This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2923610, IEEE Access

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.Doi Number

A Smart Manufacturing Service System Based

on Edge Computing, Fog Computing and Cloud
Computing
Qinglin Qi1, and Fei Tao1, Senior Member, IEEE
1
School of Automation Science and Electrical Engineering, Beihang University, Beijing 100083, China

Corresponding author: Fei Tao (e-mail: ftao@ buaa.edu.cn).

This work was supported in part by the National Natural Science Foundation of China under Grant 51875030, in part by the National Key Research and
Development Program of China under Grant 2016YFB1101700, in part by Beijing Science and Technology Project under Grant Z181100003118001, and in
part by the Academic Excellence Foundation of BUAA for PhD Students.

ABSTRACT The state-of-the-art technologies in new generation information technologies (New IT),
greatly stimulates the development of smart manufacturing. In smart manufacturing environment, more and
more devices would be connected to the Internet, so that a large volume of data can be obtained during all
phases of product lifecycle. Cloud based smart manufacturing paradigm facilitates a new variety of
applications and services to analyze a large volume of data and enable large-scale manufacturing
collaboration. However, different factors, such as the network unavailability, overfull bandwidth, and
latency time, etc., restrict its availability for high speed and low latency real time applications. Fog
computing and edge computing extended the compute, storage, and networking capabilities of cloud to the
edge, which will respond to the above issues. Based on cloud computing, fog computing, and edge
computing, in this study, a hierarchy reference architecture is introduced for smart manufacturing. The
architecture is expected to be applied in the digital twin shop floor, which opens a bright perspective of new
applications within the field of manufacturing.

INDEX TERMS Cloud computing, digital twin, edge computing, fog computing, hierarchical architecture,
smart manufacturing.

I. INTRODUCTION manufacturing [4]. However, the exponential growth of

The increasingly evolvement of new generation information data goes beyond general processing capabilities of users.
technologies (New IT), including cloud computing (CC), Cloud computing is an answer to the extensive need to
Internet of Things (IoT), big data analytics (BDA), artificial deal with data explosion in many cases [5]. Cloud computing
intelligence, cyber-physical system (CPS), etc., contributes is an Internet-based computing where the shared resources
significantly in manufacturing to achieve more efficient, (e.g., storage and computing facilities, software, data,
competitive, and smarter manufacturing [1]. Driven by New applications, etc.) are accessed and used on demand in a
IT, smart manufacturing pursues a higher degree of convenient “pay-as-you-go” manner [6]. In the cloud
intelligence with the powers of low-cost and ubiquitous computing paradigm, users get high quality of services at a
sensing, advance computing and data analytics, and cyber- lower cost. Due to the potential and practical benefits to
physical integration. The convergence of various advanced society and economy, cloud computing paradigm has
and connected devices and machines is the inevitability for attracted enormous attention from academia and industry.
smart manufacturing. Moreover, the growing ubiquity of For instance, a lot of companies, such as Amazon, Google,
IoT [2], provides promising opportunities to build powerful Salesforce, IBM, and Microsoft etc., have developed their
industrial systems and applications [3]. As a result, a large cloud computing platforms to provide services for ubiquitous
volume of various manufacturing data is generated by these users [7]. Meanwhile, combined with manufacturing, cloud
connected machines and sensors, etc. These data need to be computing gave rise to a new cloud-based manufacturing
cleaned, stored, and processed to produce information and model. All manufacturing resources and capabilities
knowledge, which are poised to be the basis for smart

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 3.0 License. For more information, see http://creativecommons.org/licenses/by/3.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2923610, IEEE Access

(MRs&Cs) are virtualized and encapsulated as services to be II. MANUFACTURING SYSTEM ARCHITECTURE UNDER
managed, allocated, and on-demand used through cloud CLOUD ENVIRONMENT AND ITS SHORTAGES
[8][9]. Bringing together IoT, CPS, BDA, and CC The deep integration of all aspects of the manufacturing
technologies, it enables smart manufacturing [10]. Moreover, process with the New IT, promotes profound changes in
cloud computing is also applied in industrial Internet of manufacturing pattern, which can be named as smart
Things (IIoT) and CPS. For example, Silva et al. [11] manufacturing [1]. Benefited from New IT, more and more
proposed a cloud-based architecture for IIoT for remote manufacturing devices are connected to the Internet, and
monitor and control industrial devices. Xiong et al. [12] more and more users participate in the manufacturing process.
designed a cloud operating system for industrial applications. As a result, manufacturing system becomes increasingly
However, data collected from machines and sensors are sent complex and dynamic, which requires resources and users
to the cloud and desired output is sent to the desire devices collaboration. Besides, data generated in manufacturing is
again, which may result in delayed response. Furthermore, also becomes increasingly richer, more diverse and complex.
massive data requires high bandwidth, which is expensive With the powerful computing, high scalability, and virtual
[13]. With network unavailability, these problems become resource services, cloud provides an efficient manner for
barriers to applications required delay constraints [14]. massive data processing and storage, and satisfies the
Fog computing and edge computing are solutions to above demand for large-scale collaboration[19].
problems, in which some compute, storage, and networking
capabilities of cloud are shifted to the edge network [15]. A. CLOUD-BASED MANUFACTURING SYSTEM
ARCHITECTURE
Through fog computing and edge computing, some data from
As one of the major enablers for the manufacturing, cloud
smart machines or sensors do not have to be stored and
computing transformed traditional manufacturing models to
processed on cloud, but on edge devices instead (e.g., fog
computing and service-oriented manufacturing model, which
routers or smart gateway/IIhub [16], etc.), which fulfills the
is becoming more interoperable, smart, adaptable and
requirements like low latency, real time response, reduction
distributed. Besides, industrial IoT (IIoT) and cyber-physical
of network traffic, etc. [17] Moreover, fog computing is also
systems (CPS) are also integrated with cloud computing to
adept in filtering raw data and reducing resource wastage
cope with the challenges with regards to massive data
[18]. As the new paradigm of computing, fog computing and
processing, service provisioning, and knowledge
edge computing offer the possibility to increase efficiencies
management, etc. [20] [21]
and develop innovative services [5], which would accelerate
development of smart manufacturing. Therefore, a reference
architecture for smart manufacturing based on edge
computing, fog computing and cloud computing, is proposed
to provide guidance for the development of smart
manufacturing. The reference architecture will be applied in
digital twin shop floor to improve different levels of
intelligence of shop floor.
The main contributions of this paper include:
(1) The manufacturing system under cloud environment
and its general architecture are reviewed, and some problems
are summarized.
(2) A hierarchy reference architecture for smart
manufacturing based on edge computing, fog computing and
cloud computing is proposed to facilitate a new variety of
applications and services.
FIGURE 1. The cloud-based manufacturing system architecture.
(3) The hierarchical architecture could be applied in the
digital twin shop-floor, which opens a bright perspective for
smart manufacturing. From multiple cloud-based manufacturing applications
The rest of this paper is organized as follows. The [21]-[25], the cloud-based manufacturing system architecture
solutions of manufacturing system under cloud environment can be summarized as shown in Figure 1. The architecture
are reviewed in Section II, and their merits and demerits are consists of two main layers, which are physical
discussed. In Section III, the reference architecture for smart manufacturing system and cloud environment. The lower
manufacturing based edge, fog, and cloud computing is layer represents the MRs&Cs, including manufacturing
presented. In Section IV, applications of the reference devices, as well as the users who participate in the
architecture in digital twin shop floor is discussed. manufacturing activities. The physical manufacturing system
Conclusion is drawn in Section V. covers all physical devices on the factory [22], even multiple
plants, including cooperative plants located in different

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 3.0 License. For more information, see http://creativecommons.org/licenses/by/3.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2923610, IEEE Access

regions [23], as well as all users, including enterprise users unacceptable Internet round-trip latency, ranging from
and customer users [24]. Whereas the upper layer represents tens to a few hundreds of milliseconds [31].
data processing, process control and services management in 4) Data validity. A lot of insignificant data (e.g., redundancy,
the cloud environment. The cloud provides non-physical noise, temporary data, etc.) are conveyed to cloud, which
services and applications for users in the manufacturing waste resources. Besides, some locally consumed data,
lifecycle. The applications and services provided by cloud is also do not need to be sent to the cloud [32]. However,
richer, as well as faster, simpler and cheaper to use. In the data filtering capacity has not received enough attention.
manufacturing system under cloud environment, 5) Security and privacy. The constant development of new
manufacturing enterprises or users do not have to invest in attack vectors (e.g., coming from the communication
expensive manufacturing resources, but buy manufacturing channels, denial-of-service (DoS) attacks, etc.) results in
capacities through the cloud platform in a “pay-as-you-go” numerous security issues [33]. Besides, when all the data
manner [26]. Cloud connects massive social manufacturing are conveyed to the cloud, they also contain privacy data,
resources together and can provide various manufacturing which increases the risk of disclosing user privacy [34].
services, enabling open collaboration and sharing [27]. 6) Inefficient interaction. The cloud-based communication
In this architecture, cloud connects all the manufacturers among manufacturers, users and machines, which may
and customers. Customers communicate with each other and locate in the same area, limits the flexibility and efficiency
manufacturers to effectively integrate social demands and in connectivity and interactive messaging.
production capacities by using the desktops or mobile As a result, despite the broad utilization of cloud
devices [21]. The physical manufacturing system with the computing in smart manufacturing, some applications
MRs&Cs, are sensed and accessed to the cloud by IoT (e.g., demand real-time sensitive, and precise reaction to events,
smart things, sensor networks, radio frequency identification cannot count on the cloud alone, due to inherent problems
(RFID), etc.). Each manufacturing unit can be considered as [30]. Taking the actual status of smart manufacturing into
a CPS. Multiple unit-level CPSs constitute the system-level consideration, some new technical concepts are expected.
CPS. The integration between the physical devices, users and
cloud is enabled by encapsulating data and functions inside III. A REFERENCE ARCHITECTURE FOR SMART
services. Moreover, data which are collected from physical MANUFACTURING SYSTEM BASED ON EDGE
COMPUTING, FOG COMPUTING and CLOUD
manufacturing system, as well as control instructions (e.g.,
COMPUTING
machining parameters, tool path, etc.) [25] from cloud, are
transmitted between the lower and upper layers through
A. FROM CLOUD COMPUTING TO FOG AND EDGE
network. Through the cloud-based solutions, manufacturing COMPUTING
enterprise are able to develop better-integrated and more Cloud computing is the computing paradigm that enables
efficient processes with the lower cost. ubiquitous, convenient, on-demand network access to a
shared pool of configurable computing resources (e.g.,
B. PROBLEMS AND SHORTAGES computing and storage facilities, applications, services, etc.)
The manufacturing system under cloud environment allows [35]. Through virtualization technology, cloud computing
higher utilization without increasing investment and shields the diversity of underlying devices and provides users
degrading performance, as well as frees manufacturers and with a variety of services in a transparent way, including IaaS
users from many details. However, there are still many (Infrastructure-as-a-Service), PaaS (Platform-as-a-Service)
problems that limit the expansion of smart manufacturing. and SaaS (Software-as-a-Service) [36]. Due to the increasing
1) Overfull bandwidth. Data generated by various number of access devices, cloud computing may face some
manufacturing resources, which may be geographically problems in the bandwidth, latency, network unavailability,
distributed, are experiencing explosive growth. These data security and privacy, etc. Fog computing is considered as an
are conveyed over the network to cloud, where data extension of cloud computing to the edge network, providing
processing is carried out [28]. The increasing volume and services (e.g., compute, storage, networking, etc.) closer to
velocity of data require high bandwidth, which is very near-user devices (e.g., network routers, various information
expensive. When the network congestion is serious, some systems, etc.), instead of sending data to cloud [37]. In fog
data may be loss. computing paradigm, data storage and processing rely more
2) Unavailability. Although the data stored in the cloud can on local devices, rather than on cloud data center. Fog
be accessed from anywhere at any time, user relies heavily computing makes the applications more convenient, meeting
on the availability of internet connection and the servers a wider range of node access. Similar to fog computing, edge
[29]. When data cannot be accessed due to the network computing also allows computation to be performed at the
unavailability, the power of the cloud becomes unusable. edge of the network, but at closer proximity to the data
3) Latency. Some real-time and concurrent scenarios require sources [38]. The difference between fog computing and
time synchronization, bringing real-time issues [30]. The edge computing is that fog computing relies on
data transmission between terminals and cloud may suffer interconnection capabilities among nodes, whereas edge

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 3.0 License. For more information, see http://creativecommons.org/licenses/by/3.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2923610, IEEE Access

computing runs in isolated edge nodes. Edge computing time changes, a great deal of various sensing devices (e.g.,
provides edge services near the source of data to meet the various smart sensors, RFID, etc.) are deployed on the
critical requirements in agile connectivity, real-time manufacturing resources. Together with actuators and drivers,
optimization, smart applications, security and privacy. As the they form the sensing system and execution system. In
supplementary, fog computing and edge computing, which addition, the environment factors should also be monitored,
provide compute, storage, and networking services between such as temperature, humidity, pressure, etc. These resources
end devices and traditional cloud computing [37], offer closely cooperate with each other to carry out various
bright prospects for smart manufacturing applications. manufacturing activities.
To extend the smart manufacturing applications under
cloud environment and offer perspectives for implementation
of solutions that require very low and predictable latency, a
reference architecture based on edge computing, fog
computing, and cloud computing for smart manufacturing is
proposed. As illustrated in Figure 2, the reference
architecture consists of the following three layers.

FIGURE 3. Edge computing based manufacturing interaction and

control.

In edge layer, CNC machine tools, smart robots, sensors,

terminals, and edge connection devices, not only produce
data, but also have the capacities of data analysis and
processing, which support edge computing and provide near-
edge smart services. At the site closest to the physical
equipment, the edge layer uses limited hardware resources
(e.g., gateway, industrial PC, PLC, IIHub, etc.) to complete
data collection, protocol conversion, data filtering, data
upload, data storage, data analysis and other operations. First
of all, the real-time status (e.g., regular physical quantity,
environmental and operating conditions) and various
parameter data of the manufacturing resources and users are
collected in real time through appropriate access methods,
FIGURE 2. The hierarchy architecture for smart manufacturing based
on edge computing, fog computing and cloud computing.
such as RFID, sensors, industrial Ethernet, et al. At the edge
level, there are many types of equipment, but the protocols
are varied. Therefore, the first-line task of the edge
B. EDGE COMPUTING BASED MANUFACTURING computing devices is interface and protocol conversion to
INTERACTION AND CONTROL LAYER convert different data formats and communication protocols
As illustrated in Figure 3, this layer consists of all the into a unified protocol. Data interaction in edge layer is
manufacturing resources involved in various manufacturing realized based on the unified protocol. Moreover, for later
activities. In a shop floor or factory, principal resources can processing, the collected data is also subjected to noise
be summarized as “Human-Machine-Material-Environment”. processing, abnormal data cleaning, if necessary, data
Human refers to workers and managers. Machine includes compression. The edge devices also have a certain data
machining equipment (e.g., machine tools, robots, etc.), storage capability (may not large) to store real-time state data,
transporting equipment (e.g., the material conveyors, as well as alarms, faults and other information. Moreover, the
automatic guided vehicles, etc.). Material consists of raw micro data center is used effectively store the real-time data
material, parts and components, semi-finished products, of the state of the production field. As the volume of data is
finished products, etc. Moreover, to keep abreast of the real- large, the edge computing devices would overwrite the

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 3.0 License. For more information, see http://creativecommons.org/licenses/by/3.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2923610, IEEE Access

original data over a certain period of time with new data. To

save data transmission costs and bandwidth, only the
necessary edge data is uploaded to upper layer. In addition,
much of the collected data may not have much value, just
reporting "normal" conditions. Therefore, before edge data is
overwritten, the edge computing devices would filter, discard,
or more efficiently reassemble the data for transmission to
the fog or cloud for storage and analysis. For the collected
real-time data, the edge computing devices can perform
simple data processing and analysis. For example, through
data analysis for various sensors, edge computing can
determine whether the manufacturing equipment state is
normal and what action it is performing. In addition, after
simple data processing and analysis, automatic feedback
control for the operation of the manufacturing equipment
according to preset rules can be realized. The feedback
control on the edge side is sufficient to ensure real-time
performance. Even in the case of offline, edge computing is
still able to provide localized application services.
On the basis of realizing the smart perception and
interconnection, this layer is to further realize the following FIGURE 4. Fog computing based manufacturing information
functions based on edge computing. 1) Man-Machine integration.

coordination. Through integrating the accuracy of machine

and the flexibility of human, manufacturing system can be As shown in Figure 4, the fog layer is a network group
quickly adjusted to adapt to the dynamic production activities. composed of interconnected fog nodes. Each fog network has
As a result, human and machines can be harmonious a fog gateway and a fog management node, as well as inner
coexistence. 2) Machine-Machine coordination. The fog nodes (e.g., router, network switch, local sever, local data
machines cooperate with each other, which could effectively center, etc.). The function of the fog gateway is to support
improve the speed and efficiency. Through machine-machine edge network access, receive data collected at the edge layer,
collaboration, manufacturing system could have better and perform further processing. The fog management node
robustness and the self-regulating capacity. As a result, can be aware of all the fog nodes in a fog network. The fog
manufacturing would be more flexible, so as to produce node communicates with the cloud data center of the external
diversified products more efficiently. 3) Machine-Material network through the fog management node. In fog layer,
coordination. Based on sensors, RFID, machine vision, etc., edge connection devices and edge computing devices
material and products can be sensed and positioned communicates with the fog gateway through a single-hop
intelligently. In the light of the edge computing, smart low-latency link. The inner fog nodes communicate with the
machines automatically execute the corresponding edge layer through the fog gateway. The main function of the
machining program. 4) Environment coordination. The inner fog node is to perform data processing (data filtering,
manufacturing environment is dynamically changed to compression, storage, encryption, etc.) on the data from the
ensure product quality and equipment safety according to the edge layer and processing service requests. When a inner fog
specific requirements. node receives the task request assigned by the fog
management node, it can perform the task locally, or may
C. FOG COMPUTING BASED MANUFACTURING send the data to other fog nodes to compute. The fog
INFORMATION INTEGRATION LAYER management node is responsible for the coordinated
This layer is responsible for medium speed and latency data assignment of task requests and computing resources within
analysis for manufacturing information integration and the fog. Under the unified management of the fog
management. The components (e.g., network routers, management node, the fog layer achieves mutual cooperation
regional servers, network switches, etc.) of edge or core between the fog nodes. In the fog layer, data from different
network, and enterprise-class management systems could be edge networks is further unified, including data formats and
used as the infrastructures for fog computing. These communication protocols. Then these data are integrated to
infrastructures can provide different compute, storage, and find valuable insights for enterprise-level applications. The
networking functions. Through fog computing, multiple fog nodes can effectively reduce the service delay and the
applications and services deployed at this layer can connect a amount of data transmission in the network. Finally, part of
large number of smart equipment and sensors, and integrate the data in fog layer is transmitted to the cloud computing
the data from them. layer, to drive the corresponding services.

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 3.0 License. For more information, see http://creativecommons.org/licenses/by/3.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2923610, IEEE Access

On the basis of information integration, this layer could and-play” components. In addition, not only providers could
achieve some management applications, such as inventory conveniently provide their services to their collaborative
management and process optimization, etc. These stakeholders, but also manufacturing enterprises could use
applications need to get insights from the integrated data. various resources in a pay-as-you-go manner. As a result,
With respect to inventory management, the volume and batch different kinds of collaboration applications, such as
of material consumed in manufacturing activities, as well as commerce cooperation, manufacturing cooperation, business
different kinds and quantity of finished products are cooperation, are achieved, greatly improving the value and
continually recorded. As a result, the managers are able to efficiency of enterprises. Besides, various users could also
keep abreast of inventory changes of material and products. participate in the innovation of product and value creation
As for the process optimization, the historical process data is with the help of cloud services.
analyzed in this layer. The correlation between different
technological parameters and the effect of these parameters
on yield and quality can be determined by analyzing the
process data. The productivity and product quality can be
improved effectively by adjusting technological processes in
relation to these parameters.
On the other hand, this layer also receives the data from
outside enterprise (such as orders, invoked services, etc.),
which are mainly from the cloud. Based on these data,
manufacturing companies carry out production planning and
resources scheduling, etc. For instance, once the
manufacturing enterprise receives the orders from customers,
production plans are developed to determine the optimal
configuration of manufacturing resources according to the
availability of resources. In addition, when some faults
occurred, the managers contact third-party maintenance
serviceman to replace or repair the malfunction equipment. FIGURE 5. Cloud based manufacturing collaboration.
As an intermediate layer, fog computing paradigm reduce
bandwidth towards the cloud, and provide computing and In addition, as shown in Figure 5, cloud layer is
storage services for the enterprise management applications. responsible for big data analysis and storage. In the process
Moreover, the data form manufacturing resources is further of manufacturing activities and enterprises collaboration, a
filtered in this layer to be forwarded to cloud. large volume of data is generated. With powerful computing
and storage capabilities, cloud layer receives and processes
C. CLOUD COMPUTING BASED MANUFACTURING massive data uploaded by fog management node from the
SERVICE COLLABORATION LAYER fog layer and edge layer. In order to extract additional value
The diversification of market and customer demands, from data, data processing models are established to analyze
requires manufacturing enterprises to have abilities to large computational complexity, low latency sensitivity and
manufacture multiple products. Moreover, with the massive data, and display the results. Taking the fault
continuous improvement of product performance and diagnosis of equipment as an example, diagnostic model is
structure, it is impossible to complete all the manufacturing established based on equipment data. The larger the amount
activities only by a single enterprise. Therefore, socialized of equipment data, the easier it is to form an accurate and
sharing of MRs&Cs, enterprise and business collaboration, as reliable diagnostic model, which requires more computing
well as the socialized user participation, become an resources and higher computing power. So the diagnostic
inevitable trend. To achieve better sharing of limited model is more suitable for forming and updating in the cloud.
resources and efficient adding of value, manufacturers should After the diagnosis model is formed, the model is
make full use of high-quality resources in the whole society downloaded to fog layer and edge layer to realize real-time
to conduct large-scale collaboration in product design, and reliable equipment diagnosis. Therefore, big data
manufacturing, marketing and services. With the analysis and mining are performed in cloud layer to form a
characteristics of ubiquitous, convenient, on-demand model. The fog layer and edge layer download the model to
resources sharing, low costs, and interoperability, cloud pave perform real-time analysis on the real-time status. During the
the way for the large-scale collaboration. In this layer, all the equipment running process, new data is generated and
MRs&Cs are managed in the form of cloud services to be uploaded to cloud again. The model in cloud layer will be
shared by various users, including the resources services, updated based on historical big data and new data. Besides,
data services, models services, knowledge services, etc. big data provides sharper insight for businesses, such as
Through services, various MRs&Cs would become “plug- market trends analysis, risk assessment, etc.

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 3.0 License. For more information, see http://creativecommons.org/licenses/by/3.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2923610, IEEE Access

IV. PERSPECTIVE FOR APPLICATION OF THE resources, i.e., “Human-Machine-Material-Environment”.

REFERENCE ARCHITECTURE IN DIGITAL TWIN SHOP- VS is a collection of various multiple-dimensional models,
FLOOR
which are the faithful mapping of PS. Ss provide system
As an effective way for cyber-physical integration, digital
support and services for DTS. Data from PS, VS and SS are
twin provides manufacturing enterprises with a new way to
integrated and fused in SDTD. Through the ultra-high-
carry out smart production and precision management [39].
fidelity mapping and real-time interaction among PS, VS, Ss
Digital twin shop-floor (DTS), as a basic unit of smart
and SDTD, the production elements management, production
manufacturing, would become imperative. The hierarchical
planning, production process control in DTS are optimized.
data processing structure of above architecture could be
As shown in Figure 7, the cloud services system is based
tremendously beneficial to DTS. The DTS and services in
on cloud computing, which connects users and the
various layer are shown in Figure 6 and Table I.
manufacturing shop-floor. Users submit their personalized
product requirements as tasks to the cloud platform. Through
task management, a complex product task is decomposed
into subtasks that can be accomplished by a single service.
According to the QoS, the manufacturing service supply-
demand matching and scheduling is carried out to select
optimal services. Through service combination and
collaboration, selected services from different enterprises are
invoked and combined to complete the task collaboratively.
As a result, personalized customization, smart design, and
supply chain collaboration, etc., are achieved. In addition, the
MRs&Cs services from DTS constantly update their status
based on the data from physical resources and virtual models,
and participate in the service collaboration. Once the services
from DTS are invoked, specific requirements would be
conveyed to enterprise or shop-floor management systems.
The management systems are based on fog computing,
which is responsible for precise M&C, and reliable operation
and maintenance. The management systems constantly
acquire the status data of the physical resources, i.e.,
“Human-Machine-Material-Environment” in the PS, to
analyze, evaluate and predict the availability and
performance of PS. For instance, through analyzing the
historical and real-time status data of machines by statistic
method, mathematical modeling, and machine learning, etc.,
the prediction models are built. According to the prediction
FIGURE 6. Digital twin shop-floor (DTS) framework [40]. models, the health conditions of machines or their
components, the tendency for machines capacities to
TABLE I
SERVICES IN VARIOUS LAYER deteriorate, and the lifespan of components could be
Edge services FOG SERVICES Cloud services quantified and visualized. Moreover, when the order is
Data sensing Data acquisition Virtualization received, taking the costs, resources performance, completion
Data cleaning Data reduction Data integration time, production efficiency and other constraints into
Edge storage Fog storage Cloud storage consideration, production plan is developed. By determining
Edge analysis Fog analysis Big data analysis the machining equipment, organization mode, processing
Iterative interaction Asset optimization Task management sequence of the workpieces, as well as the allocation scheme
Simulation Process planning Supply-demand matching of material, work in process (WIP) and other key resources in
Real-time monitoring Proactive MRO Personalized customization the premise of satisfying imperative constraints, management
Disturbance found Smart scheduling Smart design collaboration systems could optimize the performance of the shop-floor.
Real-time control Precise management Supply chain collaboration
Fog computing provides storage, compute, analysis, and
Material racking Material distribution Commerce collaboration
network support for data processing of management
applications, such as resources management, energy
As shown in Figure 6, DTS consists of four components, management, proactive MRO, etc.
which are physical shop-floor (PS), virtual shop-floor (VS), The real-time iterative optimization of shop-floor is based
shop-floor digital twin data (SDTD), and service systems (Ss) edge computing. Due to the collaboration of human,
[40]. PS includes a series of physical manufacturing machines, material, environment and other heterogeneous

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 3.0 License. For more information, see http://creativecommons.org/licenses/by/3.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2923610, IEEE Access

elements, as well as the uncontrollability of the emergency, it businesses, based on real-time and historical production data.
is likely to cause downtime, delay and other phenomena. According to the simulation of virtual models, the physical
They may reduce resources utilization, increase the resources are adjusted and controlled in real time, achieving
additional inventory and energy costs, and other penalties. optimal production. Edge computing shields the
Therefore, physical resources and virtual models interact and heterogeneity of physical interfaces and communication
cooperate with each other to optimize the production process. protocols, facilitating the real-time interaction of physical
Above of all, physical resources and virtual models receive resources and virtual models. Besides, the high speed and
the operation instructions from management systems in low latency real-time data analysis of edge computing
accord with production plan, to organize production. Then, in enables the optimal control.
the actual production process, the physical resources would The data from PS, VS, and SS are stored and processed
transmit the real-time data to the virtual models. The virtual through edge computing, fog computing and cloud
models evolve according to the real-time state of the physical computing according different time-sensitive. Through the
resources, and identify the production schedule and data cleaning, reduction, modeling, association, clustering,
disturbance factors. Coping with the disturbance, virtual integration, evolution, fusion and other operations in various
models are to optimize the production process from the layer, the fused SDTD reflect the dynamic evolution process,
perspective of all elements, whole process, and whole evolution laws, and statistical characteristics of DTS.

FIGURE 7. Digital twin shop-floor based on edge computing, fog computing and cloud computing.

[1] F. Tao, and Q. Qi, “New IT Driven Service-Oriented Smart

V. CONCLUSION Manufacturing: Framework and Characteristics,” IEEE Trans. Syst.,
Man, Cybern., Syst., vol. 49, no. 1, pp. 81-91, Jan. 2019.
Smart manufacturing system involves multiple levels. The [2] C. Zhu, V. C. Leung, L. Shu, and E. C. H. Ngai, “Green Internet of
underlying layer is the smart devices, which are the sources Things for smart world,” IEEE Access, vol. 3, pp. 2151-2162, 2015.
of data and provide edge computing. The middle layer is the [3] L. Xu, W. He, and S. Li, “Internet of things in industries: A
survey,” IEEE Trans. Ind. Inform., vol. 10, no. 4, pp. 2233-2243,
data transmission network, in which fog computing is carried 2014.
out. The top layer is the cloud, in which big data is stored and [4] F. Tao, Q. Qi, A. Liu, and A. Kusiak, “Data-driven smart
analyzed. Through the compute, storage, and networking manufacturing,” J. Manuf. Syst., vol. 48, pp. 157-169, Jul. 2018.
[5] J. Pizoń, and J. Lipski, “Perspectives for Fog Computing in
capabilities in the near-end nodes, edge computing and fog Manufacturing,” Appl. Comput. Sci., vol. 12, no. 3, pp. 37-46, 2016.
computing reduce the data sent to cloud, and the probabilities [6] Y. Wei, and M. B. Blake, “Service-oriented computing and cloud
of service downtime, ensuring the robustness of smart computing: Challenges and opportunities,” IEEE Internet Comput.,
manufacturing system. Edge computing, fog computing, and vol. 14, no. 6, pp. 72-75, 2010.
[7] Q. Zhang, L. Cheng, and R. Boutaba, “Cloud computing: state-of-
cloud computing cooperate with each other, better meeting the-art and research challenges,” J. Internet Serv. Appl., vol. 1, no.
the requirements of smart manufacturing applications. 1, pp. 7-18, 2010.
[8] F. Tao, L. Zhang, V. C. Venkatesh, Y. Luo, and Y. Cheng, “Cloud
manufacturing: A computing and service-oriented manufacturing
REFERENCES

VOLUME XX, 2017 1

This work is licensed under a Creative Commons Attribution 3.0 License. For more information, see http://creativecommons.org/licenses/by/3.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2923610, IEEE Access

model,” Proc. Inst. Mech. Eng. B J. Eng. Manuf., vol. 225, no. 10, International Conference on Sensing, Communication, and
pp. 1969–1976, Oct. 2011. Networking-Workshops (SECON Workshops), 2015, pp. 1-6.
[9] F. Tao, L. Zhang, Y. Liu, Y. Cheng, L. Wang, and X. Xu, [29] Y. Wang, “Cloud-dew architecture,” Int. J. Cloud Comput., vol. 4,
“Manufacturing service management in cloud manufacturing: no. 3, pp. 199-210, 2015.
overview and future research directions,” J. Manuf. Sci. Eng., vol. [30] M. S. de Brito, S. Hoque, R. Steinke, A. Willner, and T. Magedanz,
137, no. 4, 040912, 2015. “Application of the Fog computing paradigm to Smart Factories and
[10] Y. Liu, and X. Xu, “Industry 4.0 and Cloud Manufacturing: A cyber‐physical systems,” Trans. Emerg. Telecommun. Technol. vol.
Comparative Analysis,” J. Manuf. Sci. Eng., vol. 139, no. 3, 034701, 29, no. 4, e3184, 2018.
Mar. 2017. [31] T. Hegazy, and M. Hefeeda, “Industrial automation as a cloud
[11] A. F. D. Silva, R. L. Ohta, M. N. D. Santos, and A. P. D. Binotto, service,” IEEE Trans. Parall. Distr. Syst., vol. 26, no. 10, pp. 2750-
“A cloud-based architecture for the internet of things targeting 2763, 2015.
industrial devices remote monitoring and control,” IFAC- [32] I. Stojmenovic, “Fog computing: A cloud to the ground support for
PapersOnLine, vol. 49, no. 30, pp. 108-113, 2016. smart things and machine-to-machine networks,” In Proc. IEEE
[12] G. Xiong, T. Ji, X. Zhang, F. Zhu, and W. Liu, “Cloud operating Telecommunication Networks and Applications Conference
system for industrial application,” in Proc. IEEE Int. Conf. Service (ATNAC), 2014, pp. 117-122.
Operations and Logistics, and Informatics, 2016, pp.43-48. [33] H. P. Breivold, and K. Sandström, (2015, December). “Internet of
[13] L. Shu, Y. Zhang, Z. Zhou, M. Hauswirth, Z. Yu, and G. Hynes, Things for Industrial Automation--Challenges and Technical
“Transmitting and gathering streaming data in wireless multimedia Solutions,” In Proc. 2015 IEEE International Conference on Data
sensor networks within expected network lifetime,” Mobile Netw. Science and Data Intensive Systems (DSDIS), 2015, pp. 532-539.
Appl., vol. 13, no. 3-4, pp. 306-322, 2008. [34] P. Garcia Lopez, A. Montresor, D. Epema, A. Datta, T. Higashino,
[14] L. Shu, Y. Zhang, L. T. Yang, Y. Wang, M. Hauswirth, and N. A. Iamnitchi, M. Barcellos, p. Felber, and E. Riviere, “Edge-centric
Xiong, “TPGF: geographic routing in wireless multimedia sensor computing: Vision and challenges,” ACM SIGCOMM Comput.
networks,” Telecommun. Syst., vol. 44, no. 1-2, pp. 79-95, 2010. Commun. Rev., vol. 45, no. 5, pp. 37-42, 2015.
[15] F. Bonomi, R. Milito, J. Zhu, and S. Addepalli, “Fog computing [35] A. Botta, W. De Donato, V. Persico, and A. Pescapé, “Integration
and its role in the internet of things,” in Proc. the First Edition of of cloud computing and internet of things: a survey,” Future Gener.
the MCC Workshop on Mobile Cloud Computing, 2012, pp. 13–16. Comput. Syst., vol. 56, pp. 684-700, 2016.
[16] F. Tao, J. Cheng, and Q. Qi, “IIHub: an Industrial Internet-of- [36] K. Skala, D. Davidovic, E. Afgan, I. Sovic, and Z. Sojat, “Scalable
Things Hub Towards Smart Manufacturing Based on Cyber- distributed computing hierarchy: Cloud, fog and dew computing,”
Physical System,” IEEE Tran. Ind. Inform., vol. 14, no. 5, pp. 2271- Open Journal of Cloud Computing, vol. 2, no. 1, pp. 16-24, 2015.
2280, May. 2018. [37] T. V. N. Rao, A. Khan, M. Maschendra, and M. K. Kumar, “A
[17] D. J. Deng, S. Y. Lien, C. C. Lin, S. C. Hung, and W. B. Chen, paradigm shift from cloud to fog computing,” Int. J. Sci. Eng.
“Latency Control in Software-Defined Mobile-Edge Vehicular Comput. Technol., vol. 5, no. 11, pp. 385-389, 2015.
Networking,” IEEE Commun. Mag., vol. 55, no. 8, pp. 87-93, 2017. [38] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu, “Edge computing:
[18] I. Stojmenovic, and S. Wen, “The fog computing paradigm: Vision and challenges,” IEEE Internet Things J., vol. 3, no. 5, pp.
Scenarios and security issues,” In Proc. IEEE Federated 637-646, 2016.
Conference on Computer Science and Information Systems [39] Q. Qi, and F. Tao, “Digital Twin and Big Data Towards Smart
(FedCSIS), 2014, pp. 1-8. Manufacturing and Industry 4.0: 360 Degree Comparison,” IEEE
[19] F. Tao, Y. Cheng, L. D. Xu, L. Zhang, and B. H. Li, “CCIoT-CMfg: Access, vol. 6, pp. 3585-3593, 2018.
Cloud computing and Internet of Things-based cloud manufacturing [40] F. Tao, et al., “Theories and technologies for cyber-physical fusion
service system,” IEEE Trans. Ind. Inform., vol. 10, no. 2, pp. 1435- in digital twin shop-floor,” Computer Integrated Manufacturing
1442, May 2014. Systems, vol. 23, no. 8, 2017, pp. 1603-1611.
[20] H. Lee, “Framework and development of fault detection
classification using IoT device and cloud environment,” J. Manuf.
Syst., vol. 43, pp. 257-270, 2017. Qinglin Qi received the B.S. degree in computer
[21] Z. Shu, J. Wan, D. Zhang, and D. Li, “Cloud-integrated cyber- science and technology from Ocean University of
physical systems for complex industrial applications,” Mobile Netw. China, Qingdao, China, in 2012 and the M.S. degree
Appl., vol. 21, no. 5, pp. 865-878, 2016. in computer science and technology from Beihang
[22] O. Givehchi, H. Trsek, and J. Jasperneite, “Cloud computing for University, Bejing, China, in 2015, where he is
industrial automation systems—A comprehensive overview,” In currently pursuing the Ph.D. degree.
Proc. 2013 IEEE 18th Conference on Emerging Technologies & His research interests include service oriented
Factory Automation (ETFA), 2013, pp. 1-4. smart manufacturing, manufacturing service
[23] Z. X. Guo, W. K. Wong, and C. Guo, “A cloud-based intelligent management, and digital twin.
decision-making system for order tracking and allocation in apparel
manufacturing,” Int. J. Prod. Res., vol. 52, no. 4, pp. 1100-1115,
2014.
[24] X. V. Wang, and X. W. Xu, “An interoperable solution for Cloud
manufacturing,” Robot. CIM-INT Manuf., vol. 29, no. 4, pp. 232- Fei Tao received the B.S. and Ph.D. degree in
247, 2013. Mechanical Engineering from Wuhan University of
[25] D. Mourtzis, E. Vlachou, N. Xanthopoulos, M. Givehchi, and L. Technology in 2003 and 2008, respectively.
Wang, “Cloud-based adaptive process planning considering He is currently a Professor at the School of
availability and capabilities of machine tools,” J. Manuf. Syst., vol. Automation Science and Electrical Engineering,
39, pp. 1-8, 2016. Beihang University, Beijing, China.
[26] B. Buckholtz, I. Ragai, and L. Wang, “Cloud manufacturing: His current research interests include service
current trends and future implementations,” J. Manuf. Sci. Eng., vol. oriented smart manufacturing, manufacturing
137, no. 4, 040902, 2015. service management, sustainable manufacturing, and
[27] W. He, and L. Xu, “A state-of-the-art survey of cloud digital twin driven product
manufacturing,” Int. J. Comput. Integr. Manuf., vol. 28, no. 3, pp. design/manufacturing/service. He has authored 5 monographs and over
239-250, 2015. 100 journal papers in these fields.
[28] V. Gazis, A. Leonardi, K. Mathioudakis, K. Sasloglou, P. Kikiras,
and R. Sudhaakar, “Components of fog computing in an industrial
internet of things context,” In Proc. 12th Annual IEEE

VOLUME XX, 2017 9

This work is licensed under a Creative Commons Attribution 3.0 License. For more information, see http://creativecommons.org/licenses/by/3.0/.