Chapter 12

Digital twin-driven
additive manufacturing
Advancements and future prospects
P M Abhilash
University of Strathclyde

Jibin Boban
Mikrotools Pte Ltd.

Afzaal Ahmed
Indian Institute of Technology, Palakkad

Xichun Luo
University of Strathclyde

12.1 INTRODUCTION

Additive manufacturing (AM) involves layer-by-layer manufacturing of a

3D object from the CAD STL file. AM has drastically changed the manu-
facturing landscape through its design flexibility and near-net-shape pro-
duction capabilities. AM makes it possible to conceive new designs and
functionalities that were previously unattainable. This has been reflected
in its accelerated market growth by several folds in the past decade, and
the forecast for the near future is even better. There are several metal AM
processes to choose from based on the energy source and metal deposition
mechanism, as shown in Figure 12.1. They exhibit a range of capabilities in
terms of productivity, part quality compliance, surface integrity, mechani-
cal behavior and chemical properties. However, despite having all these
merits and flexibilities over other conventional manufacturing techniques,
the inherent complex thermo-mechanical manufacturing mechanism has
fundamentally limited the stability of the AM process. Due to its inherent
variability and uncertainties, the process has under-par repeatability [1,2].
In addition to the process instabilities, other limitations, such as unantici-
pated process failures, part defects, poor efficiency and higher production
costs, have restricted the widespread adoption of AM components in the
aerospace, defense, automobile and biomedical industries, where the accu-
racy and repeatability of components are vital.

196 DOI: 10.1201/9781003406488-12

9781032460550_C012.indd 196 28/08/23 10:22 PM

 Digital twin-driven additive manufacturing 197

Figure 12.1 Schematic and classification of various metal AM processes [3].

A digital twin (DT) is a virtual replica of a physical entity, process, or

system. In the manufacturing context, DTs are used for real-time sim-
ulation, analysis and process control toward better performance. DTs
have been built to simulate just the product, the production process, or
an entire factory environment. This is beneficial since the manufactur-
ers are now able to analyze and optimize their production systems in a
virtual environment, reducing the need for costly physical testing and
experimentation. Such capabilities of DT can be utilized toward address-
ing the process uncertainties of AM. Through real-time data acquisition
and adaptive process control, DT ensures process certainty by keeping
the process within optimal limits in real-time. Also, the DTs can replace
conventional feedback approaches with AI-driven feed-forward adaptive
process control. This means the future state of the physical AM system is
anticipated with respect to the current state, which allows the DTs to pro-
actively perform the precautionary actions as opposed to the conventional
reactive strategy of feedback systems. A DT concept aimed at addressing
AM process uncertainties is given in Figure 12.2. The process is moni-
tored through multiple sensors, and the type and extent of preventive
actions are determined by AI models. The acquired data is used for pro-
cess diagnosis and real-time updating of the virtual model. Apart from
anomaly detection and feed-forward control, DT-driven product design
has also gained a lot of research interest in recent years. This concept of
Design DT for AM emphasizes a 2-way data communication between
the digital design and physical space, to achieve tailored product features
including porosity, material properties, geometry and dimensions.
The overall capabilities of DTs toward addressing the challenges of AM
are as follows:

9781032460550_C012.indd 197 28/08/23 10:22 PM

 198 Hybrid Metal Additive Manufacturing

Figure 12.2 A digital twin-driven AM system to address inherent uncertainties [4].

• Process optimization: DT optimizes the AM build parameters for bet-

ter quality and efficiency.
• Predictive maintenance: DT monitors the AM process performance
and forecasts the required maintenance, reducing the overall down-
time and enhancing equipment lifespan.
• Quality control: DT simulates the AM build process and identifies poten-
tial issues beforehand, ensuring quality control and waste reduction.
• Remote monitoring: DT can remotely monitor the AM process,
providing data-driven real-time updates of the build progress and
performance.
• Cost optimization: Through DT simulation, the optimal design and
process parameters are identified to minimize cost without compro-
mising product quality compliance.

Overall, the DT can be used to anticipate and optimize how a 3D printed

product or AM process will behave, find future problems or failures and
boost overall effectiveness and efficacy. Upon adopting the DT technology,
high quality-low cost AM products can be manufactured with a shorter lead
time. In this chapter, the fundamentals, architecture, standards, computing
aspects and enabling technologies of AM DTs are critically examined, along
with their present challenges, future prospects and research directions. In
addition, the chapter further emphasizes enhancing the acceptability of AM
across a broader scope of industrial applications due to enhanced robustness
and efficiency by employing a DT. Although the DT-driven framework is
relevant for all types of AM processes, due to its wider research and indus-
trial significance, DT for metal AM is more stressed in this book chapter.

12.2 COMPONENTS OF DT-DRIVEN AM

A typical DT system comprises (a) physical components, (b) a virtual envi-

ronment/cyberspace and (c) a communication architecture that connects
physical and cyberspace.

9781032460550_C012.indd 198 28/08/23 10:22 PM

 Digital twin-driven additive manufacturing 199

• Physical components include the 3D printer and all other associated

hardware for data acquisition, communication, computation and
storage.
• Visualization models [2D/3D CAD model, augmented reality and
virtual reality (AR/VR) visualization], predictive models, simulation
models and process control algorithms constitute virtual space. It
can also contain auxiliary models for data (signal/image) processing,
denoising, feature selection, etc.
• Finally, the communication standards govern and manage the data
flow between cyberspace and physical space. The appropriate selec-
tion of communication standards and computing architecture is cru-
cial to developing an efficient DT for AM, given the complexity and
volume of the data generated during the process.

The overall logical representation of a DT-driven AM process is given in

Figure 12.3. The three components are described further in the following
subsections.

12.2.1 Physical space

The main element of the physical space is the metal 3D printer itself. The
raw materials (metal powder), in situ sensors, communication devices,
workstations and equipment for metrology and post-processing are the
other constituents of this domain. The right selection of hardware for
data acquisition and processing is critical since the accuracy of the data-
driven DT depends on the type, number and placement of sensors. The
selection of sensor type depends on the type of process signatures to be

Figure 12.3 Logical representation of DT-driven metal AM [5].

9781032460550_C012.indd 199 28/08/23 10:23 PM

 200 Hybrid Metal Additive Manufacturing

acquired. Sensors can be direct or indirect sensors, depending on their

sensing capabilities. The direct sensor captures the feature to be analyzed
by acquiring the feature directly. An example of direct sensing is melt pool
anomaly detection by imaging the melt pool using a high-speed camera
[6]. On the other hand, the indirect sensing method monitors a feature
by acquiring secondary signatures. For instance, using acoustic signals to
monitor melt pool anomalies. Although indirect sensing is usually easier
to set up and analyze, the accuracy can be lower than that of direct sens-
ing methods [7].
A variety of vision-based sensors are used for direct sensing, which
includes photodiodes, infrared (IR) cameras, charge-coupled device (CCD)
cameras and complementary metal oxide semiconductor (CMOS) cameras.
Photodiode and IR cameras capture thermal signals (IR rays) from the
in-process melt pool, whereas the CCD and CMOS cameras capture the
optical signals in the visible light emission and ultraviolet (UV) emissions.
Among indirect sensing, the acquisition of audio and electrical signals is
more prominent. The audio signals are captured by acoustic sensors, and
additional electrical signals such as eddy currents and instantaneous dis-
placements are captured by the current sensor and strain gauges, respec-
tively [8].

12.2.2 Virtual space

Virtual space includes the data sources, acquisition methods, AM software,
AM database, digital twin human-machine interface (DT-HMI) and knowl-
edge base, together with every other digital model, method and algorithm
required to run the DT. The domain performs key tasks such as knowledge
and data management, process planning, design assistance, tool path gen-
eration, online monitoring, diagnosis and prognosis, response prediction,
optimization, process control and security management. The computations
are based on the available design data, existing knowledge database and
real-time acquired data. Some of the critical data sources within a metal
AM process are given in Figure 12.4.
Another key aspect of the virtual domain is data visualization or rep-
resentation. Typically, 3D CAD models update themselves to match the
physical state of the system based on the real-time collected data. The
other prominent modes of data simulation and visualization are given in
Figure 12.5, which include multi-physics simulations, online or offline flaw
representation, real-time data monitoring and information analytics. In DT,
data-driven predictive models are employed to forecast the future state of
the AM process based on which feed-forward process control is performed.
Common predictive models are AI/ML models, statistical models, analyti-
cal equations, physics-based expressions, or finite element analysis (FEA)
simulations. In case where anomalies or defects are predicted, process con-
trol models will autonomously perform parameter revisions to regulate and
restore AM process stability.

9781032460550_C012.indd 200 28/08/23 10:23 PM

9781032460550_C012.indd 201
Digital twin-driven additive manufacturing
201

Figure 12.4 Critical data sources at various lifecycle stages in the metal AM process [9].

28/08/23 10:23 PM
 202 Hybrid Metal Additive Manufacturing

Figure 12.5 (a) Multi-physics simulations [11]. (b) Defect/Flaw visualization [12]. (c) Real-
time sensor data [13].

In most DTs, a highly interactive human-machine interface (HMI) is pre-

ferred for effective visualization and appropriate human intervention. In
this aspect, AR/VR have gained massive public, industrial and research
attention in the past decade. It is an advanced digital environment that
permits human interactions with virtual objects. In metal AM, such a vir-
tual environment that offers interactive training, moderated by an expert
through cloud computing has already been developed [10]. Now that the
proof of concept is established, AR/VR have massive potential to leverage
the possibilities of over-the-air (OTA) training, remote support and collab-
orative digital work environments.

12.2.3 Communication standards and

data management system
Lack of standardization is one of the key bottlenecks that restricts the
cross-compatibility of DT systems. To address this issue, the ISO 23247

9781032460550_C012.indd 202 28/08/23 10:23 PM

 Digital twin-driven additive manufacturing 203

standard was developed, which defines the general guidelines for develop-
ing a DT framework for manufacturing applications. Based on this stan-
dard, a generic DT implementation framework for wire arc AM (WAAM)
was developed by Kim et al. [14]. The proposed framework is aimed at the
customizability and interoperability of DT. To demonstrate the framework,
a case study of online anomaly detection and process control has been con-
sidered. Data collection and communication are well defined, with different
moduli for data collection, device control computing and HMI. The core
entity for data monitoring and analysis is proposed through edge, cloud, or
fog computing technologies. Cloud computing promotes centralized data
handling but is less secure, demands greater data handling and can often
result in prediction latency. On the other hand, edge computing performs
distributed data analysis and is more secure and fast. Fog computing is a
hybrid technology with a combined advantage over the other two.
Another collaborative data management system for DT-driven AM is
shown in Figure 12.6 [15]. Each module uses edge computing to perform
specific tasks and also communicates with the cloud storage for centralized
storage, retrieval and collaborative data communication. The data gener-
ated in an AM process lifecycle comes mainly from five sources: design

Figure 12.6 Concept of a data management and communication framework for

DT-driven AM [15].

9781032460550_C012.indd 203 28/08/23 10:23 PM

 204 Hybrid Metal Additive Manufacturing

data, process parameters, online monitoring data, post-processing data and

metrological data. During the design stage, the required part is designed as
a CAD file, and appropriate offline simulations are performed. The design
digital model is shared with the cloud to be retrieved as and when required
by the other modules. For example, the design data may be accessed to fix
the scan strategy during post-processing and final product quality evalua-
tion. Ren et al. [16] have also presented an edge-cloud data flow framework
for a metal AM system.

12.3 ROLE OF AI AND COMPUTING

STRUCTURE IN DT-DRIVEN AM

Artificial intelligence (AI) has a key role to play in building the DT of

AM processes. AI is extensively utilized in three aspects of AM, namely,
design, process and production. Figure 12.7 represents the AI-driven tasks
during each of these stages. Even in the past, AI models have been widely
used to enhance process efficiency and the overall performance of complex
manufacturing processes. However, thanks to several recent technological
advancements in AI, there has been exponential growth in its computing
power, efficiency and data-handling potential. Thus, AI-driven DTs can
now accurately and efficiently simulate even complex processes like AM

Figure 12.7 Application areas for ML-based DT in metal additive manufacturing [17].

9781032460550_C012.indd 204 28/08/23 10:23 PM

 Digital twin-driven additive manufacturing 205

in real time. Apart from response prediction, AI models perform online

sensor-data analytics for anomaly/defect/error detection, prediction, classi-
fication and rectification. Apart from AI technologies, appropriate selection
and usage of various computing architectures such as edge, fog and cloud
computing have a critical influence on DT performance in terms of data
latency, security and storage requirements. In short, advancements in AI
and computing structures have tremendously influenced the capabilities of
DT AM processes toward better productivity, quality and cost-effectiveness
for manufacturers. Some of the prominent AI learning schemes and com-
puting architectures are described in the following subsections in the con-
text of DT-driven AM.

12.3.1 Supervised ML for data-driven

real-time predictions
Supervised learning is an ML approach that trains a model based on a
labeled dataset. For an AM training dataset, the process parameters are
the model inputs, and the output is the desired response (like part qual-
ity, or productivity). Once trained, the model will make predictions on
unseen data. When such models are used within a DT, the AM outcomes
are predicted in real time, which allows for subsequent process adjustments
if required. Such predictive models improve process efficiency by minimiz-
ing the number of trial runs. A critical point to note here is that, for a
supervised ML model, the prediction accuracy will be very much reliant
on the quality of the training data. It is thus important to acquire a diverse
range of training data to ascertain the robustness of the model, which may
otherwise lead to poor predictions. Some of the widely used supervised ML
models are convolutional neural networks (CNN), support vector machines
(SVM) and artificial neural networks (ANN). These models are briefly dis-
cussed below:

• ANN models are well suited to model AM processes due to their

proven ability to handle non-linear data, large datasets, missing data
and noisy data. Moreover, it is a robust algorithm well suited for
real-time predictions such as anomaly detection and control systems
[18,19]. Some of the limitations are the presence of a large number of
hyperparameters to train and select, a longer training duration and
poor explainability.
• CNNs are a type of ANN that is particularly well suited for image
and video recognition tasks. These models can be used in AM DTs
for quality control and inspection, predictive maintenance, process
monitoring, simulation and virtual prototyping and automated sur-
face reconstruction. In addition, other types of neural networks can
be used in combination with CNNs to improve the performance of
the model [20].

9781032460550_C012.indd 205 28/08/23 10:23 PM

 206 Hybrid Metal Additive Manufacturing

• SVM is considered more robust against overfitting than neural net-

works, is more adaptable and has superior generalization capabili-
ties. Due to an interpretable decision surface, it can be regarded as
more explainable as well. Also, due to its computational efficiency,
the model is well suited to handle complex and data-rich processes
such as AM [21].

12.3.2 Cloud/Fog/Edge computing

Computing power, data security, data storage and data access can be put
to good use when DTs are run on cloud, fog, or edge computing platforms.
Cloud computing refers to a computing technique in which the entire
system’s data is stored in a centralized online storage space (cloud) and
retrieved from the cloud on demand. Although there are several advan-
tages, such as better resource sharing and collaboration leading to reduced
setup expenses, cloud computing is not ideally recommended for real-time
systems due to data latency, poor security and difficulties in remote access.
In contrast to this approach, edge computing architecture decentralizes the
computations by taking them nearer to the data source or the user, resulting
in quicker responses. Apart from reduced latency, edge computing offers
enhanced security and privacy, reduced operational expenses and better
reliability. Fog computing is a recent hybrid concept designed to address
the limitations of cloud computing by moving the cloud toward the edge of
the IoT network. Appropriate integration of these computing techniques is
vital to enhancing the computing capabilities of a DT platform. The usage
of these technologies in DT-driven AM is presented and discussed in this
section (Figure 12.8).
Integration of cloud and edge computing can be demonstrated via a
multi-level DT architecture for metal AM [16]. Various levels in this hier-
archy are the machine level, centralized factory (cloud level) and research
center (edge level). Optimal feature selection through dimensionality reduc-
tion is extremely vital in an edge-cloud data communication scheme. This
allows the elimination of data redundancy and reduces the latency of
the DT system. Part design data (called recipes) is communicated to the
machine from the factory cloud. During 3D printing, process signatures
acquired by sensors are analyzed to check for build errors and process

Edge-cloud data communication guidelines for metal AM systems [16].

Figure 12.8 

9781032460550_C012.indd 206 28/08/23 10:23 PM

 Digital twin-driven additive manufacturing 207

anomalies. The errors are listed and validated, and key findings are com-
municated back to the cloud and research levels. Raw data and error lists
are down-sampled during their transfer to minimize latency, storage and
bandwidth requirements. Also, the error list communication to the cloud
is performed only upon encountering a new anomaly to expand the data-
base. This information is incorporated into the subsequent part design. On
repeated occurrences of process anomalies, it is sent to the research level for
detained analysis and further decision-making, including a processing halt.
The overall data-transfer scheme is shown in Figure 12.8. A similar multi-
layer computing scheme was proposed by Guo et al. [22], but for the fused
deposition modeling (FDM) process. The device, edge and cloud layers are
responsible for data acquisition, local data processing and big data analysis,
respectively. Recently, another noteworthy cloud-edge computing scheme
for DT-driven AM was proposed by Liu et al. [15]. This data management
system comprises a centralized cloud DT and multiple-edge DTs, one per
lifecycle module. Each lifecycle module and distributed shop floors com-
municate the data to the cloud DT, where data analytics is performed. A
dedicated model handles all the data associated with product quality com-
pliance at different lifecycle stages. The collaborative cloud-edge DT data
management system enables offline CAD+STL design, layer-wise defect
detection and parameter optimization.
A comprehensive IoT-driven cloud computing platform integrating design,
3D printing and production planning is presented by Wang et al. [23].
Cloud computing integrates and manages information flow across hardware
and software modules. This not only includes the data from 3D printers,
materials, sensors and software but also expert knowledge and technical
know-how. The paradigm offers a wider scope for end users and stakehold-
ers to monitor and control the process remotely. IoT architecture facilitates
information flow across design, processing and planning, thereby reducing
product development bottlenecks. Adnan et al. [24] proposed fog computing-
based real-time closed-loop process control for AM processes. The proposed
computing architecture is a hybrid mix of cloud, edge and fog technologies.
The fog computing structure addresses the fundamental limitations of the
cloud-based platform in terms of latency, bandwidth requirements, seamless
integration with cloud services, data communication and security.

12.3.3 Unsupervised and reinforcement learning

One of the major drawbacks of supervised learning models is the require-
ment for a training dataset. Often, the AM process is stochastic and can
cause new anomalies outside the ones recorded during the training in real-
world test situations. To overcome this fundamental limitation of super-
vised ML models, unsupervised and reinforcement learning schemes are
being used. One of the key applications of unsupervised learning is defect
detection. In this regard, an effective strategy is to record the process

9781032460550_C012.indd 207 28/08/23 10:23 PM

 208 Hybrid Metal Additive Manufacturing

signatures during both defective and normal operations and then utilize
a self-organizing map (SOM)-driven online monitoring to distinguish
between various process anomalies [25]. Any appropriate AM process sig-
natures can be acquired and processed in this regard, including thermal sig-
natures, acoustic emission, or vibration data. Although clustering is a very
capable method to group process signatures/process histories based on pro-
cess signatures, the approach may be unsuitable to operate within an AM
DT, where it can suffer from prediction delays while handling the raw data
from the sensors. A more computationally efficient and robust approach to
unsupervised anomaly detection is to perform dimensionality reduction,
followed by clustering. The approach is generic and can be incorporated
with any appropriate selection of process signatures, dimensionality reduc-
tion method and clustering algorithm based on the type of AM process and
the end application. Dimensionality reduction reduces the order of a multi-
dimensional feature space, and it can significantly reduce the latency issues
during a real-time DT operation. This approach has been demonstrated
successfully for laser powder bed fusion by acquiring computer vision-based
process signatures. A t-SNE (t-distributed stochastic neighbor embedding)
dimensionality reduction algorithm is used to convert the raw multi-dimen-
sional datasets into low-dimensional 2D feature space data. The 2D data is
then clustered using K-means to visualize, detect and identify process flaws
such as porosity, undermelting and balling [6]. For unsupervised clustering,
K-means clustering is one of the simplest yet robust techniques to group the
feature sets into k groups based on geometric distance [26]. Later, a simi-
lar approach was chosen to analyze time series thermal signatures. Here
thermal processing history zones are mapped into a predefined number of
clusters using a K-means clustering algorithm, following dimensionality
reduction by symbolic aggregate approximation (SAX). The thermal zone
mapping contributes to the better interpretability of material processing,
considering various scan paths and geometries [27]. The unsupervised clus-
tering algorithm is utilized for several other applications within metal AM.
It is used to categorize wire bead consistency in WAAM [28]. Recently,
K-means clustering has been used for surface quality categorization during
direct energy deposition (DED) AM, in which surfaces are classified based
on porosity, surface morphology and deposition quality by acquiring the
emission spectrum [29].

12.4 KEY APPLICATION AREAS OF DT IN AM

12.4.1 Process condition monitoring

and quality control
The integration of ML with DTs in real-time aids in the automatic and rapid
analysis of massive data with scalable performance. Reliable predictions of
process states and defects can be accomplished through ML models that

9781032460550_C012.indd 208 28/08/23 10:23 PM

 Digital twin-driven additive manufacturing 209

ultimately pave the way for effective process monitoring of the AM process.
The DT framework proposed by Mourtzis et al. [30] for the FDM process is
noteworthy for monitoring the process and optimizing the parameters. The
DT architecture in the study is formulated to enhance AM product quality
and minimize cost in the long run. Machine operation and process moni-
toring can be realized through the developed AR-based immersive GUIs.
Later on, the engineer executes a quality assessment and stores the results for
future reference in a cloud database, which serves as an efficient resource for
process parameter optimization. Based on the process deviations observed
during online monitoring, the process parameters can be adequately con-
trolled to maintain the desired part quality.

12.4.2 Detection of process defects/anomalies

In the current industrial scenario, the AM domain is greatly dependent on
machine learning (ML)-assisted DTs for the detection of flaws or defects in
fabricated components. Despite traditional image processing methods, in
situ cloud processing techniques using ML yield viable results in the iden-
tification of AM component defects or anomalies [31,32]. A software plat-
form capable of executing multiple sub-processes in parallel enables the
in situ cloud processing of data. The ML model incorporated in the cloud
processing node facilitates the real-time identification of surface defects and
anomalies with high accuracy. Thus, DT approaches can be adopted for
online monitoring and flaw detection in the LPBF process. Yavari et al. [12]
realized successful flaw detection in LPBF SS316L impeller-shaped com-
ponents through the DT approach. The in situ measurements of melt pool
temperature are combined with a thermal computational model capable of
predicting the temperature distribution in the component. The predictions
of the model were updated simultaneously, layer by layer, with the in situ
measurements of melt pool data. The proposed approach proved successful
in detecting process parameter changes/drifts, implanted voids and delami-
nation, thereby contributing to the precise detection of flaw evolution.

12.4.3 Output prediction of AM process attributes

Enormous experiments have to be conducted to study the behavior of the
AM process, which involves time-bound effort. Such experimental studies
involve the wastage of samples in large quantities due to trial-and-error
approaches and non-destructive testing procedures [5]. DT technology
facilitates the accurate prediction of output response in the AM process,
comprising microstructural aspects, mechanical properties, surface charac-
teristics and dimensional deviations. DT makes use of the thermal, struc-
tural and material models for predicting the output responses. In light of
such models, DT provides detailed information on the evolution of defects,
the development of microstructure, temperature gradients existing over

9781032460550_C012.indd 209 28/08/23 10:23 PM

 210 Hybrid Metal Additive Manufacturing

build volume and the optimal combination of process parameters for the
best part quality. Industries demanding AM parts/components strongly rely
on DT-supported prediction models for enhanced product quality. In addi-
tion, the DT approach ensures faster prediction with minimal time and
without material waste.

12.4.4 Tool path planning and simulation

The manufacturing domain has advanced to the point where produc-
tion layout information can be communicated with the assistance of AR
between a reconfigurable AM system composed of robotic arms and the
associated DT for tool path generation [33]. The laser scan strategies in
metal AM can be optimized to fulfill faster production rates and prevent
instabilities in fabrication. Spatial relations among the various system attri-
butes comprising a part substrate, robotic arms, camera, etc. are obtained
in the form of transformation matrices to provide information regarding
production layout. The same layout information is provided as input into
the DT simulation, which ensures the convenient implementation of an
optimized layout in the physical system. The deployment of an optimized
layout can shorten the processing time and minimize the costs associated
with production.

12.4.5 Data management for manufacturing systems

From the perspective of Industry 4.0, DT can be considered a promising
technology for smart manufacturing systems and cyber-physical systems
(CPSs). The islands of information existing between various stages of the
product life cycle (due to different data types) as well as the lack of iteration/
interaction between different activities in the lifecycle can hinder proper
data management in production systems [34]. However, recent innovations
in the DT approach exhibit immense potential for supporting efficient data
management by overcoming the aforementioned challenges. Developing DT
in the cloud can open up an effective environment for data management
that supports various types of advanced data analytics, such as the real-
time status of the shop floor, machine condition monitoring and produc-
tion information [15]. Moreover, DT functions as a powerful platform for
the exchange of data/information among sensors and controllers, thereby
facilitating fruitful data visualization and analysis. Thus, the interoperabil-
ity of data can be successfully realized in AM systems using a DT-driven
approach.

12.4.6 Post-processing operations

DT technology is also adopted in the post-processing operations of AM
components. Post-treatment operations are inevitable after AM fabrication

9781032460550_C012.indd 210 28/08/23 10:23 PM

 Digital twin-driven additive manufacturing 211

to improve the part quality in terms of surface integrity, form accuracy

and mechanical properties [35,36]. The post-treatment operations can
be effectively automated by utilizing the AM dataset contributed by DT,
which comprises process parameter data, fabrication methodology and
material design. Thus, the automation efforts in post-treatment phases can
be reduced with the assistance of AM DT. Moreover, enhanced quality of
AM parts/components can be assured by integrating DTs of respective post-
processing operations with AM DTs.

12.4.7 Digital visualization of physical processes

DT enables the visualization of the AM process via the complete modeling
of the process in a virtual world. The physical process and virtual model
are linked to work together simultaneously, which enables the proper con-
trol of the process through a digital platform. The virtual demonstration
using 3D CAD model-based simulations favors a better understanding of
the state and dynamics of the actual AM process. Thus, process visual-
ization using DTs allows humans to interact with the model, analyze it
and suggest appropriate changes, thereby saving cost and time associated
with production. Moreover, industries can assess optimizations at desired
phases in the process cycle by watching DT-enabled process visualization
on screens.

12.4.8 Design optimization of product

DT facilitates the creation of a comprehensive and integrated data model
that continuously updates the attributes of the live process in real time,
thereby opening up the opportunity for effective product design. High-
performance IoT sensors, data storage and transmission systems enable the
collection of product operating data and provide feedback into the digital
product model, followed by closed-loop control of the real-time process.
The traditional way of designing products is fundamentally based on the
data obtained from simulation using historical or estimated values. Such a
procedure often leads to uncertainty in product design. Consequently, the
engineers overcome the risk in design by incorporating high safety factors,
resulting in ineffective resource utilization and the development of products
with larger dimensions. The investigation conducted by Bellalouna et al.
[37] in evaluating the efficiency of DT in optimizing product design is note-
worthy in this regard. DT technology is employed for obtaining load and
material stress data accurately from product operation in order to use the
data for efficient and load-oriented product design. In the study, the FEA
of the product is executed based on operating data rather than any histori-
cal or estimated data. Thus, dependable load analysis using operating data
aided product topology optimization and effective material utilization. The
key application areas of DT in the AM domain are portrayed in Figure 12.9.

9781032460550_C012.indd 211 28/08/23 10:23 PM

 212 Hybrid Metal Additive Manufacturing

Figure 12.9 Application areas of DT technology in AM.

12.5 CHALLENGES AND FUTURE SCOPES

Challenges: Although DT-enabled AM process control can bring promising

outcomes in the industrial domain, several challenges hinder the successful
deployment of DT technology in practical applications. Such challenging
factors are limited explainability, lack of standardization in DT structure
and data communication, privacy and security vulnerabilities, comput-
ing power requirements, prediction latency, etc. that need to be critically
addressed for proposing suitable solutions.

• System complexity and uncertainty: The DT integration on AM

is challenging due to the complexities and uncertainties associated
with the process. For a metal AM process, knowledge of uncertainty
sources is very limited so far, which in turn affects its effective model-
ing as well. Overall DT system uncertainty will exponentially shoot
up when this process-level uncertainty gets superimposed on the
intrinsic computational uncertainties [38].
• Lack of standardization: Integration of the CPS is another critical
bottleneck. The hardware and software of various components in
CPS are all heterogeneous in nature. It is a huge task for them to

9781032460550_C012.indd 212 28/08/23 10:23 PM

 Digital twin-driven additive manufacturing 213

be integrated via standard physical and communication structures

so that cross-domain synergy is achieved throughout the product
lifecycle.
Apart from the lack of standardization within a DT, another big-
gest challenge is the lack of inter-DT cross-compatibility. Each DT
is different from others in its architecture and functioning due to
the lack of standardization. Standardization facilities smoother data
communication, hardware cross compatibilities, enhanced safety
and structural integrity. The time and cost to develop DT are not
justifiable for AM if it is compatible with a single system. To tackle
this issue and widen the scope of DT implementation for AM, mul-
tiple organizations, such as ISO and IIC, are working toward a stan-
dard protocol for DTs. But this has opened up a new challenge since
several independent organizations are now working toward the same
problem without formal alignment or collaboration. Thus, even after
these standards are implemented, they may be very different from
one another, leaving the problem of interoperability unaddressed or
partially addressed [39].
• Data Challenges: Due to advancements in sensing technologies, a huge
volume of data will be acquired by real-time DTs every second. This
is especially true for an AM process since vision-based monitoring is
one of the most preferred sensing techniques. Not only does real-time
image processing demand huge computational power, but extracting
useful information from such large and complex data is also diffi-
cult. In AM, most of the physical quantities of interest, such as melt
pool temperature, heat distribution, and stress distribution, are com-
putationally intensive to simulate and predict. The cumulative effect
of computational complexities, ineffective data acquisition, storage,
analytics and communication is significant prediction latencies lead-
ing to inaccurate conclusions and delayed process controls [5]. In the
realm of AR/VR, the act of reducing the size of extensive datasets to
enhance computational efficiency frequently results in a diminished
quality of the rendered 3D experience. Selection of the most appro-
priate computing structure (edge, cloud, or fog) and dimensionality
reduction can reduce the effect of big data. In any case, real-time
data transmission through wireless mode is challenging due to the
restricted sampling rate, bandwidth and storage technology [40].
In addition, there are a lot of concerns about the privacy, security
and data-sharing aspects of DT. Since DT is an online, connected and
collaborative environment, there are plenty of privacy and security
threats. The major ones are unauthorized system access, IP hacking
and non-standard security compliance. Cross-device cloud-based
information sharing creates legal vagueness about data ownership.
The critical data-related challenges associated with DTs are consoli-
dated in Figure 12.10.

9781032460550_C012.indd 213 28/08/23 10:23 PM

 214 Hybrid Metal Additive Manufacturing

Figure 12.10 DT data-related challenges [41].

• Explainability: The AI-ML models within a DT consider multi-format

data (images, signals, videos and 2D/3D renderings) for various tasks
such as anomaly prognosis, diagnosis, response prediction, process
control, tool path generation, or in-process optimization during the
AM process. The complicated computations involved in these predic-
tions have demanded the use of advanced and sophisticated AI mod-
els. However, since these models offer a black-box-style prediction
with no feedback on the prediction logic, it is challenging to convince
the stakeholders to trust such a DT system [42]. Also, due to their
opaqueness, the current DT-driven AM systems are targeted toward
highly proficient domain experts, with regular operators with semi-
technical knowledge finding it challenging to take advantage of the
DT capabilities. Although a newer field of study is being introduced
to impart interpretability, called explainable AI (XAI), its applica-
tion scope is currently limited to very few manufacturing problems.
A typical XAI system for the AM defect prediction model is given in
Figure 12.11. That being said, the seamless integration of XAI tools
with complex and multi-model systems like DT is very challenging
and yet to be implemented.
• Scalability: The scale of DT has several contradictory points of view.
The scope of DT can be as simple as an offline system with intermit-
tent data communication. However, many argue that 2-way real-time
communication is a must for a system to be considered DT. Again,
there are contradictions on whether an online CAD visualization is
mandatory. Also, DT can be built into a simple process (online moni-
toring of the melt pool during the AM process) or product (inline

9781032460550_C012.indd 214 28/08/23 10:23 PM

 Digital twin-driven additive manufacturing 215

Figure 12.11 An explainable AI (XAI) architecture for a metal AM anomaly detection

and prediction system [43].

part anomaly detection) and can extend up to a highly complex smart

factory (starting from design all the way through metrology and
inspection). These ambiguities regarding the scope and scale of DTs
have often resulted in the development of non-robust DT systems.
Such DTs are not scalable to a smaller or larger context. The gen-
eralization and robustness of the system-level DT architecture need
further improvement to achieve varied applications, viewpoints and
contexts [44].
• Poor integration of human factors: Although completely autono-
mous systems capable of zero-defect manufacturing are the vision of
Industry 4.0, the present state of technology requires human involve-
ment at many critical junctures. However, unlike digital systems,
human operators cannot be modeled or monitored. This results in DTs
with either zero space for human intervention or DTs that get halted
until a human operator gives his inputs. A human DT is a compro-
mised solution to these extremities [45], but modeling human-specific
features such as fatigue, mood and physical attributes is extremely
challenging in an AM environment.

Future scope: In the present manufacturing context, the DT-driven AM

approach seeks significant advancement in many areas to ensure wide
acceptability and applicability in industrial applications. The potential top-
ics of fruitful research in the future consist of uncertainty reduction in DTs,
novel data management strategies, explainable AI-driven DTs, robustness
enhancement, etc.

• Elimination of uncertainty: The AM process involves extremely com-

plex metallurgical changes and thermal physics. Thus, the represen-
tation of the AM process using DT models often accepts uncertain
input data owing to the challenges in accounting for all the process

9781032460550_C012.indd 215 28/08/23 10:23 PM

 216 Hybrid Metal Additive Manufacturing

parameters and physical phenomena. Therefore, a more mature data-

driven framework with timely calibrations and seamless integration
with multiple lifecycles is required to minimize the uncertainty of the
AM DT.
• Narrowing the interoperability gaps: Immense scope persists in
reducing the interoperability gaps in DT by standardizing the DT
architecture. Extensive adoption of existing standards can favor
interoperability in the DT system. Therefore, it is meaningful to com-
pare the structural framework of various DT standards and evaluate
the corresponding alignment and contradictions for successfully over-
coming interoperability issues.
• Strategies for data management: Future research thrust can be
centered on better data-handling strategies aimed at faster data
transfer without compromising on data quality or losing out on
information. Similarly, research focus can be provided to develop
better data frameworks, architectures and computing models for
effective data utilization. Superior computationally efficient mod-
els and dimensionality reduction techniques are in huge demand
for creating a structured and systematic data management system.
Moreover, more effective tools for data analysis and management
will make DT models more scalable in a large spectrum of manu-
facturing applications.
• Process and product FP identification: Process and product FP iden-
tification is an emerging and futuristic research area with a core
emphasis on extracting the fundamental relationships between pro-
cess parameters and their functional performance. Developing such
models with the support of DT looks promising, as it enables com-
putationally efficient monitoring and quality control during the AM
process. Process FP-driven DT and real-time product FP control can
transform the way DTs are perceived until this point.
• ML and AI: ML and AI algorithms are capable of greatly enhancing
the effectiveness of DTs in the optimization and prediction of the AM
process. Research concentration can be directed toward incorporating
ML and AI into DT models to transform them into more self-learning
and intelligent models.
• Explainability of systems: Research direction toward improving the
explainability of DT-driven AM systems exhibits vast potential ahead.
DT predictions and actions for an AM process need to be explained
in a human-comprehensible manner to bring trust and transparency
to the system. In the future, deep integration of XAI with every DT
predictive model should be incorporated in such a way that the deci-
sion-making rationale at every step is made clear to the expert/stake-
holder. A high degree of system explainability assists the expert in
decision-making and implementing required system/process improve-
ment changes.

9781032460550_C012.indd 216 28/08/23 10:23 PM

 Digital twin-driven additive manufacturing 217

• Data privacy and security: A massive scope of research interest is

reflected in the areas of DT data privacy and security. Risk assess-
ment studies for DT constitute one of the first steps in this regard.
These include asking questions about the kind of data sharing and
communication necessary for the application. Data can be classified
into in-built (from AM systems and default controllers) and acquired
(from IoT sensors), followed by appropriate data governance, allocat-
ing variable data access and security. For instance, as several IPs may
not require central cloud access, IP theft can be minimized by restrict-
ing access, thereby ensuring enhanced data privacy and security.
• Robustness of DT in AM processes: The first generation of DT-driven
AM continues to be in the development phase in the manufacturing
domain. The existing DT models are limited to only a few AM meth-
ods, materials and equipment. Hence, significant research focus can
be set on improving the robustness of DT to work across various AM
machines, materials and processing technologies. Advanced material
models have to be developed for AM, which takes into consideration
factors such as defects, microstructure, part distortion and design devi-
ations that affect the mechanical properties of the component. Further,
the proper integration of AM systems with advanced algorithms and
databases can yield promising improvements in the field of AM.

12.6 CONCLUSIONS

Digital twin technology is extremely capable of transforming the AM indus-

try by addressing existing challenges and unraveling its future potential.
DT provides the manufacturers with the flexibility to simulate, predict and
control the process with minimal trial-and-error or experimental testing.
The technology will shape the future of the AM field by offering insight-
ful information on the process through better understanding, representa-
tion and analysis of the physical AM process. The integration of AI/ML
tools such as ANN, CNN and SVM further enhances the DT capabilities
through better data analytics and predictive capabilities. In short, the adap-
tation of DT will massively improve the process efficiency, productivity,
quality and cost-effectiveness of AM for manufacturers and stakeholders.
Moreover, the capability of online virtual representation, monitoring and
adaptive control will significantly leverage the reliability of the process.
Notwithstanding the fact that developing a DT for AM is still a challeng-
ing task that demands a significant volume of data, computational power
and communication resources. In addition, the scope and utilization of DT
are very much dependent on the available data and end application. In sum-
mary, DT in AM has a promising future owing to the rate of advancement
in technologies, which will enable it to overcome the current challenges and
enhance the overall repeatability, robustness and acceptability of AM.

9781032460550_C012.indd 217 28/08/23 10:23 PM

 218 Hybrid Metal Additive Manufacturing

ACKNOWLEDGMENTS

The authors gratefully acknowledge the financial support from the

UK Engineering and Physical Sciences Research Council (EPSRC, EP/
T024844/1, EP/V055208/1, W004860/1).

REFERENCES

[1] Boban, J., Ahmed, A., 2021, Improving the surface integrity and mechani-
cal properties of additive manufactured stainless steel components by wire
electrical discharge polishing, Journal of Materials Processing Technology,
291:117013, https://doi.org/10.1016/j.jmatprotec.2020.117013.
[2] Boban, J., Ahmed, A., Rahman, M. A., Rahman, M. A., 2020, Wire electrical
discharge polishing of additive manufactured metallic components, Procedia
CIRP, 87:321–326, https://doi.org/10.1016/j.procir.2020.02.023.
[3] Bandyopadhyay, A., Zhang, Y., Bose, S., 2020, Recent developments in metal
additive manufacturing, Current Opinion in Chemical Engineering, 28:96–
104, https://doi.org/10.1016/j.coche.2020.03.001.
[4] Stavropoulos, P., Papacharalampopoulos, A., Michail, C. K., Chryssolouris,
G., 2021, Robust additive manufacturing performance through a control ori-
ented digital twin, Metals, 11/5:708.
[5] Zhang, L., Chen, X., Zhou, W., Cheng, T., Chen, L., et al., 2020, Digital
twins for additive manufacturing: a state‐of‐the‐art review, Applied Sciences
(Switzerland), 10/23:1–10, https://doi.org/10.3390/app10238350.
[6] Scime, L., Beuth, J., 2019, Using machine learning to identify in-situ melt
pool signatures indicative of flaw formation in a laser powder bed fusion addi-
tive manufacturing process, Additive Manufacturing, 25:151–165, https://
doi.org/10.1016/j.addma.2018.11.010.
[7] Sturm, L. D., Albakri, M. I., Tarazaga, P. A., Williams, C. B., 2019, In situ
monitoring of material jetting additive manufacturing process via imped-
ance based measurements, Additive Manufacturing, 28:456–463, https://doi.
org/10.1016/J.ADDMA.2019.05.022.
[8] Lin, X., Zhu, K., Fuh, J. Y. H., Duan, X., 2022, Metal-based additive manu-
facturing condition monitoring methods: from measurement to control, ISA
Transactions, 120:147–166, https://doi.org/10.1016/j.isatra.2021.03.001.
[9] Liu, C., Roux, L. Le, Ji, Z., Kerfriden, P., Lacan, F., et al., 2020, Machine
Learning-enabled feedback loops for metal powder bed fusion additive
manufacturing, Procedia Computer Science, 176:2586–2595, https://doi.
org/10.1016/j.procs.2020.09.314.
[10] Mogessie, M., Wolf, S. D. V., Barbosa, M., Jones, N., McLaren, B. M., 2020,
Work-in-progress-a generalizable virtual reality training and intelligent tutor
for additive manufacturing, Proceedings of 6th International Conference
of the Immersive Learning Research Network, iLRN 2020, pp. 355–358,
DOI:10.23919/ILRN47897.2020.9155119.
[11] Phua, A., Davies, C. H. J., Delaney, G. W., 2022, A digital twin hierarchy
for metal additive manufacturing, Computers in Industry, 140:103667,
DOI:10.1016/j.compind.2022.103667.

9781032460550_C012.indd 218 28/08/23 10:23 PM

 Digital twin-driven additive manufacturing 219

[12] Yavari, R., Riensche, A., Tekerek, E., Jacquemetton, L., Halliday, H., et al.,
2021, Digitally twinned additive manufacturing: detecting flaws in laser
powder bed fusion by combining thermal simulations with in-situ meltpool
sensor data, Materials & Design, 211:110167, https://doi.org/10.1016/J.
MATDES.2021.110167.
[13] Farshidianfar, M. H., Khajepour, A., Gerlich, A. P., 2016, Effect of real-time
cooling rate on microstructure in laser additive manufacturing, Journal of
Materials Processing Technology, 231:468–478, https://doi.org/10.1016/J.
JMATPROTEC.2016.01.017.
[14] Bong Kim, D., Shao, G., Jo, G., 2022, A digital twin implementation architec-
ture for wire + arc additive manufacturing based on ISO 23247, Manufacturing
Letters, 34:1–5, https://doi.org/10.1016/j.mfglet.2022.08.008.
[15] Liu, C., Le Roux, L., Körner, C., Tabaste, O., Lacan, F., et al., 2022, Digital
twin-enabled collaborative data management for metal additive manufac-
turing systems, Journal of Manufacturing Systems, 62:857–874, https://doi.
org/10.1016/j.jmsy.2020.05.010.
[16] Ren, Y. M., Ding, Y., Zhang, Y., Christofides, P. D., 2021, A three-level hier-
achical framework for additive manufacturing, Digital Chemical Engineering,
1/June:100001, https://doi.org/10.1016/j.dche.2021.100001.
[17] Wang, C., Tan, X. P., Tor, S. B., Lim, C. S., 2020, Machine learning in additive
manufacturing: State-of-the-art and perspectives, Additive Manufacturing,
36/August:101538, https://doi.org/10.1016/j.addma.2020.101538.
[18] Abhilash, P. M., Chakradhar, D., 2021, Wire EDM failure prediction and
process control based on sensor fusion and pulse train analysis, International
Journal of Advanced Manufacturing Technology, 118/5–6:1453–1467,
https://doi.org/10.1007/s00170-021-07974-8.
[19] Abhilash, P. M., Chakradhar, D., 2022, Performance monitoring and fail-
ure prediction system for wire electric discharge machining process through
multiple sensor signals, Machining Science and Technology, 26/2:245–275,
https://doi.org/10.1080/10910344.2022.2044856.
[20] Li, X., Jia, X., Yang, Q., Lee, J., 2020, Quality analysis in metal additive
manufacturing with deep learning, Journal of Intelligent Manufacturing,
31/8:2003–2017, https://doi.org/10.1007/s10845-020-01549-2.
[21] Aoyagi, K., Wang, H., Sudo, H., Chiba, A., 2019, Simple method to construct pro-
cess maps for additive manufacturing using a support vector machine, Additive
Manufacturing, 27:353–362, https://doi.org/10.1016/J.ADDMA.2019.03.013.
[22] Guo, L., Cheng, Y., Zhang, Y., Liu, Y., Wan, C., et al., 2021, Development
of Cloud-Edge Collaborative Digital Twin System for FDM Additive
Manufacturing, IEEE International Conference on Industrial Informatics
(INDIN), 2021-July/2018:1–6, DOI:10.1109/INDIN45523.2021.9557492.
[23] Wang, Y., Lin, Y., Zhong, R. Y., Xu, X., 2019, IoT-enabled cloud-based
additive manufacturing platform to support rapid product development,
International Journal of Production Research, 57/12:3975–3991, https://doi.
org/10.1080/00207543.2018.1516905.
[24] Adnan, M., Lu, Y., Jones, A., Cheng, F. T., 2019, Application of the fog comput-
ing paradigm to additive manufacturing process monitoring and control, Solid
Freeform Fabrication 2019: Proceedings of the 30th Annual International Solid
Freeform Fabrication Symposium - An Additive Manufacturing Conference,
SFF 2019, pp. 254–267, https://doi.org/10.2139/ssrn.3785854.

9781032460550_C012.indd 219 28/08/23 10:23 PM

 220 Hybrid Metal Additive Manufacturing

[25] Wu, H., Yu, Z., Wang, Y., 2019, Experimental study of the process failure diag-
nosis in additive manufacturing based on acoustic emission, Measurement:
Journal of the International Measurement Confederation, 136:445–453,
https://doi.org/10.1016/j.measurement.2018.12.067.
[26] Abhilash, P. M., Chakradhar, D., 2022, Image processing algorithm for detec-
tion, quantification and classification of microdefects in wire electric discharge
machined precision finish cut surfaces, Journal of Micromanufacturing,
5/2:116–126, https://doi.org/10.1177/25165984211015410.
[27] Donegan, S. P., Schwalbach, E. J., Groeber, M. A., 2020, Zoning addi-
tive manufacturing process histories using unsupervised machine learn-
ing, Materials Characterization, 161:110123, https://doi.org/10.1016/j.
matchar.2020.110123.
[28] Kulkarni, A., Bhatt, P. M., Kanyuck, A., Gupta, S. K., 2021, Using unsuper-
vised learning for regulating deposition speed during robotic wire arc addi-
tive manufacturing, Proceedings of the ASME Design Engineering Technical
Conference, 2:1–13, https://doi.org/10.1115/DETC2021-71865.
[29] Ren, W., Wen, G., Zhang, Z., Mazumder, J., 2022, Quality monitoring in
additive manufacturing using emission spectroscopy and unsupervised deep
learning, Materials and Manufacturing Processes, 37/11:1339–1346, https://
doi.org/10.1080/10426914.2021.1906891.
[30] Panagiotis, S., Alexios, P., Vasilis, S., Dimitris, M., 2021, An AR based digi-
tal twin for laser based manufacturing process monitoring, Procedia CIRP,
102:258–263, https://doi.org/10.1016/j.procir.2021.09.044.
[31] Chen, Y., Peng, X., Kong, L., Dong, G., Remani, A., et al., 2021, Defect
inspection technologies for additive manufacturing, International Journal of
Extreme Manufacturing, 3/2:022002, https://doi.org/10.1088/2631-7990/
abe0d0.
[32] Abhilash, P. M., Ahmed, A., 2023, An image ‑ processing approach for
polishing metal additive manufactured components to improve the dimen-
sional accuracy and surface integrity, The International Journal of
Advanced Manufacturing Technology, 125:1–21, https://doi.org/10.1007/
s00170-023-10916-1.
[33] Cai, Y., Wang, Y., Burnett, M., 2020, Using augmented reality to build
digital twin for reconfigurable additive manufacturing system, Journal
of Manufacturing Systems, 56/April:598–604, https://doi.org/10.1016/j.
jmsy.2020.04.005.
[34] Tao, W., Lai, Z. H., Leu, M. C., Yin, Z., Qin, R., 2019, A self-aware and
active-guiding training & assistant system for worker-centered intelligent
manufacturing, Manufacturing Letters, 21:45–49, https://doi.org/10.1016/j.
mfglet.2019.08.003.
[35] Boban, J., Ahmed, A., Jithinraj, E. K., Rahman, M. A., Rahman, M., 2022,
Polishing of Additive Manufactured Metallic Components: Retrospect on
Existing Methods and Future Prospects. Springer London.
[36] Boban, J., Ahmed, A., 2022, Electric discharge assisted post-processing per-
formance of high strength-to-weight ratio alloys fabricated using metal addi-
tive manufacturing, CIRP Journal of Manufacturing Science and Technology,
39:159–174, https://doi.org/10.1016/j.cirpj.2022.08.002.
[37] Bellalouna, F., 2021, Case study for design optimization using the digital
twin approach, Procedia CIRP, 100:595–600, https://doi.org/10.1016/J.
PROCIR.2021.05.129.

9781032460550_C012.indd 220 28/08/23 10:23 PM

 Digital twin-driven additive manufacturing 221

[38] Wang, Z., Jiang, C., Liu, P., Yang, W., Zhao, Y., et al., 2020, Uncertainty
quantification and reduction in metal additive manufacturing, Computational
Materials, 6/1:1–10, https://doi.org/10.1038/s41524-020-00444-x.
[39] Jacoby, M., Usländer, T., 2020, Digital twin and internet of things-current
standards landscape, Applied Sciences, 10:6519, https://doi.org/10.3390/
APP10186519.
[40] Wu, Y., Zhang, K., Zhang, Y., 2021, Digital twin networks: a survey, IEEE
Internet of Things Journal, 8/18:13789–13804, https://doi.org/10.1109/
JIOT.2021.3079510.
[41] Coorey, G., Figtree, G. A., Fletcher, D. F., Snelson, V. J., Vernon, S. T., et al.,
2022, The health digital twin to tackle cardiovascular disease-a review of
an emerging interdisciplinary field, Digital Medicine, 5/1:1–12, https://doi.
org/10.1038/s41746-022-00640-7.
[42] Colosimo, B. M., Centofanti, F., Centofanti, F., 2022, Model interpret-
ability, explainability and trust for manufacturing 4.0. In Antonio Lepore,
Biagio Palumbo, Jean-Michel Poggi (Eds.), Interpretability for industry 4.0 ,
Statistical and Machine Learning Approaches, pp. 21–36. Springer Nature,
Cham. https://doi.org/10.1007/978-3-031-12402-0_2.
[43] Lee, J. A., Sagong, M. J., Jung, J., Kim, E. S., Kim, H. S., 2023, Explainable
machine learning for understanding and predicting geometry and defect types
in Fe-Ni alloys fabricated by laser metal deposition additive manufacturing,
Journal of Materials Research and Technology, 22:413–423, https://doi.
org/10.1016/J.JMRT.2022.11.137.
[44] Shao, G., Helu, M., 2020, Framework for a digital twin in manufacturing:
scope and requirements, Manufacturing Letters, 24:105–107, https://doi.
org/10.1016/j.mfglet.2020.04.004.
[45] Löcklin, A., Jung, T., Jazdi, N., Ruppert, T., Weyrich, M., 2021, Architecture
of a human-digital twin as common interface for operator 4.0 applications,
Procedia CIRP, 104:458–463, https://doi.org/10.1016/J.PROCIR.2021.11.077.

9781032460550_C012.indd 221 28/08/23 10:23 PM