replications is an important parameter in the simulation process when there is stochasticity.
With the right number of replications, we can be sure at 95% (with the student test) that all the
numbers generated by our simulator are in the interval: [average + half-width; average-half-width].
-How we calculate this number of replications?
To calculate the number of replications needed, we take those steps:

• First step:
we simulate our model with 5 replications, then we calculate the precision. If it is less than 5%

28
Chapter 3. Implementation for a typical application

of the average, we conclude that we have a high precision and that 5 replications are enough to
get good results.

Half W idth
P recision = (3.16)
Average

• Second step:
if we have poor precision then we calculate the needed number of replications with this approximation:

h02
n = n0 (3.17)
h2

With:
-n:the number of replications needed.
-n0:the initial number of replications used.
-h0:the halfWidth for n0 replications.
-h:the halfWidth searched.

• Third step:
We simulate with n replications, then we calculate the model precision.

3.3.2.2 Process time variability

We have used uniform distribution and we changed the coefficient of variation at each time. Here it is
the new process time for each workstation for each type of variability:

Figure 3.17: Process time using uniform distribution

We constate that more the variability is high more the process time interval is large.

Simulation and results:

-For low variability:

29
Chapter 3. Implementation for a typical application

We see that if the AGVs’ number are greater than 8, there is an overlapping. We have used output
analyzer to validate this interpretation and here are the following results:

Figure 3.18: Low variability’s results

Output analyzer for low variability:

Figure 3.19: Overlapping using output analyzer

The first line represents the comparison between the scenario where we have 8 AGVs and the other
scenario where we have 9 AGVs. The output analyzer confirms the overlapping between the two
scenarios, and we have the same conclusion between 9 and 10 AGVs. It means that there is no
significant upgrade of the makespan if we use 8 or 9 or 10 AGVs.
In addition, the results show us that there is a difference between 9 and 11 AGVs but we clearly see
that this difference in makespan do not increase considerably the system performance. As a conclusion,
it is ineffective to use more than 8 AGV in the low variability.
-For medium and high variability:

Figure 3.20: Medium variability’s results

30
Chapter 3. Implementation for a typical application

Output analyzer for medium variability:

Figure 3.21: Overlapping using output analyzer

Figure 3.22: High variability’s results

Output analyzer for high variability:

Figure 3.23: Overlapping using output analyzer

We came up with these results: In the three-variability mode there is an overlapping between the
makespan confidence interval for different number of AGVs. To conclude this overlapping with statistical
results, we have utilized output analyzer. The more variability we have the more overlapping we have.
In fact,overlapping means that there is no significant difference between the two models compared.
With these different levels of stochasticity, the number of AGVs needed has decreased from 9 to 8
AGVs.

Conclusion

For the first part, the two steps approach is a powerful method to determine the optimal number
of AGVs without wasting time or money on simulating many possibilities. For the second part, we
added process time variability. In such conditions, only simulation can give us realistic results. After
comparing the different scenarios for low, medium, and high variability, we conclude the existing of

31
Chapter 3. Implementation for a typical application

overlapping. As a result, in a variable processing time conditions, it is useless to add more AGVs when
there is overlapping.

32
General conclusion

This paper presents our first step towards digital twin and industry 4.0. In our first chapter, we
introduced the main concepts in our work. We defined what is digital twin and the difference between it
and simulation. Then, we showed the benefits of flexible manufacturing system. Finally, we introduced
the material handling system in our FMS and present the type of AGV that we will work with.
In the second chapter, we presented our simulation tool ARENA and introduce our main setting
windows. We saw that simulation is a powerful method because it reproduces real conditions and give
results that are similar to real life.
The third chapter contain the main idea of this paper: set a conflict-free routing path layout in a
flexible manufacturing system and determine the optimal number of material handling system to use,
using a two steps approach. The first approach is analytical where we made some assumptions and
simplification to calculate this number with a mathematical equation then we proceeded to the next
step which is simulation to determine the optimal number of AGVs. The first step is important
because it limits the range of possibilities and decreases the simulation time. We also compared
different dispatching rules in order to choose the best one to use. A deterministic process time was
used in all the previous work. After determining the optimal number of AGVs and the best dispatching
rule we want to study the effect of stochasticity on the system. We changed the fixed process time to
variable and did the simulation for three different coefficients of variability. We concluded that there
is overlapping between the different number of AGVs from a certain point.
This work is a simple, basic model that aims to illustrate the importance of simulation to determine
the best resource utilization in different work scenarios.
As a discussion theme for future developments, we aim to upgrade this basic model to use it in a digital
twin context with real time simulation and real-life conditions.

33
Bibliography

[1] (11/05/2021). M. grieves,“digital twin: Manufacturing excellence through virtual factory replication,”
white paper, 2014, [Online]. Available: http://www.apriso.com/library/Whitepaper_Dr_
Grieves_DigitalTwin_ManufacturingExcellence.php.

[2] (11/05/2021). E. glaessgen and d. stargel,“the digital twin paradigm for future nasa and u.s. air
force vehicles,” in proc. 53rd aiaa/asme/asce/ahs/asc struct. struct. dyn. mater. conf., 2012.,
[Online]. Available: https://arc.aiaa.org/doi/pdf/10.2514/6.2012-1818.

[3] (11/05/2021). T. gabor, l. belzner, m. kiermeier, m. t. beck, and a. neitz,“a simulation-based

architecture for smart cyber-physical systems,” in proc. ieee int. conf. autonomic comput., wurzburg,
germany, 2016, pp. 374– 379, doi: 10.1109/icac.2016.29.

[4] (11/05/2021). ) t. maurer, “what is a digital twin?” 2017., [Online]. Available: https://community.
plm . automation . siemens . com / t5 / Digital - Twin - KnowledgeBase / What - is - a - digital -
twin/ta-p/432960..

[5] (11/05/2021). Digital twin in industry: State-of-the-art fei tao , senior member, ieee, he zhang ,
ang liu, and a. y. c. nee.

[6] (4/06/2021). Kritzinger, w., m. karner, g. traar, j. henjes, and w. sihn. 2018. “digital twin in
manufacturing: A categorical literature review and classification”. in paper archive of the ifac
conference papers online. 51-11 (2018): 1016-1022.

[7] (4/05/2021). Aidan fuller 1 , (student member, ieee), zhong fan1 , (member, ieee), charles day1
, (member, ieee), and chris barlow2,digital twin: Enabling technologies, challenges and open
research.

[8] (12/05/2021), [Online]. Available: http://www.meidenkorea.com/PDF/agv.pdf.

[9] (12/05/2021), [Online]. Available: https://www.meidensha.com/catalog/PA52-3139.pdf.

[10] (22/05/2021). Tao yifei, chen lunruo, liu meihong, liu xianxi,"an estimate and simulation approach
to determining the automated guided vehicle fleet size in fms".

[11] (31/05/2021). P. j. egbelu,"the use of non-simulation approaches in estimating vehicle requirements

in an automated guided vehicle based transport system," material flow, 1987, pp. 17-32.

34