Access to this full-text is provided by IGI Global Scientific Publishing.
Content available from Journal of Organizational and End User Computing
This content is subject to copyright. Terms and conditions apply.
DOI: 10.4018/JOEUC.336482
Volume 36 • Issue 1
This article published as an Open Access article distributed under the terms of the Creative Commons Attribution License
(http://creativecommons.org/licenses/by/4.0/) which permits unrestricted use, distribution, and production in any medium,
provided the author of the original work and original publication source are properly credited.
*Corresponding Author
1
Xiangqian Wang, East China Normal University, China & Pingdingshan University, China
Haifeng Hu, Pingdingshan University, China*
Yuyao Wang, Lamar University, USA
Zhaoyu Wang, Fujian Normal University, China
Conventional automobile manufacturing plants involve intricate assembly, testing, and debugging
processes heavily reliant on manual operations. This study aims to explore the application of industrial
internet of things (IIoT) and deep learning algorithms to achieve process automation in manufacturing.
Firstly, utilizing IIoT technology, OPC UA, and point cloud fitting techniques, a comprehensive
modeling of most equipment and materials within the factory is conducted, constructing a digital
twin (DT) model as a virtual representation of actual equipment. Subsequently, the study innovatively
introduces the deep Q network algorithm, facilitating the automatic transition of the production
process and improving production efficiency. Through comparison with ten baseline models, the
proposed model demonstrates an improvement in production efficiency of at least four percentage
points compared to other models. Experimental validation confirms the effectiveness of the proposed
model in the smart factory for electric vehicle manufacturing.
Digital Twin, DQN, IIoT, Process Automation, Smart Factory, Smart Manufacturing
The establishment of intelligent factories has emerged as a significant global trend in the manufacturing
sector, aimed at enhancing production efficiency, reducing costs, and achieving more flexible and
sustainable manufacturing processes through the adoption of advanced technologies and digital
solutions. Illustrative construction cases, such as the intelligent manufacturing transformation
implemented by China’s Haier Corporation, which involved technologies like the Internet of Things
(IoT), cloud computing, and big data analytics, have resulted in the development of intelligent
home appliance manufacturing facilities. This intelligent factory construction project has elevated
the f lexibility and adaptability of production lines, facilitated customized manufacturing, reduced
product time-to-market, and strengthened market competitiveness. The process automation of electric
Volume 36 • Issue 1
2
vehicle manufacturing factories is currently a highly prominent research direction in the intelligent
manufacturing landscape (Bathla et al., 2022). The investigation into this technology encompasses
various aspects such as the enhancement of factory production efficiency, optimization of production
resource utilization (Zhang & Dilanchiev, 2022), and the promotion of environmentally friendly
manufacturing (Yang et al., 2022). First, it is poised to significantly enhance production efficiency.
Traditional automotive manufacturing plants involve intricate assembly, testing, and debugging
processes heavily reliant on manual operations. The introduction of automation technologies (Li et
al., 2022b), such as robots and intelligent assembly lines, can substantially reduce the time devoted
to manual operations and elevate the operational efficiency of production lines. The improvement
in production efficiency aids in reducing manufacturing costs, facilitating quicker and more flexible
production responses to rapidly changing market demands. Secondly, it is expected to contribute
to the improvement of the quality and consistency of electric vehicles (Jiménez‐Ramírez et al.,
2023). Given the intricate nature of the components and systems in electric vehicles, errors during
manufacturing can result in a decline in product quality. The introduction of automated processes
can mitigate human errors and variations, ensuring more precise manufacturing and consequently
enhancing overall product quality and consistency. Thirdly, it holds the potential to reduce energy
consumption (Salman et al., 2022) and carbon emissions (Kumar et al., 2022). Automation technologies
enable the optimization of production processes, precise control of energy usage, and the reduction
of unnecessary waste, thereby rendering the manufacturing process more environmentally friendly
and aligning with the overall eco-friendly philosophy of electric vehicles. Fourthly, it aids in driving
the digitization transformation of the manufacturing industry (Favoretto et al., 2022). Through
the integration of advanced technologies such as intelligent manufacturing and big data analytics,
production shop floors can achieve higher levels of digital management and monitoring. This not
only facilitates real-time tracking of production processes and optimization of resource allocation but
also enhances production efficiency through data analysis, providing a more scientifically grounded
basis for business decision-making. Therefore, research into the automation of processes in electric
vehicle manufacturing plants holds crucial reference value for a nation’s industrial strategy formulation.
With the increasing global attention on the electric vehicle industry, an in-depth investigation into the
application of artificial intelligence technologies in electric vehicle manufacturing can provide robust
support for the development of national industrial policies (Srivastava et al., 2022). On a global scale,
this contributes to elevating a nation’s industrial competitiveness and strengthening its technological
leadership in the field of electric vehicles.
Currently, deep learning technology has found numerous applications in the automation of factory
production processes (Tercan & Meisen, 2022), and these innovative applications have profound
impacts on enhancing production efficiency (Salman et al., 2022), improving product quality, reducing
costs, and driving digital transformation. Due to its capability to learn and comprehend vast amounts of
production data, deep learning enables intelligent decision-making, resource optimization, and waste
reduction through real-time monitoring and analysis of data on the production line, thereby achieving
a higher level of production efficiency. Deep learning empowers production systems with autonomous
learning and adaptability, fostering the progression of factories toward intelligent manufacturing.
This enhances the autonomy and adaptability of production systems, reducing dependence on human
intervention and achieving a more highly automated production process (Zhou et al., 2022a).
Given that deep learning technology allows real-time monitoring and control of product quality
through the analysis of sensor data and image recognition, it is often utilized to predict potential quality
issues, take preemptive measures, reduce defect rates, and enhance the stability of product quality.
At the level of each module within a factory, deep learning technology holds significant potential in
energy management subsystems. By intelligently adjusting equipment operation based on the analysis
of energy usage data in the production process, the system can optimize energy utilization, reduce
energy consumption, and achieve a more environmentally friendly and sustainable manufacturing
process (Musbah et al., 2022).
Volume 36 • Issue 1
3
Furthermore, deep learning technology contributes to monitoring and analyzing the safety
conditions within a factory (Moradi et al., 2022). By real-time identification of hazards and monitoring
employee behavior, the system can preemptively alert potential safety risks, thereby ensuring the
safety of the factory’s production processes. In summary, the application of deep learning technology
in factory production processes not only injects new vitality into traditional manufacturing but also
provides innovative directions for the future of manufacturing. Through the integration of advanced
technologies such as big data and cloud computing, it becomes possible to construct a more
intelligent, flexible, and sustainable manufacturing system, propelling the manufacturing industry
towards digitization and intelligence. The current commonly employed deep learning models for the
construction of smart factories include:
1. Convolutional Neural Network (CNN): Primarily used for image recognition and processing,
it is suitable for handling visual information in factories. In smart factories, CNNs are employed
for tasks such as product quality inspection, defect identification, and real-time monitoring on
production lines. Their advantage lies in their ability to extract features from images, enabling
efficient and accurate visual analysis (Hsu et al., 2022).
2. Recurrent Neural Network (RNN): Suited for processing sequential data, such as time series or
continuous data in processes. In smart factories, RNNs can be utilized for predicting equipment
failures, detecting anomalies in production lines, and modeling dynamic changes during the
production process. Their advantage lies in possessing memory capabilities, allowing them to
handle data with strong temporal dependencies (Kannen & Subasi, 2023).
3. Long Short-Term Memory (LSTM): A specialized type of RNN designed for handling long
sequential data, addressing the issues of vanishing, and exploding gradients in traditional RNNs.
In smart factories, LSTMs find applications in modeling time series data, such as predicting
equipment performance and optimizing energy consumption. Their advantage lies in better
capturing long-term dependencies (Wahid et al., 2022).
4. Generative Adversarial Network (GAN): Mainly used for generating new data samples,
commonly employed for data augmentation and synthesis. In smart factories, GANs can
be used to simulate production environments, generate virtual data for model training, and
enhance the generalization performance of models. Their advantage lies in generating realistic
data (Zhou et al., 2022b).
5. Reinforcement Learning (RL): Employed for decision-making, RL learns optimal strategies
through interaction with the environment. In smart factories, RL can be applied to optimize
production scheduling, formulate equipment control strategies, and optimize resource allocation.
Its advantage lies in its ability to autonomously learn and adapt to complex production
environments (Lei et al., 2023).
This study is aimed at exploring the application of Industrial Internet of Things (IIoT) (Gupta et
al., 2022) and deep learning algorithms to achieve production process automation in an established
Chinese smart factory for electric vehicles. The design rationale of the proposed method encompasses
several key steps. Firstly, leveraging IIoT technologies, OPC UA (Domínguez et al., 2022), and
point cloud fitting techniques (Fan & Zhang, 2022) to model most devices and materials within the
factory. This initial step aims to comprehensively model factory equipment by integrating IoT-based
advanced physical information gathering and Poisson surface reconstruction-based three-dimensional
point cloud technologies, constructing a digital twin (DT) model (Wang et al., 2022b). Secondly, by
using a specific mapping algorithm, the DT model serves as a virtual mapping of the actual devices,
providing real-time and highly accurate references for subsequent production process automation.
Utilizing DT technology, the factory undergoes equipment behavior modeling and real-time monitoring
of equipment status (Nie et al., 2021). By monitoring and analyzing real-time data from the digital
twin model, the system can accurately simulate and predict equipment behavior, facilitating real-time
Volume 36 • Issue 1
4
monitoring of equipment status within the factory. This capability offers crucial support for achieving
refined management and an immediate response to potential issues. Finally, within the Manufacturing
Execution System (MES) (Shojaeinasab et al., 2022), the innovative introduction of the deep Q-network
algorithm (Zeng et al., 2022) facilitates the automated transformation of the production process
and enhances production efficiency. The incorporation of the deep Q-network algorithm enables
the system to optimize decisions within the production process, realizing autonomous control and
optimization of the production process. This innovative approach provides robust technical support
for the intelligent production of the smart factory.
This study presents a novel solution through the integration of DT technology and deep learning
algorithms for the automation of production processes in a Chinese electric vehicle smart factory.
There are three main innovations:
1. Comprehensive Application of DT Models: The system innovatively employs a comprehensive
approach using Industrial Internet of Things (IIoT) technology, OPC UA, and point cloud fitting
techniques to extensively model the equipment and materials within the factory, constructing a
DT model. This integrated application establishes a highly accurate mapping relationship between
the virtual model and the actual devices. The DT model not only facilitates real-time monitoring
of device status but also provides a real-time and accurate reference for subsequent production
process automation, thus laying a solid foundation for the intelligence of the factory.
2. Application of DT Technology in Equipment Behavior Modeling and Real-Time Monitoring:
Through the application of DT technology, the system engages in equipment behavior modeling
and real-time monitoring of equipment status within the factory. The innovation lies in the system’s
ability to accurately simulate and predict equipment behavior by monitoring and analyzing real-
time data from the DT model. This capability provides crucial support for achieving refined
management and immediate responses to potential issues.
3. Introduction of the Deep Q Network Algorithm in the Manufacturing Execution System:
The introduction of the deep Q network algorithm into the Manufacturing Execution System
(MSE) represents the third innovation in this system. This algorithm innovatively achieves
the automation transformation of the production process and enhances production efficiency.
Through the deep Q network algorithm, the system can learn and optimize decisions within the
production process, realizing autonomous control and optimization. This autonomous learning
and optimization capability provides robust technical support for the intelligent production of
the smart factory, offering valuable insights for the future development of industrial intelligence.
The proposed methodology demonstrates a thoughtful and practical approach, offering valuable
insights for the intelligent transformation of the industrial sector.
This article is organized as follows: We will introduce the recently related work in Section
2. Section 3 presents the proposed methods: overview, digital twin modeling for factories based
on OPC UA and point cloud fitting, real-time monitoring of factory status based on digital twins,
automation of production processes based on digital twins, and deep learning. Section 4 introduces
the experimental part, including practical details, comparative experiments, and a case study. Section
5 includes a conclusion and an outlook.
The application of IIoT and DT technology in smart manufacturing injects new vitality into modern
manufacturing, offering enterprises a more efficient, intelligent, and sustainable production approach.
The deep integration of information technology and physical systems, as exemplified by the IIoT,
Volume 36 • Issue 1
5
facilitates interconnectivity among devices. Through real-time data collection from sensors, equipment,
and process flows, IIoT establishes a production environment characterized by real-time monitoring
and control. DT technology further enriches this concept by virtually representing physical systems,
creating a digital counterpart known as a DT model. The establishment of a DT model typically
involves digitizing, connecting, and continuously updating representations of physical equipment.
The entire application process can be delineated into steps such as data collection, data transmission,
data processing, and model updates, ensuring synchronization between the DT and the actual physical
systems (Chen et al., 2023a).
DT technology brings multiple advantages to smart manufacturing. Firstly, real-time monitoring
and data analysis empower manufacturing enterprises to swiftly respond to changes in the production
environment, enhancing production line flexibility and adaptability. Secondly, through the combined
use of IIoT and DT technology, enterprises can achieve remote monitoring and maintenance of
equipment, reducing downtime and maintenance costs. The comprehensive understanding of
equipment status provided by the DT model aids in preemptively addressing potential faults and
making informed decisions for intelligent maintenance. Currently, DT technology finds widespread
applications across various facets of the manufacturing industry. In production planning and
scheduling, enterprises can intelligently optimize the allocation of production resources through
digital modeling and real-time monitoring of the entire production process. In quality control, the
high-precision simulation of production processes by DT models facilitates real-time monitoring
and prediction of product quality, thereby enhancing the stability of product quality. Regarding
equipment maintenance, the integration of sensor data collection from IIoT and model updates from
DT technology enables remote monitoring and intelligent maintenance of equipment, reducing
downtime and improving equipment utilization rates.
Furthermore, in the development of smart factories, OPC UA (Open Platform Communications
Unified Architecture) technology is extensively applied, offering an efficient, secure, and highly
interoperable solution for industrial automation systems. Functioning as a communication protocol,
OPC UA facilitates seamless integration among diverse devices and systems in smart factories through
a unified information model and standardized data exchange mechanisms. In practical application
scenarios, OPC UA supports various communication mechanisms, including publish-subscribe and
request-response, enabling the collaborative operation of devices from different manufacturers and
enhancing the coordination and flexibility of the production process. Moreover, OPC UA provides
robust security mechanisms encompassing encryption and authentication, ensuring the confidentiality
and integrity of data, and effectively addressing the escalating network security threats in smart
factory environments. Simultaneously, OPC UA supports the standardization of information models
in smart factories, enabling devices and systems to share consistent data structures and semantics,
thereby simplifying the complexity of data interpretation and integration. This standardization not
only enhances system maintainability but also reduces the costs associated with system integration.
DT technology demonstrates outstanding application prospects in the modeling of production processes
in smart factories. Through the utilization of DT technology, enterprises can achieve a highly accurate
reproduction and simulation of physical production systems, providing real-time and comprehensive
digital representations of production processes. This capability enables enterprises to conduct real-
time monitoring, analysis, and optimization of production processes. Additionally, DT models can
be employed to simulate various production scenarios, offering decision-makers comprehensive
data support. The modeling process of DT typically includes steps such as digital representation,
connecting the model to the physical world, and real-time data updates to ensure synchronization
between the model and the actual production system. Initially, digital representation involves digitizing
physical elements such as actual equipment and process flows to construct a virtual DT model. The
connecting phase involves utilizing Industrial Internet of Things (IIoT) technology to establish a
Volume 36 • Issue 1
6
connection between the DT model and actual equipment, collecting real-time data on equipment
operations, production parameters, and more. The real-time update phase involves continuously
updating the DT model through feedback mechanisms to maintain synchronization with the actual
production process. This modeling approach enables DT models to promptly reflect changes in actual
production, providing decision-makers with an accurate data foundation.
However, there are some limitations in the current application of DT technology. Firstly, there
are constraints related to data quality and real-time capabilities. Building accurate and trustworthy DT
models necessitates high-quality production data that needs to be continuously updated to maintain the
model’s authenticity. In certain circumstances, due to limitations in data collection and transmission,
DT models may not fully and accurately reflect the actual production status, potentially impacting
decision-making accuracy. Furthermore, the implementation of DT technology requires significant
technological investment and specialized knowledge. Establishing DT models involves knowledge
from multiple domains, including IoT technology, data analytics, simulation modeling, and more.
For some small and medium-sized enterprises, this may result in high costs and resource-intensive
requirements (Li et al., 2022a).
The application of intelligent production planning algorithms in modern manufacturing holds the
promise not only to enhance production efficiency and optimize resource utilization but also to reduce
costs, providing significant competitive advantages for businesses. Common intelligent production
planning algorithms include genetic algorithms, simulated annealing, particle swarm optimization,
and ant colony optimization. From existing research, these algorithms demonstrate outstanding
performance, particularly in precision and real-time capabilities. By integrating advanced data
collection techniques and big data analytics, these algorithms can monitor and analyze key parameters
in the production process in real-time, enabling rapid responses to change in market demands and
fluctuations in the manufacturing environment. Furthermore, these algorithms exhibit intelligent and
personalized production planning capabilities. Through technologies like deep learning and machine
learning, algorithms can learn from historical data, predict market trends, and formulate more flexible
and efficient production plans, thereby improving resource utilization and efficiency. Additionally,
intelligent production planning algorithms contribute to optimizing supply chain management,
ensuring smooth circulation of raw materials, semi-finished goods, and finished products, reducing
inventory costs, and enhancing product delivery efficiency (Zhou et al., 2022a).
However, there are notable drawbacks to the application of intelligent production planning
algorithms. Firstly, the implementation of these algorithms may require substantial data and
technological investments. Obtaining high-quality production data and maintaining complex
algorithmic systems could result in high construction costs compared to potential profits. Secondly,
the robustness and stability of these algorithms need further improvement. In complex and dynamic
manufacturing environments, algorithms may be susceptible to noise and outliers, leading to
instability in production planning. Therefore, further research and innovation in intelligent production
manufacturing process automation algorithms are necessary to overcome these challenges and further
propel the development of smart manufacturing.
In the automated production factory of electric vehicles, the design and implementation of the core
software system cluster are crucial for improving production efficiency, optimizing resource utilization,
and ensuring product quality. This cluster typically includes multiple subsystems, and the systems
discussed in this article encompass the following key subsystems:
Volume 36 • Issue 1
7
1. Production planning and scheduling subsystem (Oluyisola et al., 2022): This forms the
foundation of the core software system, including the development of Manufacturing Execution
Systems (MES), Advanced Planning and Scheduling Systems (APS), Energy Management
Systems (EMS), Predictive Health Management Software (PHM), and Maintenance, Repair, and
Operations (MRO) software. This component is responsible for translating market demands and
ordering information into specific production plans and optimizing the allocation of production
resources through intelligent scheduling algorithms. The production planning and scheduling
system needs to be closely integrated with the supply chain management system to ensure accurate
supply of raw materials and components, thereby ensuring the continuous and stable operation
of the production line. Among these, the Manufacturing Execution System (MES) serves as the
central control of the entire production process. The MES system, connected to equipment control
systems and sensor networks on the production line, achieves real-time monitoring, control,
and data collection of the production process. This system has a high level of automation and is
capable of adjusting production parameters, monitoring equipment status, and providing timely
feedback to the Production Planning and Scheduling System, achieving intelligent and adaptive
production processes.
2. Enterprise management subsystem: This forms the foundation of factory management and
includes the development of Enterprise Resource Planning Systems (ERP), Supply Chain
Management Systems (SCM), Customer Relationship Management Systems (CRM), Human
Resources Management (HRM), Quality Management Systems (QMS), Asset Performance
Management Systems (APM), and other software.
3. Data analysis and big data subsystems (Wang et al., 2022a): This is an emerging
component in the core software system. By collecting, storing, and analyzing vast amounts
of data generated during the production process, the factory can gain profound insights,
optimize production processes, and improve production efficiency. Data analysis and big
data technologies also provide predictive maintenance capabilities for smart factories. By
analyzing equipment operational data, potential faults can be identified in advance, reducing
downtime and improving equipment availability.
4. Design subsystem: This subsystem is essential for ensuring the factory can manufacture
high-quality products. It includes model libraries for automobile design, process libraries,
basic knowledge libraries, and comprehensive optimization software for the entire process in
the automobile industry. It also encompasses the integrated platform software for the design,
production, and operation/maintenance of automobiles, as well as comprehensive control platform
software for automobile production. Additionally, it includes some generic software such as
Computer-Aided Design (CAD), Computer-Aided Engineering (CAE), Computer-Aided Process
Planning (CAPP), Computer-Aided Manufacturing (CAM), Electronic Design Automation (EDA),
Product Data Management (PDM), and others.
5. Quality management subsystem: This subsystem is a crucial component to ensure the quality
of electric vehicle production. The system conducts quality inspection and control at various
nodes in the production process, ensuring that products meet standard requirements. The Quality
Management System collaborates with the MES system to respond in real-time to any anomalies
in the production process, preventing an increase in defective rates.
6. Human-Machine Interaction (Bathla et al., 2022) Subsystem: This subsystem is vital for
presenting information from the entire core software system to operators in an observable and
comprehensible manner. The HMI subsystem, through intuitive graphical interfaces, displays key
information such as production plans, equipment status, and quality data, facilitating operators
in making rapid decisions.
A detailed system architecture is shown in Figure 1.
Volume 36 • Issue 1
8
This system employs OPC Unified Architecture (OPC UA) technology for real-time data collection
in the factory and constructs a digital twin model based on point cloud fitting. The entire process can
be divided into the following detailed steps: The first step involves clearly defining the objectives
and scope of shop floor digital twin modeling, including the equipment, processes to be modeled,
and the required real-time monitoring and control information. The key to this step is ensuring
clear modeling objectives, which aid in the subsequent design and implementation of the system. In
the second step, deploy OPC UA servers within the shop floor to ensure effective communication
between the servers and various types of equipment in the shop floor. Configure device interfaces to
enable the OPC UA server to obtain real-time data from the equipment, including but not limited to
temperature, humidity, and equipment status. The third step entails deploying laser scanners or other
3D sensors on the equipment in the shop floor to capture point cloud data from the surfaces of the
devices. This step needs to cover all equipment that requires modeling to ensure comprehensive 3D
information is obtained. In the fourth step, process the collected point cloud data, removing noise and
outliers. Subsequently, apply point cloud fitting algorithms to fit the processed point cloud data into
highly accurate 3D models. This system utilizes a Poisson surface reconstruction (PSR) algorithm
to achieve high-precision fitting of shop floor equipment:
M V E F=
( )
, , (1)
where
V
denotes points,
E
denotes edges, and F denotes faces. Firstly, the objective implicit function
F x
( )
is computed so that the gradient of F x
( )
at each
P
point is the normal vector
V
at that
point, and the dispersion is taken to get the Poisson equation:
∇ ⋅ ∇
( )
− ∇ ⋅ = ⇔ = ∇ ⋅F V F V0∆ (2)
The F x
( )
is represented by adaptive octree, and the marching cube is applied to extract the
isosurfaces of the function, and the resulting Mesh mesh data contains topological information,
domain information.
Figure 1. A detailed architecture of this smart manufacture system
Volume 36 • Issue 1
9
Due to occlusion, the point cloud data obtained from a certain angle is incomplete. In this study,
a machine learning algorithm is used to predict the set of points at the occluded location to maximize
the possible information complementation of the modeled object. The fifth step involves integrating
real-time data collected by OPC UA with the 3D models obtained through point cloud fitting to
construct the digital twin model of the shop floor. By mapping real-time data to the corresponding
equipment models, the digital twin model can accurately reflect real-time information on the status
and operating parameters of the equipment on the shop floor. Perform validation of the digital twin
model by comparing it with the actual scene and checking for accuracy. If discrepancies are identified,
make appropriate optimizations and adjustments to ensure the digital twin model accurately reflects
the actual shop floor situation. In the sixth step, achieve real-time synchronization between the digital
twin model and the actual shop floor equipment using OPC UA technology. Feed real-time data,
such as equipment status and production parameters, back into the digital twin model to maintain
its real-time nature. Simultaneously, if changes occur in the shop floor equipment, such as adding
new equipment or adjusting equipment positions, use point cloud fitting technology to model new
equipment or refit adjusted equipment, ensuring a high degree of synchronization between the digital
twin model and the actual shop f loor equipment. In the final step, utilize the constructed digital twin
model to develop various application scenarios, such as equipment status monitoring, production
process simulation, and fault diagnosis. Through the digital twin model, achieve comprehensive
monitoring and intelligent control of the shop floor production process, thereby enhancing production
efficiency and quality. The principle of this step is shown in figure 2.
This section describes the mapping algorithms for mapping the binary state, enumerated state, and
numeric variable state of devices within the smart factory from the physical world to the digital twin
model, respectively.
1. Binary states involve devices with Boolean indicator values that can only be true or false. The
mapping algorithm for binary states can be expressed by the following equation:
BiIND IND Val n True False n SampleN
IND
=
∈
{ }
∀ ∈
{ }
＃, , ,0 (3)
Figure 2. The principle of digital twin modeling by using OPC UA and point cloud fitting
Volume 36 • Issue 1
10
In this equation, BiIND represents the set of Boolean indicators and
Va
l n
IND [ ] is an indicative
value of the
BiIND
is at the sample time
n
, and the SampleN is the total number of sampling
points of
BiIND
. The indicated value can only be true or false.
2. Enumerating states involves devices with enumerable indication values, and all possible values
form a finite set of states. The mapping algorithm can be expressed in the following equation:
EnumIND IND Val n StatusSet s s s n
IND IND mIND
=
∈ = …
{ }
∀ ∈＃1 2
, , , , 00,SampleN
{ }
(4)
In this equation,
EnumIND
represents the set of enumerable indicators,
Va
l n
IND [ ] is the indicated
value of the
EnumIND
is at sample time 𝑛, and StatusSetIND is the finite set of all possible states
of the
EnumIDN
. The indicated values can have states different from the finite set.
3. Numerical variable states involve devices with numerical indicator values, such as a robotic arm
with rotatable joints. The mapping algorithm can be expressed in the following equation:
NumIND IND f Val n Val n Values n Sa
IND IND IND
= =
( )
∈ ∀ ∈, ,0 mmpleN
{ }
(5)
In this formulation,
NumIND
denotes the set of numerical indicators, f·
( )
denotes the mapping
method of the 𝑗th
NumIND
, and
Va
luesIND is the data range in the physical device side. The mapping
algorithm maps the data on the physical device side to the
NumIND
values on the digital twin side.
For the data obtained after mapping, in order to prevent anomalies in the data collected by the
IoT subsystem from the physical world, this system has devised anomaly detection algorithms. These
algorithms are employed to analyze and identify abnormal samples from the input time-series data
originating from the physical world. In the presence of abnormal samples, the system will choose to
either eliminate or re-sample the data.
The whole mapping relationship from the physical world to the DT model is shown in Figure 3.
This smart factory uses a deep-Q-network (DQN) algorithm for production process decision
optimization in the MES system, which is based on the principle of training an agent to make near-
optimal production decisions in real time. Incorporating the DQN algorithm for production process
scheduling in smart factories brings numerous advantages. Firstly, the DQN algorithm enables
intelligent scheduling in complex and dynamic production environments through the learning and
optimization of decision-making strategies. Secondly, the DQN algorithm can dynamically adapt
to changes in production, enhancing the flexibility and adaptability of the manufacturing process
by continuously learning and optimizing, thereby more effectively addressing uncertainties and
fluctuations in production. Additionally, the DQN algorithm maximizes production efficiency by
optimizing resource utilization and task allocation, leading to reduced production costs and improved
output quality.
The DQN algorithm is a deep reinforcement learning (RL) algorithm (Pengcheng et al., 2022)
that uses a neural network called Q-network to approximate the optimal action-value function
Volume 36 • Issue 1
11
Q s a,
( )
.
s
represents the current state of the factory and
a
represents the action to be taken. This
system
s
is set as an N-tuple:
s=
{workshop 1 production rate, workshop 2 production rate,
workshop 3 production rate...workshop N production rate}. And
a
represents the action to be
taken, and the set of actions is AÎ {workshop 1 increase production rate by 1 unit, workshop 1
decrease production rate by 1 unit, workshop 2 increase production rate by 1 unit, workshop 2
decrease production rate by 1 unit, workshop N increases production rate by 1 unit, workshop N
decreases production rate by 1 unit}.
The DQN algorithm uses two key strategies to improve performance: experience replay and fixed
parameters. Experience replay involves storing past experiences in replay memory, which is then used
to randomly sample and train the Q network. This helps to break the correlation between successive
experiences and improves the stability of learning. Fixed parameters are used to categorize the Q
network into two versions: the online Q network and the target Q network. The online network is used
to select actions in the current time step, while the target network provides target values for training.
The parameters of the target network are regularly updated using the parameters of the online network.
In summary, the fundamental process of the DQN algorithm is as follows: (1) Define the state
space of the problem
s
, representing the different states observed by the algorithm during the learning
and decision-making process. (2) Action selection: Choose an action
a
based on the current state
using a specified policy. (3) Execute action and observe reward
r
: Execute the selected action and
observe the reward returned by the environment, along with the new state next s_. This process
simulates the interaction between the agent and the environment. (4) Experience replay: Store the
executed actions and observed results in an experience replay buffer. (5) Target value computation:
Utilize a neural network to approximate the Q-value function (action-value function) and calculate
the Q-values for each possible action in the current state. This Q-value represents the long-term return
of choosing a particular action. (6) Optimization based on the loss function: Define a loss function
that measures the disparity between the model’s predicted Q-values and the target Q-values. Update
Figure 3. The mapping relationship from the physical world to the DT model
Volume 36 • Issue 1
12
the parameters of the neural network through an optimization algorithm, gradually aligning the
predicted Q-values with the target Q-values. (7) Iterate the aforementioned learning process,
continually updating the parameters of the neural network to enhance the model’s decision-making
performance in the environment.
In this smart factory decision-making algorithm for production process automation, the DQN
algorithm is combined with a digital twin (DT) to build a DQN production process optimizer. This
optimizer uses the DT as input data to the algorithm to provide real-time information about the
equipment. After the optimizer retrieves the necessary input data from the DT, real-time interaction
between the optimizer and the DT is achieved, and the optimizer subsequently trains its internal deep
neural network based on the retrieved input data.
The principle of this algorithm is shown in Figure 4.
The production process optimization algorithm based on DT and DQN algorithms is represented
in the form of pseudo-code as shown in Algorithm 1.
Algorithm 1. Process of a single DQN model training with DT data input
# Initialize parameters
env = Digital_Twin_Input()
agent = DQNAgent(state_size, action_size)
# Training DQN using Digital Twin data
Initialize num_episode
for episode in range(num_episodes):
s = env.reset()
total_reward = 0
while (Loss()> m)
# Choose action using
e
-greedy strategy
a = agent.act(s)
Figure 4. The principle of the DQN process optimum algorithm
Volume 36 • Issue 1
13
# Execute action, observe next state, reward, and done
flag
next_s, reward, done = env.step(a)
next_s = np.reshape(next_s, [1, s_size])
# Train DQN model using experience replay
agent.train(s, a, reward, next_s, done)
# Update state and total reward
s = next_s
total_reward += reward
end while
end for
We conducted a simulation experiment for the automated optimization of intelligent manufacturing
processes utilizing the Deep Q Network (DQN) model from deep reinforcement learning. The objective
of the experiment was to validate the capability of the DQN model for performing intelligent production
decision tasks within a smart factory environment, aiming to maximize production efficiency and
resource utilization. Data were extracted from a running digital twin model of an electric car automated
manufacturing plant, constituting a dataset encompassing products, equipment, and manufacturing
processes within the smart factory. This dataset provided a realistic and highly simulated environment
for the DQN model to learn and optimize intelligent decision-making.
In the experiment, Network Architecture Search (NAS) techniques were employed to optimize
the Deep Q Network model. This involved tuning the network architecture, hyperparameters such as
learning rate, discount factor, and experience replay buffer size, ensuring stable convergence during
training, and enhancing the model’s learning capabilities in complex decision environments.
It is noteworthy that, to ensure the accuracy and authenticity of the data from the digital twin
factory, the collection of factory environment data was aligned with the real-time scenarios in the
digital twin model. Additionally, the experiment paid special attention to the generalization ability of
the DQN model in complex shop floor environments, ensuring superior performance when scheduling
various production stages.
To validate the performance of the model, this study conducted a comparative analysis by
examining 10 baseline models for intelligent manufacturing task scheduling, drawn from literature
over the past three years.
Model #1 (Mzili et al., 2023): Proposed a spotted hyena optimization algorithm in order to identify
and implement optimal schedules for jobs in a flow shop environment.
Model #2 (Qiu et al., 2023): Proposed an improved memory algorithm (MA) which combines a
genetic algorithm (GA) with educational operators to solve integrated production scheduling
decision problems.
Model #3 (Azevedo et al., 2023): Developed a multi-objective optimization model for improving
production scheduling performance metrics to help managers make decisions related to job
scheduling.
Model #4 (Fontes et al., 2023): Considered the minimization of two optional performance metrics,
duration and exit time, and the optimal solution of the model is solved by the Mixed Integer
Linear Programming (MILP) algorithm.
Volume 36 • Issue 1
14
Model #5 (Bamoumen et al., 2023): Proposed an algorithm that combines the MILP algorithm and
an algorithm similar to the Greedy Randomized Adaptive Search Procedure (GRASP) to solve
the problem of automated scheduling of production processes.
Model #6 (Chen & Liu, 2023b): Combined discrete-time simulation methods to establish a delivery
date change response model and constructed dynamic scheduling rules using gene expression
programming (GEP) algorithms to realize dynamic production planning.
Model #7 (Tang et al., 2023): Proposed a heuristic algorithm based on learning mechanisms and
ant colony optimization for solving the collaborative scheduling problem.
Model #8 (Kang et al., 2023): Proposed a multi-strategy individual adaptive mutation difference
evolutionary algorithm (MSIADE) for this production scheduling problem.
Model #9 (Saqlain et al., 2023): Proposed a flexible job scheduling algorithm based on Monte Carlo
tree search for scheduling highly complex jobs in a real-time job environment.
Model #10 (Wang et al., 2023): Proposed a hybrid genetic algorithm based on variable neighborhood
search (GAVNS) for solving the production scheduling problem.
Literature #1, 2, 6, 7, 8, 10 are strategies based on genetic algorithms, literature 3, 4, 5 are
strategies based on optimization methods, and literature 9 is a strategy based on solution space search.
This study inputs production demands and resource consumption data from smart factories into
these models. Multiple models are employed to provide production planning schemes for various
shop floors within the factory during a specific period. Subsequently, diverse evaluation metrics are
utilized to compare the planning results of multiple schemes. Performance metrics included production
efficiency per unit time, resource utilization rate, energy consumption during manufacturing (RMB),
raw material consumption during manufacturing (RMB), and total daily downtime (hours). Through
these metrics, we will assess the application potential and optimization effects of the DQN model
in real factory environments.
The dataset used in this experiment is derived from a recently established smart electric vehicle
manufacturing plant. This newly constructed facility serves as a rich source of data, encompassing
various workshops within the factory. The dataset includes detailed information on production tasks,
equipment statuses, and process parameters, among other aspects. The real-time data stream from
sensors deployed throughout the factory provides insights into factors such as temperature, humidity,
and equipment states, as well as detailed information on production efficiency and output. This
comprehensive dataset from the entire factory ensures robust testing of the proposed algorithms in
a realistic and dynamic production environment.
The reason for not adopting public datasets is that intelligent manufacturing tasks often occur in
a specific environment, facing a particular production task and a set of specific production equipment.
Table 1. Attributes of the experimental dataset extracted from the smart manufacturing generation environment
Attribution name Attribution value
Number of workshops 5
Number of equipment 30
Number of products 15
Data generation time 2023.1.1 ~ 2023.1.31
Data set size 20.1G
Number of data samples 9522
Volume 36 • Issue 1
15
For such algorithms, generalization ability and robustness are not the primary considerations; stability
of the algorithm and enhancement of actual production capabilities are the most crucial factors to be
considered. Therefore, algorithm optimization for intelligent manufacturing tasks must directly use
real data from the production environment so that the trained model can achieve optimal performance
in actual production scenarios.
As we can see from the experimental results, the DQN model, combined with digital twin
technology, comprehensively considers the complex correlations in the actual production process,
making the model more intelligent and adaptable. As a result, it achieves superior performance
in production scheduling.
From the experimental results, there are several reasons why the proposed DQN model
outperforms these 10 baseline models in production scheduling.
First, compared to optimization models #3, 4, 5, the DQN model utilizes deep reinforcement
learning technology, enabling real-time and automated production task scheduling in complex
manufacturing environments and adapting flexibly to evolving production demands. The comparison
results of production efficiency per unit time and total daily downtime show this advantage in Figure 5.
Second, in contrast to evolutionary computation-based models #1, 2, 6, 7, 8 and 10, the DQN
model demonstrates superior generalization capabilities, accommodating variations in different
workshops and production stages, thereby enhancing its applicability in real factory settings. The
comparison results of energy and raw material consumption during manufacturing show this advantage
in Figure 6.
Table 2. Comparison results with ten baseline models
Model Production efficiency
per unit time
Resource
utilization
rate(%)
Energy
consumption during
manufacturing(RMB)
Raw material
consumption during
manufacturing(RMB)
Total daily
downtime(h)
Mode1 #1
(Mzili et al., 2023) 89% 75% 12.52K 344.34K 1.04
Mode1 #2
(Qiu et al., 2023) 63% 70% 18.24K 315.62K 0.75
Mode1 #3
(Azevedo et al., 2023) 96% 64% 12.76K 301.46K 0.26
Mode1 #4
(Fontes et al., 2023) 70% 64% 15.53K 291.51K 0.92
Mode1 #5
(Bamoumen et al., 2023) 74% 65% 19.92K 310.42K 1.06
Mode1 #6
(Chen & Liu, 2023b) 67% 81% 14.67K 291.53K 1.04
Mode1 #7
(Tang et al., 2023) 70% 83% 19.12K 295.62K 0.67
Mode1 #8
(Kang et al., 2023) 59% 78% 15.51K 285.93K 0.49
Mode1 #9
(Saqlain et al., 2023) 80% 90% 13.54K 272.34K 0.42
Mode1 #10
(Wang et al., 2023) 71% 83% 15.22K 289.23K 0.39
Our Model 100% 95% 11.12K 242.14K 0.13
Volume 36 • Issue 1
16
Additionally, compared to model #9, the DQN’s feature learning based on deep learning
facilitates more effective planning and decision-making for complex production tasks, improving
the overall performance of the model. The comparison results of the resource utilization rate show
this advantage in Figure 7.
Finally, with the integration of digital twin technology, the DQN model comprehensively considers
intricate correlations in actual production processes, enhancing its intelligence and adaptability,
resulting in superior performance in production scheduling.
We conducted a production data analysis in the final assembly workshop of the factory, and the
experimental results unequivocally demonstrate the superior performance of our proposed DQN
production automation scheduling model compared to other models. Through real-time scheduling of
production tasks, the DQN model effectively enhanced assembly efficiency and optimized resource
utilization. The experimental results are shown in Table 3.
The results’ differences between models in the final assembly workshop are shown graphically
in Figures 8a-8e.
As a key workshop in the assembly line of a smart factory, the optimization of production process
scheduling is especially critical in the process of efficiently assembling pre-produced parts into a
complete vehicle. From the collected experimental result data, the DQN-based production process
Figure 5. The comparison results of production efficiency per unit time and total daily downtime
Figure 6. The comparison results of energy and raw material consumption during manufacturing
Volume 36 • Issue 1
17
Figure 7. The comparison results of resource utilization rate
Table 3. Comparison results in final assembly workshop
Model
Production
efficiency
per unit
time
Resource
utilization
rate(%)
Energy
consumption during
manufacturing(1000
RMB)
Raw material
consumption during
manufacturing(1000
RMB)
Total daily
downtime(h)
Mode1 #1
(Mzili et al., 2023) 85% 75% 2.04 12.46 0.01
Mode1 #2
(Qiu et al., 2023) 76% 69% 1.64 12.37 0.02
Mode1 #3
(Azevedo et al., 2023) 68% 62% 2.04 12.36 0.03
Mode1 #4
(Fontes et al., 2023) 84% 78% 1.82 12.34 0.02
Mode1 #5
(Bamoumen et al., 2023) 94% 93% 1.69 12.33 0.01
Mode1 #6
(Chen & Liu, 2023b) 98% 93% 1.86 12.31 0.04
Mode1 #7
(Tang et al., 2023) 94% 93% 1.72 12.30 0.02
Mode1 #8
(Kang et al., 2023) 82% 75% 2.24 12.28 0.07
Mode1 #9
(Saqlain et al., 2023) 85% 85% 1.79 12.22 0.03
Mode1 #10
(Wang et al., 2023) 92% 89% 1.57 12.18 0.02
Our Model 100% 95% 1.31 12.13 0.01
Volume 36 • Issue 1
18
Figure 8a. Comparison of production efficiency per unit time
Figure 8b. Comparison of resource utilization rate
Volume 36 • Issue 1
19
Figure 8c. Comparison of energy consumption during manufacturing
Figure 8d. Comparison of raw material consumption during manufacturing
Volume 36 • Issue 1
20
automation scheduling model makes the smart factory in the final assembly shop have multiple
evaluation indexes with multiple advantages compared with the 10 baseline models.
To validate the impact of the DQN algorithm on the production capacity improvement of the smart
factory, a case study was conducted to compare the manufacturing of electric vehicles in the factory
before the system went live with the situation after the system went live. The productivity changes
of the smart factory before and after the system went live are shown in Table 4.
A graphical representation of the changes before and after the system go-live is shown in Figure 9.
Figure 8e. Comparison of total daily downtime
Table 4. The productivity changes of the smart factory before and after the system go-live
Time
Production
efficiency per
unit time
Resource
utilization
rate(%)
Energy
consumption during
manufacturing(RMB)
Raw material
consumption during
manufacturing(RMB)
Total daily
downtime(h)
Before 32% 54% 19.56K 311.31K 1.32
After 100% 95% 11.12K 242.14K 0.13
Figure 9. Graphical representation of the changes before and after the system go-live
Volume 36 • Issue 1
21
After implementing intelligent production scheduling algorithms based on deep learning and
digital twin in the smart electric vehicle factory, there has been a significant improvement in factory
production efficiency. Possible reasons for this improvement include, firstly, the application of deep
learning technology, which enhances the algorithm’s ability to understand and learn from the complex
production environment, improving its capacity to handle large-scale data and complex external factors.
This aids the model in more accurately predicting key information, such as production demands and
equipment status changes, facilitating more intelligent and real-time production task scheduling.
Secondly, the introduction of digital twin technology provides a virtual simulation environment based
on the actual factory, modeling the real production processes through digital twin models, thereby
increasing the practicality of the algorithm.
From Figure 9, it can be observed that, following the implementation of the DQN model for factory
production process scheduling, there is a notable reduction in the daily downtime of the system. The
emergence of this phenomenon can be attributed to the real-time adaptability of the DQN model.
With the dynamic changes in the production environment, the model can swiftly adjust the scheduling
scheme, mitigating issues that static scheduling, traditionally employed, is unable to flexibly address.
Consequently, this capability reduces the probability of equipment downtime.
Additionally, algorithms based on deep learning and digital twin can comprehensively consider
the complex correlations between various production processes, including collaborative operations
among devices and the impacts between different manufacturing processes. This comprehensive
consideration makes the algorithm more intelligent, allowing it to make more precise and rational
scheduling decisions, thereby maximizing production efficiency.
This study aims to address challenges in optimizing the production process of a smart factory. The
research proposes resolving issues related to process automation in intelligent manufacturing by
introducing smart algorithms and digital twin technology. The primary approach involves leveraging
deep reinforcement learning techniques, specifically an enhanced Deep Q Network (DQN) model, to
achieve real-time and automated scheduling of production tasks within the factory. The research seeks
to enhance algorithmic performance and overcome limitations by introducing innovative methods
and making full use of the advantages offered by smart algorithms and digital twin technology. The
goal is to improve production efficiency and resource utilization in the electric vehicle industry. The
effectiveness of the proposed methods is briefly demonstrated through a simulation experiment and
a case study.
In addition, although this study does not improve the DQN algorithm, the combination of the
digital twin model and the relational mapping algorithm from the physical world to the DT model
proposed in this study can be seen as an improvement to the DQN model as a solution to the task of
automated scheduling of the production process in an electric vehicle smart factory.
It can be anticipated that the integration of solutions incorporating digital twins and DQN models
will contribute to elevated levels of automation, enhanced flexibility in production scheduling,
and more sustainable operational models in the construction and operation of smart factories. This
integration is expected to further propel the digital transformation and intelligent development of
the manufacturing industry.
Currently, there is room for improvement in the performance of our algorithm, especially in
real-time and distributed scheduling. To address this deficiency, we plan to design and introduce
more optimized artificial intelligence algorithms. This will involve in-depth research into scheduling
Volume 36 • Issue 1
22
algorithms to enhance their efficiency and accuracy. Additionally, we will explore more flexible
distributed scheduling methods to adapt to complex production environments.
Another notable deficiency is the lack of robustness in anomaly monitoring in the smart factory.
Despite significant advancements in enhancing the level of automation in the system, there is still room
for improvement in anomaly monitoring and automatic recovery capabilities. To address this issue,
we plan to strengthen the design of the anomaly monitoring system and introduce more intelligent
and sensitive monitoring technologies. Simultaneously, we will focus on improving the system’s
automatic recovery capabilities in response to anomalies, aiming to minimize the need for manual
intervention. This improvement will contribute to enhancing the stability and reliability of the system.
Although the current production of electric vehicles has achieved a high level of automation,
there are still some stages that require manual intervention. To further reduce human intervention in
the system, we plan to explore more automation and intelligent solutions. This includes conducting
in-depth research on stages that currently involve manual intervention to identify potential automated
alternatives. By incorporating more advanced machine learning and automation technologies, we
aim to achieve a greater degree of autonomy in the electric vehicle production process, ultimately
enhancing overall production efficiency.
This study aims to optimize the production processes in smart factories through the integration
of intelligent algorithms and digital twin technology. By enhancing the Deep Q Network (DQN)
model within the framework of improved deep reinforcement learning, the research focuses on real-
time and automated scheduling of production tasks within the factory, aiming to enhance production
efficiency and resource utilization. The study proposes improvement strategies in terms of optimizing
algorithm performance, strengthening anomaly monitoring and automatic recovery capabilities,
as well as reducing manual interventions. In summary, this research endeavors to provide a novel
algorithm and presents an effective technological application for the field of intelligent manufacturing,
contributing positively to the development of more intelligent and efficient production processes in
the electric vehicle industry.
The authors would like to thank the editor and anonymous reviewers for their contributions toward
improving the quality of this paper.
The data used to support the findings of this study are included within the article.
The authors declare that there is no conflict of interest regarding the publication of this paper.
This research received no external funding.
We have no known conflict of interest to disclose.
Volume 36 • Issue 1
23
Azevedo, B., Montanño-Vega, R., Varela, M., & Pereira, A. (2023). Bio-inspired multi-objective algorithms
applied on production scheduling problems. International Journal of Industrial Engineering Computations,
14(2), 415–436. doi:10.5267/j.ijiec.2022.12.001
Bamoumen, M., Elfirdoussi, S., Ren, L., & Tchernev, N. (2023). An efficient GRASP-like algorithm for the multi-
product straight pipeline scheduling problem. Computers & Operations Research, 150, 106082. doi:10.1016/j.
cor.2022.106082
Bathla, G., Bhadane, K., Singh, R. K., Kumar, R., Aluvalu, R., Krishnamurthi, R., Kumar, A., Thakur, R., &
Basheer, S. (2022). Autonomous vehicles and intelligent automation: Applications, challenges, and opportunities.
Mobile Information Systems.
Chen, H., Jeremiah, S. R., Lee, C., & Park, J. H. (2023a). A Digital twin-based heuristic multi-cooperation
scheduling framework for smart manufacturing in IIoT Environment. Applied Sciences (Basel, Switzerland),
13(3), 1440. doi:10.3390/app13031440
Chen, J., & Liu, X. (2023b). GEP algorithm-based optimization method for PCs production scheduling under
due date variation. 4th International Conference on Computer Engineering and Application (ICCEA), (pp. 357-
352). IEEE. doi:10.1109/ICCEA58433.2023.10135444
Domínguez, J. A., Fuentes, R. P., Vegetti, M., Roldán, L., Gonnet, S., & Diván, M. J. (2022). Ontology
implementation of OPC UA and automationML: A review. Advanced Intelligent Technologies for Industry:
Proceedings of 2nd International Conference on Advanced Intelligent Technologies (ICAIT 2021). IEEE.
Fan, L., & Zhang, L. (2022). Multi-system fusion based on deep neural network and cloud edge computing and
its application in intelligent manufacturing. Neural Computing and Applications, 1-10.
Favoretto, C., Mendes, G. H. S., Filho, M. G., Gouvea de Oliveira, M., & Ganga, G. M. D. (2022). Digital
transformation of business model in manufacturing companies: Challenges and research agenda. Journal of
Business and Industrial Marketing, 37(4), 748–767. doi:10.1108/JBIM-10-2020-0477
Fontes, D. B., Homayouni, S. M., & Gonçalves, J. F. (2023). A hybrid particle swarm optimization and
simulated annealing algorithm for the job shop scheduling problem with transport resources. European Journal
of Operational Research, 306(3), 1140–1157. doi:10.1016/j.ejor.2022.09.006
Gupta, P., Krishna, C., Rajesh, R., Ananthakrishnan, A., Vishnuvardhan, A., Patel, S. S., Kapruan, C., Brahmbhatt,
S., Kataray, T., & Narayanan, D. (2022). Industrial internet of things in intelligent manufacturing: A review,
approaches, opportunities, open challenges, and future directions. International Journal on Interactive Design
and Manufacturing (IJIDeM), 1-23.
Hsu, C.-H., Cheng, S.-J., Chang, T.-J., Huang, Y. M., Fung, C.-P., & Chen, S. F. (2022). Low-cost and high-
efficiency electromechanical integration for smart factories of IoT with CNN and FOPID controller design
under the impact of COVID-19. Applied Sciences (Basel, Switzerland), 12(7), 3231. doi:10.3390/app12073231
Jiménez‐Ramírez, A., Chacón‐Montero, J., Wojdynsky, T., & Gonzalez Enriquez, J. (2023). Automated testing
in robotic process automation projects. Journal of Software (Malden, MA), 35(3), e2259. doi:10.1002/smr.2259
Kang, L., Liu, D., Wu, Y., & Ping, G. (2023). An improved DE algorithm for solving multi-furnace optimal
scheduling of single crystal silicon production. International Journal of Pattern Recognition and Artificial
Intelligence, 37(02), 2359001. doi:10.1142/S0218001423590012
Kannen, N., & Subasi, A. (2023). Smart factories of Industry 4.0: Determination of the effective smartphone
position for human activity recognition using deep learning. In Advanced Signal Processing for Industry 4.0,
Volume 2: Security issues, management and future opportunities (pp. 3-1-3-24). IOP Publishing Bristol, UK.
Kumar, S., Sigroha, M., Kumar, K., & Sarkar, B. (2022). Manufacturing/remanufacturing based supply chain
management under advertisements and carbon emissions process. Operations Research, 56(2), 831–851.
doi:10.1051/ro/2021189
Lei, J., Hui, J., Chang, F., Dassari, S., & Ding, K. (2023). Reinforcement learning-based dynamic production-
logistics-integrated tasks allocation in smart factories. International Journal of Production Research, 61(13),
4419–4436. doi:10.1080/00207543.2022.2142314
Volume 36 • Issue 1
24
Li, L., Lei, B., & Mao, C. (2022a). Digital twin in smart manufacturing. Journal of Industrial Information
Integration, 26, 100289. doi:10.1016/j.jii.2021.100289
Li, X., Wang, Q., Huang, S., Shi, R., Han, C., & Gao, Y. (2022b). The transfer strategy of digital information
technology for heterogeneous manufacturers. [JOEUC]. Journal of Organizational and End User Computing,
34(8), 1–22. doi:10.4018/JOEUC.306248
Moradi, R., Cofre-Martel, S., Droguett, E. L., Modarres, M., & Groth, K. M. (2022). Integration of deep learning
and Bayesian networks for condition and operation risk monitoring of complex engineering systems. Reliability
Engineering & System Safety, 222, 108433. doi:10.1016/j.ress.2022.108433
Musbah, H., Ali, G., Aly, H. H., & Little, T. A. (2022). Energy management using multi-criteria decision making
and machine learning classification algorithms for intelligent system. Electric Power Systems Research, 203,
107645. doi:10.1016/j.epsr.2021.107645
Mzili, T., Mzili, I., Riffi, M. E., Pamucar, D., Kurdi, M., & Ali, A. H. (2023). Optimizing production scheduling
with the spotted hyena algorithm: A novel approach to the flow shop problem. Reports in Mechanical Engineering,
4(1), 90–103. doi:10.31181/rme040116072023m
Nie, L., Chang, F., Chang, X., Liu, C., Jin, Y., Liu, G., Fu, J., & Han, X. (2021). A novel self-adaptive multiple
kernel learning algorithm. Journal of Jilin University Science Edition, 59(5), 1212–1218.
Oluyisola, O. E., Bhalla, S., Sgarbossa, F., & Strandhagen, J. O. (2022). Designing and developing smart
production planning and control systems in the industry 4.0 era: A methodology and case study. Journal of
Intelligent Manufacturing, 33(1), 311–332. doi:10.1007/s10845-021-01808-w
Pengcheng, Z., Shang, G., & Hongmei, Y. (2022). Spatial crowdsourcing task allocation based on multi-intelligent
body deep reinforcement learning. Journal of Jilin University Science Edition, 60(2), 321–331.
Qiu, F., Geng, N., & Wang, H. (2023). An improved memetic algorithm for integrated production scheduling
and vehicle routing decisions. Computers & Operations Research, 152, 106127. doi:10.1016/j.cor.2022.106127
Salman, R. A., Myeongbae, L., Jonghyun, L., Cho, Y., & Changsun, S. (2022). A comparative study of energy
big data analysis for product management in a smart factory. [JOEUC]. Journal of Organizational and End User
Computing, 34(2), 1–17. doi:10.4018/JOEUC.291559
Saqlain, M., Ali, S., & Lee, J. (2023). A Monte-Carlo tree search algorithm for the flexible job-shop scheduling
in manufacturing systems. Flexible Services and Manufacturing Journal, 35(2), 548–571. doi:10.1007/s10696-
021-09437-4
Shojaeinasab, A., Charter, T., Jalayer, M., Khadivi, M., Ogunfowora, O., Raiyani, N., Yaghoubi, M., & Najjaran,
H. (2022). Intelligent manufacturing execution systems: A systematic review. Journal of Manufacturing Systems,
62, 503–522. doi:10.1016/j.jmsy.2022.01.004
Srivastava, A., Kumar, R. R., Chakraborty, A., Mateen, A., & Narayanamurthy, G. (2022). Design and selection
of government policies for electric vehicles adoption: A global perspective. Transportation Research Part E,
Logistics and Transportation Review, 161, 102726. doi:10.1016/j.tre.2022.102726
Tang, L., Han, H., Tan, Z., & Jing, K. (2023). Centralized collaborative production scheduling with evaluation
of a practical order-merging strategy. International Journal of Production Research, 61(1), 282–301. doi:10.1
080/00207543.2021.1978577
Tercan, H., & Meisen, T. (2022). Machine learning and deep learning based predictive quality in manufacturing:
A systematic review. Journal of Intelligent Manufacturing, 33(7), 1879–1905. doi:10.1007/s10845-022-01963-8
Wahid, A., Breslin, J. G., & Intizar, M. A. (2022). Prediction of machine failure in industry 4.0: A hybrid CNN-
LSTM framework. Applied Sciences (Basel, Switzerland), 12(9), 4221. doi:10.3390/app12094221
Wang, J., Xu, C., Zhang, J., & Zhong, R. (2022a). Big data analytics for intelligent manufacturing systems: A
review. Journal of Manufacturing Systems, 62, 738–752. doi:10.1016/j.jmsy.2021.03.005
Wang, W., Tian, G., Zhang, H., Li, Z., & Zhang, L. (2023). A hybrid genetic algorithm with multiple decoding
methods for energy-aware remanufacturing system scheduling problem. Robotics and Computer-integrated
Manufacturing, 81, 102509. doi:10.1016/j.rcim.2022.102509
Volume 36 • Issue 1
25
Wang, Y., Kang, X., & Chen, Z. (2022b). A survey of digital twin techniques in smart manufacturing and
management of energy applications. Green Energy and Intelligent Transportation, 1(2), 100014. doi:10.1016/j.
geits.2022.100014
Yang, Z., Shang, W.-L., Zhang, H., Garg, H., & Han, C. (2022). Assessing the green distribution transformer
manufacturing process using a cloud-based q-rung orthopair fuzzy multi-criteria framework. Applied Energy,
311, 118687. doi:10.1016/j.apenergy.2022.118687
Zhang, Y., & Dilanchiev, A. (2022). Economic recovery, industrial structure and natural resource utilization
efficiency in China: Effect on green economic recovery. Resources Policy, 79, 102958. doi:10.1016/j.
resourpol.2022.102958
Zhou, L., Jiang, Z., Geng, N., Niu, Y., Cui, F., Liu, K., & Qi, N. (2022a). Production and operations management
for intelligent manufacturing: A systematic literature review. International Journal of Production Research,
60(2), 808–846. doi:10.1080/00207543.2021.2017055
Zhou, X., Hu, Y., Wu, J., Liang, W., Ma, J., & Jin, Q. (2022b). Distribution bias aware collaborative generative
adversarial network for imbalanced deep learning in industrial IoT. IEEE Transactions on Industrial Informatics,
19(1), 570–580. doi:10.1109/TII.2022.3170149
Xiangqian Wang, Professor, Master’s Degree, Graduated from XiDian University in 2009. Currently a doctoral
candidate of Education Department, East China Normal University. Worked in Pingdingshan University. Her
research interests include computer application.
Haifeng Hu, Associate professor, Master’s Degree, Graduated from XiDian University in 2010.Worked in
Pingdingshan University. His research interests include Intelligent algorithm analysis.
Yuyao Wang, currently a research assistant and doctoral student at Lamar University. His research includes
radiation effects modeling in Integrated Circuits, circuit level soft error mitigation, interconnect modeling and noise
prediction, FPGA applications, machine learning.
Zhaoyu Wang, an undergraduate student, is currently studying in the School of Optoelectronics and Information
Engineering, Fujian Normal University. Her research interests include communication engineering.
Available via license: CC BY 3.0
Content may be subject to copyright.
