Integration of Digital Twin and Deep Learning in Cyber-Physical Systems: Towards Smart Manufacturing

Abstract

Digital twin is gaining popularity due to its
significant impacts in bridging the gap between physical and
cyber worlds. As reported by Grand View Research, Inc., the
global market of digital twin is expected to reach $26.07 billion
by 2025 with a Compound Annual Growth Rate of 38.2%. The
growing adoption of Cyber-Physical System (CPS), Internet of
Things (IoT), big data analytics, and cloud computing in
manufacturing sector has paved the way for low cost and
systematic implementation of digital twin, with promising
impacts on a) product design and development, b) machine and
equipment health monitoring, and c) product support and
services. Successful implementation of digital twin would
increase transparency, cooperation, flexibility, resilience,
production speed, scalability, and manufacturing efficiency.
Realization of smart manufacturing requires collaborative and
autonomous interactions between sensing, networking, and
computational resources across manufacturing assets where data
is gathered from physical systems is utilized for extraction of
actionable insights and provision of predictive services. In this
paper, a reference architecture based on deep learning, digital
twin, and 5C-CPS is proposed to facilitate the transformation
towards smart manufacturing and Industry 4.0.

Full text

IET Collaborative Intelligent Manufacturing
Short Communication
Integration of digital twin and deep learning in
cyber-physical systems: towards smart
manufacturing
eISSN 2516-8398
Received on 28th January 2020
Revised 18th February 2020
Accepted on 26th February 2020
E-First on 9th March 2020
doi: 10.1049/iet-cim.2020.0009
www.ietdl.org
Jay Lee1, Moslem Azamfar1 , Jaskaran Singh1, Shahin Siahpour1
1NSF Industry/University Cooperative Research Center on Intelligent Maintenance Systems (IMS), University of Cincinnati, Cincinnati, OH, USA
E-mail: azamfamm@mail.uc.edu
Abstract: Digital twin (DT) is gaining popularity due to its significant impacts on bridging the gap between the physical and
cyber worlds. As reported by Grand View Research, Inc., the global market of DT is expected to reach $26.07 billion by 2025
with a Compound Annual Growth Rate of 38.2%. The growing adoption of cyber-physical system (CPS), Internet of Things, big
data analytics, and cloud computing in manufacturing sector has paved the way for low cost and systematic implementation of
DT, with promising impacts on (a) product design and development, (b) machine and equipment health monitoring, and (c)
product support and services. Successful implementation of DT would increase transparency, cooperation, flexibility, resilience,
production speed, scalability, and manufacturing efficiency. Realisation of smart manufacturing requires collaborative and
autonomous interactions between sensing, networking, and computational resources across manufacturing assets where data
is gathered from physical systems is utilised for the extraction of actionable insights and provision of predictive services. In this
study, a reference architecture based on deep learning, DT, and 5C-CPS is proposed to facilitate the transformation towards
smart manufacturing and Industry 4.0.
1 Introduction
Cyber-physical system (CPS) and digital twin (DT) are two
essential elements for the realisation of smart manufacturing
systems [1]. CPS enhances communication between smart
manufacturing entities (sensors, actuators, control, etc.) and cyber
computational resources to facilitate monitoring, data collection,
perception, analysis, and real-time control of manufacturing
resources. DT integrates historical and real-time data obtained from
physical systems with physics-based models and advanced
analytics to create digital counterparts with high integrity,
awareness, and adaptability to provide predictive services to
manufacturing entities. It enhances transparency and feasibility of
functions in CPS and facilitates real-time monitoring, simulation,
optimisation, and control of cyber-physical elements [2]. A DT-
based CPS (DT-CPS) should constantly acquire, integrate, analyse,
simulate, and synchronise data across multiple stages of the
product life cycle to provide on-demand predictive services to
different users in both physical and cyber spaces. The key
characteristics and requirements for integration of DT in CPS are
summarised below:
A. Ubiquitous connectivity and smart objects: Manufacturing
assets should be equipped with smart sensors with the
capability of real-time monitoring and data exchange with
other elements in the network. These constant data
transactions require a secure, reliable, and high-speed
platform.
B. Advanced analytics: It is essential to automate the whole
process of data preprocessing, perception, analysis, learning,
and execution without the need for extensive human
interference and manual feature engineering. This process
brings self-configure, self-adapt, and self-learning
functionalities to the manufacturing systems, which increases
productivity, speed, flexibility, and efficiency [3, 4].
C. Cooperative decision making: Data from multiple resources
and real-time limitations must be considered to achieve a
globally optimal solution. In this process, feasibility,
efficiently, and execution plans of different orders are
evaluated [5].
D. Autonomous and rapid model building and updates: Data
synchronisation and advanced model mapping between
virtual and physical systems guarantee the minimum
difference between virtual components and their physical
counterparts, which is essential for real-time control,
optimisation, forecast, etc.
E. Autonomous disturbance handling and resilience control:
Manufacturing systems need to autonomously and resiliently
respond to failures in order to prevent catastrophic
operational disruptions.
Deep learning (DL) is part of a broader family of machine learning
(ML) methods that have the capability to use raw data and
automatically provide the representations required for various
applications such as classification, regression, clustering, and
pattern recognition. DL is very powerful in discovering complex
structures in high-dimensional data and therefore, it has enormous
applications in the manufacturing domain. It allows higher levels
of abstraction without manual feature engineering and its high
performance has been validated in different domains such as
speech recognition, image processing, inventory management, and
fault detection and diagnosis.
In [6], a CPS structure (5C-CPS) was proposed that supports
plug and play smart sensor connectivity, was layered,
interconnected, and enables cyber-physical smart interactions. Built
upon 5C-CPS, an integrated model based on DT and DL is
proposed (DTDL-CPS) in this work to facilitate and realise smart
manufacturing. This architecture brings autonomous, cooperative,
self-adaptive, self-learning, and self-maintain functionalities to the
5C-CPS design and enhances transparency, resilience, and
manufacturing efficiency.
Some of the emerging technologies such as 5G, industrial AI,
blockchain, and transfer learning could significantly facilitate the
realisation of DT in manufacturing systems. Briefly, 5G reduces
the communication latency, enhances network capacity, reliability
and throughput [7]. Industrial AI [3] brings self-aware, self-adapt,
and self-configure functionalities to the system and facilitate the
implementations of DTs. Blockchain could help in enhancing peer-
to-peer interactions and removing intermediaries in the networking
process [5]. Transfer learning as discussed further in part D of
IET Collab. Intell. Manuf., 2020, Vol. 2 Iss. 1, pp. 34-36
This is an open access article published by the IET under the Creative Commons Attribution License
(http://creativecommons.org/licenses/by/3.0/)
34

Section 2, would facilitate model building and rapid
implementation of DTs.
Companies based on their needs/priorities may start with the
implementation of critical technologies and gradually incorporate
other technologies into their manufacturing systems. The proposed
architecture would help in identifying the potential applications of
different technologies that could be gradually incorporated in the
operational systems for the realisation of the full potentials of
smart manufacturing.
2 DTDL-CPS for smart manufacturing
As discussed in Section 1, a smart DT-CPS structure requires five
main characteristics. In this section, a DTDL-CPS architecture
(Fig. 1) is proposed to provide a step-by-step guideline for the
realisation of these five functionalities. A sequential workflow is
presented that clearly defines how to construct DTDL-CPS from
data acquisition to DL for automatic analysis and virtual model
construction, and to provide on-demand services. The detailed
DTDL-CPS architecture is outlined as follows:
A. Smart connection: Data is the core element for DT-based
CPS applications. It is acquired from the manufacturing shop
floor through smart sensors, actuators, controllers, etc., or
from manufacturing enterprise systems such as Enterprise
Resource Planning, Supply Chain Management,
Collaborative Manufacturing Management, and Customer
Relationship Management. However, these factors are
important for consideration: 1 – data synchronisation
between different resources, 2 – low latency data transfer to
the cloud, 3 – data security and privacy, 4 – reliability in data
transfer, 5 – low energy consumption in data transfer, and 6 –
interconnectivity between different smart objects. It is
expected that advanced communication technologies such as
5G would satisfy many of these requirements in the
manufacturing domain. 5G technology provides
communication and data transfer with ultra-high-speed, high
reliability, low energy consumption, and advanced security.
Moreover, the adoption of blockchain technology and micro-
clouds as a supplementary tool would further enhance smart
connection functionalities [5].
B. Data to information conversion: In this layer, a computation
resource such as Edge or Fog computing might be utilised to
Fig. 1 DTDL-CPS architecture for smart manufacturing systems (red arrows represent the data/information flow)
Fig. 2 Autonomous big data analytics with DL
Fig. 3 DL enabled model development and update
IET Collab. Intell. Manuf., 2020, Vol. 2 Iss. 1, pp. 34-36
This is an open access article published by the IET under the Creative Commons Attribution License
(http://creativecommons.org/licenses/by/3.0/)
35

1 – reduce the communication latency between smart objects
and cloud, 2 – preprocess data and extract required
information before sending data to the cloud, 3 – manage
interaction between smart objects in the connection layer,
and 4 – execute control orders immediately to enhance
manufacturing resilience and disturbance rejection. ML
algorithms with a medium level of computational
requirements and training time are suitable for data to
information conversion at this level.
C. Big data analytics: This layer is a centre for data storage and
advanced analytics. Manufacturing big data has three major
types, namely, 1 – structured, 2 – unstructured, and 3 – semi-
structured with unstructured data makes up to 95% of big
data [8]. DL techniques for text/image mining and signal
processing can significantly enhance knowledge discovery
from unstructured data. Moreover, DL extracts meaningful
information without the need for manual feature engineering.
Therefore, it can significantly improve autonomous
functionalities such as real-time control, optimisation, and
prognostic and health management (PHM). As is shown in
Fig. 2, a portion of real-time and historical data from both
physical systems and their virtual counterparts are fused
together and utilised for training DL models, another portion
of data is used for testing the trained model. If the
performance of the trained model is not acceptable, it is
retained until the testing results satisfy the performance
criteria. Then, the trained model is used for testing the real-
time data. Finally, after each iteration and real-time
assessment, the historical model is updated. The proposed
DL framework helps in automatic detection of data/model
drift and updating the trained DL models autonomously. This
layer also provides the necessary analytics and data for the
Virtual system layer.
D. Virtual systems: Digital Models are the most important
component in this layer to mirror the lives of their physical
counterparts. They are constantly created, monitored,
evaluated, and updated to provide services at speed close to
real time. Fig. 3 shows the proposed framework for the
development and update of a digital model. It consists of two
major parts 1 – model selection and building, 2 – model
performance monitoring, and update. For model selection,
different AI tools such as a genetic algorithm could be used
to find the most similar model from a library of already
developed models. If there was no similar model, then a
model is built with the help of advanced DL models, such as
transfer learning. Using transfer learning speeds up model
building and increases efficiency by using knowledge/data
from a base model. For models in operation, a learning agent
constantly monitors the performance of both actual and
digital models to guarantee minimal difference between
them. If any changes happen in the physical model, the
learning agent will update the parameters of the digital model
accordingly.
E. Service layer: Implementation of DTDL-CPS upon this level
creates a highly synchronised and smart CPS where
intelligence, control, visualisation, optimisation, PHM, etc.
are offered as services for implementation in the physical
system. These services cover a variety of applications, such
as proper settings for machines, orders to inventory for
product purchase, controlling production job order,
optimisation of production layout, and enhancing peer-to-
peer interactions. Visualisation as a Service facilitates users’
decision-making process.
Any request for a new service is automatically evaluated and
executed through the interconnectivity between different layers of
DTDL-CPS. Feedbacks from the physical system help in refining
services and optimisation in a semi-real-time fashion.
3 Case study
Implementation of the proposed concept for a convolutional neural
networks (CNC) machine tool utilised in a shop floor is shown in
Fig. 4, where DT and DL are used to realise self-maintainability,
self-configurability, and predictive services. DL has been adopted
to reduce dependency on the human for decision making and
provides autonomous functionalities with repeating and consistent
successes. DT is used to increase transparency in the whole
operational network. In such a design, predictions on the
Remaining Useful Life of different components are passed to the
virtual system layer wherein the real-time and historical data from
machines, inventory, suppliers, customers, and maintenance
technicians are integrated into DT models to simulate and optimise
the predictive maintenance strategies. The optimised plan, along
with supplementary visualised information, is sent back to the
physical system for implementation. This process is repeated in a
semi real-time fashion and all models work autonomously and
collaboratively to create a closed-loop predictive maintenance
framework.
4 Conclusion
This paper presents a reference architecture for the integration of
DL and DT in the 5C-CPS structure. It provides a practical
guideline for the development and realisation of smart
manufacturing with enhanced transparency, cooperation,
networking, resilience, and efficiency. It is envisioned that new
technologies such as 5G will significantly contribute to the
integration of DL and DT in the CPSs. Meanwhile, the design and
operation of manufacturing systems, sensors, actuators, and
components should be improved and gradually equipped with
proper AI agents for fast, efficient, and reliable operations and
networking.
5 References
[1] Lee, J., Lapira, E., Bagheri, B., et al.: ‘Recent advances and trends in
predictive manufacturing systems in big data environment’, Manuf. Lett.,
2013, 1, (1), pp. 38–41
[2] Tao, F., Qi, Q., Wang, L., et al.: ‘Digital twins and cyber–physical systems
toward smart manufacturing and industry 4.0: correlation and comparison’,
Engineering, 2019, 5, (4), pp. 653–661
[3] Lee, J., Singh, J., Azamfar, M.: ‘Industrial artificial intelligence’, arXiv Prepr.
arXiv, 1908.02150, 2019
[4] O'Donovan, P., Leahy, K., Bruton, K., et al.: ‘An industrial big data pipeline
for data-driven analytics maintenance applications in large-scale smart
manufacturing facilities’, J. Big Data, 2015, 2, (1), p. 25
[5] Lee, J., Azamfar, M., Singh, J.: ‘A blockchain enabled cyber-physical system
architecture for industry 4.0 manufacturing systems’, Manuf. Lett., 2019, 20,
pp. 34–39
[6] Lee, J., Bagheri, B., Kao, H.A.: ‘A cyber-physical systems architecture for
industry 4.0-based manufacturing systems’, Manuf. Lett., 2015, 3,
(December), pp. 18–23
[7] Shi, Y., Han, Q., Shen, W., et al.: ‘Potential applications of 5G
communication technologies in collaborative intelligent manufacturing’, IET
Collab. Intell. Manuf., 2019, 1, (4), pp. 109–116
[8] Gu, X., Jin, X., Ni, J., et al.: ‘Manufacturing system design for resilience’,
Procedia CIRP, 2015, 36, pp. 135–140
Fig. 4 DT enabled predictive maintenance framework for a CNC machine
tool
36 IET Collab. Intell. Manuf., 2020, Vol. 2 Iss. 1, pp. 34-36
This is an open access article published by the IET under the Creative Commons Attribution License
(http://creativecommons.org/licenses/by/3.0/)