Journal of Manufacturing Systems 44 (2017) 179190

Contents lists available at ScienceDirect

Journal of Manufacturing Systems

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

A RFID-enabled positioning system in automated guided vehicle for

smart factories
Shaoping Lu a , Chen Xu b, , Ray Y. Zhong c , Lihui Wang d
a
Department of Transport and Logistics, Shenzhen University, Shenzhen, China
b
Institute of Intelligent Computing Science, Shenzhen University, Shenzhen, China
c
Department of Mechanical Engineering, The University of Auckland, Auckland, New Zealand
d
Department of Production Engineering, KTH Royal Institute of Technology, Stockholm, Sweden

a r t i c l e i n f o a b s t r a c t

Article history: Smart factory, as one of key future for our industry, requires logistics automation within a manufacturing
Received 24 October 2016 site such as a shop oor. Automated guided vehicle (AGV) systems may be one solution, whose accuracy
Received in revised form 25 March 2017 will be inuenced by some factors. This paper presents a radio frequency identication (RFID)-enabled
Accepted 31 March 2017
positioning system in AGV for smart factory. Key impact factors on AGVs accuracy such as magnetic eld
in circular antenna, circular magnetic eld, and circular contours stability are examined quantitatively.
Keywords:
Based on the examinations, simulation studies and a testbed are carried out to evaluate the feasibility
RFID
and practicality of the proposed approach. It is observed that large diameter antennas are used in driving
AGV
Positioning
zone and small diameter antennas are used in parking zone. This approach was compared with another
CPS method using passive RFID tags and it is superior to that method with greatly reduced tags deploy-
Smart Factory ment. Observations and lessons from simulation and testbed studies could be used for guiding automatic
logistics within a smart manufacturing shop oor.
2017 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved.

1. Introduction chronization of manufacturing activities and logistics behaviors

Automation in manufacturing is demanded continuously since
it saves labors and materials to improve quality, accuracy, preci-
sion and efciency [1]. Take a smart factory for example, some
cutting-edge technologies such as cyber-physical systems (CPS) is
employed to monitor physical processes through creating a vir-
tual world so that manufacturing automation could be achieved
by modularized and structured units [2]. Manufacturing systems
using CPS technology play an important role in a smart factory.
Within such systems, information from all related production activ-
ities will be closely synchronized and monitored between physical
factory oor and cyber computational space [3]. One of key fea-
tures of smart factory is the logistics automation which enables an
efcient delivery of various components within the manufacturing
sites such as workshops.
The logistics automation could be realized by automated guided
vehicles (AGVs) which are able to move different materials like
raw-materials, work-in-progress items, and nished products in a
exible manner without human interventions [4]. The high syn-
Corresponding author.
E-mail address: xuchen@szu.edu.cn (C. Xu).
must be ensured so that the production could go smoothly. There-
fore, positioning system for AGVs is crucial since determination
of the location of an AGV will largely inuence the entire manu-
facturing system. Machines and workers may spend extra waiting
time for the arrival of materials if the AGVs positioning system
cannot work properly. Additionally, manufacturing performance
heavily relies on the logistics efciency and effectiveness within a
smart factory where highly automated machines will be equipped.
That brings great challenges for the positioning system in an AGV
because its system layout and performance will greatly affected by
manufacturing facilities and situations such as dynamic changes or
emergency production orders [5]. Currently, xed guidepaths with
magnetic bands are widely used for positioning system due to its
low cost [6]. However, this approach cannot t the manufacturing
exibility since the facilities within a smart factory will be recon-
gured and redesigned given different mass customize products.
For example, the manufacturing machines may follow production
line mode in peak-season and then follow another mode by group-
ing these machines with similar functionalities in off-season. Fixed
guidepaths thus should be revised with more efforts in terms of
labor and time cost.
As the development of Auto-ID technologies like radio fre-
quency identication (RFID), the positioning systems for AGVs for

http://dx.doi.org/10.1016/j.jmsy.2017.03.009
0278-6125/ 2017 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved.
180 S. Lu et al. / Journal of Manufacturing Systems 44 (2017) 179190
smart factories could be improved. Some primary studies show the
feasibility and practicality of using RFID for AGVs positioning in
typical manufacturing factories. For example, RFID was used for
positioning by placing readers and tags upon manufacturing envi-
ronment so as to facilitate the automatic logistics through a cloud
or remote controller [7]. Passive RFID is used in an indoor position
tracking system based on a precise tracking algorithm [8]. How-
ever, there are some research questions needed to be addressed.
Firstly, what is the key inuence on positioning accuracy in a RFID-
enabled AGV? Secondly, how to design and develop a RFID-enabled
positioning system considering the key inuences? Thirdly, how to
evaluate the designed and developed system?
In order to answer this questions, this paper introduces a RFID-
enabled positioning system for AGVs by making full use of the
near-eld coupling RFID to realize precise coverage positioning to
support ultimate automatic logistics in a smart factory. To gure out
the key inuences, this paper analyzes the magnetic eld in circu-
lar antenna, circular magnetic eld, and circular contours stability.
It is observed that a circular antenna in the YOZ plane will gener-
ate a distribution law of antenna radiation patterns. To design and
develop a RFID-enabled positioning system, the key inuences are
considered. It is found that the positioning accuracy is largely inu-
enced by the radius of the eld contour circle. Thus, large diameter
driving antenna coils are used for coarse positioning and small-
diameter stopping antenna coils are used for precise positioning
and posture correction. To test this system, this paper uses sim-
ulation study, laboratory testbed, and comparisons with another
approach to verify the proposed system.
The rest of this paper is organized as follows. Section 2 gives a
literature review about RFID technology, positioning in AGV sys-
tem, and near eld coupling (NFC) technology. Section 3 presents
some basic denitions of NFC and analyzes three key impact factors.
Section 4 reports on the design and development of a RFID-enabled
positioning system. Section 5 demonstrates the simulation studies,
a testbed, and a comparison study. Section 6 concludes this paper
by giving our ndings and future research directions.
2. Literature review
2.1. Cyber-physical systems (CPS)
CPS refers to a mechanism where physical and software com-
ponents are signicantly intertwined so that the physical objects
could be well controlled or monitored in cyber space [3]. CPS-
enabled manufacturing is able to convert typical resources such
as workers, materials, and equipment into smarter objects which
show great promise in smart factory in the future [9]. Thus, cyber-
manufacturing based on CPS concept has been placed signicant
attention so as to ultimately realize high performance and reliable
production.
CPS involves large number of transdisciplinary methodologies
such as cybernetics theory, mechanical and mechatronics, design
and process science, manufacturing systems, and computer sci-
ence. One of the key technical method is embedded systems which
enable a highly coordinated and combined relationship between
physical objects and their computational elements or services [10].
A CPS-enabled system, unlike traditional embedded systems, is
a networked interaction designed and developed with physical
input and output as well as their cyber-twined services such as
control algorithms and computational capacities. Thus, large num-
ber of sensors play important roles in CPS. For example, multiple
sensory devices are widely used in CPS to achieve different pur-
poses such as touch screens, light sensors, force sensors, etc. Typical
CPS applications have been reported by making full use of sensor-
based communication-enabled autonomous systems. Vast number
of wireless sensor networks supervise the environmental aspects
so that the information from the environment could be centrally
controlled and managed for decision-makings [11].
2.2. RFID technology
RFID technology, as one of key elements in CPS, has been widely
used due to its exible operations and high accuracy [12]. RFID
is able to track and trace objects by wireless radio waves [13].
Thus, it was used for tracking the key components in a produc-
tion system so that the data captured could be used for supporting
real-time scheduling [14]. With enormous data generated by RFID-
enabled smart manufacturing objects, data mining approach was
used to excavate valuable information and knowledge for support-
ing advanced production decision-makings. For example, standard
operation times (SOTs) and dispatching rules were mined from
large set of RFID production data to support real-time planning and
scheduling in the shop oors [15,16].
Indoor localization is a particular application of RFID technology
besides its tracking and tracing ability. Hasani et al. thus introduced
a novel hybrid conguration for indoor positioning using passive
RFID and wireless local area network (WLAN) [17]. The position-
ing system is based on a mobile device, a textile tag, and several
RFID readers for locating various objects. For indoor locating per-
sons, a foot-mounted inertial measuring units (IMU)-based position
method was proposed from several active RFID tags placed at some
locations in a building [18]. This paper integrates IMUs and RSS
(Received Signal Strengths) with a loose Kalman lter to design a
tight KF-based INS/RFID system. With different types of RFID fre-
quency, large numbers of positioning system by assistance of RFID
tags and readers have been studied, designed and developed [19].
2.3. Positioning in AGV system
Positioning plays a critical role in an AGV system. A recently
proposed coil pad design named bipolar pad (BPP) was introduced
with a 10-kHz 300-W distributed inductive power transfer (IPT)
system for AGV applications [20]. A 3D nite element modeling tool
is used to evaluate the distributed AGV positioning application. In
an industrial guidance system, path tracking is signicant. Villa-
gra et al. introduced a control law which allows AGVs to operate
tracking a predened route with industrial grade of accuracy and
reliability by integrating three control technologies: fuzzy, vector
pursuit, and atness-based control [21]. Automatic conguration of
an initial position is one of the key research areas in AGV systems.
Herrero et al. thus proposed an unconstrained optimization method
coupled with probabilistic techniques to address the problem [22].
In order to achieve effective position tracking, Cho et al. used B-
Spline surface equation based on passive UHF RFID to establish a
wireless position tracking system for indoor applications [7].
2.4. Near eld coupling (NFC) technology
NFC refers to an electromagnetic eld around an object which
carries reactive or Fresnel diffraction [23]. Wave interference is a
basic manifestation of the superposition principle in many appli-
cations. Thus, [24] demonstrated that the near-eld interference
of a circularly polarized dipole resulted in a unidirectional excita-
tion of guided electromagnetic modes in the near-eld. Near-eld
analysis of Terahertz pulse generation from photo-excited charge
density gradients was studied to show that the charge dynamics in
the plane of the surface was able to radiate substantially stronger
THz pulses than the charge dynamics in the direction normal to the
surface [25].
Due to its advantages, NFC technology has been widely used. Hsu
and Huang [26] reported a Koch-shaped log-periodic dipole array
(LPDA) antenna for universal UHF RFID handheld reader. In this
 S. Lu et al. / Journal of Manufacturing Systems 44 (2017) 179190 181

Fig. 2. Precise coverage using regular triangle.

Fig. 1. Reading effective area on tags.

communication, the LPDA antenna covers the RFID band of oper-

ations which is better than 80% total radiation efciency. For the
near-eld measurement with selective feature extraction, a novel
3D radar cross section feature extraction technique was proposed
[27]. This approach is veried by using a synthetic aperture radar
facility over the 812-GHz range. An adaptation of magnetically
coupled resonators of wireless power transfer was introduced in
terms of theoretical model and experiments [28]. The experimental
results show that the method can achieve a near-constant efciency
of over 70% for a range of 070 cm. Fig. 3. Magnetic analysis.

From reviewing the related work, several research gaps could be

identied. Firstly, RFID technology is mainly used in manufactur-  the vertices
system [30],  ofthe inscribed
 triangle coordinates are

ing as well as logistics and supply chain management for tracking A(0, ), B 3
, 12  ,C 3
, 12  .
2 2
and tracing various items. RFID for positioning in AGV systems is
scarcely reported. Secondly, AGV positioning is relied on different
3.2. Analysis of magnetic eld in circular antenna
sensors. The accuracy should be improved in a cost-efcient way
in industrial applications. However, some key impact factors on its
According to Biot-Savart law, the magnetic ux density (MFD)
positioning accuracy are limitedly studied. Thirdly, NFC technol-
dB which is generated by current carrying conductor in vacuum at p
ogy has been widely used due to its good performance and high
point is proportional to the current element Idl [31]. The MFD then
reliability. The integration of near-eld coupling with RFID in AGV
could be calculated by:
positioning may provide a high-performance and low-cost solu-
tion. In order to fulll the gaps, this paper proposes a RFID-enabled 0 Idl sin
dB = (1)
positioning system in AGV to improve the accuracy by examining 4r 2
some key impact factors such as magnetic eld, circular magnetic
where 0 = 4 107 N A2 is vacuum permeability. Due to this
eld, and circular contours stability.
current segment, dB is perpendicular to the plane dened by Idl
and r. The MFD B generated by any shape current-carrying wire at
3. Positioning in a RFID-enabled AGV
point p is equal to the sum of dB:
3.1. Denitions
 
0 Idl sin
B= dB = (2)
4r 2
A near-eld antenna plays important role in AGV positioning. L
In this paper, an antenna that generates a stable and precise mag- Fig. 3 shows a circle of radius R horizontally in the plane XOY.
netic contour circle is horizontally placed in the bottom of an AGV Assuming p is the projection of p in the XOY plane. is the angle
(as shown in Fig. 1). The magnetic line is a closed curve that con- which consist of current vector in point Q and direction vector from
stantly changes its direction. When tags are placed horizontally, point Q to P.  is the angle which consists of magnetic eld direction
Fig. 1 indicates the effective area (shadowed eld) where a tag can and the Z axis. is the angle between OQ and OX. is the angle
be read. When a tag is placed vertically, the effective area is marked between OP and OQ, then we can get:
with dash lines. Thus, we may conclude that: (1) a reader should be 
installed horizontally under an AGV; and (2) tags should be placed PQ = (x R cos )2 + (y R sin )2 (3)
horizontally on the ground. 
This paper uses circular antenna coils since it can generate a OP  = x 2 + y2 (4)
circle coverage eld [29]. Regular triangle is used to design the tags
which are arrayed into polygon in equal spacing. As shown in Fig. 2, OQ = R (5)
the basic principles are as follows: the regular triangle ABC has According to the cosine theorem, we can get:
the edge length S, and its circumcircle
radius is . The relationship
(P Q ) = (OP  ) + (OQ )2 2(OQ ) (OP  ) cos
 2 2
between the S and  is S = 3. To establish a Cartesian coordinate (6)
182 S. Lu et al. / Journal of Manufacturing Systems 44 (2017) 179190

(OP  )2 + (OQ )2 (P  Q )2
cos =
2(OQ )(OP  )
(7)
x2 + y2 + r 2 (x R cos )2 (y R sin )2
= 
2R x 2 + y2

Then, (7) could be simplied as:

x cos + y sin
cos =  (8)
x2 + y2

sin = 1 cos2

x2 + y2 x2 cos2 2xy cos sin y2 sin2
= (9)
x 2 + y2
x sin y cos
=  Fig. 4. Different patterns in YOZ.
x 2 + y2
  
 R x2 + y2 sin
d

= (OP  ) sin
2+
QB 0
 B =
4
IR   3/2 (16)
x2 + y2 (x sin y cos )
(x2 + y2 ) 2R x2 + y2 sin
+ R2 + z 2
=  (10)
x 2 + y2
According to the properties of integral of periodic function, (16)
= x sin y cos
could be converted into:
PB   
cos  =
PB  2 R x2 + y2 sin d
 0
B = IR  3/2
2
(QP  ) (QB)
2 4 
0
= 2 2 (x2 + y2 ) 2R x2 + y2 sin + R2 + z 2
(QP) (QB)
 (11) (17)
2 2 2
(x R cos ) + (y R sin ) (x sin y cos )
= 2 2 2
(x R cos ) + (y R sin ) + z2 (x sin y cos )
R (x cos + y sin ) Based on (17), two signicant observations can be obtained:
= 
(x cos + y sin R2 )
(1) MFD in vertical direction is related to R, Z (the distance from
PB the point to the torus), and x2 + y2 (the distance of point p to
sin =
PQ circular center).
 (12) (2) MFD in vertical direction on the Z-axis is symmetry. With I, R
(x cos + y sin R2 ) + z 2
=  and Z are conrmed, the antenna radiation patterns is a circle
(x R cos )2 + (y R sin )2 + z 2 around the Z-axis.
Thus, the circular ring could be solved by:
3.4. Analysis of the circular contours stability
 2
0 I R cos  sin d
B = Let x = 0, different patterns of antenna radiation in YOZ plane are
4 r2
0 2 (13) as shown in Fig. 4. Using (17), when the coil is placed horizontally,
0 I R (R (x cos + y sin )) d its coverage radius Y is stable in the plane which is at a distance Z
=
4 3/2 from the coil.
0 ((x R cos )2 + (y R sin )2 + z 2 )
According to (17), let x = y = 0, the MFD at center of the circle is:

3.3. Analysis of circular magnetic eld

 2
0 I R2 d
B = 
4 3/2
For simplicity without losing generality, Auxiliary Trigonomet- 0 (R2 cos2 + R2 sin2 + z 2 )
 2
ric Function is used in the analysis [32]. Assume that: 0 I R2 d
 =  (18)
4 3
x cos + y sin = x2 + y2 sin( + ) (14) 0 (R2 + z 2 )
0 I R2
= 
According to (14), Eq. (13) could be converted into: 2 (R2 + z 2 )
3

 2
0 I R (R (x cos + y sin )) d Considering the impact on the turn number of the coil, according
B =
4  2 2
3/2 to H = BN, we can calculate the MFD from the center of the coil by:
(x R cos ) + (y R sin ) + z2
 
0

 (15)  I N R2
0
2 R x2 + y2 sin( + ) d H= 0 (19)
=
4
IR   3/2 2 (R2 + z 2 )
3
0
x 2 + y2 2R x2 + y2 sin( + ) + R2 + z 2
The range of magnetic eld strength of the reader-writer should
be 0.15 A/m < H < 5 A/m. The average of H is:
where is a constant, which is independent with . On the basis
of the integral invariant, let = ( + ), (15) can be transformed Hmax + Hmin
Havg = (20)
to: 2
 S. Lu et al. / Journal of Manufacturing Systems 44 (2017) 179190 183

Fig. 5. Analysis results.

Fig. 6. Deployment of tags under different approaches.

From (19), the turn number of the coil could be worked out by:

2
3
2H R2 + zmin
N=
0 I R2
Assume that a minimum clearance of z = 0.1 m between an AGV
and the ground, the coil radius R = 0.30 m and the current I = 1 A,
Fig. 5(a) shows the MFD inuenced by Z based on (21). It could be
observed that when the radius is shorter, the MFD varies signi-
cantly with the increasing Z. As the radius is longer, the MFD varies
with the decreasing of Z.
Fig. 5(b) shows the MFD inuenced by radius R = 0.1 m, 0.2 m,
0.3 m, 0.4 m and 0.5 m. When R is 0.1 m, eld strength reaches its
peak in R < 0.15 m. As the radius R increasing, the eld strength
decreases in R > 0.6 m. As the distance increasing, the peak of eld
strength moves rightward.
4. Design and development of RFID-enabled positioning
system
(21) 4.1. Selection of regular polygon
The sum of inside corners of regular N-gon is (n 2) 180 . A
N-gon which can be closely aligned follows the condition:
(n 2) 180
m = 360 (22)
n
From (22), there are three situations to assure that m and n to be
positive integers: when m = 3, n = 6, it is a regular hexagon. When
m = 4, n = 4, it is a square. When m = 6, n = 3, it is a regular trian-
gle. One regular hexagon consists of six regular triangles, each of
which has three vertices. In the splicing process, six triangles share
one vertex. That means each one occupies 0.5 vertex. If the radius
184 S. Lu et al. / Journal of Manufacturing Systems 44 (2017) 179190

of circumcircle is R, the triangle is: (3 3R2 )/2. The density of the
triangle array is:

3 3R2 3
1 = 0.5/ = (23)
2 9R2
For square case, each square has four vertices and four square
share one vertex. So, each square occupies one vertex. If the radius
of circumcircle is R, the square is: 2R2 . We can calculate the square
node density by:
1
2 = 1/(2R2 ) = (24)
2R2

3
Since < 12 , the node density of square array is higher. Thus,
9R2 2R
the triangular array is chosen for deploying the tags.

4.2. Distances among tags

After selection of triangular array, Fig. 6 shows a honeycomb Fig. 7. Impact analysis of radius on accuracy.
presentation as the tags position change. Several situations are
examined: (1) A blind spot exists in the center (Fig. 6(b)). No tags
2  
could be detected in that area. (2) Only one tag could be detected 3  2 2
x+  + y+ = ( + ) (26)
(Fig. 6(c)). (3) Only two tags can be detected (Fig. 6(d)). (4) Three 2 2
tags can be detected (Fig. 6(e)). (5) More than 3 tags can be detected
2  
(Fig. 6(f)). Assume the coverage radius is  and the triangle edge 3  2 2
x  + y+ = ( + ) (27)
length is s, it could be observed that: 2 2

From (25) and (26), the intersection coordinates between circle

(1) When s > 2, there are some blind spots.
A and circle B could be obtained as follows:
(2) When s = 2, only
two tags can be detected.
(3) When 2 > 3 1 2
s > 3, two tags can be detected. x1 =  3 + 24 + 122
(4) When s = 3, three tags can be detected and precise position- 4 4

ing could
be achieved. 3 1 2 (28)
(5) When 3 > s > , one tag can be detected. x2 = + 3 + 24 + 122
4 4
(6) When s = , only one tag can be detected.
(7) When s < , more than 3 tags can be detected. 3
y= x
3
Though more than three tags can be detected when s < ,the The coordinate of F is:

1 2
density and cost is too high for practical implementation. s = 3
is selected to achieve a precise positioning. 3
x = + 3 + 24 + 122
4 4
(29)
4.3. Analysis of radius impacts y = 1  3 32 + 24 + 122

4 12
In order to ensure a regular triangle covered by a circular From (26) and (27), the coordinate of G is:
antenna, the circle radius is extended from  to ( + ). Fig. 7 shows
an analysis of the radius impacts on accuracy. Three arcs with cen- x = 0
ter at point A, B, and C within the radius of +  are presented. The 3 2  (30)
shadow is the center of the movement range. A vertex is contributed
y = ( + )
2

4 2
to two circles which are the radius of + .
From Fig. 7, the circles with the center point in A, B and C could From (25) and (27), the coordinate of H is:

be formulated as:
3 1 2
2
x =  3 + 24 + 122
x2 + (y )2 = ( + ) (25) 4 4
(31)

y = 1  3 32 + 24 + 122

4 12
The distances among F, G, and H could be worked out:

2
 2

3 1 2 1 3 3
DGF :  3 + 24 + 122 + 2 + 8 + 42  + 32 + 24 + 122
4 4 2 4 12 (32)

1 3
= 32 + 24 + 122 
2 2

2
2

3 1 2 1 2 3 3
DHF :  3 + 24 + 122 +  + 8 + 4 + 
2 3 + 24 + 12
2 2
4 4 2 4 12 (33)

1 2 3
= 3 + 24 + 122 
2 2
 S. Lu et al. / Journal of Manufacturing Systems 44 (2017) 179190 185

Fig. 8. Analysis of  and .

Fig. 9. Tags placement impacts on accuracy.

1 2 3 its deviation R is:
DHF : 3 + 24 + 122  (34)
2 2 
 x + x + x 2  y + y + y 2
1 2 3 1 2 3
R = +
1 2 3 3 3
GF = HG = HF = 3 + 24 + 122  (35) 
2 2 1 2 2
= (x1 + x2 + x3 ) + (y1 + y2 + y3 )
3
Let  = 0.110 and = 0.00010.01, their relations are shown in    
1
Fig. 8. When = 0, the distance is zero. That means the deviation of = xi2 + yi2 + 2 xi xj + 2 yi yj
3
the circle center is minimized. When < 0.003 m, the deviation is
less than 0.01 m. If < 0.001 m, the deviation can be controlled less
1i3

1i3

1i,j3

1i,j3

1
than 0.004 m. = (xi2 + yi2 ) + 2 xi xj + 2 yi yj
3
1i3 1i,j3 1i,j3
  
1
32 + 2 xi xj + 2 yi yj
4.4. Impacts of tags placement 3
1i,j3 1i,j3
1 2
Tags are important factors to realize accurate positioning. There- 3 + 62
3
fore, in an AGV system, it is critical to design the tags placement
to meet different positioning requirements. In the same MFD, it is
found, in previous section, that the inter coil can have more elec- (37)
tromagnetic induction than the outer coil. The magnetic induction
coil could be presented as:
From (37), the maximum deviation of the gravity center is deter-
mined by the vertex. It could be concluded that the deviation is not
k0 s N 2 S inuenced by the coil radius but relies on the tags deviation. If the
L= (36)
l tags are installed within 0.5 mm, the positioning accuracy will be
controlled within 0.5 mm.
where 0 is the vacuum permeability. s is the relative permeabil- In order to design a exible positioning system for AGVs
ity of the magnetic core (s = 1 when coil is hollow). N2 is the square automatic material delivery within smart factories, tag array is
of the number of coils turns. S is the cross-sectional area of the coil. then deployed on the ground. Antennas for readers are installed
l is the length of the coil. k is a factor determining by the radius and under the AGVs One driving antenna and 4 parking antennas are
length of the coil. used. Large-diameter antenna is chosen for the driving antenna
Cartesian coordinate system is used for examining the tags and small-diameter antenna is used for parking antenna. Large-
placement on positioning accuracy due to its ability that species diameter antenna can guide AVGs through the sparse arrayed area.
each point uniquely in a plane by a pair of numerical coordinates. When reaching the stopping zones, small-diameter antenna are
Let O indicates the center of a circle that is the coordinate origin. used for precise positioning. Thus, tags are placed in two fash-
A(x1 , y1 ), B(x2 , y2 ), C(x3 , y3 ) are the vertices of ABC as shown in ions. Driving zone is arrayed by sparse fashion and stopping zone
Fig. 9. is arrayed by serried fashion.
Suppose that the vertices ofABC move (x  1 , y1 ), (x2 ,
y2 ), (x3 , y3 ) to A B C , x12 + y12 , x22 + y22 and 4.5. Design of antenna and tags

x32 + y32
are smaller than . That means the range of deviation
When designing the antenna and tags for smart factories, the
of each vertices is determined. Their coordinates are A (x1 + x1 ,
AGV physical structure, working background and the stability of the
y1 + y1 ), B (x2 + x2 , y2 + y 
2 ) and C (x3 + x3 , y3 + y3). The
x1 +x2 +x3 y1 +y2 +y3
magnetic eld should be considered. The antenna diameter needs
center of gravity G is G 3
, 3
and to be as large as possible to reduce the quantity of tags on one hand.
186 S. Lu et al. / Journal of Manufacturing Systems 44 (2017) 179190

evenly spaced triangles. The radius of the large antenna decides

their spacing. Large antenna radius is R = 0.5 m and the coverage
radius is 0.45 m. The side length of triangles circumcircle is 0.779 m.
To ensure the coverage area, 0.75 m is selected. The deployment of
tags in serried zone needs to meet the requirement of the sub-
antennas coverage. Assume the radius of the antenna is 0.1 m and
the coverage radius is 0.112 m, then the coverage radius is set as
0.110 m to meet < 0.003 m.

5. System simulation and implementation

This section reports on the system simulation and implemen-

tation studies for evaluating the proposed approach. Simulation
study is carried out in Matlab software for a virtual smart factory
environment and system implementation is based on a laboratory
testbed that uses a smart manufacturing shop oor for example to
test the feasibility and practicality of the proposed RFID-enabled
positioning approach.

5.1. System initiative

Fig. 10. Antenna coverage analysis.

The system initiative aims to build up a virtual smart environ-

On the other hand, the diameter should be smaller than the width
ment where AGVs could be simulated and examined. It includes
of an AGV. According to the previous analysis, a blind area will be
the tags array for driving, detecting status and recording the actual
created when the diameter is too large. Thus, the radius is set as
track. The setting of AGV system includes two steps so as to create a
0.5 m.
RFID-enabled smart environment where readers can sense the tags
According to the analysis, smaller diameter antennas are more
automatically. (1) Installation of tags array: set the tags in regular
sensitive to the change of magnetic eld thus it could achieve bet-
triangles then initial tags coordinate in matrix Amx and Amy . (2)
ter performance in precise positioning. But if the diameter is too
Simulation initiative: precongure the simulation parameters such
small, the contours zone will be decreased unstably. Therefore, the
as the driving direction, step length, starting and ending point, etc.
diameter between 0.1 m and 0.3 m is suitable and this paper uses
The simulation parameters are set as follows: the coverage antenna
0.1 m for our research.
radius is 0.45 m, the circumcircle radius of triangle tag is 0.44 m, the
Considering an AVG may stop at the edges, four stopping anten-
AGVs speed is 0.2 m/s, and the step length is 0.05s. The start point
nas are divided into two groups and placed on both sides of the
is (0.2, 0.2). The transit point is Tag No.8822, and the end is Tag
centerline. Fig. 10 shows a triangle tag array in uniform distribu-
No.8967.
tion. The red lled circle is a certain antenna. Assume that the
length of triangle edge is s and its radius of circumscribed circle
5.2. Simulation study
is , the distances from gravity to the vertices and edges are: 
and 0.5 respectively. In Fig. 10, the blue, green and yellow dash
Two simulation studies are carried out in this research. They are
circles present the arrangements of antennas parallel. The distance
AGV testing without and with disturbances.
between two antennas is an even number so that the triangle can be
precisely cover by two antennas. The distance between two circular
5.2.1. Simulation 1: AGV testing without disturbance
centers is:
Assuming that an AGV can follow the track without any random
n 3n disturbances. Fig. 11 presents the simulation results showing the
s= , (n = 2, 4, 6, . . .) (38)
2 2 AGV trajectory and its direction changes. From Fig. 11, it is observed
Since the distance between two closed triangles gravity center there are two direction changes. One happens because the AGV
is , then the distance between two triangles gravity centers along is out of the boundary as shown in (b). The other occurs in the
the direction of the opposite vertical angle is 2. So the distance of transfer point as shown in (c). In the direction changes, the slopes
triangles gravity center in the same direction is: are adjusted from 3.7239 to 3.7943 and 3.7943 to 0.0284 respec-
tively. With the changes, the AGV approaches to the predened
3 destination precisely.
(n 1), (n = 1, 3, 5. . .) (39)
2
The distance of triangles gravity center in opposite direction is: 5.2.2. Simulation 2: AGV testing with disturbances
3  An AGV will be disturbed in real-life smart factory due to the

2 n 1 , edge frequent movement of manufacturing machines, materials, and

  , (n = 2, 4, 6, . . .) (40) workers. In this simulation, random disturbances with 00.01 m

3 n 2 , vertex deviations are used for testing the positioning approach. The sim-
2 ulation results are shown in Fig. 12.
Assume that the geometric center of an AGV is the origin coordi- From Fig. 12, it is observed that the AGV can still reach the pre-
nate, the coordinates of driving antenna and the stopping antenna dened destination after adding some random disturbances. From
are set as: the experiment, the slope is adjusted three times: (1) from 3.7239
5   17
 5 9
 to 3.7943; (2) from 3.7943 to 0.0190; (3) from 0.0190 to 0.0287.
A s, 5 , B 4s,  , C(4s, 9), D s,  That means the AGV adjusts its directions 3 times two of which is
2 2 2 2
caused by running out of boundary.
Since an AGV follows the planned route in driving zone without From the above two simulation studies, when the precise posi-
serried tags and precise positioning, serried tags are deployed in tioning antenna covers the targeted tag, 1 or 2 tags will be detected
 S. Lu et al. / Journal of Manufacturing Systems 44 (2017) 179190 187

Fig. 11. Simulation results without disturbance.

Fig. 12. VGA trajectory with disturbances.

Table 1
Comparing results.

Actual Planned Deviations

x1 0.0845 0.0857 0.0012

y1 0.0490 0.0495 0.0005
x2 3.7524 3.7495 0.0029
y2 8.3655 8.3555 0.01
x3 14.3356 14.3455 0.0099
y3 47.6522 47.6625 0.0103
x4 25.6448 25.6505 0.0057
y4 58.7555 58.6075 0.148
x5 46.4564 46.3595 0.0969
y5 59.6878 59.7055 0.0177

Fig. 13. VGA movement.
by the AGV. Coordinates of three targeted tag group are: A(0,
0), B(0.0857, 0.1485), and C(0.1715, 0). The circumcircle radius is
0.099 m and the initial coordinate of the antenna is (0.01, 0.05).
The coverage radius is 0.1 m.
Fig. 13 shows the VGA movement through vertical view by
enlarging the changes at each time in simulation 2. The changes
are enabled by three stages: (a) presents the rst stage with one
time slope adjustment; (b) is the second stage without slope adjust-
ment; and (c) is the third stage with four times slop adjustment.
Through the three stages, it could be observed that the VGA is able
to approach the dened destination (Tag No.8967).
Table 1 shows a result by comparing the actual and planned
coordinates from simulation 2. Five groups of coordinates are ran-
domly selected for the comparison analysis. From Table 1, it could
be observed that the deviations in vertical axis (Y) are larger than
that from horizontal axis (X). The average deviations from both axes
are 0.0373 m and 0.02332 m. The biggest deviation is y4 where the
AGV just passed the turning point. It could be found that the accu-
racy of positioning is lower at turning points comparing with other
points on the running track. That because the signal density will be
decreased at the turning points when the readers antenna sensing
the tags. In order to improve the accuracy, more dense tags array
could be placed at these points. But the cost will be a little bit higher.
188 S. Lu et al. / Journal of Manufacturing Systems 44 (2017) 179190

Fig. 14. A testbed implementation in laboratory.

5.3. Laboratory testbed Table 2

Results from testbed.

Fig. 14 presents a smart factory testbed for evaluating the fea- Pre-C Actual-R
sibility and practicality of the proposed positioning approach. It
S-1 S-2
covers an area about 20 m2 where typical manufacturing objects
are equipped with RFID devices so that they are converted into No. A No. A No. B No. C

smart manufacturing objects which can sense and interact with (0, 0.8) (0.02, 0.81) (0.05, 0.73) (0.06, 0.82) (0.05, 0.88)
each other. Three AGVs are labeled as No. A, No. B, and No. C for the (0, 0.2) (0.02, 0.19) (0.06, 0.15) (0.05, 0.26) (0.06, 0.26)
(1.0, 0) (1.0, 0.02) (1.25, 0.36) (1.36, 0.42) (1.35, 0.45)
testing. First of all, RFID readers are installed at the bottom of AGVs.
(3.0, 0) (2.90, 0.02) (2.88, 0.32) (3.16, 0.35) (3.10, 0.25)
Secondly, tags are placed on the ground with the distance of 0.75 m (5.0, 0) (4.90, 0.01) (5.25, 0.26) (5.04, 0.24) (5.10, 0.25)
among each other. Third, two testing scenarios are designed: sin-
gle AGV (S-1) and multiple AGV (S-2) scenarios. For each scenario, 5
sets of coordinates are compared between the pre-congured posi- values 0.32 and 0.41 in x and y-axis happens when the AGVs passing
tions and actual running positions (Actual-R). The pre-congured a quarter bend. Thus, a control algorithm with parallel processing
(Pre-C) positions are (0, 0.8), (0, 0.2), (1.0, 0), (3.0, 0), and (5.0, ability would be worked out to improve the accuracy.
0). Table 2 presents the validation results from these two scenarios.
From Table 2, only one AGV is tested for S-1. Comparing the Pre- 5.4. Comparison study
C and Actual-R coordinates, it is found that the average deviations
of x and y-axis are 0.048 and 0.014 respectively. For S-2, three AGVs This approach was compared with a published paper using
are used for the testing. The average deviations are (0.146, 0.212), sparsely distributed passive RFID tags for efcient object localiza-
(0.134, 0.218), and (0.132, 0.218) respectively. It is found that the tion [33]. In their paper the moving step is 10 cm and they use grid
deviations in S-2 are much bigger than that in S-1. The main reason pattern for the testing. Table 3 presents the comparison study on
is that the coordination of multiple AGVs in the working area will 10 m2 , 15 m2 and 20 m2 scenarios. We compare the RFID trajectory
reduce the accuracy due to the limited calculation capacity. Addi- along X, Y, and both X and Y axis which are followed the experiments
tionally, it is also observed that the greatest deviation with average in the published paper.

Table 3
Comparisons (Unit: m, I-our approach; II-approach proposed by [33]).

Scenario Number of tags X axis Y axis X and Y

I II I II I II I II

10 m 2 100 21 0.8 1.0 0.5 0.6 (0.8, 0.5) (0.8, 0.6)

1.9 2.0 1.1 1.0 (0.8, 1.4) (0.8, 1.6)
2.8 3.0 1.7 1.8 (1.2, 1.4) (1.2, 1.6)
15 m2 150 28 1.8 2.0 0.8 1.0 (1.5, 1.2) (1.6, 1.4)
2.8 3.0 1.4 0.6 (2.5, 2.0) (2.6, 2.0)
3.8 4.0 2.4 2.5 (3.8, 2.0) (3.8, 2.0)
20 m2 2000 42 2.8 3.0 0.5 0.6 (2.5, 1.0) (2.5, 1.0)
4.9 5.0 1.2 1.0 (3.9, 1.0) (4.0, 1.0)
7.8 8.0 2.5 2.6 (6.5, 1.6) (6.5, 1.8)
 S. Lu et al. / Journal of Manufacturing Systems 44 (2017) 179190 189

From Table 3, it could be observed that the approach proposed in
this paper uses much less RFID tags for achieving the similar posi-
tioning accuracy. The price of tags in this experiment is $0.1USD.
The cost is much higher from II because the tags distance is 0.1 m
and the readers covering range is 0.24 m2 . While in our proposed
approach, the tags distance is 0.75 m after analysing the features
of circular magnetic eld which is the optimal value in this case.
The readers coverage is 0.56 m2 . While the AGV moving alone X
and Y axis the differences are around 0.10.2 m. The positioning
differences are not obvious in these cases. When it is moving along
both X and Y axis, the trajectories from I and II is measured in pre-
cision regarding three scenarios. The reduced tags are superior to
the approach II with achieving the similar positioning accuracy.
6. Conclusions
This paper introduces a RFID-enabled positioning approach in an
Automated Guided Vehicle (AGV) system using near-led coupling
mechanism for smart factories where automatic and smart logis-
tics could be achieved. Key impact factors on positioning accuracy
such as magnetic eld in circular antenna, circular magnetic eld,
and circular contours stability are examined in both theoretical and
practical perspectives. The proposed approach is then veried in
simulation studies and a smart factory testbed.
Several contributions are signicant from this paper. Firstly, dif-
ferent diameter antennas are tested in the sparse and intensive tags
area. Sparse tags layout is tested for large antennas in the driving
zone. Intensive tags layout is tested for small antennas in the park-
ing zone. The key ndings are signicant for smart factory where
automatic logistics with AGVs could be ultimately achieved. Sec-
ondly, large diameter antennas are used in driving zone and small
diameter antennas are used in parking zone. Two parking anten-
nas group can adjust vehicle directions. Finally, observations and
lessons from simulation, testbed, and comparison studies could be
used for guiding automatic logistics within a smart manufacturing
shop oor where RFID devices could be deployed systematically.
Some limitations of this paper will be addressed in our future
research. First, the driven speed, the size of AGV, the environment,
the computational capability, and a lot of other factors may inu-
ence the positioning accuracy of the RFID system. In the future,
more impact factors will be studied. Second, the system will be
tested in industrial level for example a real factory where CPS
devices will be deployed to create a smart factory. Insights and
lessons from the smart factory implementations will be explored to
a case study from which practitioners may get some useful informa-
tion when they contemplate an AGV-based solution for automatic
logistics.
Acknowledgment
This work was supported by National Natural Science Foun-
dation of China (Grant no. 61472257, 51405307, and 61401283),
Guangdong Technology Plan (Grant no. 2012B010600016),
Guangdong Technology Service Development Plan (Grant no.
2013B040403005), PhD Supervisor Project from Department
of Education China (20134408110001), Shenzhen Governmen-
tal Project (GJHS20140418183120414) and Guangdong High
Education Institution project (2013CXZDC008).
Appendix

In order to solve the (17), Mathematica is used for getting the

result as follows:

If r 2 + y2 + z 2 =
/ 2ry&&r 2 + 2ry + y2 + z 2 =
/ 0&&
   
r 2 + y2 + z 2 r 2 + y2 + z 2
Re 2||2 + Re
0||
ry ry
   
2 2 2 2 2 2
Im r 2ry + y + z =/ 0&&Im r + 2ry + y + z =/ 0&& ,


ry   ry
 
2 2
r +y +z 2 2 2
r +y +z 2
Re = 0||Im =
/ 0
ry ry
1
 
r r2 + y2 + z2 r 2 + 2ry + y2 + z 2
 4ry
 1

(3r 2 y2 z 2 )EllipticE , + 
4 r 2 2ry + y2 + z 2
r 2 + y2 + z 2

 r 2 + 2ry + y2 + z 2

  
r 2 + y2 + z 2 r 2 + y2 + z 2 4ry r 2 + y2 + z 2
4r 2 EllipticE , + ,
r 2 2ry + y2 + z 2 r 2 2ry + y2 + z 2 4 r 2 + 2ry + y2 + z 2 r 2 + 2ry + y2 + z 2
 3 
4ry
(r 2 + y2 + z 2 )EllipticE , 2 + (r 2 + 2ry + y2 + z 2 )

  4 r  2ry + y2 +z 2 
 4ry 3 4ry
EllipticF , 2 + EllipticF , 2
4 r 2ry + y2 + z 2 4 r 2ry + y2 + z 2

r y sin x
Integrate[ , {x, 0, 2},
3/2 3/2
(r 2 + y2 + z 2 2ry sin x) (r 2 + y2 + z 2 2ry sin x)
Assumptions r 2 + y2 + z 2 = 2ry||r 2 + 2ry + y2 + z 2 = 0||
 
r 2 2ry + y2 + z 2
Im = 0||
ry
     
2 2 2 2 2 2 2 2 2
Re r + y + z < 2&&2 + Re r + y + z > 0&& Im r + 2ry + y + z = 0||
ry ry ry

    
2
r + y2 + z 2 r 2 + y2 + z 2
Im = 0&&Re =
/ 0
ry ry
190 S. Lu et al. / Journal of Manufacturing Systems 44 (2017) 179190
References
[1] Ramesh R, Jyothirmai S, Lavanya K. Intelligent automation of design and
manufacturing in machine tools using an open architecture motion controller.
J Manuf Syst 2013;32(1):24859.
[2] Ivanov D, Dolgui A, Sokolov B, et al. A dynamic model and an algorithm for
short-term supply chain scheduling in the smart factory industry 4. 0. Int J
Prod Res 2016;2:386402.
[3] Lee J, Bagheri B, Kao H-A. A cyber-physical systems architecture for industry
4.0-based manufacturing systems. Manuf Lett 2015;3:1823.
[4] Um I, Cheon H, Lee H. The simulation design and analysis of a exible
manufacturing system with automated guided vehicle system. J Manuf Syst
2009;28(4):11522.
[5] Yu W, Egbelu PJ. Design of a variable path tandem layout for automated
guided vehicle systems. J Manuf Syst 2001;20(5):30519.
[6] Xing W, Peihuang L, Jun Y, et al. Intersection recognition and guide-path
selection for a vision-based AGV in a bidirectional ow network. Int J Adv
Robot Syst 2014;11(3):39.
[7] Cho JH, Cho M-W. Effective position tracking using B-spline surface equation
based on wireless sensor networks and passive UHF-RFID. IEEE Trans Instrum
Meas 2013;62(9):245664.
[8] Saab SS, Nakad ZS. A standalone RFID indoor positioning system using passive
tags. IEEE Trans Ind Electron 2011;58(5):196170.
[9] Wang LH, Trngren M, Onori M. Current status and advancement of
cyber-physical systems in manufacturing. J Manuf Syst 2015;37(Part
2):51727.
[10] Tan Y, Goddard S, Perez LC. A prototype architecture for cyber-physical
systems. ACM Sigbed Rev 2008;5(1):26.
[11] Ali S, Qaisar SB, Saeed H, et al. Network challenges for cyber physical systems
with tiny wireless devices: a case study on reliable pipeline condition
monitoring. Sensors 2015;15(4):7172205.
[12] Ngai E, Chau D, Poon J, et al. Implementing an RFID-based manufacturing
process management system: lessons learned and success factors. J Eng
Technol Manag 2011;29(1):11230.
[13] Mejjaouli S, Babiceanu RF. RFID-wireless sensor networks integration:
decision models and optimization of logistics systems operations. J Manuf
Syst 2015;35:23445.
[14] Zhong RY, Dai Q, Qu T, et al. RFID-enabled real-time manufacturing execution
system for mass-customization production. Robot Comput-Integr Manuf
2013;29(2):28392.
[15] Zhong RY, Huang GQ, Dai QY, et al. Mining SOTs and dispatching rules from
RFID-enabled real-time shopoor production data. J Intell Manuf
2014;25(4):82543.
[16] Zhong RY, Huang GQ, Lan SL, et al. A big data approach for logistics trajectory
discovery from RFID-enabled production data. Int J Prod Econ
2015;165:26072.
[17] Hasani M, Talvitie J, Sydnheimo L, et al. Hybrid WLAN-RFID indoor
localization solution utilizing textile tag IEEE Antennas. Wirel Propag Lett
2015;14:135861.
[18] Ahmad MY, Mohan AS. Novel bridge-loop reader for positioning with HF RFID
under sparse tag grid. IEEE Trans Ind Electron 2014;61(1):55566.
[19] DiGiampaolo E, Martinelli F. A passive UHF-RFID system for the localization of
an indoor autonomous vehicle. IEEE Trans Ind Electron 2012;59(10):396170.
[20] Porter JD, Billo RE, Mickle MH. A standard test protocol for evaluation of radio
frequency identication systems for supply chain applications. J Manuf Syst
2004;23(1):4655.
[21] Villagra J, Herrero-Prez D. A comparison of control techniques for robust
docking maneuvers of an AGV. IEEE Trans Control Syst Technol
2012;20(4):111623.
[22] Herrero D, Villagr J, Martnez H. Self-conguration of waypoints for docking
maneuvers of exible automated guided vehicles. IEEE Trans Autom Sci Eng
2013;10(2):4705.
[23] Lee J, Nam S. Fundamental aspects of near-eld coupling small antennas for
wireless power transfer. IEEE Trans Antennas Propag 2010;58(11):34429.
[24] Rodrguez-Fortuno FJ, Marino G, Ginzburg P, et al. Near-eld interference for
the unidirectional excitation of electromagnetic guided modes. Science
2013;340(6130):32830.
[25] Maina JY, Mickle MH, Lovell MR, Schaefer LA. Application of CDMA for
anti-collision and increased read efciency of multiple RFID tags. J Manuf Syst
2007;26(1):3743.
[26] Hsu H-T, Huang T-J. A koch-shaped log-periodic dipole array (LPDA) antenna
for universal ultra-high-frequency (UHF) radio frequency identication (RFID)
handheld reader. IEEE Trans Antennas Propag 2013;61(9):48526.
[27] Roy Chowdhury D, Xu N, Zhang W, Singh R. Resonance tuning due to Coulomb
interaction in strong near-eld coupled metamaterials. J Appl Phys
2015;118(2):023104.
[28] -l. Wan M, He J-N, -l. Song Y, Zhou F-Q. Electromagnetically induced
transparency and absorption in plasmonic metasurfaces based on near-eld
coupling. Phys Lett A 2015;379(30):17915.
[29] Nguyen MQ, Hughes Z, Woods P, et al. Field distribution models of spiral coil
for misalignment analysis in wireless power transfer systems. IEEE Trans
Microw Theory Tech 2014;62(4):92030.
[30] Niehrs C. On growth and form: a Cartesian coordinate system of Wnt and BMP
signaling species bilaterian body axes. Development 2010;137(6):84557.
[31] Majidi S. A comparison between the knowledge organization of university
physics teachers and the textbooks they use for their teaching purposes:
Biot-Savart law and Ampres law. Int J Sci Math Educ 2014;12(6):1281314.
[32] nsal , Ma W-X. Linear superposition principle of hyperbolic and
trigonometric function solutions to generalized bilinear equations. Comput
Math Appl 2016;71(6):12427.
[33] Yang P, Wu W, Moniri M, Chibelushi CC. Efcient object localization using
sparsely distributed passive RFID tags IEEE. Trans Ind Electron
2013;60(12):591424.