Journal of

Marine Science
and Engineering

Article
Simulation Modelling for Automated Guided Vehicle
Introduction to the Loading Process of Ro-Ro Ships
Sanghyung Park 1 , Jeho Hwang 2 , Hangjin Yang 3 and Sihyun Kim 2, *

1 KMI-KMOU Cooperative Program, Korea Maritime and Ocean University, Busan 49112, Korea;
sang@g.kmou.ac.kr
2 Department of Logistics, Korea Maritime and Ocean University, Busan 49112, Korea;
hwangjeho9288@gmail.com
3 Department of International Trade, Gyeongsang National University, Jinju 52828, Korea; tooto21@hanmail.net
* Correspondence: sihyunkim@kmou.ac.kr

Abstract: This paper aims to introduce the adaptation of automated guided vehicles (AGVs) in the
car-loading process of Ro-Ro ships compared with the current loading process. This study analyzed
the applicable scenarios for the AGVs’ adaptation in a Ro-Ro port, employing Arena simulation
to compare the productivity of the loading processes. The results revealed that the adaptation of
the AGVs in the car-loading process of the Ro-Ro ships improves productivity and solves several
problems of the current loading process. With 21 or more AGVs, the entire processing time is similar
to or less than the current loading process, whereas, after 40 AGVs, it stayed the same. Furthermore, as
the number of AGVs increases, the transfer time decreases, but the queue becomes longer. Identifying
the effect of the AGV adaptation, this study provides valuable insights for developing the various





 traffic situations in Ro-Ro port operations.
Citation: Park, S.; Hwang, J.; Yang,
H.; Kim, S. Simulation Modelling for Keywords: automation; automated guided vehicles; Ro-Ro port; vehicle loading process; productivity
Automated Guided Vehicle
Introduction to the Loading Process
of Ro-Ro Ships. J. Mar. Sci. Eng. 2021,
9, 441. https://doi.org/10.3390/ 1. Introduction
jmse9040441 With an increase in global seaborne trade and the volume in port terminals, cargo
loading/unloading operations have become important in reducing vessel turnaround
Academic Editors: Guilherme Vaz
time [1]. To increase the port operational efficiency, automated guided vehicles (AGV) have
and Alessandro Ridolfi
been employed at a number of terminals, including the Europe Combined Terminal (ECT)
in Rotterdam and the Container Terminal Altenwerder in Hamburg, among others [2]. The
Received: 25 March 2021
adaptation of these AGVs is expected to increase in the future because these unmanned
Accepted: 16 April 2021
vehicles can reduce safety concerns, labor shortages, and operational costs [3]. Therefore,
Published: 19 April 2021
global container terminals strive to expand the scope of the automation system, with
machines carrying out their own loadings, and several studies related to AGVs introduction
Publisher’s Note: MDPI stays neutral
have been conducted. For instance, the effectiveness measure of a container terminal
with regard to jurisdictional claims in
published maps and institutional affil-
from a number of employed AGVs was simulated and pointed out the importance of
iations.
the appropriate choice of both numbers of employed AGVs and the AGVs dispatching
rule [4]. Although prior studies focused on container cargo handling, few studies have been
conducted on car carriers, the Ro-Ro (roll-on and roll-off) system for the loading/unloading
process. Generally, the cargo loading/unloading operation at Ro-Ro terminals requires
more stevedores than other cargo terminals as the vehicles depend on drivers for loading
Copyright: © 2021 by the authors.
and unloading. A gang of stevedores is composed of a supervisor (foreman), a hatch boss,
Licensee MDPI, Basel, Switzerland.
a signalman, a key car, a driver, a van driver, and a lasher. The number of total gangs
This article is an open access article
distributed under the terms and
is decided according to the ship capacity. Several groups of drivers drive and park their
conditions of the Creative Commons
cars on board while lashers work on the deck. Due to the nature of these tasks, they are
Attribution (CC BY) license (https://
challenging to carry out in an unmanned or mechanized form, so the current Ro-Ro system
creativecommons.org/licenses/by/ is heavily dependent on human resources.
4.0/).

J. Mar. Sci. Eng. 2021, 9, 441. https://doi.org/10.3390/jmse9040441 https://www.mdpi.com/journal/jmse

J. Mar. Sci. Eng. 2021, 9, 441 2 of 12

Scholars have noted two problems with the current loading process: labor shortages
and car accidents that could lead to vehicle damage, mainly during the loading/unloading
process [5]. These marine accidents happen because of human error (~85%) [6]. Cars falling
into the sea accounted for 80% of accidents during loading/unloading operations, and life
losses have occurred in 20% of these accidents [7]. These accidents are a severe loss for port
operations. In the Ro-Ro terminal, it is the nature of cargo that cannot be stacked to require
the terminal to proceed with the operation, with automation, in that way overcoming such
problems and improving the operation process, which is needed rapidly.
This paper introduces the potentials of using AGVs in the car loading process of
Ro-Ro ports to address these problems and improve productivity. Arena software was
chosen to simulate the current loading process and the AGV loading process. Simulated
processes were compared through various scenarios to analyze the productivity. Among
the simulation module functions, “tnow” was used to measure the processing time within
specific sections. After modeling the simulation, more productivity was found with the
employment of a specific number of AGVs, along with several improvements. First, in
terms of terminal operators, personnel expenses can be reduced. The simulation revealed
that approximately 21 AGVs equated to three stevedore gangs’ productivity, increasing the
ROI by 34%. Moreover, productivity can be increased as the AGVs are available almost
all day, except for its battery-charging time, while the typical work is less than 12 h a day.
Lastly, the simulation has demonstrated that the adaptation of AGVs improves productivity
and several problems, with significant implications for automating Ro-Ro port operations.

2. Literature Review
2.1. Previous Research on Productivity in the Loading Process in a Port
The service time, mainly about (un)loading time, provided by the terminal is one of
the leading quality indicators when evaluating the Ro-Ro port performance and the margin
the terminal has left to improve [8]. Generally, a worker’s proficiency in cargo handling
and the operational method at the yard area are critical factors for improving the service
time, and they could decide port competitiveness [9]. With regard to improving the Ro-Ro
terminal operational method, a simulation model is often developed to identify possible
bottlenecks for decision support [10,11]. A formulation approach, such as a metaheuristic
algorithm, also was experimented with to optimize the yard management problem in
a Ro-Ro terminal [12]. However, when it comes to improving workers’ productivity at
extreme industry sites, such as the Ro-Ro terminal, several problems must be handled.
Extreme environmental conditions, a noisy environment, and lack of resources and facilities
are reported to be the leading causes of fatigue, back pain, upper body pain, and headache,
resulting in loss of worker productivity and reduced health and safety in the industry [13].
Unless systematically trained, stevedores are continuously recruited, and the ergonomic
design of the work systems is established in the terminal, which one can hardly expect
from the Ro-Ro port; thus, it is difficult to demand more human resource productivity.
Therefore, this study introduces an adaptation of an AGV to a Ro-Ro port that is highly
reliant on human resources, to improve its productivity.

2.2. Automated Guided Vehicle Introduction to the Loading Processes of Pure Car Carriers
Ro-Ro ships are cargo ships intended to transport all kinds of rolling vehicles; the
two most common types are PCC (Pure Car Carriers) and PCTC (Pure Car and Truck
Carriers) [14]. PCC is specially designed to transport only cars capable of 2000–4000 units
at a time, and PCTC is optimized for transporting both cars and other heavy units with a
payload of 4000–8000 units [15].
As shown in Figure 1, for loading a PCTC, stevedores are usually sent by shuttle van
to the yard, from where they drive new cars one by one onto the ship via its stern ramp.
Once drivers reach the assigned parking sector, they leave the car, and the van takes them
back to the yard to pick up another lot while key cars park the cars on the spot. At the
same time, a gang of lashers fastens the cars to the ground in the vessel.
 As shown in Figure 1, for loading a PCTC, stevedores are usually sent by shuttle van
to the yard, from where they drive new cars one by one onto the ship via its stern ramp.
J. Mar. Sci. Eng. 2021, Once
9, 441 drivers reach the assigned parking sector, they leave the car, and the van takes them 3 of 12
back to the yard to pick up another lot while key cars park the cars on the spot. At the
same time, a gang of lashers fastens the cars to the ground in the vessel.

Figure 1. Current loading process

Figure 1. (source:
Current theprocess
loading Authors).
(source: the Authors).

To improve productivity,
To improve productivity, container
container terminals terminalshave
worldwide worldwide have increasingly
increasingly adapted adapted
AGVs. In theAGVs.
case ofIn the case terminals,
container of container terminals,
AGVs can beAGVs
movedcan in be moved
several in several
ways, ways, depending
depending
on theand
on the technology, technology,
nowadays, andmany
nowadays, many
container container
terminals terminals
have chosen have
AGVschosen AGVs designed
designed
to navigate by the line-following principle using technologies
to navigate by the line-following principle using technologies from embedded guided from embedded guided
wires, paint stripes, magnetic tape, laser guidance, and inertial
wires, paint stripes, magnetic tape, laser guidance, and inertial navigation [16]. However, navigation [16]. However, it
is challenging
it is challenging to applydevice
to apply guidance guidance device technologies
technologies in Ro-Ro operations
in Ro-Ro operations since the vessel
since the ves-
sel position onposition on landatchanges
land changes at every
every port. Thus,port. Thus, environments,
in variable in variable environments,
such as Ro-Ro such as Ro-Ro
ports, the introduction
ports, the introduction of AGVs thatofcan AGVs that can
navigate navigate
without the without the line-following
line-following principle principle
must be considered.
must be considered. The development
The development of an unmanned
of an unmanned forklift navigating
forklift navigating without without
the the line-
following principle has been discussed [17,18]. Thus far, French
line-following principle has been discussed [17,18]. Thus far, French startup Stanley Ro- startup Stanley Robotics is
the leading company that has developed an AGV capable
botics is the leading company that has developed an AGV capable of moving and parking of moving and parking cars both
outside
cars both outside andand indoors.
indoors. ThisThis autonomous
autonomous vehicle
vehicle uses GNSS,
uses GNSS, cameras,
cameras, and LiDAR-based
and LiDAR-
simultaneous
based simultaneous localization
localization and mapping
and mapping (SLAM) (SLAM)
to findtothefind the and
path pathmoveand move
cars, cars, with a
maximum speed of 10 km/h. At selected airports in Europe,
with a maximum speed of 10 km/h. At selected airports in Europe, including the Gatwick including the Gatwick (UK)
and Lyon Saint-Exupéry (France) airports, the robot
(UK) and Lyon Saint-Exupéry (France) airports, the robot is operational [19]. is operational [19].
The input data Theforinput data for
simulation wassimulation
motivated was by anmotivated
application byofantheapplication of the latest car
latest car carri-
able AGV technology. The following section discusses the simulation methodologymethodology
carriable AGV technology. The following section discusses the simulation
adapted
adapted for this study. for this study.

3. Simulation Methodology
3. Simulation Methodology
Arena is a DES (Discrete Event Simulation) software [20], and several studies
Arena is a DES (Discrete Event Simulation) software [20], and several studies have
haveemployed it to imitate complex port operations, such as Muravev et al. [11] and
employed it Kotachi
to imitate et complex port operations,
al. [21]. Other such as Muravev
simulation software, et al. [11]and
such as Anylogic andFlexsim
Kotachiflow process,
et al. [21]. Other
also can be used for modeling concerning problems. Subtle differences incan
simulation software, such as Anylogic and Flexsim flow process, also the results can
be used for modeling
be given due concerning problems.
to random numbersSubtle differences
created in the results
by the different can be given
mechanisms of the programs.
due to random Thenumbers
benefitscreated
of thesebysimulations
the different mechanisms
will of the programs.
enable researchers to estimate The port
bene-performance,
fits of these simulations
overcomingwill theenable researchers
mathematical to estimate
limitations portoptimization
of the performance, overcoming
technique [22], and Rusca
the mathematical limitations of the optimization technique [22], and Rusca
et al. [23] have tested a container terminal. The simulation also identifies et al. [23] havepossible bottle-
tested a container
necksterminal. The simulation
in the process and supports alsoproductivity
identifies possible bottlenecks
improvement [10,24].in the pro- this study
Therefore,
cess and supports
uses Arenaproductivity improvement
14.0 for model building,[10,24]. Therefore,
productivity this study
measure, uses Arena
and analysis.
14.0 for model building, productivity measure, and analysis.
The concept of simulation models was developed based on the existing loading system.
The concept of simulation
The concept modelsmodels
of simulation was developed based
is to imitate on the existing
the current loading loading
system and sys-AGV loading
tem. The concept of simulation
system, respectively. models is to imitatemodel
The simulation the current
coversloading system
the loading and AGV
process from the yard to
loading system, respectively.
parking areas in The simulation
the vessel. model 3.1,
In Section covers
thethe loading
sample deckprocess
with its from the areas and
parking
yard to parking areas in the vessel. In Section 3.1, the sample deck with its parking
vehicles traveling speed in each section is introduced. Section 3.3 demonstrates the two areas
loading systems developed by Arena software.
Sci. Eng. 2021, 9, x FOR PEER REVIEW 4 of 13

and vehicles traveling speed in each section is introduced. Section 3.3 demonstrates the
two
J. Mar. Sci. Eng. 2021, loading
9, 441 systems developed by Arena software. 4 of 12

3.1. Simulation Model

3.1.1. Parking Area
3.1. Simulation Model
The nominal vehicle
3.1.1. Parkingcarrying
Area capacity is the largest number of standard-sized vehi-
cles that a ship canThe load [14], which is generally represented by R/T, so this study was
nominal vehicle carrying capacity is the largest number of standard-sized vehicles
motivated to simulate the vessel's
that a ship can load [14],nominal
whichvehicle carrying
is generally capacity.by
represented R/T originated
R/T, fromwas motivated
so this study
RT43, a 1966 Toyota Corolla, the first mass-produced car shipped in car carriers.
to simulate the vessel’s nominal vehicle carrying capacity. R/T originated It is now from RT43, a
widely used to1966represent the largest number of standard-sized vehicles that Ro-Ro ships
Toyota Corolla, the first mass-produced car shipped in car carriers. It is now widely
can load. Eachuseddeck'sto surface
represent was
thedivided into a rectangular
largest number shape tovehicles
of standard-sized representthatparking
Ro-Ro ships can load.
sectors to use aEach
triangular distribution in the simulation. Each parking
deck’s surface was divided into a rectangular shape to representsector's unit was parking sectors to
calculated as 4.125
use amtriangular
× 1.550 m,distribution
consideringinathe space between Each
simulation. individual
parkingRT43 cars when
sector’s unit was calculated
lashed. As shown in Figure
as 4.125 m × 2, Deck
1.550 m,6considering
presents an aexample to demonstrate
space between individualtheRT43
deck's divi-
cars when lashed. As
sions, as it has shown
the most general structure of the 13 decks.
in Figure 2, Deck 6 presents an example to demonstrate the deck’s divisions, as it
has the most general structure of the 13 decks.

Figure 2. Deck 6 layout (source: the Authors).

Figure 2. Deck 6 layout (source: the Authors).
3.1.2. Speed Variations and Safe Following Distance
3.1.2. Speed Variations and Safe
As shown in Following
Table 1, theDistance
sectional vehicle speed is set based on the Hyundai Glovis
transportation and handling
As shown in Table 1, the sectional vehicle manual.
speed isInset
the conventional
based approach,
on the Hyundai it is challenging
Glovis
transportation to expect
and a constant
handling speed
manual. from
In the 7000 cars since
conventional speeditvariations
approach, occurtoamong drivers.
is challenging
Nevertheless,
expect a constant speed fromrobots can since
7000 cars attainspeed
this. variations occur among drivers. Nev-
ertheless, robots can attain this.
Table 1. Speed variations and safe following distance.
Table 1. Speed variations and safe following distance.
External
Yard Stern Ramp Parking Area Safe Distance
Ramp
External Parking Safe Dis-
Yard Stern
6.8–8.0 kmRamp
Drivers Ramp
21.25–25.0 km 21.25–25.0Area
km 5–10 tance
mph 15 m
(5 mph)
21.25–25.0
AGVs 106.8–8.0
km km 1021.25–25.0
km 10 5km
mph–10 10 km 6m
Drivers 15 m
km (5 mph) km mph
AGVs 3.1.3.10 km
Sectional 10 km
Vehicle Travel Time10 km 10 km 6m
Max, mid, and min vehicle travel times to each parking sector are calculated based
on travel distance and car speed. Drivers’ parking and traveling processes within the
3.1.3 Sectional ramp
Vehicle Travel
area Time
are expressed by a normal distribution. However, a robot’s parking and trav-
eling processes are expressed by a constant value. Furthermore, lashing and inspection
processes are not involved in the simulation as they are considered as another process
conducted simultaneously.
 J. Mar. Sci. Eng. 2021, 9, 441 5 of 12

3.2. Assumptions of the Simulation Model

Three different arena simulation models were developed; current loading process,
AGV loading process scenario 1, and AGV loading process scenario 2. Some assumptions
were made to reduce unnecessary details and compare the models under the same situation.
The model has been constructed based on the following assumptions:
• The loading/unloading place is identical, and the loading cargo is the same as the
R/T cars, assuming that all 7352 cars are loaded.
• The surfaces of the deck pillars are not considered, as the space between each pillar is
large, and its surface is small.
• The vehicles depart from the yard to the ship at the same time.
• The specific stowage plan considering the balance of the ship is not detailed.
r. Sci. Eng. 2021, 9, x FOR PEER REVIEW 6 of 13
These assumptions were chosen for the following reasons. First, cars and loading
places with different sizes and loads can measure the loading time differently. Second, we
have focused on testing the vessel loading system in an unconstrained situation in the port
Step 5: After a batch isThird,
environment. made with eight entities,
we focused a temporary
on comparing batch goes
two loading through
systems the
in the same stowage
internal rampplan,
process "deck
so the U tostowage
specific U-1" or plan
"deckwas
D to D+1", which is the opposite process
unnecessary.
of Step 3.
3.3. Arena Simulation Modeling
* U: 6,7,8,9,10,11,12,13
* D: 1,2,3,4 This study analyzed three different simulation models, respectively: the current
Step 6: Once the process
loading batch arrives at3),
(Figure theAGV
yard,loading
it is separated
process into a single
scenario entity 4),
1 (Figure again
and AGV load-
through the Separate module,
ing process and the
scenario entity passes
2 (Figure 5). The through
entirethe Release
process is module
detailedtoindepart
six steps to aid a
the next driver teams.
better understanding.

Figure 3. Current loading

Figure process loading
3. Current model (source:
process the Authors).
model (source: the Authors).

3.3.2. AGV Loading Process, Scenario 1

Step 1: The AGV loading process scenario 1 is divided into two groups through the
Decide module. The first group of entities is limited by the Seize module to allow re-
sources by a specific number, representing the number of AGVs. Another group is limited
by the Seize module to allow resources with the same number of resources as Group 1, to
simulate the deck 5 loading process. In Step 1, the entity departs from the yard and enters
the ship to reach the deck 5 starting point through the "yard to stern ramp" and "stern to
deck 5" processes.
Step 2: After arrival at the deck 5 starting point, the group first moves to process
"deck 5 to 6 start point", and when all the loading processes between decks 6 to 13 are
completed, the group moves to the process "deck 5 to 4 start point".
Step 3: Upon arrival at the starting point, the AGV passes through the internal ramp
 J. Mar. Sci. Eng. 2021, 9, x FOR PEER REVIEW 7 of 13

J. Mar. Sci. Eng. 2021, 9, 441 Step 6: Finally, the AGV returns to the yard and then passes through the "release"
6 of 12
module to send the next AGV.

Sci. Eng. 2021, 9, x FOR PEER REVIEW 8 of 13

Figure
Figure 4.
4. AGV
AGV scenario
scenario 11 simulation
simulation model
model (source:
(source: the
the Authors).
Authors).

3.3.3. AGV Loading Process, Scenario 2

Step 1: As shown in Figure 5, in scenario 2, the entities are divided into three groups
through the Decide modules. The first group is limited by the Seize module to allow re-
sources for a specific number of AGVs and goes through all the processes between decks
6 and 13, while the second group goes through all the processes between decks 1 and 4,
and the third group consists of a merge of the other groups to work on deck 5 at the end.
In Step 1, the entity departs from the yard and enters the ship to reach the deck 5 starting
point by the "yard to stern ramp" and "stern to deck 5" processes.
Step 2: After the arrival at the deck 5 starting point, the first and second groups move
to the processes "deck 5 to 6 start point" and "deck 5 to 4 start point", respectively.
Step 3: Upon arrival to the "deck 5 to 6 start point", the first group passes through the
internal ramp process "deck U-1 to U". Simultaneously, the second group goes through
the internal ramp process "deck D to D-1".
* U: 6,7,8,9,10,11,12,13
* D: 2,3,4
Step 4: When the entity reaches deck N, it moves to the assigned parking sector
through the process "deck N sector X". Each AGV then discharges the car through the
parking process "deck N parking" and returns to the front of the inner ramp through the
process "deck N sector X R".
* N: 1~13
Figure 5. AGV scenario
Figure 5.2AGV
simulation model
scenario (source: model
2 simulation the Authors).
(source: the Authors).
Step 5: Once the AGV arrives at "deck N sector X R", it goes through the process "deck
4. SimulationU to U-1" or "deck D to D+1" to return to the yard.
Results
3.3.1. Current Loading Process
Step 6: The AGV returns to the yard through the opposite process of Step 1, and the
4.1. Current LoadingStep
Process Simulation
1: The Create Results represents the arrival of 7532 cars to the yard. Through
entity passes through themodule
Release module to send the next AGVs.
Figure 6 shows the simulation
the Decide modules, theresults of the
entire current
entities are loading
divided process.
into threeThe simulation
groups. The first group of
was conducted entities is limited
with five by the The
replications. Seizeentire
module to allowtime
processing 16 resources,
was definedrepresenting
as "E idle two teams of
time measure", drivers
decks and
6 to going through
13 as "idle timeall the processes
measure", decksbetween
1 to 4 asdecks 6 andmeasure
"idle time 13. The 2",
second group is
allowed
and work on deck 5 asby eight
"deck resources
5 idle passing through
time measurement". all the
Since processes
"idle betweenwhich
time measure", decks 4 and 1, and
the third group
is the upside processing time, consists
and "idleoftime
a merge of the2",other
measure whichtwois groups and finishes
the downside the work on deck 5.
processing
In Step 1,
time, were conducted the entity departs
simultaneously, the from thetook
one that yard127,598
and enters the ship
s longer to reach
affected the deck 5 starting
the entire
point by the “Yard to stern ramp” and “stern to deck 5” processes.
processing time. Then, the entire processing time was 137,598 s after the final processing
time on deck 5 was added.
J. Mar. Sci. Eng. 2021, 9, 441 7 of 12

Step 2: After arrival at the deck five starting point, the first and second groups move
to the processes “deck 5 to 6 start point” and “deck 5 to 4 start point”, respectively.
Step 3: From “deck 5 to 6 start point”, the first group passes through the internal ramp
process “deck U-1 to U”. Simultaneously, from “deck 5 to 4 start point”, the second group
goes through the internal ramp process “deck D+1 to D”.
* U: 6,7,8,9,10,11,12,13
* D: 1,2,3,4
Step 4: When the entity reaches deck N, it moves to the assigned parking sector
through the process “deck N sector X”. At the parking sector, the entity is duplicated by a
separate module and goes through the “deck N Parking” and “batch” processes, one each,
to simulate the situation key car’s parking and driver’s grouping to the shuttle van.
* N: 1~13
Step 5: After a batch is made with eight entities, a temporary batch goes through the
internal ramp process “deck U to U-1” or “deck D to D+1”, which is the opposite process
of Step 3.
* U: 6,7,8,9,10,11,12,13
* D: 1,2,3,4
Step 6: Once the batch arrives at the yard, it is separated into a single entity again
through the Separate module, and the entity passes through the Release module to depart
the next driver teams.

3.3.2. AGV Loading Process, Scenario 1

Step 1: The AGV loading process scenario 1 is divided into two groups through the
Decide module. The first group of entities is limited by the Seize module to allow resources
by a specific number, representing the number of AGVs. Another group is limited by
the Seize module to allow resources with the same number of resources as Group 1, to
simulate the deck 5 loading process. In Step 1, the entity departs from the yard and enters
the ship to reach the deck 5 starting point through the “yard to stern ramp” and “stern to
deck 5” processes.
Step 2: After arrival at the deck 5 starting point, the group first moves to process “deck
5 to 6 start point”, and when all the loading processes between decks 6 to 13 are completed,
the group moves to the process “deck 5 to 4 start point”.
Step 3: Upon arrival at the starting point, the AGV passes through the internal ramp
process “deck U-1 to U” and later passes through “deck D+1 to D”.
* U: 6,7,8,9,10,11,12,13
* D: 1,2,3,4
Step 4: When the entity reaches deck N, it moves to the assigned parking sector
through the process “deck N sector X.” Each AGV then discharges the car through the
parking process “deck N parking” and returns to the front of the inner ramp through the
process “deck N sector X R.”
* N: 1~13
Step 5: AGVs go to the process “deck U to U-1” or “deck D to D+1” to return to
the yard.
Step 6: Finally, the AGV returns to the yard and then passes through the “release”
module to send the next AGV.

3.3.3. AGV Loading Process, Scenario 2

Step 1: As shown in Figure 5, in scenario 2, the entities are divided into three groups
through the Decide modules. The first group is limited by the Seize module to allow
resources for a specific number of AGVs and goes through all the processes between decks
6 and 13, while the second group goes through all the processes between decks 1 and 4,
and the third group consists of a merge of the other groups to work on deck 5 at the end.
In Step 1, the entity departs from the yard and enters the ship to reach the deck 5 starting
point by the “yard to stern ramp” and “stern to deck 5” processes.
J. Mar. Sci. Eng. 2021, 9, 441 8 of 12

Step 2: After the arrival at the deck 5 starting point, the first and second groups move
to the processes “deck 5 to 6 start point” and “deck 5 to 4 start point”, respectively.
Step 3: Upon arrival to the “deck 5 to 6 start point”, the first group passes through the
internal ramp process “deck U-1 to U”. Simultaneously, the second group goes through the
internal ramp process “deck D to D-1”.
* U: 6,7,8,9,10,11,12,13
* D: 2,3,4
Step 4: When the entity reaches deck N, it moves to the assigned parking sector
through the process “deck N sector X”. Each AGV then discharges the car through the
parking process “deck N parking” and returns to the front of the inner ramp through the
process “deck N sector X R”.
* N: 1~13
Step 5: Once the AGV arrives at “deck N sector X R”, it goes through the process
“deck U to U-1” or “deck D to D+1” to return to the yard.
Step
Figure 6: The
5. AGV AGV 2returns
scenario to the
simulation yard
model through
(source: the the opposite process of Step 1, and the
Authors).
entity passes through the Release module to send the next AGVs.
4. Simulation Results
4. Simulation Results
4.1. Current
4.1. Current Loading
Loading Process
Process Simulation
Simulation Results
Results
Figure 66 shows
Figure shows thethe simulation
simulation results
results of
of the
the current
current loading
loading process.
process. The
The simulation
simulation
was conducted with five replications. The entire processing time
was conducted with five replications. The entire processing time was defined as was defined as “E
"E idle
idle
time measure", decks 6 to 13 as "idle time measure", decks 1 to 4 as "idle time
time measure”, decks 6 to 13 as “idle time measure”, decks 1 to 4 as “idle time measure 2”, measure 2",
and work
and work on
on deck
deck 55 as
as“deck
"deck 55 idle
idle time
timemeasurement”.
measurement". Since
Since “idle
"idle time
time measure”,
measure", which
which
is the
is the upside
upside processing
processingtime,time,and
and“idle
"idle time
time measure
measure 2”,
2", which
which isis the
the downside processing
downside processing
time, were
time, were conducted
conducted simultaneously,
simultaneously,the theone
onethat
thattook
took 127,598
127,598 ss longer
longer affected
affected the
the entire
entire
processing time. Then, the entire processing time was 137,598 s after the
processing time. Then, the entire processing time was 137,598 s after the final processingfinal processing
time on deck 5 was added.
added.

Figure 6. Current loading process

process simulation
simulation result
result (source:
(source: the
the Authors).
Authors).

4.2.
4.2. Simulation
Simulation Validation
Validation
Before we used the simulation model to test the AGV loading system, we confirmed
Before we used the simulation model to test the AGV loading system, we confirmed
the validation against the real data from the actual loading process. First, we modeled the
the validation against the real data from the actual loading process. First, we modeled the
current loading system based on the actual loading process. Then we ran the simulation
current loading system based on the actual loading process. Then we ran the simulation
using the values (vehicle speeds, sources limit) from the Hyundai Glovis transportation and
using the values (vehicle speeds, sources limit) from the Hyundai Glovis transportation
handling manual and the same number of drivers participating in the actual loading process.
The simulation took 1,338,652 s to complete the loading process, equal to 96 vehicles
load per 90 min. According to a survey from the Pyeongtaek Port, the average vehicle
load recorded 115 vehicles per 90 min. The comparison with real data shows that the
simulation’s current loading system takes more time for loads. The discrepancy between
J. Mar. Sci. Eng. 2021, 9, 441 9 of 12

the simulation result and the real data was attributed to running the simulation based on
terminal transportation and handling manual. The actual loading process is carried out
with acceleration beyond the speed limit as drivers are put under pressure to complete
tasks quickly, so we used our simulation model to measure the standard loading time as the
manual. To ensure results validity, we reported the output of the simulation’s confidence
interval at the 95% (α = 0.05) confidence level.

4.3. AGV Loading Process, Scenario 1, Simulation Results

The simulation was run with 8–40 AGVs. Table 2 shows the simulated AGVs’ loading
processing time and diminishing rate. In scenario 1, all the AGVs are dispatched for loading
vehicles from upside to downside collectively, so the downside processing is followed after
the upside processing. Their loading times are shown on the table, respectively. When
more than 22 AGVs were employed, the entire processing time was lower than the current
loading process. As presented in the simulation results of 38–40 AGVs, once reaching the
highest value, increasing the number of AGVs decreased productivity or maintained the
same level with increased queueing time on a ramp. To resolve the queue, the operation
was separated into two parts by separating the AGVs into two groups in scenario 2.

Table 2. AGV loading process scenario 1 simulation results.

Upside Processing Downside Processing Entire Processing Diminishing

Total AGVs
Time (s) Time (s) Time (s) Rate (%)
8 268,320 59,140 351,604.00
9 238,589 52,697 312,880.00 −11.01
10 215,141 47,532 281,963.00 −9.88
11 195,882 43,326 257,026.00 −8.84
12 180,435 39,640 236,340.00 −8.05
13 166,358 36,746 218,204.00 −7.67
14 155,110 34,068 203,093.00 −6.93
15 145,133 31,924 190,098.00 −6.40
16 136,211 29,896 178,535.00 −6.08
17 128,661 28,360 168,641.00 −5.54
18 122,084 26,858 160,105.00 −5.06
19 116,115 25,551 152,229.00 −4.92
20 110,702 24,411 145,265.33 −4.57
21 105,924 23,371 138,889.33 −4.39
22 101,942 22,459 133,789.00 −3.67
23 97,901 21,691 128,719.00 −3.79
24 94,618 20,926 124,489.00 −3.29
25 91,852 20,390 120,983.00 −2.82
26 89,290 19,911 117,781.00 −2.65
27 87,394 19,390 115,240.00 −2.16
28 86,020 19,206 113,620.00 −1.41
29 84,910 19,014 112,294.00 −1.17
30 83,890 18,867 111,085.00 −1.08
31 83,287 18,851 110,438.00 −0.58
32 82,736 18,921 110,006.00 −0.39
33 82,295 18,910 109,513.00 −0.45
34 81,874 18,852 109,041.00 −0.43
35 81,769 18,920 108,997.00 −0.04
36 81,469 18,863 108,679.33 −0.29
37 81,428 18,882 108,645.00 −0.03
38 81,314 18,866 108,506.00 −0.13
39 81,411 18,824 108,506.00 0.00
40 81,333 18,913 108,560.00 0.05

4.4. AGV Loading Process, Scenario 2, Simulation Results

Table 3 shows the results of scenario 2. A total of 21 AGVs equated to the productivity
of the current loading process. Queueing time on a ramp was reduced significantly.
J. Mar. Sci. Eng. 2021, 9, 441 10 of 12

Moreover, the overall productivity of scenario 2 was higher than scenario 1. However,
raising the number of AGVs in the upside or downside, beyond a certain level, decreased
productivity, so it was essential to balance the number of AGVs between the two parts. As
shown in the simulation results of 39–42 AGVs, upon reaching the lowest time value, more
AGVs did not decrease the time spent in the process because of the maximum capacity of
the stern ramp.

Table 3. AGV loading process scenario 2 simulation results.

Upside Downside Entire

Total AGVs AGVs Diminishing
Processing Processing Processing
AGVs (Upside) (Downside) Rate (%)
Time (s) Time (s) Time (s)
8 6 2 327,041.33 236,749.67 350,821.00
9 7 2 290,852.67 236,634.00 312,006.33 −11.06
10 8 2 262,208.33 237,603.33 281,278.00 −9.85
11 9 2 238,292.67 236,166.00 255,694.67 −9.10
12 9 3 218,935.67 158,367.33 234,860.00 −8.15
13 10 3 202,296.00 157,852.00 217,084.33 −7.57
14 11 3 188,178.33 158,389.67 201,906.00 −6.99
15 12 3 176,167.00 158,476.00 189,083.00 −6.35
16 13 3 165,282.00 158,691.33 177,424.00 −6.17
17 13 4 156,143.00 119,141.00 167,635.00 −5.52
18 14 4 147,691.00 119,484.00 158,610.00 −5.38
19 15 4 140,284.00 119,861.00 150,695.00 −4.99
20 16 4 133,788.00 120,084.00 143,779.00 −4.59
21 17 4 127,758.00 120,283.00 137,361.00 −4.46
22 17 5 122,303.67 96,386.33 131,604.33 −4.19
23 18 5 117,629.33 96,698.00 126,654.00 −3.76
24 19 5 113,248.00 97,187.67 122,076.00 −3.61
25 20 5 109,384.00 97,694.33 118,052.67 −3.30
26 21 5 105,717.67 98,064.00 114,253.67 −3.22
27 21 6 102,380.00 81,852.00 110,847.00 −2.98
28 22 6 99,578.33 82,506.00 107,942.33 −2.62
29 22 7 97,063.00 71,270.00 105,388.00 −2.37
30 22 8 94,666.00 62,894.00 102,990.00 −2.28
31 22 9 92,839.00 56,447.00 101,140.00 −1.80
32 22 10 91,044.00 51,348.00 99,339.00 −1.78
33 23 10 89,658.00 52,295.00 97,909.00 −1.44
34 24 10 88,635.00 53,079.00 96,908.00 −1.02
35 24 11 87,582.00 48,946.00 95,879.00 −1.06
36 25 11 86,885.00 50,259.00 95,189.00 −0.72
37 26 11 86,817.00 51,722.00 95,163.00 −0.03
38 26 12 86,174.00 48,307.00 94,512.00 −0.68
39 27 12 86,203.00 50,312.00 94,498.00 −0.01
40 27 13 86,372.00 47,675.00 94,690.00 0.20
41 28 13 86,217.00 49,287.00 94,535.00 −0.16
42 28 14 86,397.00 47,174.00 94,660.00 0.13

5. Discussion and Conclusions

The results of the simulation revealed many advantages of adapting AGVs in the
loading process. First, in terms of port operators, the driver-related personnel costs can
be cut. To date, the cost of a single AGV remains unclear, but experts presume it to cost
about $233,939 [25]. Despite the expensive vehicle capital, the AGV loading system has the
potential to reduce 60% of the total operating cost in 10 years.
Furthermore, 21 AGVs equate to 35 stevedores, including drivers, key cars, signalers,
and van drivers, so the ROI of employing AGVs can be calculated at 34%. Moreover,
average repairs cost around $250 per car damaged [5]. As physical damages during the
loading/unloading process happen because of human failure, the employment of AGVs
is expected to reduce hidden costs. Second, possible operating hours can be increased.
J. Mar. Sci. Eng. 2021, 9, 441 11 of 12

The working hours of the Ro-Ro port stevedore team are 1.5 h × 6, less than 12 h. In the
simulation, the entire processing time of the current loading process was 136,522.00 s, but
the actual work time charge can be three days. However, AGVs can operate for almost
24 h. Therefore, even if the productivity of the two types of loading processes is equal in
the simulation, AGVs can provide more productivity due to the available operating hours.
The simulation results suggest the employment of AGVs can improve productivity and
solve several problems, which has significant implications for automating the operations
of the Ro-Ro ports.
Several limitations were noted during this study. These limitations provide directions
for future studies. First, this simulation was constructed under the assumption that the
vessel’s nominal vehicle capacity is fully loaded with the actual number of stevedores
participating. However, because the actual size of the cars to be loaded differs from the
standard, the total number of vehicles that can be loaded probably differ from the nominal
vehicle capacity. Therefore, the study must be developed to be applied to more varying
vehicles. Second, the departure of the initial AGVs is simultaneous, potentially increasing
waiting times at the external ramp. Further studies must be conducted by considering
departure schedules, giving priority to specified groups in the yard. Lastly, the AGV
loading process results revealed that as the number of AGVs increased, queuing times
on a ramp increased, affecting the entire processing time. In order to resolve the queue,
scenario 2 was implemented with a plan to separate the queue in a ramp, and it has proven
to improve productivity and reduced queueing time. However, as the final queueing time
arises from the stern ramp, the maximum number of AGVs was determined by the capacity
of the stern ramp. Therefore, a bidirectional movement plan for AGVs within the stern
ramp must be considered to improve final productivity.

Author Contributions: Conceptualization, S.K. and S.P.; methodology, S.P.; validation, S.P., H.Y.
and J.H.; formal analysis, S.P.; data curation, S.P. and J.H.; writing—original draft preparation,
S.P.; writing—review and editing, S.K.; visualization, S.P., H.Y. and J.H.; supervision, S.K.; project
administration, S.K.; funding acquisition, S.K. All authors have read and agreed to the published
version of the manuscript.
Funding: This work was supported by the National Research Foundation of Korea (NRF) grant
funded by the Korea government (MSIT) (No. NRF-2019R1G1A1087736).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data presented in this study are available on request from the
corresponding author.
Acknowledgments: The authors would like to thank the anonymous reviewers and the handling
editors for their constructive comments that greatly improved this article from its original form.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Kim, S.H.; Kang, D.W.; Dinwoodie, J. Competitiveness in a multipolar port system: Striving for regional gateway status in
Northeast Asia. Asian J. Shipp. Logist. 2016, 32, 119–126. [CrossRef]
2. Luo, J.; Wu, Y. Modelling of dual-cycle strategy for container storage and vehicle scheduling problems at automated container
terminals. Transp. Res. Part E Logist. Transp. Rev. 2015, 79, 49–64. [CrossRef]
3. Aguiar, G.T.; Oliveira, G.A.; Tan, K.H.; Kazantsev, N.; Setti, D. Sustainable implementation success factors of AGVs in the
Brazilian industry supply chain management. Procedia Manuf. 2019, 39, 1577–1586. [CrossRef]
4. Pjevčević, D.; Vladisavljević, I.; Vukadinović, K.; Teodorović, D. Application of DEA to the analysis of AGV fleet operations in a
port container terminal. Procedia Soc. Behav. Sci. 2011, 20, 816–825. [CrossRef]
5. Holewg, M.; Miemczyk, J. Logistics in the “three-day car” age: Assessing the responsiveness of vehicle distribution logistics in
the UK. Int. J. Phys. Distrib. Logist. Manag. 2002, 32, 829–850. [CrossRef]
6. Chan, S.R.; Hamid, N.A.; Mokhtar, K. A Theoretical review of human error in maritime accidents. Am. Sci. Publ. 2016, 22,
2109–2112. [CrossRef]
7. Emecen, K.E.G. Analysis of accidents at the quayside operations in the Turkish port. Int. J. Res. Eng. Technol. 2016, 5, 1–5.
J. Mar. Sci. Eng. 2021, 9, 441 12 of 12

8. Morales, F.P.; Saurí, S. Quality indicators and capacity calculation for RoRo terminals. Transp. Plan. Technol. 2010, 33,
695–717. [CrossRef]
9. Kim, H.Y. A comparative analysis of the efficiency of automobile export ports in Korea and Japan. J. Korea Port Econ. Assoc. 2017,
33, 73–82. [CrossRef]
10. Keceli, Y.; Aksoy, S.; Aydogdu, Y.V. A simulation model for decision support in Ro-Ro terminal operations. Int. J. Logist. Syst.
Manag. 2013, 15, 338–358. [CrossRef]
11. Muravev, D.; Aksoy, S.; Rakhmangulov, A.; Aydogdu, V. Comparing model development in discrete event simulation on Ro-Ro
terminal example. Int. J. Logist. Syst. Manag. 2016, 24, 283–297.
12. Cordeau, J.F.; Laporte, G.; Moccia, L.; Sorrentino, G. Optimizing yard assignment in an automotive transshipment terminal. Eur.
J. Oper. Res. 2011, 215, 149–160. [CrossRef]
13. Shikdar, A.A.; Sawaqed, N.M. Worker productivity, and occupational health and safety issues in selected industries. Comput. Ind.
Eng. 2003, 45, 563–572. [CrossRef]
14. Marine Insight. Available online: https://www.marineinsight.com/types-of-ships/different-types-of-roll-on-roll-off-ships/
(accessed on 2 April 2020).
15. Todorrov, D.M. Ro-Ro Handbook: A Practical Guide to Roll-On Roll-Off Cargo Ships; Schiffer: Atglen, PA, USA, 2016.
16. FLEXQUBE. Available online: https://www.flexqube.com/news/what-agv/ (accessed on 21 February 2020).
17. Tamba, T.A.; Hong, B.; Hong, K.-S. A path following control of an unmanned autonomous forklift. Int. J. Control. Autom. Syst.
2009, 7, 113–122. [CrossRef]
18. Pradnya, T.C.; Ganesh, R. An autonomous industrial load carrying vehicle. Adv. Electron. Electr. Eng. 2014, 4, 169–178.
19. NEW ATLAS. Available online: https://newatlas.com/stanley-robotics-stan-robot-parking-valet/58204/ (accessed on
25 March 2020).
20. Kamrani, M.; Abadi, S.M.H.E.; Golroudbary, S.R. Traffic simulation of two adjacent unsignalized T-junctions during rush hours
using Arena software. Simul. Model. Pract. Theory 2014, 49, 167–179. [CrossRef]
21. Kotachi, M.; Rabadi, G.; Obeid, M.F. Simulation modeling and analysis of complex port operations with multimodal transportation.
Procedia Comput. Sci. 2013, 20, 229–234. [CrossRef]
22. Cartenì, A.; Luca, S.D. Tactical and strategic planning for a container terminal: Modelling issues within a discrete event simulation
approach. Simul. Model. Pract. Theory 2012, 21, 123–145. [CrossRef]
23. Rusca, F.V.; Popa, M.; Rosca, E.; Rosca, M.A. Simulation model for maritime container terminal. Transp. Probl. 2018, 13,
47–54. [CrossRef]
24. James, J.; John, B.; Rengaraj, M. Productivity Improvement by Enhancing the Bottleneck Station in an Alternator Production Plant
with Layout Improvement and Its Cost Analysis. Int. J. Sci. Res. Publ. 2013, 3, 1–7.
25. KURBADS. Available online: https://www.kurbads.lv/en/welcome-stanley-future-robot-at-work-today/ (accessed on 9 April 2020).