Tech Science Press

DOI: 10.32604/iasc.2024.043854

ARTICLE

A Study on Optimizing the Double-Spine Type Flow Path Design for the
Overhead Transportation System Using Tabu Search Algorithm

Nguyen Huu Loc Khuu1,2,3 , Thuy Duy Truong1,2,3 , Quoc Dien Le1,2,3 , Tran Thanh Cong Vu1,2,3 ,
Hoa Binh Tran1,2,3 and Tuong Quan Vo1,2,3, *
1
Department of Mechatronics, Faculty of Mechanical Engineering, Ho Chi Minh City University of Technology (HCMUT),
Ho Chi Minh City, Vietnam
2
Bach Khoa Research Center for Manufacturing Engineering, Ho Chi Minh City University of Technology (HCMUT),
Ho Chi Minh City, Vietnam
3
Vietnam National University Ho Chi Minh City, Ho Chi Minh City, Vietnam
*Corresponding Author: Tuong Quan Vo. Email: [email protected]
Received: 14 July 2023 Accepted: 05 February 2024 Published: 21 May 2024

ABSTRACT
Optimizing Flow Path Design (FPD) is a popular research area in transportation system design, but its application
to Overhead Transportation Systems (OTSs) has been limited. This study focuses on optimizing a double-spine
flow path design for OTSs with 10 stations by minimizing the total travel distance for both loaded and empty
flows. We employ transportation methods, specifically the North-West Corner and Stepping-Stone methods, to
determine empty vehicle travel flows. Additionally, the Tabu Search (TS) algorithm is applied to branch the 10
stations into two main layout branches. The results obtained from our proposed method demonstrate a reduction
in the objective function value compared to the initial feasible solution. Furthermore, we explore how changes in the
parameters of the TS algorithm affect the optimal result. We validate the feasibility of our approach by comparing
it with relevant literature and conducting additional tests on layouts with 20 and 30 stations.

KEYWORDS
Overhead transportation systems; tabu search; double-spine layout; transportation method; empty travel; flow path
design

1 Introduction
Vehicles or shuttles find common application across various industrial sectors for resolving
internal transportation challenges. In the past, vehicle-based transportation was primarily associated
with facility layouts. However, today, numerous research centers necessitate a zero-footprint approach
to maintain clean rooms, and many facilities boast intricate machinery layouts that require an efficient
materials handling system. Consequently, there is a growing need for OTSs to facilitate the movement
of materials between stations. Numerous research efforts and systems pertain to OTSs, including
Telelift’s transport system, OTSs designed for semiconductor fabrication plants [1–5], multi-station,
multi-container transport systems [6], and others. OTSs are designed to operate independently of

This work is licensed under a Creative Commons Attribution 4.0 International License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the
original work is properly cited.
256 IASC, 2024, vol.39, no.2

fleet layout considerations [5], making them ideal for managing materials without disrupting floor
activities. Furthermore, OTSs can alleviate staff shortages by automating material transport, allowing
personnel to focus on their core tasks rather than manual material handling [7].
Although the transport system is known as OTSs, it can still be considered a familiar transporta-
tion system. Consequently, its design can be categorized as a type of Automated Guided Vehicle system
(AGVs). Related research has identified several problems in designing transportation systems [8–
10], including issues such as: (1) facility layout design (FLD); (2) estimating the number of vehicles
required; (3) vehicle scheduling; (4) idle-vehicle positioning; (5) battery management; (6) vehicle
routing; (7) deadlock resolution. Each of these issues plays a vital role in the design and control of the
transportation system, with FLD being of strategic importance. FLD involves choosing the system’s
layout type, pick-up and drop-off points, and Flow Path Design (FPD) [10]. The performance of the
transportation system is profoundly affected by FPD due to its significant influence on routing and
dispatching rules. Conversely, a poorly designed flow path can lead to congestion and deadlock during
deliveries [11].
The concept of Flow Path Design (FPD) was initially introduced in 1987 by Gaskin et al. [12].
This introduction sparked the development of numerous studies aimed at optimizing Flow Path Design
(FPD). From related research, several basic features of how to optimize flow path design have emerged,
including (1) layout; (2) flow path type; (3) optimization problems; (4) performance criteria; (5) input;
and (6) methods to solve the objective function. The three most used layouts for designing flow
paths are the conventional, single loop, and tandem layouts. Additionally, the rectilinear flow layout
holds promise for future expansion of Overhead Transportation Systems (OTSs) [5], while the block
layout format is currently under study [13–16]. Selecting the flow path type involves determining the
direction for each path, which can be unidirectional, bidirectional, or multiple-lane flow path types. It
is worth noting that bidirectional flow paths can reduce the total travel distance of vehicles; however,
it is important to be aware that they may also lead to congestion and operational deadlock during
transportation [8–10]. Most researchers construct objective functions to present the problem, aiming
to optimize one or more criteria related to system operation. The most common method for modeling
these objectives is mathematical programming. Performance criteria used for optimization include
the total distance of both loaded and empty flows; the shortest path; the fixed cost such as path
construction, space cost, and cost of control [17]; the total moving distance between the transfer
points and workstations (tandem type); the total distance of congestion; and more. Optimization
problems typically require two parameters: The facility layout and the from-to chart of material flows.
Finally, methods for solving objective functions can be mathematical, simulation-based, or heuristic.
The following tables (Tables 1 and 2) below provide summaries of studies employing various methods
related to optimizing Facility Layout Design (FLD), in accordance with the basic features mentioned
above.

Table 1: Differences between the related works about the layouts; the types of flow paths and the
optimization problems
Authors The layouts The types of flow The optimization problems
paths
Suh et al. [4] Rectilinear Spine type Mixed integer linear
programming
(Continued)
IASC, 2024, vol.39, no.2 257

Table 1 (continued)
Authors The layouts The types of flow The optimization problems
paths
Ting et al. [5] Rectilinear Spine type Mixed integer linear
programming
Gaskins et al. [12] Conventional Unidirectional Zero-one integer linear
programming
Nishi et al. [15] Block Unidirectional and Mixed integer linear
bidirectional programming
Hamzheei et al. [16] Block Bidirectional Integer linear
programming
Kim et al. [17] Conventional Unidirectional Mixed integer linear
programming
Guan et al. [18] Conventional Unidirectional Zero-one integer linear
programming
Kaspi et al. [19] Conventional Unidirectional Zero-one integer linear
programming
Kaspi et al. [20] Conventional Unidirectional Mixed integer linear
programming
Seo et al. [21] Conventional Unidirectional Integer linear
programming
Rubaszewski et al. [22] Conventional Unidirectional Zero-one integer linear
programming
Banerjee et al. [23] Single loop Unidirectional Integer linear
programming
Huang [24] Tandem Unidirectional and Integer linear
bidirectional programming
Akiyama et al. [25] Block Bidirectional Mixed Integer linear
programming
Rubaszewski et al. [26] Conventional Unidirectional Integer linear
programming
Xiao et al. [27] Conventional Unidirectional Zero-one integer linear
programming
Shanmugasundaram Conventional Unidirectional Integer linear
et al. [28] programming
Pourvaziri et al. [29] Aisle structure Bidirectional Mixed integer linear
programming
258 IASC, 2024, vol.39, no.2

Table 2: Differences among related works pertain to performance criteria and solution techniques
Authors The performance criteria Solution technique
Suh et al. [4] The total distance of The genetic algorithm (GA)
congestion
Ting et al. [5] The total travel distance of the The haft-sum method and
loaded vehicle optimal allocation solution
Gaskins et al. [12] The total travel distance of the Multipurpose Optimization
loaded vehicle Systems computer package
Nishi et al. [15] The total of both fixed and The cell-based local search
travel costs heuristics
Hamzheei et al. [16] The total length of the flow An ant colony-based
path algorithm
Kim et al. [17] The total of both fixed and The branch and bound
travel costs procedure
Guan et al. [18] The total travel distance of The discrete
both loaded and empty electromagnetism-like
vehicles mechanism
Kaspi et al. [19] The total travel distance of the The branch and bound
loaded vehicles procedure
Kaspi et al. [20] The total travel distance of The branch and bound
both loaded and empty procedure
vehicles
Seo et al. [21] The total travel distance of the The Tabu search algorithm
loaded vehicles
Rubaszewski et al. [22] The total travel distance of The genetic algorithm (GA)
both loaded and empty
vehicles
Banerjee et al. [23] The total travel distance of the The genetic algorithm (GA)
loaded vehicles
Huang [24] The total moving distance The solution algorithm of the
between the transfer points minimal spanning tree [30]
and workstations
Akiyama et al. [25] The total of both fixed and Sub-tour elimination method
travel costs
Rubaszewski et al. [26] The total travel distance of The heuristic methods such as
both loaded and empty Iterated Local Search (ILS);
vehicles Muli-Start Local Search
(MLS); Bee Algorithm (BA)
and Ant Colony Optimization
(ACO)
Xiao et al. [27] The maximum completion A dual-population
time of all handling tasks collaborative evolutionary
(makespan) genetic algorithm (CEGA)
(Continued)
IASC, 2024, vol.39, no.2 259

Table 2 (continued)
Authors The performance criteria Solution technique
Shanmugasundaram The total travel distance of The genetic algorithm (GA)
et al. [28] both loaded and unloaded and The branch and bound
vehicles procedures
Pourvaziri et al. [29] - The number, position, and An improved branch and cut
width of the aisles algorithm
- The position of the entrance
and exit doors and how to
connect them to the aisles

After conducting a literature review, it becomes evident that there is a wide range of methods
available for optimizing Flow Path Design (FPD). Some commonly used methods to optimize FPD
include exact mathematical analysis models, such as branch and bound, or simulation methods.
However, it can be time-consuming when dealing with scaled-up problems with a large number of
variables [15,18]. On the other hand, most related studies tend to focus solely on optimizing the loaded
travel distance of vehicles, particularly for conventional types with layouts attached to the ground.
However, there are certain critical issues that have not received as much research attention, namely,
(1) optimizing the FPD for OTSs, and (2) minimizing the total travel distance of empty flows. Hence,
this research aims to bridge this gap by proposing a metaheuristic method along with transportation
methods to optimize the FPD for OTSs. The primary objective is to minimize the combined travel
distance of both loaded and empty flows, ensuring efficient cargo delivery from pick-up stations
to delivery stations. This system is designed to allow single-load transportation of vehicles along a
designated flow path. For future expansion considerations, the double-spine layout type has been
chosen. The proposed approach leverages the Tabu Search algorithm, the North-West Corner method,
the Stepping-Stone Method, and the Rectilinear Minisum Location Problem as the appropriate tools
to address these optimization problems.
The organization of this manuscript is as follows. The proposed problem is described in Section 2.
In Section 3, the numerical analysis will be discussed. Then Section 4 will assess the results. The
conclusion of the research will be discussed in Section 5.

2 Problem Description
2.1 Optimize the Double-Spine Type Layout Design
The spine layout is commonly used for systems with future expansion needs, offering two main
forms: Single-spine and double-spine. In the single-spine layout, there is one main branch (which can
be either X-spine or Y-spine), with load ports or stations connecting to the spine as sub-branches.
Similarly, the double-spine layout comprises two perpendicular single main branches (the X-spine and
Y-spine). Each station will only connect to one of the two main spines (not both simultaneously).
Figs. 1a and 1b below introduce the basic single-spine and double-spine layouts, respectively.
260 IASC, 2024, vol.39, no.2

y y
X-spine
(ai,bi)
Y-spine
(aj,bj)

y0
x0 x x

Station i

(a)

y
X-spine

y0

x0 x
Y-spine

(b)

Figure 1: (a) The single-spine layout (X and Y types), (b) The double spine layout

The spine model is positioned in the layout referred to as the rectilinear layout. In this layout,
the X-spine is positioned at coordinates of x = x0 , and the Y-spine is positioned at coordinates y =
y0 . Each station is represented as a point with known coordinates, denoted as Pi (ai , bi ), as shown in
Fig. 1a. The station positions in the coordinate system are illustrated in Fig. 2, which is employed in
this research.
10 meters

10 meters
y
5
6
1
10
3
8

2
7
9
4

O x

Figure 2: The 10 stations are arranged in a rectilinear layout

IASC, 2024, vol.39, no.2 261

As mentioned earlier, each point is only connected to one main branch. Therefore, the stations
are divided into two separate sets, respectively:
SX : The set of points Pi associated with the X-spine.
SY : The set of points Pj associated with the Y-spine.
Based on Figs. 1a and 1b, we derive the general formula for the rectilinear distance between two
points Pi and Pj as follows:
   
dij = |ai − x0 | + bi − bj  + x0 − aj : When Pi and Pj associated with the X-spine.
   
dij = |bi − y0 | + ai − aj  + y0 − bj : When Pi and Pj associated with the Y-spine.
   
dij = |ai − x0 | + |bi − y0 | + x0 − aj  + bj − y0 : When Pi associated with the X-spine and Pj
associated with the Y-spine.
Denoting that the material flows from the Pi to Pj as fij . Because the empty travel flows (the travel
that the vehicles do not deliver the loads) are considered, we denote eij to represent the empty travel
from Pi to Pj . The transportation method can be applied to optimal the empty travel [31].
The integer programming model is employed to formulate the objective function for minimizing
the total travel distance among the 10 stations. However, it is essential to divide the objective function
into three parts: (1) the paths that include flows among the points in SX ; (2) the paths that encompass
flows among the points in SY ; and (3) the paths that involve flows among the points in both sets SX
and SY .
The objective function aims to minimize the total travel distance of both loaded and empty flows
among the points in SX :
       
fij dij + eij dij = fij + eij |ai − x0 | + bi − bj  + x0 − aj  (1)
i=SX j=SX i=SX j=SX i=SX j=SX

The objective function aims to minimize the total travel distance of both loaded and empty flows
among the points in SY :
       
fij dij + eij dij = fij + eij |bi − y0 | + ai − aj  + y0 − bj  (2)
i=SY j=SY i=SY j=SY i=SY j=SY

The objective function aims to minimize the total travel distance of both loaded and empty flows
among the points from two sets SX and SY :
   
fij dij + fij dij + eij dij + eij dij
i=SX j=SY i=SY j=SX i=SX j=SY i=SY j=SX
     
= fij + eij |ai − x0 | + |bi − y0 | + x0 − aj  + bj − y0 
i=SX j=SY
     
+ fij + eij |ai − x0 | + |bi − y0 | + x0 − aj  + bj − y0  (3)
i=SY j=SX

According to Ting et al. [5] and Tompkins et al. [32], the optimization of x0 and y0 , as well as the
branching of stations into two sets SX and SY can be solved separately. Consequently, three distinct
subproblems need to be addressed: Optimizing the x0 and y0 values, and branching the stations.
262 IASC, 2024, vol.39, no.2

Subproblem 1: Optimize x0 —This involves considering the flow between the points in SX as Eq. (1)
and the flow among the points from both sets SX and SY as Eq. (3):
   
Zx0 = fij + eij |ai − x0 | + x0 − aj 
i=SX j=SX
   
+ fij + eij |ai − x0 | + x0 − aj 
i=SX j=SY
   
+ fij + eij |ai − x0 | + x0 − aj 
i=SY j=SX


n

n
   
= fij + eij |ai − x0 | + x0 − aj 
i=1 j=1


n

= (wi + zi ) |ai − x0 | (4)

i=1

Subproblem 2: Optimize y0 —This involves considering the flow between the points in SY as Eq. (2)
and the flow among the points from both sets SX and SY as Eq. (3):
   
Zy0 = fij + eij |bi − y0 | + y0 − bj 
i=SY j=SY
   
+ fij + eij |bi − y0 | + y0 − bj 
i=SX j=SY
   
+ fij + eij |bi − y0 | + y0 − bj 
i=SY j=SX


n
n
   
= fij + eij |bi − y0 | + y0 − bj 
i=1 j=1

n

= (wi + zi ) |bi − y0 | (5)

i=1

Subproblem 3: Branching the stations–This subproblem involves considering all the flows in
Eqs. (1)–(3):
     
Zx0 y0 = fij + eij |ai − x0 | + bi − bj  + x0 − aj 
i=SX j=SX
     
+ fij + eij |bi − y0 | + ai − aj  + y0 − bj 
i=SY j=SY
     
+ fij + eij |ai − x0 | + |bi − y0 | + x0 − aj  + bj − y0 
i=SX j=SY
     
+ fij + eij |ai − x0 | + |bi − y0 | + x0 − aj  + bj − y0  (6)
i=SY j=SX
IASC, 2024, vol.39, no.2 263

As mentioned earlier, the objective functions represented by Eqs. (4) and (5) can be solved
independently. The Rectilinear Minisum Location Problem [32] can be applied to optimize these two
values, x0 and y0 . It is important to note that wi represents the total loaded travel flows to and from
station i, and zi represents
 the total empty travel flows to and from station i. Additionally, because
ai − aj  and bi − bj  are constant, they can be removed from Eqs. (4) and (5) during the analysis.

2.2 The Tabu Search (TS) Algorithm and the Transportation Methods
The TS algorithm is a metaheuristic method that relies on an initial feasible solution. Once
the initial feasible solution is obtained, a search algorithm is employed to identify the admissible
neighborhood and replace the current solution with the best-improving neighborhood. Notably, one
unique feature of TS is its ability to accept non-improving moves [33]. In this study, the TS algorithm
will be characterized by two factors: Short-term memory and long-term memory. The TS algorithm
is applied to handle the objective function represented by Eq. (6).
The function represented by Eq. (6) is converted into a binary integer programming form, as
demonstrated in Eq. (7). In Eq. (7), dijx and dijy denote the distance between two stations intra X-spine
and Y-spine, while dijz and dijw represent the distance between two stations inter X-spine and Y-spine.
The formulas for dijx , dijy , dijz , dijw are provided in Eqs. (8) to (11).

n

n
  
Z= fij + eij dijx + dijy + dijz + dijw (7)
i=1 j=1

Two decision variables xi and yi , where i ∈ {1, 2, 3, . . . , 10}, correspond to the branches that
connect to X-spine and Y-spine, respectively. When xi = 1, it signifies a branch from station i that
links with the X-spine, and when yi = 1, it indicates a branch from station i that connects to the Y-
spine, and vice versa. To ensure that each station is exclusively associated with either the X-spine or
Y-spine, a constraint is necessary, as depicted in Eq. (12).
Thus:
     
dijx = xi + xj − 1 |ai − x0 | + bi − bj  + x0 − aj  ≥ 0 (8)
     
dijy = yi + yj − 1 |bi − y0 | + ai − aj  + y0 − bj  ≥ 0 (9)
     
dijz = xi + yj − 1 |ai − x0 | + |bi − y0 | + x0 − aj  + bj − y0  ≥ 0 (10)
     
dijw = yi + xj − 1 |ai − x0 | + |bi − y0 | + x0 − aj  + bj − y0  ≥ 0 (11)
xi + yi = 1 (12)

To execute the TS, an initial solution that represents the existence of associated branches with
a binary string C must be determined. In this binary string, the bit 0 or 1 signifies the presence of
branches from stations to either the X-spine or Y-spine. A value of 1 indicates that the station is
connected to the X-spine, corresponding to xi = 1 and yi = 0. Conversely, a value of 0 denotes that
the station is linked to the Y-spine, resulting in xi = 0 and yi = 1.
Assuming the initial possible solution as the current and the best solution to the problem, the next
stage of TS will involve searching for neighbors by making changes bit-by-bit to the binary string C to
achieve admissible move. In TS, “move” refers to transitioning from the current solution to the best-
improved solution. For all admissible neighbors, the total travel distance is computed using Eq. (7) to
identify the best-improving move, which is then executed a move.
264 IASC, 2024, vol.39, no.2

Since this research considers empty travel flows, it is necessary to determine the empty travel
matrix. According to Muckstadt and  Maxwell [34], the net vehicle flow of each station, denoted as
NF (i), is defined as NF (i) = fji − fij . Here, fji represents the total travel distance of loaded flows
 j j j
to station i, and fij represents the total travel distance of loaded flows from station i. On the other
j
hand, the supply constraint for each delivery station, denoted as Sm , and the demand constraint for
each pick-up station, denoted as Rl , are defined in Eqs. (13) and (14) based on the net flow equation.
In this study, the transportation method is utilized to optimize the empty travel matrix for this system.
⎧   
⎨ flm − fml flm − fml > 0
Sm = l l
l
l
(13)
⎩0 flm − fml ≤ 0
l l
⎧   
⎨ flm − fml flm − fml > 0
Rl = m m

m

m
(14)
⎩0 flm − fml ≤ 0
m m

The variable eij denotes the quantity of empty travel flows of the vehicle from station
i to station j.
According to Muckstadt et al. [34], if the net vehicle flow NF (i) is non-negative, then eij = NF (i)
  j

and eji = 0. Conversely, if the net vehicle flow NF (i) is negative, then eji = |NF (i)| and eij = 0.
j
 j
 j
In these equations, eij is the total of empty travel flows from node i, and eji is the total of empty
j j
travel flows to node i.
The transportation method requires a distance matrix among the 10 stations and the net vehicle
flow for each station. Using this distance matrix and the net vehicle flows, the matrix of empty travel
flows can be calculated using transportation methods. Two methods are applied, namely the North-
West Corner method and the Stepping-Stone Method [31]. The North-West Corner method can be
summarized as follows:
Step 1: At the position (1, 1)—the North-West Corner, allocate as much as possible the flow value.
It depends on the supply constraint—Sm , and the demand constraint—Rl . The assignment of supply
and demand constraints of each station is Sm(1) to Sm(n) , and R(1)
l to R(n)
l .

Step 2: There are 3 cases.

Step 2.a: If Sm(1) < R(1)
l , the value
 at position (1)
 (1, 1) is set to Sm . Consequently, column 1 is
disregarded. Meanwhile, Rl becomes Rl − Sm , and position (1, 2) is taken into consideration.
(1) (1) (1)

Step 2.b: If Sm(1) > R(1)

l , the value
 at position
 (1, 1) is set to R(1)
l . Consequently, the row 1 is
disregarded. Meanwhile, Sm(1) becomes Sm(1) − R(1)
l , and position (2, 1) is taken into consideration.
Step 2.c: If Sm(1) = R(1) (1) (1)
l , the value at position (1, 1) is set to both Sm and Rl . Consequently, both
column 1 and row 1 are disregarded. Next, position (2, 2) is taken into consideration.
Step 3: Continue this process until reaching the position (n, n), which is the South-East Corner.
The initial feasible matrix of empty travel flows can be obtained using the North-West Corner
method. Subsequently, by applying the Stepping-Stone Method, the matrix of empty travel flows with
the minimum total travel distance can be determined. The conditions for using this method are: (1)
IASC, 2024, vol.39, no.2 265

the number of allocations in the initial feasible solution is 2n + 1, where n is the number of stations;
and (2) the allocations are independent. The Stepping-Stone Method can be summarized as follows:
Step 1: Define the initial feasible matrix of empty travel flows using the North-West Corner
method.
Step 2: Examine the conditions mentioned above for the initial feasible matrix. Following that,
evaluate all unoccupied positions and replace them with other occupied positions to achieve a more
optimal solution. The replacement process can be handled as follows:
Step 2.1: Select an unoccupied position.
Step 2.2: Determine a closed loop through at least 3 occupied positions starting from this chosen
position. The direction of the closed loop is not significant as it yields the same solution.
Step 2.3: At each corner of the closed loop, assign either a (+) or (–) sign, beginning with a (+)
for the unoccupied position.
Step 2.4: Each corner of the closed loop corresponds to a cost (distance between the 2 stations).
Calculate the net changes in cost for the closed loop using the assigned signs at each corner. The net
change in cost is given by c∗1 − c2 + c3 − c4 , where c∗1 is the cost of the unoccupied position.
Step 2.5: Repeat steps 2.1 to 2.4 for all unoccupied positions.
Step 3: The matrix of empty travel is considered optimal if all net changes in cost are greater than
or equal to zero, and the process concludes at this point. Otherwise, proceed to step 4.
Step 4: If there is more than one net change in cost that is negative, select the closed loop with the
most negative value. If multiple net changes in costs have the same negative value, choose the closed
loop that allows for more flow transition at the minimum cost.
Step 5: Within the chosen closed loop, replace the flow of an unoccupied position with a value
from an occupied position, maximizing this replacement as much as possible.
Step 6: Return to step 2 and continue the process until all net change values are greater than or
equal to zero.
As mentioned earlier, we will utilize two fundamental aspects of the TS: Short-term memory and
long-term memory. In this research, static memory with a fixed Tabu tenure T is employed for short-
term memory. This implies that if station a has a value b (0 or 1) denoted as [a, b] since the latest
move, then after the next T iteration, the value [a, b] will not be accepted. Short-term memory is thus
utilized to prevent the solution from getting stuck in local optima and encourages exploration [33].
On the other hand, long-term memory involves counting the frequency of repetition of [a, b], denoted
as F [a, b]. This process of tracking the entire search is known as long-term memory. In the case of
non-improving moves, the values of the objective function Z will increase proportionally with the
repetition frequency F [a, b] and a penalty factor ρ. This reduces the weight given to checking non-
improving moves in subsequent iterations, promoting diversity in the TS algorithm [33]. The iterative
process continues until the stopping conditions are met. As suggested by Seo et al. [21], these stopping
conditions can include reaching the maximum number of iterations or failing to improve the solution
beyond a certain threshold within a set number of iterations. These two criteria are also applied in this
research. Additionally, an aspiration criterion will be used to accept neighbors in the Tabu list if the
solution is better than the current best solution [33].
The TS algorithm applied for optimizing the FPD can be illustrated as follows:
266 IASC, 2024, vol.39, no.2

Step 1: Find an initial possible solution and assign it to both the current solution and the current
best solution, denoting Cbest = Ccur = Cinit . Calculate Zinit based on the value of Cinit , then assign it to
Zcur and Zbest . Begin the process with an iteration count of 0, denoted as niteration = 0 and nnonimpr = 0.
Step 2: Change bit-by-bit of Ccur to generate neighbor solutions. Evaluate all neighbors using short-
term and long-term memory to determine admissible neighbors:
Step 2.1: Calculate Zneighbor for each neighbor.
Step 2.2: If the number of iterations since the latest move to the neighbor is less than the Tabu
tenure value T, and the aspiration criterion is not met (Zneighbor > Zbest ), reject the neighbor and return
to the beginning of step 2. Otherwise, include the neighbor in the set of admissible neighbors and
proceed to step 2.3. This step utilizes short-term memory.
Step 2.3: If the Zneighbor − Zbest ≥ 0, set Zneighbor = Zneighbor + ρ.F [a, b]. This means that for non-
improving moves, the value of the objective function Z will increase proportionally with the repetition
frequency of [a, b] and the penalty factor ρ. This step involves long-term memory.
Step 3: Among the set of admissible neighbors, select the neighbor with the minimum objective
function Z as the current solution. Update Ccur = Cbestimproveneighbor and Zcur = Zbestimproveneighbor . Also, update
F [a, b] = F [a, b] + 1 accordingly. If Zbestimproveneighbor < Zbest , assign Zbest = Zbestimproveneighbor and Cbest =
Cbestimproveneighbor .
Step 4: If nnonimpr ≤ 0 and Zbestimproveneighbor − Zcur < 0, the move is considered an improving moves, and
nnonimpr = nnonimpr − 1. Else if nnonimpr ≥ 0 and the Zbestimproveneighbor − Zcur > 0, the move is a non-improving
move, and nnonimpr = nnonimpr + 1. These actions account for consecutive improving and non-improving
moves. If the move is neither consecutive improving nor consecutive non-improving, set nnonimpr = 0.
Step 5–Checking the stop condition: If niteration ≥ Niteration or nnonimpr ≥ Nnonimpr , terminate the search
process. Otherwise, set niteration = niteration + 1 and return to step 2.

3 Calculation and Numerical Analysis for Optimizing the Double-Spine Type Flow Path Design
The from-to chart of material flows matrix is presented in Table 3.

Table 3: The from-to chart of material flows of the layout

Station 1 2 3 4 5 6 7 8 9 10 To
1 20 10 30 45 25 30 10
2 10 20 15 10 5 10
3 30 10 10 25 30 5 5
4 25 10 5 10 15 25 30
5 15 10 5 20 15 5 10
6 20 15 40 30 15 10
7 25 5 10 20 15 30 15
8 15 20 25 30 40 5 20
9 10 30 10 15 35 5 20
10 5 30 25 10 5 40 15
From
IASC, 2024, vol.39, no.2 267

The coordinates of the 10 stations, denoted as (ai , bi ), are listed in Table 4. It is important to note
that in Fig. 2, each unit of length corresponds to 10 meters, simplifying calculations by using this unit.

Table 4: The coordinates of each station in the layout

Stations 1 2 3 4 5 6 7 8 9 10
ai 1 3 4 6 8 10 13 16 17 18
bi 10 4 8 1 12 11 3 7 2 9

Utilizing the data from Table 3, the net vehicle flow matrix is computed based on the vehicle flows
traveling to and from station i, as shown in Table 5.

Table 5 : The net flow matrix

Stations 1 2 3 4 5 6 7 8 9 10 Total
Total TO 125 130 120 165 110 185 65 105 150 60 1215
Total FROM 170 70 115 120 80 130 120 155 125 130 1215
Net Flow −45 60 5 45 30 55 −55 −50 25 −70 0

Derived from Table 5, the values for Sm and Rl are determined for each station, as depicted in
Table 6.

Table 6: The supply and demand constraints of each station

Stations 1 2 3 4 5 6 7 8 9 10 Total
Sm 0 60 5 45 30 55 0 0 25 0 220
Rl 45 0 0 0 0 0 55 50 0 70 220

The two objective functions presented in Eqs. (4) and (5) can be addressed using “The Rectilinear
Minisum Location Problem”. This method considers the total weight of both loaded and empty travel
flows. The total weight is calculated as the sum of the flows to and from each station. Consequently,
Table 7 has been generated, which illustrates the total weights of loaded vehicles and the total weight
of empty vehicles at each station, based on the data from Tables 5 and 6.

Table 7: The total weight of loaded flows and the total weight of empty flows at each station
Stations 1 2 3 4 5 6 7 8 9 10

fij 170 70 115 120 80 130 120 155 125 130
j
eij 0 60 5 45 30 55 0 0 25 0
j

fij + eij 170 130 120 165 110 185 120 155 150 130
j j

(Continued)
268 IASC, 2024, vol.39, no.2

Table 7 (continued)
Stations 1 2 3 4 5 6 7 8 9 10

fji 125 130 120 165 110 185 65 105 150 60
j
eji 45 0 0 0 0 0 55 50 0 70
j

fji + eji 170 130 120 165 110 185 120 155 150 130
j j
Total weight 340 260 240 330 220 370 240 310 300 260

According to Tompkins et al. [32], for functions like Eqs. (4) and (5) with piecewise linear
structures, the optimal value of x0 should be such that no more than half of the total weight lies to the
left of x0 , and no more than half of the total weight lies to the right of x0 . Similarly, the optimal value
of y0 should be such that no more than half of the total weight is above y0 , and no more than half
of the total weight is below y0 . Furthermore, each optimal solution corresponds to the coordinates of
one station among the 10 stations. In other words, x0 should be equal to one of the ai values, and y0
should be equal to one of the bi values.
The following tables, Tables 8 and 9, provide the calculations for determining the optimal x0
coordinate and the optimum y0 coordinate, respectively.

Table 8: Solving the optimal x0 coordinate


n
Stations (i) Coordinate (ai ) Total weight (Wi ) Wi
i=1

1 1 340 340
3 3 240 580
2 4 260 840
4 6 330 1170
5 8 220 1390 < 1435
6 10 370 1760 > 1435
7 13 240 240
8 16 310 2310
9 17 300 2610
10 18 260 2870

Table 9: Solving the optimal y0 coordinate


n
Stations (i) Coordinate (bi ) Total weight (Wi ) Wi
i=1

4 1 330 330
9 2 300 630
(Continued)
IASC, 2024, vol.39, no.2 269

Table 9 (continued)

n
Stations (i) Coordinate (bi ) Total weight (Wi ) Wi
i=1

7 3 240 870
2 4 260 1130 < 1435
8 7 310 1440 > 1435
3 8 240 1680
10 9 260 1940
1 10 340 2280
6 11 370 2650
5 12 220 2870

Table 8 displays the sum of the weights as 2870 and half of this sum as 1435. Based on these
values, the optimal x0 coordinate is determined to be a6 , which corresponds to station 6. Similarly,
Table 9 reveals that the optimal y0 coordinate is b8 , representing station 8. This configuration results
in the placement of the X-spine at station 6 and the Y-spine at station 8, as illustrated in Fig. 3.
X-spine
y
5
6
1
10
3
8

2 Y-spine
7
9
4

O x

Figure 3: The two optimal spines in the rectilinear layout

Once the coordinates of the two spines, x0 and y0 , have been determined, the next step is to branch
the 8 stations to these two spines in a way that each station is connected to only one spine. To achieve
this, an initial feasible solution is selected for the TS algorithm. The initial feasible solution is presented
as Cinit = x10 x9 x8 x7 x6 x5 x4 x3 x2 x1 . The TS algorithm parameters include a Tabu tenure T = 5, a penalty
factor ρ = 50, a maximum iteration count Niteration = 100, and a maximum number of consecutive non-
improving moves Nnonimpr = 10. These parameters are determined through a trial-and-error process.
A MATLAB program is used to find the optimal solution of the objective function Eq. (7) using the
TS algorithm. Assure that, the initial feasible solution is set to Cinit = 1101111111. With the Cinit
established, the distance matrix among the 10 stations is calculated, as shown in Table 10.
Using the transportation method, the matrix of empty travel flows is optimized based on the
matrix of loaded travel flows and the distance matrix among the 10 stations. As a result, Table 11
illustrates the empty travel flows among the 10 stations for the initial feasible solution Cinit .
270 IASC, 2024, vol.39, no.2

Table 10: The distance matrix among 10 stations based on C init

Station 1 2 3 4 5 6 7 8 9 10
1 0 22 17 22 13 10 19 18 24 18
2 22 0 17 14 17 1 11 16 16 20
3 17 17 0 17 12 9 14 13 19 15
4 22 14 17 0 17 14 9 16 12 20
5 13 17 12 17 0 3 14 13 19 13
6 10 14 9 14 3 0 11 10 16 10
7 19 11 14 9 14 11 0 13 11 17
8 18 16 13 16 13 10 13 0 18 16
9 24 16 19 12 19 16 11 18 0 22
10 18 20 15 20 13 10 17 16 22 0

Table 11: The matrix of empty travel among 10 stations of C init

Station 1 2 3 4 5 6 7 8 9 10
1 0 0 0 0 0 0 0 0 0 0
2 0 0 0 0 0 0 0 35 0 25
3 0 0 0 0 0 0 0 0 0 5
4 0 0 0 0 0 0 45 0 0 0
5 30 0 0 0 0 0 0 0 0 0
6 15 0 0 0 0 0 0 0 0 40
7 0 0 0 0 0 0 0 0 0 0
8 0 0 0 0 0 0 0 0 0 0
9 0 0 0 0 0 0 10 15 0 0
10 0 0 0 0 0 0 0 0 0 0

After 33 iterations of the TS algorithm, the optimal solution has been obtained with Cbest =
0000110000 and Zbest = 19360. The empty travel flows for the optimal solution are shown in
Table 12. When compared to the initial feasible solution, the optimal solution, which minimizes
the objective function Eq. (7), represents as an improvement ΔZ = 1590. The flow path layout for
Cbest = 0000110000 is depicted in Fig. 4.
IASC, 2024, vol.39, no.2 271

Table 12: The matrix of empty travel of the vehicles based on C best
Station 1 2 3 4 5 6 7 8 9 10
1 0 0 0 0 0 0 0 0 0 0
2 45 0 0 0 0 0 15 0 0 0
3 0 0 0 0 0 0 5 0 0 0
4 0 0 0 0 0 0 35 10 0 0
5 0 0 0 0 0 0 0 30 0 0
6 0 0 0 0 0 0 0 10 0 45
7 0 0 0 0 0 0 0 0 0 0
8 0 0 0 0 0 0 0 0 0 0
9 0 0 0 0 0 0 0 0 0 25
10 0 0 0 0 0 0 0 0 0 0

X-spine

5
6
1
10
3
8

2 Y-spine
7
9
4

Figure 4: The optimal flow path after 33 iteration times

4 Evaluate the Numerical Analysis Results

The results obtained from the proposed method indicate a reduction in the objective function
value when compared to the initial feasible solution. This reduction amounts to ΔZ = 1590, which
corresponds to a distance reduction of Simpr = 15900(m). Fig. 5 below illustrates the graph depicting
the objective function values across 100 iteration times. At the 33th iteration time, the value of the
best objective function remains unchanged. Therefore, the optimal solution, Cbest = 0000110000, is
achieved at the 33th iteration time, with an objective function value of Zbest = 19360.
The number of iterations required to obtain the optimal solution can vary depending on the Tabu
tenure and penalty factor used in the TS algorithm. In Fig. 6, it is shown that with T = 4, the optimal
solution is reached after 35 iteration times. Conversely, with T = 12, it takes 111 iterations to reach
the optimal solution. Similarly, with T = 13, it takes 225 iterations, and with T = 14, it still takes after
225 iterations to reach the optimal solution. In Fig. 7, the impact of the penalty factor on the number
of iterations is demonstrated. When ρ = 100, the optimal solution is achieved after 29 iterations. With
ρ = 200, it takes 19 iterations to reach the optimal solution. When ρ = 300, it takes 23 iterations, and
with ρ = 400, it still takes 23 iterations to reach the optimal solution. Since the number of iterations
272 IASC, 2024, vol.39, no.2

directly affects the running time of the search process, it is important to carefully consider the choice
of Tabu tenure and penalty factor when applying the TS algorithm. Finding an appropriate balance
between these parameters is crucial for efficient optimization.

Figure 5: The optimal solution by using the TS algorithm

Figure 6: The optimal solutions when changing the Tabu tenure

To assess the feasibility of the proposed metaheuristic method, it was applied to the layout with
10 stations, whose research was conducted by Ting et al. [5], without considering the empty travel
flows. The algorithm was configured with the following parameters: Tabu tenure T = 5, penalty factor
ρ = 5000, maximum iteration Niteration = 100, and maximum consecutive non-improving move Nnonimpr =
100. The initial feasible solution was set as Cinit = 1101111111 and the initial objective function value
IASC, 2024, vol.39, no.2 273

was Zinit = 3825367. After 10 iteration times, the result is Cbest = 0000100000 and Zbest = 3039339. The
result means that all 10 stations only connect with the Y-spine, and the Y-spine is located at station 8.
This result is consistent with the findings of the research conducted by Ting et al. [5], demonstrating
the appropriateness of the proposed method for optimizing spine flow path design. The result is shown
in Fig. 8.

Figure 7: The optimal solutions when changing the penalty factor

Figure 8: The optimal solution of the example in research [5]

In addition, two examples of layouts with 20 stations and 30 stations were generated using our
proposed method. First, with the 20-station layout, the algorithm was configured with the following
parameters: Tabu tenure T = 5, penalty factor ρ = 1000, the maximum iteration Niteration = 100, and
the maximum of the consecutive non-improving move Nnonimpr = 100. The optimal solution, Cbest =
274 IASC, 2024, vol.39, no.2

01111011110001110101, with an objective function value of Zbest = 645908, was achieved after 74
iterations, starting with Cinit = 11111111111111110111 and Zinit = 656150, as shown in Fig. 9. In this
layout, the X-spine and Y-spine are located at station 7 and station 4, respectively.

Figure 9: The optimal solution of the example with 20 stations

On the other hand, with the 30 stations layout, the algorithm was configured with the following
parameters: Tabu tenure T = 5, penalty factor ρ = 5000, the maximum iteration Niteration = 100, and
the maximum of the consecutive non-improving move Nnonimpr = 100. The optimal solution, Cbest =
111111110111111101011111011101, with an objective function value of Zbest = 2058714, was achieved
after 34 iterations, starting with Cbest = 111111111111111111111111011111 and Zinit = 2068350, as
demonstrated in Fig. 10. In this example, the X-spine and Y-spine are located at station 3 and station
6, respectively. The from-to chart of material flows and the coordinates table of the two examples are
shown in Appendix A (Tables 13 to 16).

Figure 10: The optimal solution of the example with 30 stations

IASC, 2024, vol.39, no.2 275

5 Conclusions
The primary objective of this research is to explore the design of OTSs, considering various critical
factors outlined in the Introduction section. Among these factors, optimizing the FPD is a strategic
factor that should be considered carefully. Furthermore, since this is an OTSs, the spine layout is
chosen to ensure future scalability. Hence, this study places a specific emphasis on the optimization
of a double-spine flow path design to minimize the total travel distance for both loaded and empty
flows. Additionally, this research takes empty travel flows into account, a factor frequently overlooked
in previous work. To address this, the transportation methods, including the North-West Corner and
the Stepping-Stone Method, are applied to optimize empty travel flows. The Rectilinear Minisum
Location Problem is also utilized to determine the locations of the two main branches, X-spine and
Y-spine. Finally, the problem of branching the 10 stations into the two main branches is resolved using
the TS algorithm. Key elements of this method involve specific coefficients like Tabu tenure, penalty
factor, and stop criteria, which significantly impact its effectiveness.
Based on the optimal results presented in Section 4, it is evident that the suggested methods are
suitable for optimizing the spine flow path design. Furthermore, to validate our methods, comparisons
and examples have been conducted. The results demonstrate that the optimal solution’s value can be
reduced, and the iteration count in the search process can be influenced by adjusting the coefficients
used in the TS algorithm. In this study, the coefficients, including Tabu tenure and penalty factor,
were determined through a trial-and-error method. However, future research could explore methods
for optimizing these coefficients using techniques such as Neural Networks, Genetic Algorithms,
and others. Additionally, practical experiments should be conducted in the near future to assess the
feasibility and effectiveness of the proposed solution in real-world scenarios.

Acknowledgement: This research is funded by Ho Chi Minh City University of Technology (HCMUT),
VNU-HCM under Grant Number B2021-20-04. We acknowledge Ho Chi Minh City University of
Technology (HCMUT), VNU-HCM for supporting this study.

Funding Statement: This research is funded by Ho Chi Minh City University of Technology
(HCMUT), VNU-HCM under Grant Number B2021-20-04.

Author Contributions: Mr. Nguyen Huu Loc Khuu is the main author who wrote the manuscript. He
is also the main author who did the survey about this research and proposed the method to develop
this research. Mr. Thuy Duy Truong and Mr. Quoc Dien Le are the authors who contributed to the
exploration of all the research and also proposed the ideas for the research problems. Mr. Tran Thanh
Cong Vu, and Mr. Hoa Binh Tran designed, fabricated the mechanical part of this proposed issue
and preparing for the experiments in the near future of this system. Assoc. Prof. Tuong Quan Vo
is the supervisor of this research topic. He proposes the ideas to the authors to reach the necessary
problems about this system. And he is also correct about the scientific and technological issues of this
manuscript.

Availability of Data and Materials: The whole data sets proposed in this manuscript are not publicly
available, but they are available from the corresponding author on reasonable request.

Conflicts of Interest: The authors declare that they have no conflicts of interest to report regarding the
present study.
276 IASC, 2024, vol.39, no.2

References
[1] B. I. Kim, S. Oh, J. Shin, M. Jung, J. Chae and S. Lee, “Effectiveness of vehicle reassignment in a large-
scale overhead hoist transport system,” Int. J. Produc. Res., vol. 45, no. 4, pp. 789–802, Feb. 2007. doi:
10.1080/00207540600675819.
[2] A. Zhakov et al., “Application of ANN for fault detection in overhead transport systems for semiconductor
fab,” IEEE Trans. Semicond. Manuf., vol. 33, no. 3, pp. 337–345, Aug. 2020. doi: 10.1109/TSM.66.
[3] C. H. Duo, “Modelling and performance evaluation of an overhead hoist transport system in a 300 mm
fabrication plant,” Int. J. Adv. Manuf. Technol., vol. 20, pp. 153–161, 2002. doi: 10.1007/s001700200137.
[4] Y. J. Suh and J. Y. Choi, “Efficient Fab facility layout with spine structure using genetic algorithm under
various material-handling considerations,” Int. J. Prod. Res., vol. 60, no. 9, pp. 2816–2829, 2022. doi:
10.1080/00207543.2021.1904159.
[5] J. H. Ting and J. M. A. Tanchoco, “Optimal bidirectional spine layout for overhead material handling
systems,” IEEE Trans. Semiconduct. Manuf., vol. 14, no. 1, pp. 57–64, 2001. doi: 10.1109/66.909655.
[6] N. H. L. Khuu, V. A. Pham, T. T. C. Vu, V. T. B. Dao, T. D. Nguyen and V. Quan Tuong, “A study on
design and control of the multi-station multi-container transportation system,” Appl. Sci., vol. 12, no. 5,
pp. 2686–2713, 2022. doi: 10.3390/app12052686.
[7] W. A. Chen, R. D. de Koster, and Y. Gong, “Performance evaluation of automated medicine
delivery systems,” Transp. Res. Part E: Logist. Transp. Rev., vol. 147, pp. 102242, Mar. 2021. doi:
10.1016/j.tre.2021.102242.
[8] I. F. A. Vis, “Survey of research in the design and control of automated guided vehicle systems,” Eur. J.
Oper. Res., vol. 170, no. 3, pp. 677–709, May 2006. doi: 10.1016/j.ejor.2004.09.020.
[9] T. A. Le and M. B. M. de Koster, “A review of design and control of automated guided vehicle systems,”
Eur J. Oper Res, vol. 171, no. 1, pp. 1–23, May 2006. doi: 10.1016/j.ejor.2005.01.036.
[10] P. R. Gutta, V. S. Chinthala, R. V. Manchoju, V. C. MVN, and R. Purohit, “A review on facility layout
design of an automated guided vehicle in flexible manufacturing system,” Mater. Today: Proc., vol. 5, no.
2, pp. 3981–3986, 2018.
[11] E. B. Hoff and B. R. Sarker, “An overview of path design and dispatching methods for automated guided
vehicles,” Integr. Manuf. Syst., vol. 9, no. 5, pp. 296–307, 1998. doi: 10.1108/09576069810230400.
[12] R. J. Gaskins and J. M. A. Tanchoco, “Flow path design for automated guided vehicle systems,” Int. J.
Prod. Res., vol. 25, no. 5, pp. 667–676, 1987. doi: 10.1080/00207548708919869.
[13] R. F. Zanjirani, G. Laporte, and M. Sharifyazdi, “A practical exact algorithm for the shortest loop
design problem in a block layout,” Int. J. Prod. Res., vol. 43, no. 9, pp. 1879–1887, May 2005. doi:
10.1080/00207540500031980.
[14] K. Eshghi and M. Kazemi, “Ant colony algorithm for the shortest loop design problem,” Comput. Ind.
Eng., vol. 50, no. 4, pp. 358–366, Aug. 2006. doi: 10.1016/j.cie.2005.05.003.
[15] T. Nishi, S. Akiyama, T. Higashi, and K. Kumagai, “Cell-based local search heuristics for guide path design
of automated guided vehicle systems with dynamic multicommodity flow,” IEEE Trans. Autom. Sci. Eng.,
vol. 17, no. 2, pp. 966–980, Apr. 2020. doi: 10.1109/TASE.8856.
[16] M. Hamzheei, R. Z. Farahani, and H. B. Rashidi, “An ant colony-based algorithm for finding the shortest
bidirectional path for automated guided vehicles in a block layout,” Int. J. Adv. Manuf. Technol., vol. 64,
pp. 399–409, Jan. 2013. doi: 10.1007/s00170-012-3999-1.
[17] K. H. Kim and J. M. A. Tanchoco, “Economical design of material flow paths,” Int. J. Prod. Res., vol. 31,
no. 6, pp. 1387–1407, 1993. doi: 10.1080/00207549308956797.
[18] X. Guan, X. Dai, and J. Li, “Revised electromagnetism-like mechanism for flow path design
of unidirectional AGV systems,” Int. J. Prod. Res., vol. 49, no. 2, pp. 401–429, Jan. 2011. doi:
10.1080/00207540903490155.
[19] M. Kaspi and J. M. A. Tanchoco, “Optimal flow path design of unidirectional AGV systems,” Int. J. Prod.
Res., vol. 28, no. 6, pp. 1023–1030, 1990. doi: 10.1080/00207549008942772.
IASC, 2024, vol.39, no.2 277

[20] M. Kaspi, U. Kesselman, and J. M. A. Tanchoco, “Optimal solution for the flow path design problem
of a balanced unidirectional AGV system,” Int. J. Prod. Res., vol. 40, no. 2, pp. 389–401, 2002. doi:
10.1080/00207540110079761.
[21] Y. Seo, C. Lee, and C. Moon, “Tabu search algorithm for flexible flow path design of unidirectional
automated-guided vehicle systems,” OR Spectrum, vol. 29, pp. 471–487, Jul. 2007.
[22] J. Rubaszewski, A. Yalaoui, L. Amodeo, and S. Fuchs, “Efficient genetic algorithm for unidirectional flow
path design,” IFAC Proc., vol. 45, no. 6, pp. 883–888, 2012.
[23] P. Banerjee and Y. Zhou, “Facilities layout design optimization with single loop material flow path
configuration,” The Int. J. Prod. Res., vol. 33, no. 1, pp. 183–203, 1995. doi: 10.1080/00207549508930143.
[24] C. Huang, “Design of material transportation system for tandem automated guided vehicle systems,” Int.
J. Prod. Res., vol. 35, no. 4, pp. 943–953, 1997. doi: 10.1080/002075497195461.
[25] S. Akiyama, T. Nishi, T. Higashi, K. Kumagai, and M. Hashizume, “A multi-commodity flow model for
guide path layout design of AGV systems,” in 2017 IEEE Int. Conf. Indust. Eng. Eng. Manag. (IEEM),
2017, pp. 1251–1255.
[26] J. Rubaszewski, A. Yalaoui, and L. Amodeo, “Solving unidirectional flow path design problems using
metaheuristics,” in Metaheuristic for Production Systems, Cham: Springer, Operations Research/Computer
Science Interfaces Series, 2016, vol. 60, pp. 25–56. doi: 10.1007/978-3-319-23350-5.
[27] H. Xiao, X. Wu, Y. Zeng, and J. Zhai, “A CEGA-based optimization approach for integrated designing of
a unidirectional guide-path network and scheduling of AGVs,” Math. Problems Eng., vol. 2020, no. 2, pp.
1–16, 2020.
[28] N. Shanmugasundaram, K. Sushita, S. P. Kumar, and E. N. Ganesh, “Genetic algorithm-based road
network design for optimising the vehicle travel distance,” Int. J. Veh. Inf. Commun. Syst., vol. 4, no. 4,
pp. 344–354, 2019.
[29] H. Pourvaziri, H. Pierreval, and H. Marian, “Integrating facility layout design and aisle structure in
manufacturing systems: Formulation and exact solution,” Eur J. Oper. Res., vol. 290, no. 2, pp. 499–513,
2021. doi: 10.1016/j.ejor.2020.08.012.
[30] F. S. Hiller and G. J. Lieberman, “The minimum spanning tree problem,” in Introduction to Operations
Research, 7th ed. New York: Mc Graw-Hill, 1990, pp. 415–420.
[31] C. F. Palmer and A. E. Innes, “The transportation method,” in Operational Research by Example, UK:
Macmillan Education, 1980, pp. 121–142.
[32] J. A. Tompkins, J. A. White, Y. A. Bozer, and J. M. A. Tanchoco, “Rectilinear-distance facility location
problems,” in Facilities Planning, 4th ed. Hoboken, New Jersey: John Wiley & Sons, 2010, pp. 520–521.
[33] M. Laguna, R. Martí, P. Pardalos, and M. Resende, “Tabu search,” in Handbook of Heuristics, Berlin,
Germany, Springer, 2018, pp. 741–758.
[34] W. L. Maxwell and J. A. Muckstadt, “Design of automatic guided vehicle systems,” AIIE Trans., vol. 14,
no. 2, pp. 114–124, 1982. doi: 10.1080/05695558208975046.

Appendix A.
The following tables show the from-to chart and the coordinates of the layout with 20 stations
and 30 stations.

Table 13: The from-to chart of material flows of the layout with 20 stations
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1 0 21 36 66 4 23 11 33 61 33 60 11 4 24 24 67 56 57 57 93
2 73 0 75 63 18 62 21 62 85 95 1 76 95 27 0 81 70 72 47 96
3 91 83 0 17 55 61 6 55 96 22 34 89 44 29 5 54 8 93 4 26
(Continued)
278 IASC, 2024, vol.39, no.2

Table 13 (continued)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
4 63 10 5 0 90 6 81 54 99 40 58 61 72 88 84 31 26 31 9 86
5 1 1 4 99 0 26 84 76 57 40 27 39 10 48 73 36 82 67 75 81
6 20 72 26 0 50 0 13 67 23 27 83 95 76 68 44 68 78 43 18 5
7 14 63 5 29 28 40 0 8 24 30 96 82 48 87 82 49 36 68 15 26
8 43 61 24 6 21 57 3 0 92 79 10 21 15 62 35 71 57 25 36 79
9 25 48 40 36 85 99 1 1 0 18 61 80 19 71 76 29 26 57 6 94
10 41 80 79 82 43 75 37 0 40 0 84 4 6 15 32 18 31 20 41 22
11 61 60 7 3 89 67 54 26 16 68 0 99 83 82 11 14 73 4 77 4
12 84 60 11 4 17 96 42 78 49 24 41 0 9 49 63 19 79 86 32 64
13 23 17 79 17 9 14 5 13 88 17 86 25 0 99 17 87 21 14 79 57
14 11 16 33 96 6 78 58 77 99 16 73 64 91 0 49 74 13 79 31 96
15 10 11 8 80 98 83 40 79 53 51 1 69 14 28 0 92 93 87 73 81
16 90 25 88 69 97 71 81 36 47 62 28 20 73 16 81 0 4 79 47 66
17 15 49 32 84 20 60 36 46 92 23 95 41 47 37 30 33 0 65 21 93
18 89 69 1 67 67 26 83 44 90 9 71 58 42 61 70 1 25 0 92 96
19 17 0 43 29 20 98 84 47 98 1 14 0 89 33 91 74 45 55 0 34
20 28 96 48 48 61 3 94 80 38 75 65 1 25 37 77 42 3 36 37 0

Table 14: The coordinates of each station in the layout 20 stations

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
ai 11 8 37 2 35 36 25 3 4 22 26 34 31 25 15 29 21 27 33 6
bi 4 27 3 20 16 29 1 6 35 30 10 28 13 25 11 2 5 26 22 33

Table 15: The from-to chart of material flows of the layout with 30 stations
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
1 0 7 13 86 82 34 58 96 84 14 70 91 57 36 14 68 69 96 93 63 85 35 68 33 50 71 67 25 71 54
2 6 0 15 15 54 69 4 65 71 39 37 95 62 9 49 47 80 50 30 26 91 0 74 76 15 94 39 70 57 51
3 15 33 0 50 59 79 67 83 67 34 3 12 90 87 46 42 53 57 95 19 70 5 9 93 90 8 98 0 52 99
4 12 19 50 0 28 73 47 37 27 8 17 80 70 69 14 82 24 97 0 15 66 13 30 96 57 22 16 99 39 59
5 6 38 59 88 0 38 46 4 16 28 32 6 86 88 64 10 9 57 12 90 29 40 44 77 77 9 11 83 94 6
6 84 67 83 65 72 0 28 90 94 85 40 3 65 99 74 45 33 41 25 79 36 2 55 82 67 71 99 46 31 40
7 42 37 15 2 7 59 0 5 5 62 98 38 36 54 59 62 52 11 38 50 89 90 65 99 39 99 6 10 54 93
8 98 96 86 57 10 59 75 0 79 76 75 80 32 36 58 78 79 63 11 38 26 37 36 43 45 9 25 9 59 31
9 89 8 36 20 61 63 16 97 0 84 45 44 77 99 48 37 65 14 88 6 12 20 13 78 13 86 47 93 79 78
10 84 89 92 67 51 97 77 65 21 0 21 44 70 56 26 40 88 16 65 73 58 61 43 7 51 28 33 75 86 62
11 9 29 33 85 85 98 4 57 53 10 0 60 31 80 37 19 10 67 81 12 96 3 47 71 26 39 29 83 96 42
12 84 1 79 51 30 93 14 31 98 33 92 0 84 91 29 26 71 53 16 55 11 67 6 50 99 10 56 94 18 83
13 92 39 33 65 69 59 45 85 46 59 47 18 0 27 4 42 62 32 45 81 70 70 90 17 97 97 46 13 15 34
14 5 79 35 44 24 93 78 90 44 93 86 29 43 0 20 35 56 95 66 7 22 86 51 84 73 66 22 58 85 88
15 59 54 27 99 59 77 32 32 64 93 46 3 38 35 0 75 27 77 15 53 65 63 27 30 29 66 40 22 25 38
(Continued)
IASC, 2024, vol.39, no.2 279

Table 15 (continued)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
16 18 17 4 59 51 6 1 13 57 86 6 35 83 32 16 0 30 19 56 33 38 44 8 55 17 43 70 64 86 29
17 38 57 64 82 8 19 31 53 72 85 57 5 12 25 11 37 0 94 81 84 13 51 62 44 84 28 9 58 51 32
18 22 37 36 19 84 33 75 54 72 55 0 20 90 48 16 34 5 0 22 15 20 10 8 41 82 90 37 85 28 31
19 58 84 86 68 71 51 65 8 41 23 26 7 22 53 9 96 20 62 0 79 3 95 20 75 33 66 48 97 70 1
20 70 55 74 28 28 22 76 98 29 31 57 54 62 37 36 32 58 76 86 0 7 19 66 8 35 56 72 27 0 6
21 43 76 44 65 49 62 81 3 77 19 41 96 72 62 78 87 59 98 46 86 0 43 16 92 96 14 18 39 11 77
22 97 94 80 18 35 24 34 59 3 22 93 66 94 91 39 76 70 48 5 38 44 0 57 16 26 81 43 41 50 14
23 70 22 90 59 82 68 98 42 21 74 29 73 35 60 50 66 86 32 12 80 16 91 0 54 93 16 59 32 7 77
24 18 78 15 12 42 65 38 16 18 24 0 91 84 9 1 73 2 78 92 10 87 53 71 0 7 80 1 38 96 27
25 54 9 66 54 42 77 8 68 27 67 67 14 80 19 91 71 16 81 33 85 43 87 10 71 0 6 62 7 70 42
26 35 5 99 71 14 98 56 75 14 12 10 3 32 16 16 18 62 43 27 14 32 13 64 51 75 0 48 31 11 26
27 81 96 52 35 5 15 78 70 14 94 8 34 10 64 84 58 32 11 99 98 36 36 14 12 48 12 0 92 29 28
28 61 66 70 99 33 17 40 18 0 57 81 48 83 16 0 64 77 24 0 14 6 25 50 80 50 81 76 0 10 5
29 12 97 10 37 62 54 62 13 99 18 94 94 11 53 44 63 70 39 77 28 87 72 42 87 40 48 13 79 0 53
30 44 80 89 72 75 14 90 44 18 71 5 19 13 83 92 26 42 71 63 47 69 73 5 23 77 35 77 3 82 0

Table 16: The coordinates of each station in the layout 30 stations

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
ai 19 52 27 22 23 39 1 37 7 15 59 56 32 53 18 16 20 12 29 11 31 46 13 28 35 5 45 24 38 8
bi 2 36 56 34 53 31 30 5 38 1 44 33 3 26 39 42 51 9 37 20 10 29 32 40 8 18 58 19 60 22