ENABLING WIRELESS COMMUNICATIONS AND NETWORKING

TECHNOLOGIES FOR EDGE COMPUTING

Edge Computing in IoT-Based Manufacturing

Baotong Chen, Jiafu Wan, Antonio Celesti, Di Li, Haider Abbas, and Qin Zhang

Abstract for Internet of Things (IoT) data analytics. Its use With the focus on the
case can be found in intelligent manufacturing. concept of edge com-
Edge computing extends the capabilities of Condry et al. [5] introduced a model using smart puting, the authors
computation, network connection, and storage edge IoT devices for safer, rapid response with
from the cloud to the edge of the network. It industry IoT control operations. As IoT appli- propose an architecture
enables the application of business logic between cation to the industrial domain is spreading, of edge computing for
the downstream data of the cloud service and the Georgakopoulos et al. [6] proposed a roadmap IoT-based manufacturing.
upstream data of the Internet of Things (IoT). In combining cloud edge computing for IoT-based They analyze the role of
the field of Industrial IoT, edge computing pro- manufacturing. Suganuma et al. [7] proposed edge computing from
vides added benefits of agility, real-time process- multiagent-based flexible edge computing archi-
ing, and autonomy to create value for intelligent tecture for a large-scale IoT system. Byers [8] pro- four aspects including
manufacturing. With the focus on the concept of vides a review of the architecture, use case, and edge equipment, net-
edge computing, this article proposes an architec- requirements for fog-enabled IoT networks. Both work communication,
ture of edge computing for IoT-based manufactur- of them are important references for the deploy- information fusion, and
ing. It also analyzes the role of edge computing ment of edge computing in IoT-based manufac- cooperative mechanism
from four aspects including edge equipment, net- turing. Satyanarayanan [9] pointed out that the
work communication, information fusion, and emergence of edge computing has enabled the with cloud computing.
cooperative mechanism with cloud computing. cloud services for mobile computing, scalability, Finally, they give a case
Finally, we give a case study to implement the and privacy policy. However, the reference archi- study to implement active
active maintenance based on a prototype plat- tecture for IoT-based manufacturing has not yet maintenance based on a
form. This article aims to provide a technical ref- been proposed. Edge computing and its role in prototype platform.
erence for the deployment of edge computing in the Industrial Internet of Things (IIoT) application
the smart factory. should be established.
Industrial applications in the domain of IoT
Introduction require characteristics such as location aware-
In the context of intelligent manufacturing, the ness and low latency. Edge computing, which is
proliferation of terminal network devices has given implemented at the peripheral devices on an IoT
rise to new challenges for operation and mainte- network, has the ability to support these require-
nance, scalability, and reliability of the data cen- ments of mobility and geographic independence.
ters. The growth of edge computing has moved Computing’s physical proximity to a mobile
the computing from centralized data centers to device makes it easier to achieve low end-to-end
the periphery of the network. Edge computing latency, high bandwidth, and low jitter to services
aims to address these challenges by creating an located at the network edge. Edge computing is
open platform with the capability of integrating being introduced into IoT-based manufacturing.
core capabilities such as networking, computing, The scope of edge computing extends the range
storage, and application. It has enabled intelligent of an IoT embedded device with capabilities of
services close to the manufacturing unit to meet limited storage, cache, and processing to the
the key requirements such as agile connection, edge near the terminal infrastructure (e.g., Rasp-
data analytics via edge nodes, highly responsive berry Pi and Arduino board). The contributions of
cloud services, and privacy-policy strategy [1, 2]. this article can be summarized as follows:
Edge computing can make full use of embedded •In view of the typical applications of edge
computing capabilities of field devices to achieve computing in the domains of intelligent manu-
equipment autonomy based on distributed infor- facturing, this article proposes a novel system
mation processing. Furthermore, edge computing architecture from four perspectives: the device
supports the need of digital manufacturing enter- domain, the network domain, the data domain,
prises for rapid configuration of the smart factory, and the application domain.
which must adapt to personal demands of users •With the focus on the development of
and dynamic changes in production conditions. edge-specific field devices, this article discusses
Research on edge computing is progressing how to achieve efficient information interaction,
rapidly. Due to the limitation of the cloud plat- heterogeneous data fusion, and a cooperative
form, Lopez et al. [3] considered that edge com- mechanism with the cloud platform. Some related
puting is needed for moving the data and services use cases are also presented for IoT-based manu-
from centralized nodes to the edge of the net- facturing.
work. Patel et al. [4] outlined the novel approach •Based on the typical characteristics of edge
of using the intelligent edge (e.g., Raspberry Pi) computing in IoT-based manufacturing, this

Baotong Chen, Jiafu Wan, Di Li, and Qin Zhang are with South China University of Technology; Antonio Celesti is with the University of Messina; Digital Object Identifier:
Haider Abbas is with Florida Institute of Technology. 10.1109/MCOM.2018.1701231

IEEE Communications Magazine • September 2018 0163-6804/18/$25.00 © 2018 IEEE 103

Authorized licensed use limited to: Georgia Institute of Technology. Downloaded on February 03,2021 at 23:13:24 UTC from IEEE Xplore. Restrictions apply.
 Due to the lack of the
Edge computing
flexible configurable Device-to-edge
Data account & Digital twins

Data domain
middleware, it is chal- data analysis Sensors-to-edge
D2D data signal
lenging to implement Edge data Distributed
storage intelligence SDN-based interaction
a flexible adjustment OPC UA C/S transfer
to the equipment in Full customization &
integration capabilities
order to cope with the
Smart factory

Network domain
changing manufactur- Route &
Controller data cache OPC UA
ing status. However, client
the popularity of IIoT Network- Gateway
requires a more open management
mode in the industrial
Sensors, actuators, & control

field, especially for

Operation Autonomy Security Perception D2D
safe remote access,
Smart Real-time RFID
periodic maintenance, Sensors monitor RFID reader
node
and industrial big data Operation Local Workpiece Products
analysis. equipment activities process parts
Application domain

Figure 1. Architecture of an edge computing platform in IoT-based manufacturing.

research develops a laboratory prototype plat- •The network domain connects the field equip-
form, and conducts experiments for the active ment to the data platform in a flat manner. In
maintenance of equipment in smart factory. The IoT-based manufacturing, the network domain
experiments demonstrate the advantages of edge uses software defined networking (SDN) [11] to
computing for business agility and bandwidth achieve separation between the network trans-
optimization. mission and the control. Due to the necessary
The rest of the article is organized as follows. time sensitivity of the task data, a time-sensitive
We propose a system architecture for the appli- network (TSN) protocol is applied to process the
cation of edge computing in IoT-based manufac- sequence of network information and to provide
turing. We present the role of edge computing the general standards for maintaining and manag-
in manufacturing scenarios involving devices, net- ing the sensitive time nodes.
works, data, and the manufacturing cloud. We •The data domain in the data origin pro-
present the combination of edge computing with vides services such as data cleaning and feature
the typical industrial applications to achieve active extraction, which improves the availability of the
maintenance of the production line. The final sec- heterogeneous industrial data. This also allows the
tion concludes our work. implementation of pre-defined responses that can
be based on the real-time IoT data. The abstract
System Architecture data at the device terminal end is provided to the
With the goal of proposing a novel application remote service center for the virtualization of the
scenario for edge computing, we present the sys- manufacturing resources.
tem architecture of edge computing in an IoT- •The application domain inherits the applica-
based manufacturing scenario. As shown in Fig. tions of the manufacturing cloud at the network
1, the architecture is divided into four fields: the edge. It integrates the key technologies among
device domain, the network domain, the data the network, data, computing, and control. The
domain, and the application domain. application domain enables edge computing to
•The device domain is either embedded in, or provide general, flexible, and interoperable intel-
located close to, the field devices such as sensors, ligent applications. Components such as service
meters, robots, and machine tools. The device composition, based on the requirements of the
domain should support a flexible communication manufacturing process, allow the dynamic man-
infrastructure and be able to establish standard- agement and optimal scheduling of field equip-
ized communication models to support various ment.
types of communication protocols. Nodes in an The device domain supports real-time inter-
edge computing network must have the ability connections and the deployment of intelligent
to compute and store data to dynamically adjust applications to field devices. It provides an import-
the execution strategy of the industrial equipment ant foundation for the upper layers of the appli-
based on sensor inputs. The information model is cation. The network domain is useful for multiple
built on the edge computing nodes, and includes purposes such as the real-time transmission of
mainstream protocols such as OLE for Process heterogeneous industrial data, the control of com-
Control Unified Architecture (OPC UA) [10] and plex network states, and the convenient access
Data Distributed Service (DDS). It is deployed to to manufacturing resources. It provides a plat-
realize the unified semantics of information inter- form for performing inter-connection, information
action, and ensure data security and privacy. interaction, and data fusion on the edge comput-

104 IEEE Communications Magazine • September 2018

Authorized licensed use limited to: Georgia Institute of Technology. Downloaded on February 03,2021 at 23:13:24 UTC from IEEE Xplore. Restrictions apply.
ing nodes. The data domain also provides data
Wireless
optimization services. One of its main tasks is to access point (eNB)
ensure consistency and integrity of the data. The
application domain provides intelligent applica- D2D
tion services for edge computing, which allows communication
Edge router
the independent implementation of local business OPC UA
logic at the peripheral devices. The domain also communication
provides open interfaces for the device domain
and the network domain to realize edge industry
Edge gateway
application. The proposed edge computing archi-
tecture, presented in Fig. 1, makes full use of the
D2D transfer signal
embedded computing capabilities and ensures Switchboard
OPC-based transfer
the autonomy of the system and the manufac- Web server
turing equipment while following a distributed TCP/IP networks
computing paradigm. Furthermore, the edge com- Data Service server
Request signal
puting framework cooperates with the remote server
data center to realize intelligent terminals for the
manufacturing system. Figure 2. Efficient information interaction in IoT-based manufacturing.

The Roles of Edge Computing in Application

IoT-Based Manufacturing Process visibility Fault prediction Digital twins
In view of the dimensions of the manufacturing
system, edge computing can be located either
at the equipment level, at the control level, or at
Data fusion
the workshop level. It may support individual stag-
es such as system integration, interconnection, Multi-device data
Deployment of algorithms
for MA or AI
Edge support

information fusion, or the entire life cycle of the acquisition

manufacturing process. The deployment of edge
computing transforms the system to allow flexibil-
Semantic interoperability Fault detection standard
ity in adapting to changes in the resources of the
production line, which can in turn promote the
development for IoT-based manufacturing.
Definition of basic functionality of edge computing nodes
Edge-Specific Device Supporting
Network Middleware Field device Edge nodes
The whole IoT-based manufacturing architecture Raspberry Arduino
Actuator Sensor DCS
is like a pyramid that can be simply divided into pie board board
the device layer, the control layer, and the net-
work layer. There is a strong interaction between Figure 3. Architecture of an edge-enhanced data fusion system.
the underlying manufacturing equipment and the
affiliated services. Due to the lack of flexible con-
figurable middleware, it is challenging to imple- business logic analysis and autonomous comput-
ment flexible adjustment to the equipment in ing, but also to give them the capability to opti-
order to cope with the changing manufacturing mize and adjust their execution strategies in real
status. However, the popularity of IIoT requires a time. Edge computing nodes also make it pos-
more open mode in the industrial field, especially sible to implement active maintenance at the
for safe remote access, periodic maintenance, edge of IIoT. It enables the acquisition of real-
and industrial big data analysis. The emergence time and accurate state of the equipment. In the
of edge computing attempts to deal with this same way, some of the diagnostic tasks run on
challenge by transforming the computing model the manufacturing cloud can be offloaded to the
of intelligent manufacturing from a centralized edge computing nodes, which decreases the time
control model to a distributed processing model. delay needed in performing these diagnoses, and
Additionally, the edge computing nodes support enables proactive maintenance. This decreases
the plug and play of new equipment and rapid the maintenance time and reduces the application
replacement in case of failure. One challenge in monitoring pressure of manufacturing cloud. In
the IIoT framework is the ubiquitous compatibility general, edge computing nodes are integrated
problem due to the existence of a diverse number into IIoT, making the embedded control more
of protocols. Therefore, to deal with this diversity, powerful and improving the scalability of the net-
edge computing nodes provide multiple access work.
modes within their modular network interfaces.
Furthermore, the format of transmission payload Edge-Enhanced Information Interaction
is modified to adapt to different communica- IIoT is deployed to achieve seamless commu-
tion protocols and to facilitate easy information nication between manufacturing resources and
exchange between the equipment. It also allows different control systems. Edge computing in
the flexible adjustment of production plans, and IIoT systems provides important support for the
supports rapid deployments of new processes. applications at the network edge. As shown in
Edge computing improves the terminal intel- Fig. 2, the OPC UA provides protocols and ser-
ligence of manufacturing equipment. The goal vices [10], which helps to realize an integrated
is to not only allow the IIoT devices to perform information model independent of the network

IEEE Communications Magazine • September 2018 105

Authorized licensed use limited to: Georgia Institute of Technology. Downloaded on February 03,2021 at 23:13:24 UTC from IEEE Xplore. Restrictions apply.
 layer to decrease the network uncertainty of the
underlying architecture. Edge computing is intro-
Cloud Manufacturing
cloud
duced in these scenarios to act as the time-aware
computing
scheduler. It ensures that the network is sensitive
Big data analysis Knowledge mining to the time factor requirements of the different
data tasks. Edge computing provides real-time
Edge
performance on the time node and processes
Edge cloud services
Cooperation computing Low jitter sequences of information interaction.
framework Mobile edge computing (MEC) is a network-
- +
ARDUINO

High bandwidth Edge gateway Low latency ing application and edge service based on intel-
ligent interaction at the equipment terminal. The
deployment strategy of MEC has the advantages
Network of low latency, high bandwidth, and a lightweight
edge device
network. MEC has the ability to provide more
Sensors PLC Smart meter Zigbee RFID accurate services in real time by accessing wire-
less network information and location information
Figure 4. Cooperation mechanism in IoT-based manufacturing. [13]. In addition, device-to-device (D2D) com-
munication technology has also been extended
to the industrial field [14]. This allows convenient
resource allocation by means of the base station
Robot PLC in the smart factory. Mobile equipment can inter-
act with each other through the wireless network
infrastructure. Furthermore, in the case of no net-
work coverage, devices can piggyback their con-
nections off nearby devices, which can form local
subnetworks, and then access the core network
through these intermediate connections.

Edge-Enhanced Data Fusion and Advanced Analytics

Fusion of different types and sources of data is an
important step for data acquisition and intelligent
Raspberry Pi
control in the management of complex industri-
al processes. Edge computing can enhance the
capability of data fusion at the network edge.
Figure 3 shows the architecture of an edge-en-
Not packed candy box
hanced data fusion framework. Edge computing
can make full use of the sensor resources at the
peripheral network devices. The complementa-
Industrial device driver ry and redundant information obtained from the
sensors in space and time are combined accord-
ing to pre-defined optimization criteria and algo-
Figure 5. A prototype platform with edge computing equipment for IoT-based rithms. This improves the consistency of the
manufacturing. interpretation and the analysis of the detected
event. Based on a knowledge base, a self-learn-
facilities. This allows the system to achieve ubiq- ing mechanism can perform dynamic reasoning
uitous perception and semantic interoperability of and generate responses. Edge computing nodes
manufacturing resources. Edge computing nodes can be deployed to extract features from the sen-
are being implemented to the smart factory for sor signals by applying techniques such as time
the fast configuration of the OPC UA technol- series analysis, frequency analysis, and wavelet
ogy. The server of OPC UA can be built with analysis. An artificial intelligence (AI) or machine
edge computing nodes, thus removing the infra- learning (ML) model can be run on the edge
structure obstacles that are present in traditional computing nodes to utilize the characteristic data
manufacturing equipment. The emergence of the to make predictions and to subsequently update
application of information technologies (e.g., SDN the knowledge base. Dynamic reasoning at the
and network functions virtualization) in intelligent edge can be performed according to the updated
manufacturing has helped to break the limitation knowledge and received data. This allows intelli-
of traditional network architecture. These systems gent decision making at the network edge.
therefore meet the requirements of IIoT by pro- The implementation of fine-grained data acqui-
viding massive connections, automatic operation sition is easy to realize at the peripheral equip-
and maintenance, and advanced network security ment, and the different streams of the original
policies [11, 12]. The deployment of edge com- data can be directly fused. The fused data is more
puting is generally in the form of a virtual machine informative and synthetic than the original inputs.
or a software container on the gateway, which It is possible to provide IoT-based computational
decouples network connections and supports resources in a similar vein as the operating sys-
third-party applications. While SDN makes the tem or software containers offered on the edge
industrial network highly reconfigurable, it cannot nodes. The manufacturing process can be visual-
meet the requirement of low latency. Recently, ized based on advanced data visualization tech-
time-sensitive networks (TSNs) have been sug- nologies. This would allow the system to forecast
gested as a new solution for real-time IIoT. It is failures and perform active maintenance. Addi-
a set of protocol clusters located at the data link tionally, detection devices are used to collect all

106 IEEE Communications Magazine • September 2018

Authorized licensed use limited to: Georgia Institute of Technology. Downloaded on February 03,2021 at 23:13:24 UTC from IEEE Xplore. Restrictions apply.
 20
Centralized task scheduling mechanism

The average upload speed of backbone

45.6 18
The time length of robots operation (h)
45 Self-organized task scheduling mechanism
39.5 16

network speed (Mbps)

35.2 14
35
30.6 12
26.4 10
25 8
16.8 20.5
6
15 11.2 20.1 4
6.3 2 Centralized task scheduling mechanism
6.1 10.2 Self-organized task scheduling mechanism
5 0
500 800 1500 2000 2500 3000 0 0.5 1 1.5 2 2.5 5 20 40 80 120 200 400
(a) (b)

Figure 6. Experimental results of the equipment scheduling mechanisms: a) order quantity in a candy packing line; b) running time of
the candy packing line (min).

the equipment execution data, extract features ing the requirements for processing capacity on
based on the data fusion strategies implemented the cloud is extremely time- and cost-intensive.
at the edge computing, and abstract the functions However, edge computing makes it easier to
of the equipment to construct a digital twin of the implement cloud services with low latency, high
intelligent equipment. From the perspective of bandwidth, and low jitter. The cloud platform
each server, edge computing mainly implements can be a tremendous tool for the smart factory
intelligent local distribution of computing resourc- to process massive data. For instance, it provides
es in the IIoT. It maps the intelligent equipment a scientific basis for production scheduling and
in virtualized manufacturing environments and market demand forecast. The framework of cloud
supports data centers to build digital plants. The computing can be used to dispatch and distribute
digital twins can then be used for direct product edge computing nodes according to global oper-
manufacturing, process optimization, and cost ational requirements, to equip IIoT nodes with
reduction. edge decision capabilities. Meanwhile, the task of
the edge computing nodes is to provide local data
Cooperation Mechanism between storage. In the case of connection failures to the
Cloud Computing and Edge Computing cloud (e.g., power failure), the data can temporar-
The combination between cloud computing and ily be cached at the edge. Once the connection is
edge computing enables the digital transforma- restored, the locally stored data is optimized and
tion of manufacturing. As shown in Fig. 4, such automatically synchronized with the cloud, which
a cooperation mechanism can be described as ensures full mapping of the equipment.
follows.
•Since the focus of the system is on providing A Case Study: Active Maintenance
real-time or short-term data analysis, edge com- Edge computing can bring significant advan-
puting can better support the real-time processing tages in terms of efficiency and cost reduction
and performance evaluation of local businesses. in Industry 4.0. In fact, activities such as active
This reduces network pressure, and improves data maintenance are being migrated to the network
security and privacy protection. edge to create novel industrial applications,
•Cloud computing, on the other hand, does which are changing the service mode. In the
not focus on real-time or short-cycle data analy- case study, we build distributed data process-
sis. Its focus is to implement big data analysis and ing in the cloud using Hadoop architecture. The
knowledge mining from the data obtained from Hadoop Distributed File System (HDFS) and
edge computing networks. It plays an important Hadoop MapReduce were used to perform real-
role in periodic maintenance, decision support, time data analysis and data mining of the local
and other activities that do not necessarily need database. The big data consisted of machine sta-
to be performed in real time. tus data and machine log data. All the data is
•Since edge computing is performed close to combined to generate a reasoning model, which
the execution unit, the edge computing nodes is loaded on a local Linux system of a Raspberry
act as collection units of heterogeneous data Pi. An OPC UA server on Raspberry Pi is applied
and move complex optimization problems to the to vendor-neutral transmission of pre-processing
cloud. They supply the cloud computing frame- information, or raw data from a programmable
works with the necessary high value data that can logic controller (PLC) with safety and reliability.
better support the cloud application (e.g., big data A semantic model is also built to integrate mul-
analysis). The decision model through the optimi- tiple data sources to produce more consistent,
zation of the business rules can be established on accurate, and useful information. This data fusion
the edge nodes for improving the efficiency and technology is used to input the acquired data
performance of the cloud. feature to the reasoning model.
As the number of sensors embedded in In our laboratory, a variety of customized
devices increases, the amount of data grows candy products are packed in the production line.
exponentially. Implementing real-time control Customer order service is provided using a pri-
loop feedback becomes more complex, and the vate cloud. An ad hoc network between the edge
analysis of all the data streams to generate deci- computing nodes is implemented to realize high-
sions becomes more time-consuming. Extend- speed and stable communication. The integra-

IEEE Communications Magazine • September 2018 107

Authorized licensed use limited to: Georgia Institute of Technology. Downloaded on February 03,2021 at 23:13:24 UTC from IEEE Xplore. Restrictions apply.
 Edge computing is tion of the Ethernet with DDS standard protocol edge computing in IIoT meets the requirements
is deployed to achieve efficient information inter- of real time for lightweight intelligent manufac-
a new computing action. As shown in Fig. 5, the robots get candy turing, and increases the agility and security of
paradigm that allows packing tasks associated with the cloud, then suck the network. This article presents a system archi-
the selected candy and put it into the not packed tecture for implementing edge computing in IIoT
the integration of core candy box. During the operation, the robots are applications. The role of edge computing for
capabilities such as net- represented as network nodes in the backbone IoT-based manufacturing is analyzed. The final
work, computing, and network. If any node is overloaded or fails, the experiments show that the self-organized task
task of that node will be switched to an adja- scheduling mechanism with edge computing
storage closer to the cent node. Therein, a multi-agent system is intro- provides obvious advantages in terms of busi-
peripheral equipment. duced to enhance the autonomous negotiation ness agility and bandwidth optimization com-
capability in the smart factory. In this way, each pared to traditional approaches. Furthermore, as
The application of edge robot can be regarded as an agent. The activities the computational capabilities in the cloud do
computing in IIoT meets of an agent are autonomous and independent. not include increased response latency or net-
the requirements of Their own goals and actions are not restricted work security, even limited computing resources
by others. The multi-agent system is employed deployed at the network edge will help in the
real time for lightweight to complete the production tasks with the task service quality of industrial use cases in IoT-
intelligent manufactur- procedures assigned to different agents, and we based manufacturing.
introduce Contract Net Protocol (CNP) [15] to
ing, and increases the allocate each task through the open tender, bid- Acknowledgment
agility and security of ding, and winning modes. The agents negotiate Dr. Jiafu Wan is the corresponding author. All
the network. and resolve the conflicts by means of competition of the authors contributed equally to this work.
and consultation. Thus, the order task completed This work was supported in part by the Nation-
in the self-organized task mechanism. On the con- al Key Research and Development Program of
trary, the instructions to a subordinate robot are China (No. 2017YFE0101000), the Science and
distributed by the central server. This may lead to Technology Program of Guangzhou, China (No.
execution delay and network congestion by the 201802030005), the Science and Technology
centralized task scheduling mechanism. In this Planning Project of Guangdong Province (No.
way, the cloud platform actually does not apply 2017B090913002), and the Key Program of the
in this case. The comparisons between the self-or- Natural Science Foundation of Guangdong Prov-
ganized task mechanism and the centralized task ince (No. 2017B030311008).
scheduling mechanism are presented as follows.
For a given order quantity, we compared the References
operating time between the self-organized task [1] W. Shi and S. Dustdar, “The Promise of Edge Computing,”
mechanism and the centralized task scheduling Computer, vol. 49, no. 5, 2016, pp. 78–81.
[2] M. Mukherjee et al., “Security and Privacy in Fog Computing:
mechanism. As shown in Fig. 6a, for an increase in Challenges,” IEEE Access, vol. 5, 2017, pp. 19,293–19,304.
the order quantity, the experimental results show [3] P. G. Lopez et al., “Edge-Centric Computing: Vision and
that the self-organized task mechanism based on Challenges,” ACM Sigcomm Comp. Commun. Rev., vol. 45,
edge computing is more agile and efficient, espe- no. 5, 2015, pp. 37–42.
[4] P. Patel, M. I. Ali, and A. Sheth, “On Using the Intelligent
cially after 2000 orders. Additionally, as shown in Edge for IoT Analytics,” IEEE Intelligent Systems, vol. 32, no.
Fig. 6b, we used an Internet speed meter to com- 5, 2017, pp. 64–69.
pare the bandwidth of the core network for the [5] M. W. Condry and C. B. Nelson, “Using Smart Edge IoT
two mechanisms. When the production line was Devices for Safer, Rapid Response with Industry IoT Control
Operations,” Proc. IEEE, vol. 104, no. 5, 2016, pp. 938–46.
stable, the speed of the production line backbone [6] D. Georgakopoulos et al., “Internet of Things and Edge
network based on the centralized task scheduling Cloud Computing Roadmap for Manufacturing,” IEEE Cloud
mechanism was 16~17 Mb/s. Once the Raspberry Computing, vol. 3, no. 4, 2016, pp. 66–73.
Pi devices were deployed, the speed of the back- [7] T. Suganuma et al., “Multiagent-Based Flexible Edge Comput-
ing Architecture for IoT,” IEEE Network, vol. 32, no. 1, Jan./
bone network based on the self-organized task Feb. 2018, pp. 16–23.
scheduling mechanism was between 5~6 Mb/s. [8] C. C. Byers, “Architectural Imperatives for Fog Computing:
This showed an overall 60 percent reduction in Use Cases, Requirements, and Architectural Techniques for
network speed. The proposed self-organized task Fog-Enabled IoT Networks,” IEEE Commun. Mag., vol. 55,
no. 8, Aug. 2017, pp. 14–20.
scheduling mechanism focuses on the dynamic [9] M. Satyanarayanan, “The Emergence of Edge Computing,”
scheduling problem for IoT-based manufactur- Computer, vol. 50, no. 1, 2017, pp. 30–39.
ing. In this case study, Raspberry Pi as the edge [10] A. Gıırbea et al., “Design and Implementation of an OLE for
computing node has enough processing ability for Process Control Unified Architecture Aggregating Server for
A Group of Flexible Manufacturing Systems,” Software Lett.,
data fusion and cooperation with working part- vol. 5, no. 4, 2011, pp. 406–14.
ners. Cloud services can be transferred to the net- [11] M. Imran et al., “Software-Defined Optical Burst Switching
work edge. The service delay is reduced, and the for HPC and Cloud Computing Data Centers,” IEEE/OSA J.
transmission of large amounts of data is avoided. Optical Commun. & Net., vol. 8, no. 8, 2016, pp. 610–20.
[12] J. Wan et al., “Software-Defined Industrial Internet of Things
This is the device domain and data domain appli- in the Context of Industry 4.0,” IEEE Sensors J., vol. 16, no.
cation of edge computing in the reference archi- 20, 2016, pp. 7373–80.
tecture. Furthermore, edge computing will play an [13] J. Liu et al., “A Scalable and Quick-Response Software
important role in IoT-based manufacturing, such Defined Vehicular Network Assisted by Mobile Edge Com-
puting,” IEEE Commun. Mag., vol. 55, no. 7, July 2017, pp.
as thermal sensors, SDN, and network security. 94–100.
[14] E. Ahmed et al., “Social-Aware Resource Allocation and
Conclusions Optimization for D2D Communication,” IEEE Wireless Com-
Edge computing is a new computing paradigm mun., vol. 24, no. 3, June 2017, pp. 122–29.
[15] S. Wang et al., “Towards Smart Factory for Industry 4.0:
that allows the integration of core capabilities A Self-Organized Multi-Agent System with Big Data Based
such as network, computing, and storage closer Feedback and Coordination,” Computer Networks, vol. 101,
to the peripheral equipment. The application of 2016, pp. 158–68.

108 IEEE Communications Magazine • September 2018

Authorized licensed use limited to: Georgia Institute of Technology. Downloaded on February 03,2021 at 23:13:24 UTC from IEEE Xplore. Restrictions apply.
 Biographies books. His main research interests include distributed systems,
cloud computing, and IoT service federation and security.
B aotong C hen ([email protected]) received his B.A.
degree in mechanical engineering from Nanyang Institute of Di Li ([email protected]) is a professor in the School of Mechan-
Technology, China, in 2014. He is currently pursuing a Ph.D.de- ical and Automotive Engineering, SCUT. She has directed 50+
gree in the School of Mechanical and Automotive Engineering, research projects, including ones funded by the National Nat-
South China University of Technology (SCUT). His research ural Science Foundation of China, among others. Thus far, she
interests include cyber-physical systems, industrial wireless net- has authored/co-authored 180+ scientific papers. Her research
works, the Internet of Things, and embedded control systems. interests include embedded systems, computer vision, and
cyber-physical systems.
Jiafu Wan ([email protected]) has been a professor in the
School of Mechanical & Automotive Engineering,SCUT since Haider Abbas [SM]([email protected]) is a cyber security pro-
September 2015. Thus far, he has published more than 150 fessional who received professional training and certifications
scientific papers, including 90+ SCI-indexed papers, 30+ IEEE from Massachusetts Institute of Technology, Stockholm Univer-
transactions/journal papers, 14 ESI Highly Cited Papers, and sity, Sweden, IBM, and EC-Council. His professional career con-
4 ESI Hot Papers. His research interests include cyber-physi- sists of activities ranging from R&D and industry consultations,
cal systems, Industry 4.0, smart factory, industrial big data, and through multi-national research projects, research fellowships,
industrial robots. doctoral advisory services, international journal editorships, con-
ferences chairmanships, and invited/keynote speaker.
Antonio Celesti ([email protected]) is an adjunct professor and
member of the Scientific Research Organizational Unit at the Qin Zhang ([email protected]) received her Ph.D. degree
University of Messina, Italy. He received his Ph.D. in advanced from SCUT in 2013. She is currently a lecturer with the School
technology for information engineering in 2012 from the Uni- of Computer Science and Engineering, SCUT. Her research
versity of Messina. He is a co-author of more than 130 papers interests include embedded systems, industrial robotics, pattern
published in international journals, conference proceedings, and recognition, and image analysis.

IEEE Communications Magazine • September 2018 109

Authorized licensed use limited to: Georgia Institute of Technology. Downloaded on February 03,2021 at 23:13:24 UTC from IEEE Xplore. Restrictions apply.