Journal of Intelligent Manufacturing (2023) 34:21–55

https://doi.org/10.1007/s10845-022-02029-5

Machine learning techniques in additive manufacturing: a state

of the art review on design, processes and production control
Sachin Kumar1 · T. Gopi2 · N. Harikeerthana3 · Munish Kumar Gupta4 · Vidit Gaur5 · Grzegorz M. Krolczyk4 ·
ChuanSong Wu6

Received: 19 April 2022 / Accepted: 14 September 2022 / Published online: 10 October 2022
© The Author(s) 2022

Abstract
For several industries, the traditional manufacturing processes are time-consuming and uneconomical due to the absence of
the right tool to produce the products. In a couple of years, machine learning (ML) algorithms have become more prevalent
in manufacturing to develop items and products with reduced labor cost, time, and effort. Digitalization with cutting-edge
manufacturing methods and massive data availability have further boosted the necessity and interest in integrating ML
and optimization techniques to enhance product quality. ML integrated manufacturing methods increase acceptance of new
approaches, save time, energy, and resources, and avoid waste. ML integrated assembly processes help creating what is
known as smart manufacturing, where technology automatically adjusts any errors in real-time to prevent any spillage.
Though manufacturing sectors use different techniques and tools for computing, recent methods such as the ML and data
mining techniques are instrumental in solving challenging industrial and research problems. Therefore, this paper discusses
the current state of ML technique, focusing on modern manufacturing methods i.e., additive manufacturing. The various
categories especially focus on design, processes and production control of additive manufacturing are described in the form
of state of the art review.

Keywords Manufacturing · Industry 4.0 · Machine learning · Additive manufacturing · Smart manufacturing

Introduction
B Sachin Kumar
[email protected] Ever since the evolution of humankind, technology has also
B Munish Kumar Gupta
undergone evolving at its own pace. Man used stones to light
[email protected] up the fire to make up for the absence of the mighty sun.
T. Gopi
Now the technology has evolved even to replicate the sun’s
[email protected] power itself (Sivaram, 2018). Such technologies have revolu-
N. Harikeerthana
tionized the entire world. Starting with the steam engines in
[email protected]
Vidit Gaur
3 Department of Mechanical Engineering, Nitte Meenakshi
[email protected]
Institute of Technology Bengaluru, Bangalore, Karnataka
Grzegorz M. Krolczyk 560054, India
[email protected]
4 Faculty of Mechanical Engineering, Opole University of
ChuanSong Wu Technology, 76 Proszkowska St., 45-758 Opole, Poland
[email protected]
5 Department of Mechanical and Industrial Engineering, Indian
1 Department of Mechanical Engineering, Indian Institute of Institute of Technology (IIT) Roorkee, Roorke, Uttarakhand
Science (IISc) Bengaluru, Bangalore, Karnataka 560012, 247667, India
India 6 MOE Key Lab for Liquid-Solid Structure Evolution and
2 Department of Mechanical Engineering, Indian Institute of Materials Processing, Institute of Materials Joining,
Technology (IIT) Palakkad, Palakkad, Kerala 678557, India Shandong University Jinan, Jinan 250061, China

123
22 Journal of Intelligent Manufacturing (2023) 34:21–55

the eighteenth century, the industries worldwide have under- processes to produce higher-quality goods at reduced bud-
gone a substantial self- up-gradation. The nineteenth century gets (Chu et al., 2014; Gorecky et al., 2014; Jin et al., 2020;
witnessed a revolution because of one of humankind’s best Kumar & Kishor, 2021; Paturi & Cheruku, 2021; Tizghadam
inventions, the electricity. With this invention, factories et al., 2019; Vaidya et al., 2018). This new revolution of
are better equipped with self-operating mechanisms, auto- the twenty-first century will increase productivity, shift eco-
driving motors, etc. This revolution ignited the spark for the nomics, enrich industrial growth, and modify the workforce
next, which occurred very soon in the late 1960s because of profile (Gorecky et al., 2014; Vaidya et al., 2018; Wang
the invention of computers. Later on, it was universally recog- et al., 2015). Figure 1 explains the digitization process of
nized as the 3rd revolution of the industries (Ali et al., 2021; Industry 4.0, including the functional domains of its major
Hudson, 1982; Singh et al., 2013). The concept of automa- components. The various latest technologies are involved
tion began in this revolution which immensely helped mass in implementing and enabling digitization. Some of these
production (David, 2017; Dunk, 1992; Kuric et al., 2018; technologies are robots in industry, industrial automation,
Wu et al., 2015a). With the innovation of computers, sev- cloud computing, internet of things and services and AI are
eral manufacturing concepts were progressed, cumulating highly used in industrial applications. Most of these tech-
the production effectiveness (Azzone & Bertele, 2007; Clegg nologies have completed their saturation stages in terms of
et al., 2010; Gorecky et al., 2014; Hudson, 1982; Singh et al., integration in digitization (Nascimento et al., 2018; Yin et al.,
2013; Wagner et al., 2008). 2017).
It’s almost 50 years since the 3rd industrial revolution, and The term Industry 4.0 collectively refers to a wide range
the manufacturing sectors worldwide are preparing for the of current concepts, whose clear classification concerning a
next. It is the 4th Industrial revolution, a modern gear driven discipline as well as their precise distinction is not possible
by the internet and the use of computers (Frank et al., 2019; in individual cases. In the following fundamental concepts
Hofmann & Rüsch, 2017; Vaidya et al., 2018). In industry are listed:
4.0, the advantages over 3.0 were summed, and the comput-
ers were connected to facilitate mutual communication with • Smart Factory: Manufacturing will completely be
the provision to take decisions without human participation equipped with sensors, actors, and autonomous systems.
(Dalenogare et al., 2018; Frank et al., 2019; Ghobakhloo, By using “smart technology” related to holistically digi-
2020). Cutting-edge technologies like cyber-physical sys- talized models of products and factories (digital factory)
tems, the Internet of Things (IoT), and the Internet of Systems and an application of various technologies of Ubiquitous
are the major cause for Industry 4.0. The concept of the intel- Computing, so-called “Smart Factories” develop which are
ligent factory has now become a reality (Dalenogare et al., autonomously controlled (Lucke et al., 2008).
2018; Gorecky et al., 2014). Due to artificial intelligence • Cyber-physical Systems: The physical and the digital level
and easy access to more data, smart machines are getting merge. If this covers the level of production as well as that
smarter day by day. This helps industries turn out to be more of the products, systems emerge whose physical and digi-
well-organized, efficient, and optimized (Ghobakhloo, 2020; tal representation cannot be differentiated in a reasonable
Gorecky et al., 2014). Ultimately, it’s machines network, way anymore. An example can be observed in the area of
which are digitally connected for sharing the information preventive maintenance: Process parameters (stress, pro-
that results in the true power of Industry (Gorecky et al., ductive time etc.) of mechanical components underlying
2014; Vaidya et al., 2018; Wang et al., 2015). Nascent innova- a (physical) wear and tear are recorded digitally. The real
tion in technology is advancing rapidly, with an ultimate aim condition of the system results from the physical object
to help humanity in all possible ways. Industrial processes and its digital process parameters.
are also advancing in parallel with the advent of new tech- • Self-organization: Existing manufacturing systems are
nologies. Thus all the manufacturing sectors must implement becoming increasingly decentralized. This comes along
Industry 4.0 technologies such as cyber-physical systems with a decomposition of classic production hierarchy and
(CSP), big data analytics, 3-D printing, IoT, Artificial Intelli- a change towards decentralized self-organization.
gence (AI), Additive Manufacturing (AM) (Guo et al., 2020; • New systems in distribution and procurement: Distribution
Zhou et al., 2021), and Machine Learning (ML), etc., All and procurement will increasingly be individualized. Con-
these techniques held the boundless potential for sustainable nected processes will be handled by using various different
manufacturing (Chu et al., 2014; Jin et al., 2020; Kumar & channels.
Kishor, 2021; Paturi & Cheruku, 2021; Tizghadam et al., • New systems in the development of products and services:
2019). Product and service development will be individualized.
Industry 4.0 is a transformation of the manufacturing fac- In this context, approaches of open innovation and product
tories that makes it possible to gather and analyse data across intelligence as well as product memory are of outstanding
machines, enabling faster, more flexible, and more efficient importance.

123
Journal of Intelligent Manufacturing (2023) 34:21–55 23

Fig. 1 Digitization in manufacturing (Ghobakhloo, 2020)

• Adaptation to human needs: New manufacturing systems systems, an attractive and at the same time challenging task
should be designed to follow human needs instead of the for the academic discipline of business and information sys-
reverse. tems engineering BISE arises, which can secure and further
• Corporate Social Responsibility: Sustainability and develop the competitiveness of industrial enterprises.
resource-efficiency are increasingly in the focus of the Zhou et al. (2020) proposed a PD-type iterative learning
design of industrial manufacturing processes. These fac- algorithm to identify a category of distinct spatially unified
tors are fundamental framework conditions for succeeding systems with unstructured indecision. As per stability theory
products. of repetitive process, suitable measures for system’s steadi-
ness are provided as linear matrix inequalities (LMIs). To
end, efficacy of modelled algorithm is tested using simulation
“Industry 4.0” describes different – primarily IT driven – of ladder circuits. Xin et al. (2022) suggested a novel online
changes in manufacturing systems. These developments do integral reinforcement learning algorithm for resolving muti-
not only have technological but furthermore versatile organi- player non-zero sum games. They proved the convergence of
zational implications. As a result, a change from product- to the iterative algorithm and the results showed the effective-
service-orientation even in traditional industries, is expected. ness and feasibility of this design method. Stojanovic et al.
Second, an appearance of new types of enterprises can (2020) projected two classes of approaches for estimating
be anticipated which adopt new specific roles within the the joint parameter-state robustness of linear stochastic mod-
manufacturing process resp. the value-creation networks. els, collectively having major faults and non-Gaussian noises
For instance, it is possible that, comparable to brokers and which was duly supported by the experimental results.
clearing-points in the branch of financial services, analog The concept of industry 4.0 can be made possible
types of enterprises will also appear within the industry. only through nine new technologies, as mentioned in
With the planning, analysis, modeling, design, imple- Fig. 2a. These, when interlinked together, can standardize
mentation and maintenance (in short: the development) of the conventional process to the level of industry 4.0. Mass
such highly complex, dynamic, and integrated information

123
24 Journal of Intelligent Manufacturing (2023) 34:21–55

Fig. 2 a Nine technologies

transforming the industrial
product design (Industry & 4.0
Relate - & Why Manufacturers
Should Care n.d., 2021).
b Breakeven comparison
Additive manufacturing Vs
Conventional manufacturing
(CART, 2021). c Major
challenges faced by the
manufacturing sector

customization and personalization are the two influential multi-national companies, small-scale industries and SMEs
features in implementing Industry 4.0 guidelines. AM, along still are far from implementing Industry 4.0 concepts. The
with the AI & ML methods, has excellent potential for final foremost challenge lies in the installation cost and the knowl-
product personalization. Yet, there are a few difficulties with edge to work with, which is hard to accommodate for low
AM implementation in mass production. For example, the scale production units. The manufacturing people firmly
cost associated with AM implementation is significantly believe that industry 4.0 will rule all the sectors in this century
huge after a certain point compared to the conventional irrespective of the prevailing situation; therefore, large scale
machining process (Fig. 2b). and multinational, national companies have shown massive
Besides, there exists some hesitance among the industries interest to implement the same with no time loss.
for AM implementation because of its inability to produce
large-sized components of desired shape and strength. Much
research has been oriented to tackle various issues facing
implementing the AM techniques directly for mass produc- Motivation and background
tion; however, it is still under exploration. The concepts of
Industry 4.0 will be of a significant boost to all the manufac- The central focus of this article is the estimation of the level of
turing industries, especially AM and ML-based, to eradicate interest and efforts required in incorporating ML techniques
the difficulties in implementing the new technologies. The into the present manufacturing industries. Furthermore, by
combined effect and advantage of these technologies can identifying subpopulations of related and relevant literature,
yield success in multiple manufacturing domains. Though this work aims to identify critical areas of the ML technolo-
all these technologies are interdependent, they can handle gies application. Also, the paper identifies the relevant gaps
the situation independently and increase the profit mar- in the deployment of ML techniques which are presented as
gin of the manufacturing. Though it is in practice in some future research scopes. The technology available now enables
us to design and develop products as per the industrial needs.

123
Journal of Intelligent Manufacturing (2023) 34:21–55 25

With the advent of digital media, it has become a piece of cake are evident to have an optimum solution that is dependent
to search and download the research literature, which was on the variable-derived necessities. Major hurdles of ML
quite cumbersome in manual methods (Chonde, 2016). Var- integration in manufacturing have been/are on the verge of
ious search engines and repositories such as google scholar being overcome; however, their optimum association is still
and ScienceDirect are available to receive detailed infor- remained to work out. This paper also covers major ML
mation by entering the keywords of simple phrases. These techniques implementable to manufacturing, including their
customized searches are relied on the input words and do strengths and limitations. A close tuning of ML’s with the
not fetch any underlying mechanisms or concepts involved. possible product type and its future perspective should be
In this review work, Latent Semantic Analysis (LSA) was addressed before making any judgment. Still, there are sev-
used to develop the relation between sentences and docu- eral issues left unsolved that should be taken care of before
ments (Dumais, 2004). This emphasizes not just simple word implementing ML.
matching as with traditional searches but core concept justi- Thus this article is framed to address the potential domains
fication. of ML in AM processes including, the need and challenges
The enormous application domains and ease of invention associated with the AM, ML and its algorithms, their appli-
in ML techniques, which were once considered not appro- cations and limitations, ML in design of AM processes
priable or feasible, have permitted their enhanced integration (topological and material design), ML for AM processes
and realization with the modern cutting-edge epoch. With (process parameter optimization, process monitoring, defects
their spread in manufacturing, industries also welcomed ML assessment, quality prediction, close loop control, geometric
to gain economic benefits. Presently, the manufacturing sec- deviation control and cost estimation), and ML for AM pro-
tor is undergoing major changes due to great demand to duction (planning, quality control, data security, dimensional
implement intelligent Manufacturing and Industry 4.0. How- deviation management).
ever, still, there is major hesitation from several sectors,
especially mid and lower cap, due to fear of cost or training
incurred and their implementation. Thus, another motivation Challenges of the manufacturing domain
of this article is to summarize the decade’s publications to
facilitate the quantification of efforts added forward for inte- Manufacturing industries are evolved over the centuries, and
grating intelligent techniques such as ML in manufacturing. it is an integral part of a country’s economy. Few countries
Major ML application areas and popular ML algorithms are with a well-matured economy experienced reduced contri-
detailed in different sections, also highlighting ML’s limita- bution toward their GDP through the manufacturing process
tions. over the last decades. We can observe that efforts are con-
A critical examination of focus areas, and major gaps stantly put to revamp the manufacturing sectors and boost
(what advancement or research and development can be pro- the economy. In 2014, the US started an initiative ‘Executive
posed) should be taken into account so that literature/research Actions to Strengthen Advanced Manufacturing in Amer-
gaps can be filled and ML can have its widespread application ica’ to improve the country’s employment rate and support
not only in manufacturing but in other domains also. Effi- the manufacturing factories (Anderson, 2011). The Euro-
cient integration of ML is also possible when its algorithm pean Union came up with ‘Factories of the Future’ program
is judged based on real-life issues which are challenging, to identify industries of the future and nurture them in an
time-consuming, and expensive to tackle otherwise. In order optimistic way (Mavrikios et al., 2011). The manufactur-
to cater above-mentioned motivations, a detailed literature ing sectors face various problems nowadays compared to
analysis was made with a major focus on presenting recent the past, wheter it is welding, 3D printing, forming and cast-
state of the art for ML algorithms and analyzing the cross- ing (Kumar, 2016; Kumar & Wu, 2018, 2021a; Kumar et al.,
domain possibilities for ML product life cycle. 2017, 2019, 2020a). Environmental pollution, financial cri-
Typically three motivational paradigms are taken: sis, and geopolitical factors are being taken very seriously.
At present, multiple research reports are available in open
• Which ML algorithm is selected and on what parameters? literature addressing the major challenges of the manufac-
• ML method’s application frequency on a manufacturing turing sector to advance in global standards. The conclusion
process and its level? of a few is given in Fig. 2c. The above-discussed challenges
• Possible shortcomings which remain untouched, and what explain the ongoing trend of the manufacturing factories.
advancement can be made/suggested for ML’s better inte- These challenges make the business complicated and fragile.
gration with its cross-examination? The obvious complication is deeply inborn in industrial pro-
cess itself and the company’s business processes (Wiendahl
Intelligence is a primary need for learning, which is an & Scholtissek, 1994). The dynamic business scenario of man-
essential feature of smart manufacturing. The ML algorithms ufacturing industries is prominently influenced by several

123
26 Journal of Intelligent Manufacturing (2023) 34:21–55

uncertainties that complicate existing factors (Monostori, Experiment-based UQ of AM process

2003). To overcome all these complexities, sophisticated
techniques like ML and AI can be brought into the con- For UQ at the process level, AM experiments are performed
text. These techniques, which are completely driven by the repeatedly at different process parameter settings. Based on
data sources, can resolve key industrial issues. They may find the data of QoIs collected at different input settings from
extremely intelligent and non-linear data and transform it into physical experiments, the effects of process parameters on
a complex model that can be utilized for predictions, recogni- the quality of manufactured products are analyzed using sta-
tion, categorization, analysis, and projection by transforming tistical analysis (Kamath, 2016). For instance, Delgado et al.
raw input. (2012) used analysis of variance (ANOVA) to evaluate the
effects of scan speed, layer thickness and building orientation
Uncertainity in AM on dimensional error and surface roughness. Raghunath and
Pandey (Raghunath & Pandey, 2007) investigated the influ-
One of the major barriers that hinder the realization of the ence of process variables such as laser power, beam speed,
significant potential of AM techniques is the variation in the hatch spacing, and scan length on the shrinkage of the product
quality of the manufactured parts. Uncertainty quantification using signal-to-noise (S/N) ratio and ANOVA methods. By
(UQ) and un- certainty management (UM) can resolve this identifying the factors that have the most significant effects
challenge based on the modeling and simulation of the AM on the variation of product quality, the quality of AM can
process. From powder bed forming to melting and solidi- be improved by implementing quality control on the influen-
fication, various sources of uncertainty are involved in the tial factors. The Taguchi method has also been employed to
processes. These sources of uncertainty result in variability design experiments for the uncertainty analysis of AM (Garg
in the quality of the manufactured component. The qual- et al., 2014; Raghunath & Pandey, 2007).
ity variation hinders consistent manufacturing of products
with guaranteed high quality. This becomes a major hur- UQ of melting pool
dle for the wide application of AM techniques, especially
in the manufacturing of metal components. To achieve the As one of the most important models in the AM process,
quality control of the AM process, a good understanding of UQ of melting pool is of great interest to the researchers.
the uncertainty sources in each step of the AM process and For example, Schaaf performed uncertainty and sensitivity
their effects on product quality is needed. Uncertainty quan- analysis for the melting pool model to identify the most
tification (UQ) is a process of investigating the effects of sensitive parameters in the model (Schaaf, 1999). Anderson
uncertainty sources (aleatory and epistemic) on the quanti- proposed to use DAKOTA (Swiler et al., 2017) and ALE3D
ties of interest (QoIs) (Hu & Mahadevan, 2017; Hu et al., software to explore UQ of the melting process (Anderson &
2016). Even though UQ for models of physical hardware has Delplanque, 2015). Most recently, Lopez et al. (2016) per-
been intensively studied during the past decades and con- formed UQ for the metal melting pool model based on a
tinue to address important research questions, UQ in AM thermal model developed by Devesse et al. (2014). They iden-
is still at its early stage. Only a few examples have been tified four sources of uncertainty in the melting pool model,
reported in the literature (Lopez et al., 2016). In addition, namely model assumptions, unknown simulation parameters,
currently reported UQ methods for AM are mainly based numerical approximations, and measurement error (Adam-
on experiments and are performed at the process level. This czak et al., 2014). In order to reduce the uncertainty, they
will result in excessive material wastage, increased product incorporated the online measurement data into the model.
development cost, and delay in the product development pro- Based on these efforts, the effects of uncertain parameters on
cess (Hu & Mahadevan, 2017) because UQ usually requires the shape of the melting pool are studied.
numerous experiments and process optimization and UQ are
implemented in a double loop framework (i.e., UQ needs to UQ of solidification
be performed repeatedly when the process is changed). To
ensure the best quality of AM product, UQ and UM have Along with the research efforts in UQ of the melting pool
gained increasing attention in recent years. Current research model, efforts have also been devoted to UQ of the solidifi-
efforts in UQ and UM of AM processes can be roughly clas- cation model in recent years. Ma et al. (2015) used design
sified into three groups: (1) UQ of AM using experiments, of experiments and FE models to identify the critical vari-
(2) UQ of melting pool model, and (3) UQ of solidification ables in laser powder bed fusion. Loughnane (Loughnane,
(microstructure) model. 2015) has developed a UQ framework for microstructure
characterization in AM. This framework accounts for sources
of characterization errors, which are modeled using phan-
toms. Based on the modeling of error sources, effects of the

123
Journal of Intelligent Manufacturing (2023) 34:21–55 27

errors on the microstructure statistics are analyzed. Statisti- algorithm (GA) for solving a hybrid flow scheduling pro-
cal analysis and virtual modeling tools are also developed gram. The developed model gave improved results than the
for the analysis of the microstructure. Park et al. (2014) procedure adopted in the industries. Using support vector
used a homogenization method to investigate the effect of regression (SVR), Ahmed et al. (2020) estimated the splic-
microstructure on the mechanical properties in the macro- ing intensity of an unrestrained beam sample. This was one of
model. Cai and Mahadevan (2016) studied the effect of the first ML studies in architecture and civil materials man-
cooling rate on the microstructure and considered various agement. The possibilities of employing the Naïve Bayes
sources of uncertainty during the process of solidification. classification approach for deterioration detection in con-
The above literature review indicates that UQ and UM of struction were investigated by Addin et al. (2007). They
AM process are still at its early stages. demonstrated that machine learning might be used in mate-
rial science and design. Peters et al. (2007) used the random
forest algorithm to model the distribution of Eco hydrology
ML in manufacturing in ecological modelling. In ergonomics, the logistic regres-
sion technique was used by McFadden (1997) to forecast
The manufacturing industry produces a vast amount of data the pilot-error incidences of US airline pilots. Dubey et al.
every day (Chand & Davis, 2010). These data compromise (2015) demonstrated the ANN technique’s application to find
various formats, for example Monitoring information from the power delivery in pressurized heavy water reactors in
the manufacturing line, meteorological specifications, pro- nuclear engineering. The above-discussed techniques were
cess performance, machining time, and machine tool settings, also explored in various fields of engineering, which opened
to name a few examples (Davis et al., 2015). Different coun- the path to significant developments to solve complex real-
tries have used unlike names for this process; for example, time problems (Alade et al., 2020; Ali et al., 2021; Calvé &
Germany uses Industry 4.0, the USA uses Smart Manufac- Savoy, 2000; Hapfelmeier & Ulm, 2014; Piro et al., 2012;
turing, while in South Korea, it is known as smart Factory. Ramachandran et al., 2020; Seibi & Al-Alawi, 1997).
The vast amount of research publications increases the mas- Typically, the two most important aims for any industrial
sive amount of data, sometimes called Big Data (Lee et al., company are product performance and volume. To sustain
2013). Such data helps to improve the process performance healthy competition among the competitors, a firm should
by giving active feedback to the machine. The extracted use- produce sophisticated quality products on a mass scale.
ful information from the Big Data helps to expand the process Therefore, research of computing techniques such as ML and
and product quality sustainably (Elangovan et al., 2015). AI in the manufacturing domains has become very signifi-
However, the negative impact of such a huge amount of data cant in a couple of years. For the past 50 years, AI has helped
will confuse or lead to a false conclusion. If the system used to improve the process by constantly developing ML models
to manage such massive data is well-established, it is always a (Zhang & Huang, 1995). Manav et al. (2018) solved turn-
boon to the manufacturing industries. It can also be noted that ing process optimization using the GA model. Weiwen et al.
the availability of such a reliable data system helps improve (2018) developed an SVR model for sensing the breakouts for
the process quality, cost reduction, understanding of the cus- high-speed small hole drilling EDM. Sukumar et al. (2014)
tomers’ expectations, and analysing business complexity and developed the ANN model to optimize the process condi-
dynamics involved (Davis et al., 2015; Loyer et al., 2016). tions for face milling of AI alloy. Yi et al. (2020) employed
Industrial production, engineering services, materials and the random forest algorithm for performance evaluation in
processes, environmental simulation, aerospace, computers the fused deposition modeling process. In a similar context,
and privacy, nuclear physics, thermal engineering, elec- Imran et al. (2017) employed genetic algorithm techniques
tronics and communication, automotive industry, chemical for the cellular manufacturing systems. Similarly, massive
engineering, ergonomics, bio-medical, pharmaceutics, and research work has been exercised in past few decades to
commercial are some of the fields where ML has attracted improve process and cost reduction in the manufacturing
many researchers and investigators from around the world industries (D’Addona & Antonelli, 2019; Knoll et al., 2019;
(Fahle et al., 2020; Jin et al., 2020; Rai et al., 2021; Sharp Kreutz et al., 2019; Schreiber et al., 2019; Wu et al., 2017).
et al., 2018; Wang et al., 2020; Wuest et al., 2016). On Because each algorithm has advantages and disadvantages,
top of all, ML in manufacturing is an area where numer- the model’s performance is heavily reliant on the data used
ous research works are carried out. ML techniques are used to construct the model. Singh et al. (2013) created an SVR
for design, management, scheduling, material resource plan- method to assess layer stresses during the hydro-mechanical
ning, capacity analysis, quality control (Peres et al., 2019), deep design method. They examined the efficiency of the
maintenance, and automation, etc. (Fahle et al., 2020; Jin procedure using the ANN algorithm with the finite element
et al., 2020; Rai et al., 2021; Sharp et al., 2018; Wang et al., approach. Niu et al. (2017) created an SVR method to assess
2020; Wuest et al., 2016). Rolf et al. (2020) used the genetic layer stresses during the hydro—mechanical deep design

123
28 Journal of Intelligent Manufacturing (2023) 34:21–55

Fig. 3 a Pie chart representing the contribution of different ML models in recent years. b Number of journals year wise which are taken for the
study. c Chart representing the application ML in different manufacturing processes (Paturi & Cheruku, 2021)

method. They examined the efficiency of the procedure using ML concepts might be inherited in production and other sec-
the ANN algorithm with the finite element approach. Niu tors. ML has found its surge of application in manufacturing
et al. (2017) conducted comparison research to assess the twice. The first time occurred in the early 1980s when the
emission characteristics of a CRDI-assisted marine diesel computer and the internet were invented. Because of some
engine using ANN and support vector machines (SVM- practical difficulty, ML could not deliver its full potential in
Regression). Similar comparison studies have been done in problem-solving, because of which the industrial adoption of
different fields of the manufacturing industries (Acayaba & this technology was significantly poor. But after the invention
Escalona, 2015; Çaydaş & Ekici, 2012; Gokulachandran & of coding platforms with improved user interface and well-
Mohandas, 2015; Jurkovic et al., 2018; Tian & Luo, 2020). advanced computer processors, ML has once again started
A comprehensive comparison is made by Paturi et al. (2021) playing a significant role in almost all industries (Thoben
for ML applicability in different industries. The summary of et al., 2017). The focus of major industries and the research
their study is shown in Fig. 3a-c. group is mainly on developing domain-specific ML models
Though Industrial application tops the chart, ML tech- based on references from past works. It has been postulated
niques are widely used in significant manufacturing pro- that the cross-domain models pulled very little attention from
cesses like welding, grinding, and AM (Jia et al., 2021). ML ML models. This would be a handy tool to connect and inter-
in AM is one of the attractive research in AM as it forms the pret the data throughout the life cycle of any product.
stepping stone for the next industrial revolution. The range ML platform has remained separated throughout the
and architecture of the ISA-95 framework are highlighted product life cycle, including conceptualization, planning,
in Fig. 4, which also outlines key application sectors where fabrication, inspection, and maintenance. A huge junk of

123
Journal of Intelligent Manufacturing (2023) 34:21–55 29

Fig. 4 Theorized application

domains and prospects of ML
application and beyond
(Johnsson et al., 2006; Sharp
et al., 2018)

data is being generated every day with increased applica- algorithms is to find patterns or regularities that explain rela-
tion and adoption of modern manufacturing concepts such as tionships between the various causative parameters involved
the Industrial (IoT), a subsystem of Industry 4.0, and intelli- in the process. Due to the challenges of a quickly changing,
gent manufacturing. The question that remains unanswered complex manufacturing setting, machine learning (ML) as
is how well such data can be connected to developing an part of AI has the ability to understand and evolve. Hence,
ML system for manufacturing. In applications such as Total the developer has the freedom of analyzing without expect-
Conceptual models, Design for Lean Thinking, and Design ing the consequences of the situation. As a result, ML makes
for Manufacturing, understanding the numerous product life a compelling case for its implementation in manufacturing
cycle stages is critical (Garbie, 2013; Silva et al., 2014). All compared to other prevailing models. A significant power of
types of cost growth are mostly the result of decision-making. ML models is that it automatically learns from and adapt to
It was discovered that lowering the negative impacts of deci- changing situations (Lu, 1990).
sions earlier in the lifecycle could benefit a manufacturing A major challenge of ML integration with manufactur-
system’s price and performance. For these kinds of decision- ing processes is how to collect the appropriate data? It poses
making, the production needs a thorough grasp of the entire a major limitation as data’s quality, quantity and composi-
lifespan and how one action affects the other. tion are variable and depends on the end-term requirement.
Several ML models are developed to tackle massive data Thus type of data available to feed the algorithms is crucial
(Multivariate Statistical Methods in Quality Management and casts a major influence on output performance. Several
xxxx). However, factors like probable over-fitting must be times, high dimensional data may contain huge immaterial
considered in the implementation process (Widodo & Yang, and obsolete information, which can widely impact the per-
2007). Several options are available for reducing dimensions formance of the ML’s algorithm used (Yu & Liu, 2003). In the
if it ascertains to be an issue, even though it is improba- present era, major ML techniques process the data containing
ble because of the influence of the algorithms used (Pham continuous and nominal figures. The degree of influence is
& Afify, 2005). Using ML in manufacturing can be vital primarily based on the multiple variables, including the algo-
in extracting outlines from available data that can estimate rithm as well as mode of variables and their types. Hence it
the possible output (Nilsson & Nilsson, 1996). This new can be regarded as major challenge for several researchers
technique could help process owners make better decisions in the context of ML in manufacturing. Multiple times the
or automatically improve the system for a better marginal security concerns and industrial competition also restrict to
profit in the business. Lastly, the objective of specific ML gathering of the required quality data and hamper the pro-
cess of ML integration. It is important to note that in multiple

123
30 Journal of Intelligent Manufacturing (2023) 34:21–55

instances, compared to conventional methods, ML permits advancements in technology. The potential of machine learn-
better analysis of the information feed with it with consid- ing to create value out of data has made it appealing for
erably low need of raw data availability; still, its successful businesses in many different industries. Most machine learn-
integration with the system needs special attention. As men- ing products are designed and implemented with off-the-shelf
tioned by Hoffmann (Hoffmann, 1990), ML requires a lot machine learning algorithms with some tuning and minor
of time to collect the data, while in conventional methods, changes.
numerous effort is vested in extracting the useful informa- There is a wide variety of machine learning algorithms
tion from raw data. After data is collected, its preprocessing that can be grouped in three main categories:
is crucial as output will be largely impacted based on which
preprocessing algorithm is chosen. It may cause a challenge
for the training of set of algorithms used. It is pretty common Supervised machine learning
in various manufacturing processes that some atrributes or
their values are missing or absent from the master data sheet. Supervised learning (SL) permits a software application to
Thus the absent values cause major hurdles for successful learn a training data set to correctly identify unlabelled data
and reliable integration of ML in manufacting. in the test set (Learned-Miller, 2014). The ML model adjusts
Another important challenge with ML integration is the weightage of the fed input variables until a proper fit of
choosing suitable techniques or algorithms for a particular the data is attained. SL algorithms can solve almost all the
problem. Though there are multiple sets of ‘general ML tech- numerical problems in engineering. In SL, every input data
niques, specialized problems need unique methods to tackle is labeled with an output Y, and the training dataset contains
the issue and may have their own pros and cons (Hoffmann, numerous input–output combinations. Every input, such as
1990). X1; X2;…;Xn, is a vector that contains all of the attributes
As of now, a few common approaches are in trend for that may impact the performance. Every output can be an
choosing an appropriate ML algorithm for common prob- objective classification (good or unsatisfactory), classifica-
lems, which are as follow: tion as the ML category, or objective parameter (porosity
and strength), and regression as the ML class. The datasets
• First is to determine the data availability and its origin of can take many forms, including photos, audio samples, and
source (reliable data sources are preferred). Next is how text. An objective function, called the cost function, is used
it is framed, whenever it is labelled or unlabelled to mak- to calculate the error among the predicted and actual output
ing a choice between a supervised, unsupervised, or RL values. The trial phase can yield an unbiased valuation of
approach. the correctness of the model with previously unseen extra
• Second is, suitability of the ML algorithm with the problem information, known as a test set.
type and its definition. Special attention should be made At present multiple SL algorithms have been invented by
on data structure and its quantity required for training and the researchers and each of them has its special applications,
assessment. advantage, and disadvantage. However, choosing the best of
• Third is, past history of the ML algorithms applied under them for different applications is a daunting task.
similar condition to check their response time and output.
The research problem where it is applicable should have
similar background and applicability. Statistical Learning Theory (SLT)

Next critical challenge for successful ML integration It is a most suitable, widely adopted ML algorithm for
is “data interpretation”. There are multiple variables, that solving manufacturing problems. As per the SL theory, the
should be taken into to account for reliable data analysis, for algorithm training is to educate it (without being explicitly
example, format of output, type of algorithm, variables set- programmed) for selecting a function to establish a relation
ting, intended outcome, data, and its’ preprocessing. With the between inputs and output. Primary emphasis of SLT lies on
outcomes analysis, certain different boundaries might have the “extent of efficiently choosing output for an unseen previ-
more influence. ous inputs” (Evgeniou et al., 2000). Considering the base of
SLT algorithm, other practical algorithms are derived such
as NNs, SVMs, and Bayesian modeling (Evgeniou et al.,
Common ML algorithms applicable 2002). With minor changes, the SLT can fit well on multiple
to manufacturing application manufacturing applications; hence it is its major advantage.
SLT also permits to work with less number of specimens or
The prevalence of machine learning has been increasing samples needed for analysis. A major limitation of SLT is its
tremendously in recent years due to the high demand and overfitting with some realization problems (Evgeniou et al.,

123
Journal of Intelligent Manufacturing (2023) 34:21–55 31

2002). Additionally, using SLT, the computational complex- that SVM is a pretty robust and very accurate ML technique
ity is not entirely avoided but eliminated by relaxing design which is also well matched for structure–activity relation-
queries (Koltchinskii et al., 2001). ship analysis. It may be regarded as an applied technique for
STL (Wuest et al., 2014). Another fact is that it signifies the
Bayesian networks (BNs) decision boundary using a subset of the training examples,
known as the support vectors.
It is a graphical model which does describe the probabil-
ity relationship among multiple parameters. BNs can be Ensemble methods
regarded most renowned applications of SLT. Naïve Bayesian
Networks is a modified but modest form of BNs, which It combines a weighted committee of learners to solve a clas-
is comprised of directed acyclic graphs. BM claims a few sification or regression problem. The committee or ensemble
advantages, such as limited storage requirements, ability to contains several base learners like NNs, trees, or nearest
be used as an incremental learner, and ease to analyze the neighbors. At several instances, the base learners can be from
output. BNs has limitation that the tolerance for superfluous the algorithm family, which is referred as a homogeneous
and inter-reliant attributes is minimal (Joshi et al., 2019). ensemble. At the same time, a heterogeneous example is con-
structed by compounding base learners of different types.
Instance-based learning (IBL)
Deep machine learning
IBL (Kang & Cho, 2008), also referred to as Memory-Based
Reasoning (MBR) (Kang & Cho, 2008) is typically based This is a new field of ML with the capability to process the
on k-nearest neighbor (k-NN) classifiers. Though a satisfac- data in several processing layer toward highly non-linear and
tory level of accuracy with good stability can be achieved in complex feature representations. It is primarily governed by
IBL/MBR techniques, they have not proven the best match the computer vision and language processing domain but has
for ML integration (Dutt & Gonzalez, 2012). The primary significant capability for data-driven manufacturing applica-
reason for IBL/MBR’s exclusion from their implementation tions. Deep Convolutional Neural Networks (ConvNets) have
may be, among other things, their difficulty in setting the established their astonishing prediction ability in a wide spec-
attribute weight vector in little-known domains. trum of computer vision. Compared to standard NNs, where
each neuron from layer n is connected to all neurons in layer
NN or artificial neural networks (n − 1), a ConvNet is built by numerous filter stages with a
restricted view and hence most suited for image, video, and
These techniques are enthused by brain’s functionality. Con- volumetric data. From layer to layer, a ConvNet transforms
sideirng the fact that brain can intelligently complete many the output of the previous layer in a higher abstraction by
tasks (e.g. vision, speech recognition and analysis), same can applying non-linear activation.
be beneficial when dealing with engineering problems when
successfully transferred to ML systems (Nilsson & Nilsson,
1996). NN permits an artificial system to accomplish unsu- ML in additive manufacturing
pervised, reinforcement and supervised learning tasks (e.g.
pattern recognition) by simulating decentralized ‘computa- Production industries are creating large volumes of data on
tion’ of the central nervous system by parallel processing. NN the manufacturing line in this 4th industrial revolution epoch,
finds its application in multiple manufacturing sectors such often called as "Industry 4.0," and AM /3D printing indus-
as in the semiconductor industry and in diverse problems try is no exemption. AM technologies are crucial elements
such as process control. Manallack and Livingstone (Man- of Industry 4.0 concept (Alabi, 2018; Xing et al., 2020).
allack & Livingstone, 1999) suggested that NN algorithms AM denotes a group of manufacturing technologies in which
can yield best results in most cases however had the cons materials are joined directly to produce components based on
of overfitting the fed data, and it is major limitation of NN 3D modelling data. AM is a technique of creating physical
in real-life applications. Additionally, NN makes the model items from 3D modelling by layering or solidifying a mate-
more complicated and suffers from intolerance concerning rial (Guo & Leu, 2013; Kulkarni et al., 2000). AM has been
missing values and heavy computational time. found successful in developing 3D structures of various mate-
rials, as mentioned in Table 1. Any improvements to present
SVMs AM processes in design, production, and process require sig-
nificant operators’ and designers’ knowledge (Wang et al.,
SVMs were added as novel ML techniques aimed at two- 2020). Design, process, and manufacturing turn out to be fur-
group classification problems. Burbidge et al. (2001) claimed ther complicated to reap AM’s benefits (Jia et al., 2021; Wang

123
32 Journal of Intelligent Manufacturing (2023) 34:21–55

Table 1 Commonly used metals/alloys for commercial use of AM (Frazier, 2014)

Titanium Aluminum Tool steels Super alloys Stainless steel Refractory

Ti-6Al-4 V Al–Si–Mg H13 IN625 316 & 316L MoRe

ELI Ti 6061 Cermets IN718 420 Ta–W
CP Ti Stellite 347 CoCr
γ-TiAl PH 17–4 Alumina

et al., 2020). Any major alteration in the design demands in- damage to the skin and breathing problems. Table 2 listed
depth knowledge of variables and their corresponding effects the various chemicals utilized in AM processes, as well as
on the part behaviour. As a result, these objectives frequently their hazardous and ecological consequences (Huang et al.,
entail significant time and/or computation trade-offs. 2012).
In AM parameters, a limited orthogonalization is reported: In this article, the phrases 3D printing and AM are
Boosting the extrusion temperatures, for example, could used interchangeably to describe technologies capable of
enhance layer adhesion but also cumulate the stringing. Fine- making three-dimensional physical items based on three-
tuning workflow variables for particular parts or innovative dimensional digital models by layer to layer addon, curative,
materials may be proven time-consuming and uneconom- or other component processing. These technologies suppress
ical (Qi et al., 2019). Component regularity is crucial in traditional methods of extracting material from a structure
industries in which the AM integration is favoured, for exam- to produce an anticipated output shape, e.g., drilling, turn-
ple, aerospace, however, deviation in component grade of an ing, or other subtractive fabrication methods (Beaman et al.,
equipment is a hurdle to wider acceptance. The administra- 2020). While suitable to all major industries, the aerospace,
tion and evaluation of vast volumes of data and knowledge automotive, and medical sectors have been at the forefront
are all part of these difficulties. Such issues can be alleviated of AM development. The main driver of the aerospace and
by using ML algorithms to reduce the quantity of human automotive industries is minimizing component weight with-
or computational work needed for achieving satisfying out- out compromising performance. Medical applications of AM
comes. Baumann et al. (2017) provided the basis for this have a broader range of motives, while patient customization,
introduction by mentioning AM is a rapid prototyping (RP) enhanced biocompatibility, and performance are all common
technology but has far more potential than merely prototyp- themes. Consumer products commonly employ AM tech-
ing. As a result, there is no need to construct any production nique, with mass customization and light-weighting being
tools ahead of time. Furthermore, material waste is decreased two prominent motives (Gibson et al., 2021). Apart from
because the process is additive rather than subtractive. On the predefined sequence of producing predictions through
the downside, the cost per created product is significantly data modelling, experts are investigating creative and novel
higher than for mass-produced items; hence it’s best for lim- techniques to incorporate ML algorithms with the AM. ML
ited batches (Morrison, 2015). Based on the process used, techniques, apps, and frameworks are being used by AM
the quality of additively generated objects can differ and professionals to improve its quality, optimise production pro-
be inferior to that of bulk or tailor machined things. This cesses, and reduce costs. With this wide concept in account,
is usually done in discrete planar layers, although there are ML may be characterised as supervised, unsupervised, or
other non-planar techniques (Ahlers et al., 2019). AM offers reinforcement learning (Sutton & Barto, 2015).
a range of benefits compared to conventional methods, such AM has existed since nearly 1980, but it was only recently
as adaptation of mass-part parts and increased complexity in that the revolution gained popularity and budgets after the key
macro, meso, and micro scales and complicated components patent expired, notably for consumer-grade 3D printing tech-
with complicated structures and designs. As a result, AM nology. ML has been studied since around 1960 (Widrow &
has generated a high degree of research in industrial and aca- Lehr, 1990) and is based on natural notions such as the per-
demic environments worldwide in recent years. Despite these ceptron (Rosenblatt, 1958). Still, it has lately gained traction
advantages, there are some disadvantages, such as a funda- due to outstanding results produced by research organizations
mental lack of consistency (Dowling et al., 2020), which has and commercial enterprises like Google. Different studies are
made certification problematic in several industries (Thomp- now being specialized to additive production research, which
son et al., 2016). Another downside is scarcity of proper and can address fundamentals and complex challenges through
adequate acquaintance of design guidelines, despite various ML approaches. Research on ML has been conducted at
pros as discussed above (Thompson et al., 2016). Neverthe- this age, and it has been discovered that this is the era of
less, they are known to produce some side effects such as

123
Journal of Intelligent Manufacturing (2023) 34:21–55 33

Table 2 Hazardous and ecological impact of different chemicals used in AM processes (Huang et al., 2012)

S No AM Chemical/solvent Emissions Hazards of usage Biodegradability

process

1 SLA Propylene carbonate CO2 ,CO,SOx Minor system toxicity was Readily biodegradable (more
reported in rats than 80% degraded in
10 days)
Urethane resins Excessive ingestion may Not reported harmful to
result vomiting atmosphere
Tripropylene glycol Minor irritation after eye Can be biodegraded by 50%
contact in just 8.7 days, and by
No other effect to skin 81.9% over a 28-day test
period
Isopropanol Irritation and burning in Has a potential to acutely
eyes decrease oxygen from
and occasionally corneal aqueous systems
injuries; irritation
and soreness on skin and
prolonged
exposure may have
dermatitis
2 SLS Polyamide resin CO2 No serious issues were Form inflammable mixture in
seen while contact with some
handling or exposure to chemicals or long exposure
this chemical to air
Acrylonitrile butadiene Molten plastic may Since it is insoluble in water,
styrene develop lethal burns, its eco-toxicity is low
processing fumes may
result eye irritation and
choking of the
respiratory tract
3 LENS Photopolymers CO2 ,CO,SOx Inhalation may cause No hazardous decomposition
ulceras, throat burning products
and coughing; skin and
eyes may have redness,
irritation and swelling
4 FDM Propylene glycol CO2 ,CO,SOx ,PMc,NOx Irritation in eyes, skin, No hazardous decomposition
monomethylether nose, throat; headache, products
nausea, dizziness,
drowsiness,
incoordination; vomiting,
diarrhea

database creation since a big quantity of knowledge is gener- produce a large quantity of bug-free components or finished
ated everyday across numerous networks, production, online products that satisfy high failure criteria also constitutes his
networks, pharmaceuticals, aviation, 3D printing, automo- challenges. Parameter variations during construction were
biles, and telecommunications are just a few examples. recently revealed as factors for the creation of defects in
ML is a discipline of AI that enables a device to train from a the AM sector. This could help the AM industry to iden-
dataset obtained from various sources and perform intelligent tify potential areas of trouble with the finished product of
activities, such as conducting complex processes by gather- the equipment and even perhaps produce more strong build-
ing accessible data, rather than following a pre-programmed ing procedures and cost-cutting strategies. Computer vision,
technique (Craig et al., xxxx). ML mainly deals with massive prediction, and information retrieval are three major compo-
data quantities. The sector has inevitably resorted to mas- nents that find their direct applications in AM processes. With
ter learning approaches because of the vast amount of data the implementation of cutting-edge techniques in graphics,
collected throughout the AM build process. As the supply hardware has allowed their in-depth research, allowing quick
sector increases, the difficulty of constructing industrial addi- optimization of ML algorithms on massive data (Shinde &
tive machinery or 3D entry-level desktop printing devices to

123
34 Journal of Intelligent Manufacturing (2023) 34:21–55

Fig. 5 a Demonstrates a taxonomy of machine learning applications in output layer that is fully interconnected. (Grierson & Quayle, 2021).
the AM domain (Meng et al., 2020). X1 ; X2 ; …;Xn represents the input Neurons are symbolized by circles, which apply activation functions to
vector, which includes various input characteristics, and Y represents the total sum of the activation and weight pairs from the previous layer
the output. b An NN having one input layer, 2 hidden layers, and one

Shah, 2018). As a result of these advancements, ML solu- the abovementioned backpropagation process. At the same
tions can now be used in AM for better productivity. As a time, deep learning is relied on an ANN with multiple lay-
result, the focus of this section is on AM technology and ers, including hidden layers (i.e., layers that are neither
how it might be used in conjunction with ML. input nor output), and also convolutional and pooling lay-
Figure 5a depicts the ML taxonomy and the corresponding ers that are only regionally connected (i.e., not all neurons
implementations in AM. A well-known ML approach is Arti- in one layer are linked to neurons in the next layer). Neu-
ficial Neural Network (ANN). Its prominence has grown in rons may be arranged in layers. By layering and connecting
tandem with the advancement of processing power, notably these levels with neighboring layers, convolution neural net-
with the adoption of GPUs and their simultaneous compu- works (CNNs) are constructed. In the existing AM industry,
tational power for easy mathematical analysis. An ANN is variability in manufactured components’ functionality that
comprised of different layers, each containing a set of fun- relies significantly on numerous processing variables, such
damental neurons (Fig. 5b). The perceptron, designed by as printing speed and film thickness, is a key hurdle. Vari-
Rosenblatt in 1958, was one of the first studies on ANN ous review articles have looked into the relationship between
(Rosenblatt, 1958). The Heaviside step function serves as process, structure, and property (Kumar & Kar, 2021; Kumar
an activation function, calculates response by input values, & Kishor, 2021; Singh et al., 2017). Experimenting or run-
and defines the neuron by its own weight(s) in that context. ning high-fidelity simulations is one option to deal with this
A layer’s output is the sum of all of its neurons. During problem, acquire reliable data, and aid in adjusting process-
the training phase of a multilayer ANN, backpropagation ing parameters. Still, both are time intensive or pricey, or
may be employed for increasing each neuron’s weight. It both. The use of in situ monitoring systems is another way to
can be obtained by manipulating loss function’s gradient, ensure part quality and process dependability, but an effec-
representing a distance between the ANN’s output and the tive defect detection approach based on in situ data such as
training data, and then propagating the errors backward. pictures is required. A robust and accurate data processing
The subsequent network may be a (classical) feedforward and data extraction tool is required in both directions. These
neural network (FNN) if the network layers are ordered challenges are being sorted out by ML, a branch of AI. With a
sequentially, i.e., if no backward connection exists between trustworthy dataset, ML algorithms can acquire information
separate layers; otherwise a recurrent neural network (RNN). from the training set and create conclusions relying on that
When training RNNs, long short-term memory networks information. On another side, trained ML algorithms might
(LSTMs) deal with the issue of disappearing gradients in have predictions and identify the ideal operational settings.

123
Journal of Intelligent Manufacturing (2023) 34:21–55 35

On the other side, ML algorithms use in-situ data to detect ML methodology for AM design feature suggestions while
defects in real-time. More ML applications, such as geomet- the designing phase used a clustering algorithm. Figure 7a
ric variation control, cost estimate, and quality analysis, have shows a typical case study in this context. In contrast, this
been documented in recent studies (Jin et al., 2020; Razvi work used no TO techniques and instead substituted the
et al., 2019). In general, ML applications can be considered heavy pieces with lightweight components retrieved from
a type of data processing. Therefore ML integration can be a set in the initial model. The CNN was utilized to skill
regarded as an important part of Industry 4.0. the in-between topologies acquired using typical TO tech-
As per ASTM F42 (ISO & ASTM5, 2900 - xxxx), various niques in order to solve a mechanical malfunction. The TO
AM methods can be classified into major groups, as shown algorithm was interrupted after only a few iterations to antic-
in Fig. 6a. Materials jetting (Wang et al., 2018a) and stereo- ipate the optimal designs at an interim phase. With some
lithography (Lee et al., 2013) are also the classifications of unusual pixel-wise tweaks, the learned CNN algorithm may
AM processes; however, they will not be discussed here in anticipate eventual topology optimisation by nearly twenty
detail. To better comprehend the advantages of ML in man- times quicker than standard simplified isotropic material with
ufacturing processes, three broad categories are proposed penalization (SIMP). Developed network may be employed
and shown in Fig. 6b. The aim here is to demonstrate how for handling heat flux issues, and when applying threshold-
these groups affect successful AM integration and informa- ing, and exceeded SIMP in terms of performance as well
tion protection planning and scheduling. These processes will as numerical precision. This establishes CNN model’s broad
be dealt with in detail in the forthcoming section. As com- generalizability without requiring knowledge of the prob-
plied by Shinde and Shah (2018), major application domains lem’s nature. This technology was expanded by Banga et al.
of ML technique can be grouped as computer vision, pre- (2018) to construct 3D structures. When the FEM-based
diction, semantic analysis, natural language processing, and SIMP technique was used exclusively, it could anticipate the
information retrieval (Fig. 6c). final designs fairly soon, with an average numerical preci-
sion of 96.2% and a savings in time of 40% once trained.
Machine learning in design for additive The generative adversarial network (GAN) can predict the
manufacturing (DfAM) optimized structures without SIMP iterations. GAN is a gen-
erative programming approach that incorporates advanced
DfAM is a subset of Design for Manufacturing and Assembly deep learning techniques (Goodfellow et al., 2014). The well-
(DfMA), although it differs from conventional DfMA in sev- learned GAN model may create a huge sum of unrecognized
eral ways. Because AM can produce complicated structures design having intricate geometry which fit design specifi-
that are impossible to fabricate using traditional manufac- cations based on the limitations and variables. Figure 7b-d
turing processes, designers are rethinking the conventional shows several examples of these studies. All of the train-
DfMA process used in AM (Thompson et al., 2016). AM also ing data was generated via the usual TO approach, which
avoids the assembly phase because it may fabricate the entire should be stressed. As a result, while ML cannot com-
product in a single step. DfAM (Ponche et al., 2014; Thomp- pletely replace the classic TO technique, it can be utilized to
son et al., 2016) is a novel word that considers the unique reduce the number of iterations and accelerate the optimiza-
possibilities of AM as well as the differences between tra- tion method. Furthermore, the ML-centered TO technique
ditional manufacturing and AM approaches. The processing might be utilised for a speedy, approximate forecast of initial
of DfAM comes under the following categories: information too. But at the other side, the above-mentioned
undertakings have yet to be conducted in additive manufac-
Topology design turing.

Topology Optimization (TO) (Bendsøe, 1999) is a method

for designing structures that optimizes material distribution Material design
inside a design region whilst taking into account spe-
cific stresses and restraints.TO procedures typically include Materials experts and researchers have created a variety
multiple designs and prototyping repetitions, making them of metamaterials, which are composites with distinguish-
technologically intensive, especially for mass scale and dif- able properties. Manually developing metamaterials using
ficult to made components. Once the ML models have been the Edisonian method is extremely difficult and laborious.
correctly learned, they could provide favorable ideas without The synthesis of metamaterials can be considerably accel-
starting over, allowing the ML-centric technique to comple- erated using modern ML approaches (Gu et al., 2018).
ment the classic TO approach. Unfortunately, there is still a Thanks to recent developments in ML, material specialists
paucity of research on employing ML for topology design for and researchers can now predict material properties to invent
AM applications. Yao et al. (2017) presented an integrated new metamaterials. Furthermore, as several researchers have

123
36 Journal of Intelligent Manufacturing (2023) 34:21–55

Fig. 6 a Categorisation of typical AM processes according to ASTM F42. b Applications of ML techniques in numerous AM domains. c Major
application domains for ML

proved, AM techniques can materialize previously impossi- kinds of unit cells on an 8 by 8 lattice structure, accounting
ble ideas to construct. The synergistic potential of cutting- for fewer than 10–8% of all possible combinations. Convolu-
edge ML in materials and AM techniques is largely untapped. tional neural networks (CNN) were then utilized for training
Chen et al. (2018) devised a fully automated method for a database that included mechanical parameters calculated
identifying ideal metamaterial designs, which were then using the FEM, yielding innovative microstructural patterns
experimentally confirmed using the PEBA2301 elastic mate- for a composite metamaterial twice strong and forty times
rial and a selective laser sintering (SLS) procedure (Fig. 8). hard. The multi-material jetting AM method was used to val-
Given the intended elastic material parameters, such as idate their ideas (Fig. 9). One notable difference is that FEM
Young’s modulus, Poisson’s ratio, and shear modulus, the simulation required around five days to calculate mechanical
system is expected to produce a customized microstruc- characteristics. Still, CNN only took ten hours to learn and a
ture that meets the specification using ML technique. Gu few seconds to generate the output.
et al. (2018) constructed 100,000 microstructures using three

123
Journal of Intelligent Manufacturing (2023) 34:21–55 37

Fig. 7 a Existing target components design versus re-designed SLM- et al., 2020). The expected results are compared to real data obtained by
made components after a guideline for hybrid machine learning (Wang traditional TO procedures(Sosnovik & Oseledets, 2019). c The similar
et al., 2020; Yao et al., 2017). b CNN was utilised to predict the best approach is utilised for 3D structures(Banga et al., 2018). d GAN was
structures from then intermediate topologies that served as inputs (Wang used to create TO structures (Rawat & Shen, 2018)

Machine learning for additive manufacturing "resultant processing data," in common PSP connections. It
processes is possible to discriminate between data accessible before-
hand and after processing.
ML is a tool for data processing. The data that can be eval-
uated and used for the Process- Structure–Property (PSP) • Extruder temperature in ME, laser power in L-PBF, print-
relation chain are presented in Fig. 10. The term "process" ing pace, and film thickness all have a significant effect on
is separated into two words , "processing parameter" and printed items and therefore influence their reliability and
productivity;

123
38 Journal of Intelligent Manufacturing (2023) 34:21–55

Fig. 8 Illustrates a computational workflow for identifying extremal microstructure families based on elastic material parameters (Desai et al., 2018;
Wang et al., 2020)

Fig. 9 The optimisation of the microstructure by ML could help to build stronger and more durable materials (Gu et al., 2018; Wang et al., 2020)

Fig. 10 Shows the link between process, structure, and property in AM are shown by highlighted text. The source and terminus of each
additive manufacturing. The available data for machine learning is rep- arrow reflect the source and outgoing values, respectively (Meng et al.,
resented by the text in the boxes. Some known ML implementations in 2020)

123
Journal of Intelligent Manufacturing (2023) 34:21–55 39

• The intended structure has a major impact on production empirical observations, the powder-based and wire-based
charges and geometric variation in produced items; DED (wire-based) processes were predicted using a multi-
• The obtained in situ images and acoustic emissions (AE) layer perceptron (MLP). The process variables were thus
can be used for the identification and type of problems in inexorably linked to the melt pool morphologies. It thus
real-time by surveillance systems. indicates a particular geometry can be attained by adjust-
ing parametric combinations in the opposite direction. In
As a consequence, ML models trained on a given dataset the 3D response maps to melt pool depths vs. process set-
comprising a minimum of two categories of linked data in tings, Tapia et al. (2017) used the Gaussian Process-base
PSP network will be able to derive conclusions from this (GP) substitute model as shown in Fig. 11a-c. As a result,
input. It is the most common approach to ML model imple- parametric combinations can be defined to eliminate forma-
mentation. tion of keyhole melting. A mixture of one empirical dataset
and two additional literature data sets have been used for the
Process parameter optimization 139 data sets. Several ad hoc filters have been implemented
to decrease abnormalities, resulting in 96 valid data points.
Developers will not know the grade of a part developed with Their 6.023 μm preview error is acceptable since it was
a particular set of processing parameters until it is manufac- equivalent to data gathering faults. The porosity of AM-built
tured. Consequently, several steps must be taken to ensure parts is another primary concern at the mesoscale. Total den-
product performance, such as printing specimens and validat- sity is the prime purpose in metal AM since the mechanical
ing their efficiency, making the design costly, time-intensive, behavior of components is heavily affected by porosity, espe-
as well as unpredictable. Therefore, a direct link between cially by fatigue. Multi-Layer Perceptron (MLP) can depict
governing variables and product performance can be highly complex non-linear interactions with little insight into how
advantageous. Tests and computations are valuable tools for predictions are produced. Moreover, the unsafe nature of the
establishing a connection, however, getting optimal variables prediction outputs may be usually evaluated, but when the
is difficult when multiple input parameters are tangled. ML same amount of input information is given, it is substantially
techniques can be utilized in the form of substitute models to costly than the MLP. As a result, MLP and GP combined with
enhance operational efficiencies (Wang et al., 2018b, 2019). Bayesian approaches were used to estimate porosity in selec-
To additively build new materials, process parameter devel- tive laser melting (SLM) based on combinations of process
opment and optimization have traditionally been carried out factors, as shown in Fig. 11(d)–(e). The PLA sample open
using the design of experiments or simulation methods. How- porosity during SLS processing was estimated using SVM
ever, in the case of metal AM, developing an experimental and MLP algorithms.
strategy often necessitates a lengthy and costly investigation Multi-gene Genetic Programming (MGGP), though lim-
process (Wang et al., 2018b, 2019). ited by the system’s generalization, is a cutting-edge tech-
A physical-based simulation can be used to demonstrate nique that may robotically develop the shape. The printing
the theoretical underpinnings for the production of various of a 58 wt% HA + 42 wt% PA powder mix utilizing SLS was
features during handlings. However, macro-scale models, done to achieve the necessary open porosity values through
such as FEM, may have inaccuracies with experimental out- adjustment of process parameters. Ensemble-based MGGP
comes owing to decreased assumptions. Single tracks or a with a better generalizing capacity.
few tracks and layers are frequently the subjects of increas- ML technique can also be used to investigate the macro-
ingly more advanced approaches, such as computational fluid scale characteristics of AM-built objects. A fuzzy inference
dynamics. This makes predicting the mechanical character- system based on adaptive systems (ANFIS) can usually only
istics of pieces on a macro scale or in a continuum complex. handle partial values. As a result, since there are so many
As a result, several researchers have looked at the poten- unknowns in the fatigue process, it’s beneficial for analyzing
tial of using ML to overcome the issues mentioned above fatigue attributes. Zhang et al. (2019) acquired 139 SS316 L
in metal AM process optimization, as shown in Table 3. ML fatigue data that had been manufactured under 18 different
was found to be primarily used as a link to two levels of qual- treatment configurations on the same SLM equipment. The
ity criteria for significant process parameters: mesoscale and ANFIS was effectively used with the ’process based’ model
macroscale. Moreover, some scholars have employed process and the ’property model’ to predict high-cyclical fatigue with
maps as a means of finding process frames. These process an average root squared error of 11–16%. However, when
maps can be a valuable tool for further analysis. Single routes they used the training set with the 66 data points to estimate
are the main construction elements in the mesoscale of high- fatigue life, the performance of their algorithms was reduced,
energy AM. Topology of the weld pool can have a substantial owing to the variability of machine-to-mechanical systems.
influence on the finished quality of product, for example, It is therefore advised to use both empirical and bibliograph-
shape, continuation and consistency. Because of insufficient ical inputs in model training in order to increase its ability for

123
 40

Table 3 ML approaches for optimizing parameters in the AM process (Wang et al., 2020)

AM Materials Builds Sensors Input Features ML Methods Output targets References

Processes

123
Acoustic –Based Monitoring
SLM SS304 Single tracks Microphone Acoustic signals DBN,MLP,SVM Classify melting states Ye et al., 2018a)
SLM SS 10 × 10 x 20mm3 Fibre Bragg grating Acoustic signals Special CNN Classify of build Shevchik et al., 2018)
cubes sensor qualities
FDM ABS - Acoustic emission Acoustic signals K-means clustering Identify failure mode Wu & Yu, 2016)
sensor
FDM - - Acoustic emission Acoustic signals Hidden Identify extruder state Wu et al., 2016a)
sensor semi-Markov
model
Optical-Based Monitoring
SLM SS316 L Single tracks High-speed camera Layer-wise images of SVM,CNN Detect anomalies of Zhang et al., 2018)
melt pools, plume melt tracks
and spatter
SLM Zinc 5 × 5 x 5mm3 cubes IR camera IR images(plumes and Unsupervised ML Detect unstable melting Grasso et al., 2018)
laser-heated zones) conditions
SLM In718 Single Tracks, High-speed camera, In-situ and ex-situ SVM Detect keyhole Scime & Beuth, 2019)
unsupported optical morphologies of the porosities and balling
overhangs microscope(ex-situ) melt pool instabilities
SLM SS Step cylinder Digital single lens reflex Layer-wise images SVM Detect and locate Gobert et al., 2018)
camera, CT under 8 lighting anomalies
scan(ex-situ) conditions, ex-situ
CT scan data
SLM In625 40.5 “ unsupported Photodetector, high- Intensity, morphology, MLP,SVM,KNN Distinguish between the Montazeri & Rao,
overhangs speed camera, IR thermal profile of overhang and bulk 2018)
camera melt pools build states
SLM SS304 Single tracks Near –IR camera Plume and spatter DBN,CNN,MLP Classify melting states Ye et al., 2018b)
signatures
SLM SS304 8.5 × 8.5 x 4mm3 High- speed camera Images and locations DBN Classify MP images Kwon et al., 2018)
cubes of melt pools concerning laser
power
SLM In718 Tensile Bars Visible light and IR Plasma emission and Gaussian mixture Detect faulty bars Okaro et al., 2019)
photodiode sensors thermal radiation of model
melt pools
SLM SS316 L Ø16 × 44 mm3 High- speed camera Intensity profile of K-means clustering Detect and locate M. G, V. L, Q. S, B.M
cylinder,50 × 50 × melt pools defects due to C. xxxx)
50 mm3 lattice overheating
structure
Journal of Intelligent Manufacturing (2023) 34:21–55
 Table 3 (continued)

AM Materials Builds Sensors Input Features ML Methods Output targets References

Processes

SLM - Hollow cylinders Optical camera Layer-wise surface RF,SVM Detect elevated regions Jacobsmühlen & zur,
images before and after laser exposure 2015)
after powder coating
SLM In718 Ø40 × 20 mm3 Digital single-lens reflex Layer-wise surface CNN,SVM To recognize defects Caggiano et al., 2019)
cylinder camera images before and induced by process
after powder coating non-conformities
DED Ti-6Al-4 V Thin walls Pyrometer, IR camera, Thermal profile and SOM Detect the location and Jafari-Marandi et al.,
CT scan(ex-situ) location of the melt size of pores 2019)
pool
DED Ti-6Al-4 V Thin walls Pyrometer, IR camera, Morphological and SOM Detect location of pores Khanzadeh et al., 2019)
CT scan(ex-situ) thermal
Journal of Intelligent Manufacturing (2023) 34:21–55

characteristics of the
melt pool
FDM PLA,ABS DNA model Optical camera Images of parts at SVM Classify good and Delli & Chang, 2018)
specified defective parts
checkpoints
FDM - Parts with different Cameras mounted on Simulated images KNN, RF, Detect malicious infill Wu et al., 2017)
types of infills both extruder and from software, real unsupervised ML structure
frame of printer images from camera

123
41
42 Journal of Intelligent Manufacturing (2023) 34:21–55

Fig. 11 A GP-based model for predicting mesoscale features of SLM- depth based on experiments (c) and modelling. PH samples from SS17-
fabricated parts (Wang et al., 2020). Samples of SS316 L(Tapia et al., 4 (Tapia et al., 2016). d Test coupons that are exactly as they were built.
2017): a A single track optical micrograph. B Prediction of melt pool e The observation’s spatial behaviour throughout the process parameter
grid. f Forecast of porosity for every power-speed combination

generalization. Wang et al. (2018b) argued that examining the dynamics, creep, and restorative properties were thoroughly
top surface morphology might assist in narrowing down the assessed in estimating material tensile parameters of PLA
electron beam melting process window (EBM). SVM, for and PC-APS materials.
example, works well when the proximity between classes
is apparent, but it is simple to overfit. As a result, Aoyagi
et al. (2019) proposed a straightforward method for con-
Process monitoring
structing EBM flowcharts using only 11 datasets. It should
also be noted that SVM was used in this study solely to fit the
While parameter optimization can aid in process predictabil-
data to identify the decision boundaries. Because the training
ity, it cannot totally eradicate failures (Kwon et al., 2018).
dataset was so small, assigning a training dataset to evalu-
Process monitoring approaches that can detect build failures
ate the model’s accuracy was problematic. Recurrent neural
and defects are required since print problems account for a
networks (RNN) are being used to anticipate time series. To
considerable portion of the cost of AM parts. Various ML
determine the high thermal history of complicated compo-
solutions have attempted to handle this challenge, and they
nents of the DED process as established by Mozaffar et al.
are categorized into two groups depending on input data type:
(2018), RNN was therefore utilized for training FEM data in
optical and acoustic. The most frequently applied monitoring
view of the time dependence of the inputs. In addition, both
solutions are those that utilize information from digital, high-
MLP and SVM have tried to forecast the thin wall deposits
speed, or infrared cameras. The melt pool is a typical target in
for DED.
PBF operations, at which majority of monitoring investiga-
The study focused on the mechanical characteristics of
tion are oriented. Kwon et al. ( 2018) used melt pool thermal
macroscopic dimensions in the AM extrusion material pro-
data to train CNN-based software for distinguishing high to
cess. For process parameters that had been carefully studied,
low-quality constructions having failure rate less than 1.1%,
FDM includes the thickness layer, temperature, and struc-
potentially saving time and money. Zhang et al. (2020b)
ture guidance. Here, the MLP is utmost widely employed
uncovered that integrating melt pool, plume, and spatter data
methodology. An adequately trained MLP is preferable to
to categorize component quality yields best results. A form
the precision and prediction of the system’s nonlinear data.
of NN known as a long–short term memory network has been
The usage of the compressive strength, wear rate, elasticity
determined much better prediction (Zhang et al., 2021). Other

123
Journal of Intelligent Manufacturing (2023) 34:21–55 43

AM techniques, such as binder jetting and material extru- powder distribution can cause a variety of problems, includ-
sion, have also benefited from optical monitoring. Gunther ing warping and swelling, which can cause the entire build
et al. (2020) employed an optical tool for analyzing defects in to fail. Re-coater striking curled-up or humped compo-
binder jet parts. The techniques used here were not disclosed, nents, re-coater dragging impurities, re-coater blade damage,
and there was no discussion of the model’s accuracy. Opti- debris over powder bed are all examples of powder spread-
cal tracking has been used in material extrusion to detect ing problems. Furthermore, eliminating the requirement for
defects in real-time. Wu et al. (2016b) employed a classi- human-made detectors for specific abnormalities is highly
fication method for detecting infill print faults in material desirable. To that purpose, a mechanism for detecting and
extrusion, giving them more confidence in final product’s classifying powder spreading faults autonomously while
quality. The study had a 95% accuracy rate; however, it didn’t build was introduced. Scime and Beuth used modern com-
consider other essential quality metrics like precision and puter vision techniques such as k-means clustering (Scime
recall. Li et al. (2021) determined dimensional deviation with & Beuth, 2018a) and multi-scale CNN (Scime & Beuth,
zero mean error and a standard deviation of 0.02 mm using 2018b) to learn the algorithm accurately. They also categorize
in situ optical monitoring of a material extrusion process. powder-bed picture patches into seven kinds using photos
Sun et al. (2021) employed adaptive fault detection and root- recorded throughout the SLM process. This technology also
cause analysis using moving window KPCA and information allows for in-process repair of flaws in the AM process when
geometric causal inference. They noted that this scheme had a feedback control system is used.
good performance in reducing the faulty false alarms and
missed detection rates and locating fault root-cause. Said
et al. (2020) added a new Fault Detection method applicable Defect detection, quality prediction, and closed-loop
to the process monitoring using Kernel partial least squares control
(KPLS) in static and dynamic forms. They noted that results
obtained from reduced kernel partial least squares (KRPLS) In situ monitoring devices have advanced to the point that
have demonstrated the efficiency of the developed technique real-time data can be collected for defect detection and
in terms of false alarm rate, good detection rate and compu- closed-loop control in AM. ML models may employ actual
tation time, compared with the conventional fault detection information like spectroscopy, images, AE, and computed
KPLS. Lee et al. (2020) proposed Kernel principal compo- tomography (CT) in a variety of ways, as shown below.
nent analysis and found it can effectively monitor the tool
wear. The proposed method can effectively integrate multi-
sensor information and synthesize the data to estimate the • Identify the information with error (potentially with defect
state of the process. categories) or not by experimental tests or human exper-
A more recent and comparatively cheaper means to mon- tise. Then, train supervised learning models for detecting
itor the build throughout the printing process is called defects and performance projection in real-time, which is a
acoustic monitoring. These methods depend on acoustic sig- frequent implementation of ML classification techniques.
nals related to part porosity and melt states in PBF processes • Execute cluster assessment with unsupervised learning
and material extrusion process failures. In contrast to optical approaches for abnormal data to be clustered and detect
monitoring, acoustic monitoring systems have cheaper sen- faults without labelling.
sors. The ML algorithms used range from supervised CNNs • Build the ML regression models utilizing data from some
to clustering solutions. Acoustic monitoring effectively high- real-time adjustable process conditions to modify these
lights problem builds with reduced requirement for post-print processing parameters in real-time. The third way is illus-
examination and testing, with confidence levels of up to trated by voltage level control in the MJ process. Their
89% for porosity classification and 94% for melt-pool-related process control structure is made up of three basic com-
defects. Extrusion procedures have also been subjected to ponents, as indicated in Fig. 12. A charge-coupled device
acoustic monitoring. Wu et al. (Wu et al., 2015b) employed (CCD) camera captures the dynamic photos of the droplet
acoustic monitoring and an SVM classifier to assess if the first. Secondly, the images are used to extract four droplet
extruder was pushing out material with 100% accuracy. The properties (satellite, ligament, quantity, and pace), which
SVM could detect extruder obstructions with 92% accuracy are then coupled with the current–voltage to create a neu-
(normal, semi-blocked, or blocked). ral network (NN) ML algorithm. Finally, the trained ML
model is used to determine the appropriate voltage level
Powder spreading characterization and send it to the voltage modification system, which reg-
ulates the droplet jetting pattern.
The degree of consistency for powder distribution in the
PBF process is critical to quality of final products. Improper

123
44 Journal of Intelligent Manufacturing (2023) 34:21–55

Fig. 12 The MJ program’s

closed-loop voltage regulation
architecture (Meng et al., 2020;
Wang et al., 2018a)

Fig. 13 The geometric error compensation approach for Ti-6Al-4 V in The distortion in the CAD model is reversed to compensate for errors.
the L-PBF process is shown (Meng et al., 2020). The input data consist CAMP-BD is a convolutional and ANN for predicting additive manu-
of a the thermal history and b the processing parameters. The pre- facturing using big data (Francis & Letters, 2019)
dicted distortion using the deep learning model is c the output data.

Geometric deviation control compensation framework for the L-PBF process using a con-
volutional neural network (CNN) ML model. The trained ML
AM parts commonly have low geometric precision and sur- model can anticipate distortion by considering thermal data
face integrity(Grasso, 2017). These geometric flaws obstruct and a few process variables for feeding and distortion as an
AM’s use in a variety of fields, including aerospace and outcome, which is then fed back into the CAD model for
pharma. Under such scenario, ML models can recognize error detection and correction. The geometric accuracy of
geometric faults, quantify the deviation, and make recom- items manufactured using the adjusted CAD model will be
mendations for correcting the flaws. As shown in Fig. 13, greatly enhanced due to this method.
Francis et al.(Francis & Letters, 2019) built a geometric error

123
Journal of Intelligent Manufacturing (2023) 34:21–55 45

Fig. 14 A framework for cost analysis on account of geometry and process likenesses (Chan & Lu, 2018; Meng et al., 2020)

Cost estimation Machine learning for additive manufacturing

production
Manufacturers, clients, and other supply chain stakeholders
need to know how much printing costs and how long it takes Additive manufacturing planning
to print something. Although the dimensions of the proposed
shape can be used to estimate them, an additional precise Since AM is regarded a costly production procedure at this
and effective cost approximation technique is still necessary. level, many end-users need a substantial return. A delicate
Chan et al. (2018) recently published an application of cost pre-production strategy is needed for the production process,
estimation. The cost estimation framework they proposed is from the CAD design to the finished product quality test. As
shown in Fig. 14: a result, some projects have used ML to aid in AM planning.
Tang et al. (2016) showed that for FDM-print grid structures,
• A customer presents a production order that includes a the manufacturability of a component could be evaluated uti-
three-dimensional model; lizing ML in pre-manufacturing. Moreover, the multimodal
• The input vector is formed from the three-dimensional learning system combining CNN and MLP had been built
model and fed into trained ML algorithms for cost estima- based on concepts, materials, and process parameters such
tion using clustering analysis based on similar workloads. that a metal part could be produced successfully by SLM.
• If such customer requests it or data required for training Likewise, Lu (xxxx) has utilized SVM to boost the reliabil-
the ML algorithm is insufficient, the 3D model will be ity of a 3D printable testing model to determine if the AM
delivered to modeling techniques to estimate costs, which process was suitable for the design. Moreover, in contrast to
will be used as training source for ML algorithms. the previous estimators created on SLS machines, MLP was
• After integrating the ML forecasts, the gross expected cost designed to decrease the error rate from 20 to 35% by 2 to
is calculated; 15%.
• The customer gets the final estimate.

123
46 Journal of Intelligent Manufacturing (2023) 34:21–55

Fig. 15 ML approach to minimize SLM treatment thermal distortion (Chowdhury et al., 2018; Wang et al., 2020)

Additive manufacturing quality control algorithms to recognize deformed portions quickly before
powder coating.
A significant barrier to AM quality assurance is the vari- By incorporating different sensors and cameras, in-
ability in the production quality between different machines process supervision is used to increase the quality of addi-
in the same process, or even from construction to produc- tively manufactured components, as aforementioned. Signal
tion. Variations in dimensional precision, density, process emissions, mostly visual and auditory may be recorded and
stability, and material properties may result from the dis- analysed to acquaint multiple ML systems for managing the
parity (Kumar & Wu, 2020, 2021b; Kumar et al., 2020b). 3D build. ML can help diagnosis printing status and failure
As a result, for establishing quality in AM parts, numerous mechanisms, melting state, porosity detection, tensile prop-
studies are endeavored for utilizing the ML algorithms. Three erty estimation, and surface roughness estimation in AM.
methods for eliminating geometric errors are rescaling whole
components, altering their primary CAD, and employing the Additive manufacturing data security
quality measures. Prior to manufacturing, MLP or CNN may
be employed to predict scaling ratio and change total size Intellectual property (IP) protection are very important and
of the components. Shape-dependent geometric variations industries pay a huge amount in that. The cyber and physical
because of heat stress could be modelled using the ML algo- realms are the two important characteristics of digital man-
rithms, allowing for appropriate geometric adjustments in the ufacturing, as depicted in Fig. 16a. Usually, the data thefts
CAD model. MLP was utilized to correct geometric distor- are most commonly occur in the cyber domain; this could
tion to decrease the thermal consequences induced by SLM also happen in the real-life realm (such as side channels)
(Chowdhury et al., 2018). The FEM modeling output was since AM technology releases several signals when gener-
produced for forecasting the distorted regions and altering ating 3D builds. ML techniques could be used to monitor
existing CAD data (Fig. 15). Noriega et al. (2013) employed outgoing channels and extract CAD data in IP surveillance.
a similar method in FDM printing, replacing simulated data To now, ML has only been able to record acoustic signals
with test results. For providing process control, SOM may during printing in order to structure 3d models from side
interlink distinct geometric aberrations for specific process- channels. Transducers could capture acoustic signals from
ing parameters. In comparison to many current supervised stepper motors in an FDM, as seen in Fig. 16b. It’s similar
ML algorithms, added with a laser scanner this method can to G-code in that it leaks data for the FDM process, includ-
cut short required cloud data necessary for measuring the ing axis motion, nozzle velocity, temperature, and material
dimensional correctness of additively manufactured com- discharge. The retrieved features of audio files can be used
ponents. Additionally, the single tracks may be changed at to develop ML algorithms in Mohammad et al. (2016) to
the macro scale by modifying DED process parameters to rebuild a model with a forecast efficiency of 78% axis and
minimize geometric flaws. Surface pictures taken after each a predicted inaccuracy of 18%. To capture the audio data in
created layer in the PBF process is laser exposed can train ML an IP-theft case, the burglar even can position his cell phone
near the device. This information was used by Hojjati et al.

123
Journal of Intelligent Manufacturing (2023) 34:21–55 47

Fig. 16 a Cyber and physical

invasion during AM system
development process. b Side
Channel Acoustic Model Attack
(Al Faruque et al., 2016; Wang
et al., 2020)

(2016) to properly rebuild an aircraft model around one mm suboptimal conditions in the process. By changing the part’s
in length with one-degree angular error. size for material extrusion, Noriega et al. (2013) used NNs to
compensate for dimensional variation. Their two NN imple-
mentations reduced deviance by 50% for outer dimensions
Printability and dimensional deviation management and 30% for internal dimensions, respectively. Charalampous
et al. (2021) used the AM of T4 spinal vertebrae as a case
Models can be created for identifying component printabil- study and attained 25% decrease in dimensional deviance
ity in material extrusion processes and PBF, using the CNNs for 1:1 scale and 33% reduction at 3:1. Multiple literature
and the SVMs. The usage of NNs can cut print time esti- has investigated how to limit geometrical deviance in DED
mates for PBF operations from 20–35 to 2–15%, allowing and binder jetting. Shen et al. (2018) conducted an analysis in
for better equipment control. The material, machine, and which CNNs were used to forecast geometrical deviance and
film production are all three-dimensional elements in AM recompense it by tilting, resizing, and turning the CAD geom-
systems; for example, in the process of conversion of a etry of a dental crown that will be made using binder jetting
CAD model to an STL file format, the resolution will be procedures. During the operational study, the CNN casted-off
reduced. In PBF, ML was used to rectify these to initially a voxel-based method, where each voxel was graded as reli-
optimise the orientation of the components, reduce noncon- able or defective. The F1 scores, which are a single-value
formity because of machine, and adjust the computer design metric added for assessing the prediction accuracy, were
for material thermal effects. To analyse and evaluate point- then calculated. The F1 scores for both the predictive and
cloud data for dimensional deviance of components produced compensatory models averaged 94%. Despite this, no actual
by extrusion procedures, Khanzedah et al. (2019) have put samples were produced, making the conclusions unreliable.
an uncontrolled learning algorithm, a self-organizing map. The most typical method of dimensional deviation correction
Their implementation allowed for the classification of par- in DED is to use process parameter optimization to optimise
tial differences into discrete clusters on account of severity
of the existing differences that permitted the identification of

123
48 Journal of Intelligent Manufacturing (2023) 34:21–55

the geometric features of separate channels. It partially cor- concentrated soon on novel materials, reasonable produc-
rects component and machinery issues, but not document tion plans, and computerized in-process feedback systems
formatting issues. Caiazzo and Caggiano (2024) established to AM that can further enhance smart AM.
optimization model for analysis of results. References (Al • Researchers can apply advanced machine learning tech-
Faruque et al., 2016) and (Hojjati et al., 2016) discovered that niques to time series analysis, conceptual advancements
material extrusion machines are vulnerable to intellectual and simulation, model-based fabrication, production data
property theft. The noise produced in stepper motors might be processing and retrieval, analytical thinking, medical and
collected and treated to infer characteristics of the 3D builds, biological records, data-driven clustering algorithms, fault
according to Al Faruque et al. (2016). This data was utilised and outlier identification, and in-situ and performance
to rebuild the CAD geometry that was being manufactured tracking.
in a physical-to-cyber-theft. These studies employed non- • Overall, ML has increased the likelihood of AM imple-
descript supervised classification and regression models for mentation and enhanced its perceived value. Most ML
predicting elements added with their quantities. With a mean solutions for AM, on the other hand, are not tested suffi-
axis prediction and overall length prediction error 78 and ciently to be applied in real-world problems. As a solution,
18%, respectively, the geometry was efficiently rebuilt. IP current studies should aim on enhancing the applicabil-
theft is a feature of AM, influenced by ML, contrasting to the ity of these tools for actual industrial issues and offering
other uses mentioned. If AM, specifically material extrusion, practical industrial examples to boost confidence in their
is made feasible for encryption methods, this must be solved. efficacy.
• Considering the restrictions of the feedstock, the opera-
tion settings, and the prior processing outcomes, one might
optimise for best succeeding machine variables assigned
Conclusion and future research
to obtain greatest feasible quality standards, as done by
Weiss et al. (2014). Pondering further, this technique
Based on many studies, the world is experiencing a flood
might be extended from part-specific phases to machine-
of data that generates large amounts of data every day from
and facility-specific phases, enhancing overall productiv-
different sources, for example (urban planning statistics, sen-
ity while accounting for resources, power, and constraints.
sor readings, environmental parameters, financial records,
• Another potential study issue could be equipment simplic-
AM/3D printer data, medical information, data on mobility
ity and the utilisation of higher raw element limitations.
etc.). It is vital to gather relevant insights and information
ML–based optimization strategies may cope to greater
from the diverse information available on many networks.
requirements of the processing phases, providing consis-
ML algorithms can be used to examine new research areas by
tent quality while lowering machinery and raw cost of
representing and extracting insights from heterogeneous data
materials.
utilising available data. In research on Intelligent Manufac-
turing by Kang et al. (2016), AM and ML have been regarded
as important innovations for Industry 4.0 also known as the
4th industrial revolution.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing, adap-
tation, distribution and reproduction in any medium or format, as
• ML techniques and their applications in various key AM
long as you give appropriate credit to the original author(s) and the
processes have been thoroughly examined in this study. In source, provide a link to the Creative Commons licence, and indi-
research, ML is found effective in improving 3D build’s cate if changes were made. The images or other third party material
quality. In AM, ML is integrated to improve tool productiv- in this article are included in the article’s Creative Commons licence,
unless indicated otherwise in a credit line to the material. If material
ity, research novel materials, and find property–structure
is not included in the article’s Creative Commons licence and your
relationships. The great majority of existing ML appli- intended use is not permitted by statutory regulation or exceeds the
cations in manufacturing fields are heavily focused on permitted use, you will need to obtain permission directly from the copy-
processing-related procedures such as process parameter right holder. To view a copy of this licence, visit http://creativecomm
ons.org/licenses/by/4.0/.
optimization, as seen. These can be used to tweak oper-
ation parameters for one or more performance standards.
Even though these optimization algorithms are machine-
oriented, there is a lack of work that attempts to develop References
better approaches. Irrespective of whether the final optimi-
Acayaba, G. M. A., & de Escalona, P. M. (2015). Prediction of surface
sation technique is conventional or ML-based, a huge num- roughness in low speed turning of AISI316 austenitic stainless
ber of tests will be required without such breakthroughs. steel. CIRP Journal of Manufacturing Science Technology, 11,
Therefore, most of ML ongoing research are expected to be 62–67. https://doi.org/10.1016/j.cirpj.2015.08.004

123
Journal of Intelligent Manufacturing (2023) 34:21–55 49

Adamczak, S., Bochnia, J., & Kaczmarska, B. (2014). Estimating the data analysis. Computers & Chemistry, 26, 5–14. https://doi.org/
uncertainty of tensile strength measurement for a photocured mate- 10.1016/S0097-8485(01)00094-8
rial produced by additive manufacturing. Metrological Measuring CART – Regression Tree from scratch with a hands-on example(in R)
System, 21, 553–560. https://doi.org/10.2478/mms-2014-0047 – Insight – Data Science Society, IMI, New Delhi n.d. https://
Addin, O., Sapuan, S. M., Mahdi, E., & Othman, M. (2007). A Naïve- insightimi.wordpress.com/2020/03/15/cart-regression-tree-from-
Bayes classifier for damage detection in engineering materials. scratch-with-a-hands-on-examplein-r/ (accessed July 16, 2021).
Materials and Design, 28, 2379–2386. https://doi.org/10.1016/j. Caggiano, A., Zhang, J., Alfieri, V., Caiazzo, F., Gao, R., & Teti,
matdes.2006.07.018 R. (2019). Machine learning-based image processing for on-line
Ahlers, D., Wasserfall, F., Hendrich, N., & Zhang, J. (2019). 3D printing defect recognition in additive manufacturing. CIRP Annals, 68,
of nonplanar layers for smooth surface generation. IEEE Inter- 451–454. https://doi.org/10.1016/J.CIRP.2019.03.021
national Conference Automative Science Enginerring. https://doi. Cai, G., & Mahadevan, S. (2016). Uncertainty quantification of manu-
org/10.1109/COASE.2019.8843116 facturing process effects on macroscale material properties. Inter-
Ahmad, M. S., Adnan, S. M., Zaidi, S., & Bhargava, P. (2020). A national Journal for Multiscale Computational Engineering, 14,
novel support vector regression (SVR) model for the prediction 191–213. https://doi.org/10.1615/IntJMultCompEng.2016015552
of splice strength of the unconfined beam specimens. Construc- Caiazzo, F., & Caggiano, A. (2018). Laser Direct metal deposition of
tion Building Materials, 248, 118475. https://doi.org/10.1016/j. 2024 al alloy: Trace geometry prediction via machine learning.
conbuildmat.2020.118475 Materials, 11, 444. https://doi.org/10.3390/MA11030444
Al Faruque, M. A., Chhetri, S. R., Canedo, A., Wan, J. (2016). Acoustic Çaydaş, U., & Ekici, S. (2012). Support vector machines models for
Side-Channel Attacks on Additive Manufacturing Systems. 2016 surface roughness prediction in CNC turning of AISI 304 austenitic
ACM/IEEE 7th Int Conf Cyber-Physical Syst ICCPS 2016 - Pro- stainless steel. Journal of Intelligent Manufacturing, 23, 639–650.
ceedings 2016. https://doi.org/10.1109/ICCPS.2016.7479068. https://doi.org/10.1007/s10845-010-0415-2
Alabi, M. O. (2018). Big data, 3D printing technology, and industry of Chan, S., & Lu, Y. (2018). Data-driven cost estimation for additive
the future. International Journal of Big Data and Anal Healthcare, manufacturing in cybermanufacturing. Amsterdam: Elsevier.
2, 1–20. https://doi.org/10.4018/ijbdah.2017070101 Chand, S., & Davis, J. (2010). What is smart manufacturing. Time Mag-
Alade, I. O., Rahman, M. A. A., & Saleh, T. A. (2020). An approach azine Wrapper, 7, 28–33.
to predict the isobaric specific heat capacity of nitrides/ethylene Charalampous, P., Kostavelis, I., Kontodina, T., & Tzovaras, D. (2021).
glycol-based nanofluids using support vector regression. Journal Learning-based error modeling in FDM 3D printing process. Rapid
of Energy Storage, 29, 101313. https://doi.org/10.1016/j.est.2020. Prototyping Journal, 27, 507–517. https://doi.org/10.1108/RPJ-
101313 03-2020-0046
Ali, W., Shamsuddin, S. M., & Ismail, A. S. (2012). Intelligent Naïve Chonde, S. (2016). A methodology of machine learning in automated
Bayes-based approaches for Web proxy caching. Knowledge- entity summarization. Pennsylvania State University.
Based System, 31, 162–175. https://doi.org/10.1016/j.knosys. Chowdhury, S., Mhapsekar, K., & Anand, S. (2018). Part build orienta-
2012.02.015 tion optimization and neural network-based geometry compensa-
Ali, N. H. M., Ahmad, F., Abidin, N. I., Suhaili, S., Rahman, M. A. A., tion for additive manufacturing process. Journal of Manufacturing
Harun, H., et al. (2021). Agile Project Management Software for Science and Engineering, 1, 140.
Construction and Management Industries (pp. 101–111). https:// Chu, W. W. S., Kim, C. S. C., Lee, H. T. H., Choi, J. O. J., Park, J. I.
doi.org/10.1007/978-981-16-0742-4_7 J., Song, J. H., et al. (2014). Hybrid manufacturing in micro/nano
Anderson, A. (2011). Report to the President on Ensuring American scale: A review. International Journal of Precision Engineering
Leadership in Advanced Manufacturing. Exec Off Pres. and Manufacturing - Green Technology, 1, 75–92. https://doi.org/
Anderson, A., & Delplanque, J.-P. (2015). Development of Physics- 10.1007/s40684-014-0012-5
Based Numerical Models for Uncertainty Quantification of Selec- Clegg, B. A., Heggestad, E. D., & Blalock, L. D. (2010). The
tive Laser Melting Processes - 2015 Annual Progress Report. influences of automation and trainee aptitude on training effec-
Livermore, CA (United States). https://doi.org/10.2172/1226942. tiveness. Proceedings of the Human Factors and Ergonomics
Aoyagi, K., Wang, H., Sudo, H., & Chiba, A. (2019). Simple method to Society Annual Meeting, 54, 2329–2332. https://doi.org/10.1177/
construct process maps for additive manufacturing using a support 154193121005402723
vector machine. Additive Manufacturing, 27, 353–362. https://doi. Craig, C., N. McCarthy, J., Montgomery, T. H., & Fourniol, F.
org/10.1016/J.ADDMA.2019.03.013 MACHINE LEARNING: THE POWER AND PROMISE OF
Azzone, G., & Bertele, U. (2007). Measuring the economic effective- COMPUTERS THAT LEARN BY EXAMPLE. 1st ed. n.d.
ness of flexible automation: A new approach. International Journal Dalenogare, L. S., Benitez, G. B., Ayala, N. F., & Frank, A. G. (2018).
of Production Research., 27, 735–746. https://doi.org/10.1080/ The expected contribution of Industry 4.0 technologies for indus-
00207548908942583 trial performance. International Journal of Production Economics,
Banga, S., Gehani, H., Bhilare, S. (2018). SP preprint arXiv, 2018 204, 383–394. https://doi.org/10.1016/J.IJPE.2018.08.019
undefined. 3d topology optimization using convolutional neural David, F. N. (2017). Forces of production: A social history of industrial
networks. ArxivOrg n.d. automation. Forces Prod. https://doi.org/10.4324/9780203791806
Baumann, F., Scholz, J., & Fleischer, J. (2017). Investigation of a new Davis, J., Edgar, T., Graybill, R., Korambath, P., Schott, B., Swink,
approach for additively manufactured continuous fiber-reinforced D., et al. (2015). Smart manufacturing. Annual Review of Chem-
polymers. Procedia CIRP, 66, 323–328. https://doi.org/10.1016/J. ical and Biomolecular Engineering, 6, 141–160. https://doi.org/
PROCIR.2017.03.276 10.1146/annurev-chembioeng-061114-123255
Beaman, J. J., Bourell, D. L., Seepersad, C. C., & Kovar, D. (2020). Delgado, J., Ciurana, J., & Rodríguez, C. A. (2012). Influence of pro-
Additive manufacturing review: Early past to current practice. cess parameters on part quality and mechanical properties for
Journal of Manufacturing Science and Engineering Transactions DMLS and SLM with iron-based materials (pp. 601–610). Berlin:
ASME. https://doi.org/10.1115/1.4048193/1086507 Springer.
Bendsøe, M. (1999). Material interpolation schemes in topology opti- Delli, U., & Chang, S. (2018). Automated process monitoring in
mization. Amsterdam: Springer. 3D printing using supervised machine learning. Procedia Manu-
Burbidge, R., Trotter, M., Buxton, B., & Holden, S. (2001). Drug design facturing, 26, 865–870. https://doi.org/10.1016/J.PROMFG.2018.
by machine learning: Support vector machines for pharmaceutical 07.111

123
50 Journal of Intelligent Manufacturing (2023) 34:21–55

Desai, C., Skouras, M., Zhu, B., & Matusik, W. (2018). Computational Gobert, C., Reutzel, E. W., Petrich, J., Nassar, A. R., & Phoha, S. (2018).
discovery of extremalmicrostructure families. Science Advaces, 1, Application of supervised machine learning for defect detection
7. during metallic powder bed fusion additive manufacturing using
Devesse, W., De Baere, D., & Guillaume, P. (2014). The isotherm migra- high resolution imaging. Additive Manufacturing, 21, 517–528.
tion method in spherical coordinates with a moving heat source. https://doi.org/10.1016/J.ADDMA.2018.04.005
International Journal of Heat and Mass Transfer, 75, 726–735. Gokulachandran, J., & Mohandas, K. (2015). Comparative study of
https://doi.org/10.1016/j.ijheatmasstransfer.2014.04.015 two soft computing techniques for the prediction of remaining
Dowling, L., Kennedy, J., O’Shaughnessy, S., & Trimble, D. (2020). A useful life of cutting tools. Journal of Intelligent Manufacturing,
review of critical repeatability and reproducibility issues in powder 26, 255–268. https://doi.org/10.1007/s10845-013-0778-2
bed fusion. Materials and Design, 186, 108346. https://doi.org/10. Goodfellow, I., Jean Pouget-Abadie, M. M., Xu, B., David Warde-
1016/j.matdes.2019.108346 Farley, S. O., Courville, A., & Bengio, Y. (2014). Generative
Dumais, S. T. (2004). Latent semantic analysis. Annual Review of Infor- adversarial nets. Advanced Neural Information Processing Sys-
mation Science and Technology, 38, 188–230. https://doi.org/10. tems, 1, 27.
1002/aris.1440380105 Gorecky, D., Schmitt, M., Loskyll, M., & Zühlke, D. (2014). Human-
Dunk, A. S. (1992). Reliance on budgetary control, manufacturing pro- machine-interaction in the industry 40 era. In: Proceedings of
cess automation and production subunit performance: A research 2014 12th IEEE International Conference of Industrial Informat-
note. Accounting, Organ Society, 17, 195–203. https://doi.org/10. ics INDIN 2014:289–294. https://doi.org/10.1109/INDIN.2014.6
1016/0361-3682(92)90020-S 945523
Dutt, V., & Gonzalez, C. (2012). Making Instance-based Learning The- Grasso, M., Demir, A. G., Previtali, B., & Colosimo, B. M. (2018).
ory usable and understandable: The Instance-based Learning Tool. In situ monitoring of selective laser melting of zinc powder via
Comput Human Behav, 28, 1227–1240. https://doi.org/10.1016/J. infrared imaging of the process plume. Robot Computer Integrat-
CHB.2012.02.006 ing Manufacturing, 49, 229–239. https://doi.org/10.1016/J.RCIM.
D’Addona, D. M., & Antonelli, D. (2019). Application of numerical 2017.07.001
simulation for the estimation of die life after repeated hot forging Grasso, M., Laguzza, V., Semeraro, Q., & Colosimo, B. M. (2017).
work cycles. Procedia CIRP, 79, 632–637. https://doi.org/10.1016/ In-process monitoring of selective laser melting: spatial detection
j.procir.2019.02.063 of defects via image data analysis. American Society Mechanical
Elangovan, M., Sakthivel, N. R., Saravanamurugan, S., Nair, B. B., Engineering, 2017, 139.
& Sugumaran, V. (2015). Machine learning approach to the Grasso, M., Technology, B.C.-M.S. (2017). Process defects and in situ
prediction of surface roughness using statistical features of vibra- monitoring methods in metal powder bed fusion: A review. Iop-
tion signal acquired in turning. Procedia Computer Science, 50, scienceIopOrg n.d.
282–288. Grierson, D. R., & Quayle, S. D. (2021). Machine learning for addi-
Evgeniou, T., Poggio, T., Pontil, M., & Verri, A. (2002). Regularization tive manufacturing. Encyclopedia, 3, 1541–1556. https://doi.org/
and statistical learning theory for data analysis. Computational 10.1016/j.matt.2020.08.023
Statistics & Data Analysis, 38, 421–432. https://doi.org/10.1016/ Gu, G. X., Chen, C. T., Richmond, D. J., & Buehler, M. J. (2018).
S0167-9473(01)00069-X Bioinspired hierarchical composite design using machine learn-
Evgeniou, T., Pontil, M., & Poggio, T. (2000). Statistical learning the- ing: Simulation, additive manufacturing, and experiment. Material
ory: A primer. International Journal of Computer Vision, 38, 9–13. Horizons, 5, 939–945. https://doi.org/10.1039/C8MH00653A
https://doi.org/10.1023/A:1008110632619 Gunther, D., Pirehgalin, M. F., Weis, I., Vogel-Heuser, B. (2020). Con-
Fahle, S., Prinz, C., & Kuhlenkötter, B. (2020). Systematic review dition monitoring for the Binder Jetting AM-process with machine
on machine learning (ML) methods for manufacturing processes learning approaches. Proceedings - 2020 IEEE Conference Indus-
– Identifying artificial intelligence (AI) methods for field appli- trial Cyberphysical Systems ICPS 2020 2020:417–20. https://doi.
cation. Procedia CIRP, 93, 413–418. https://doi.org/10.1016/J. org/10.1109/ICPS48405.2020.9274716.
PROCIR.2020.04.109 Guo, M., Jia, C., Zhou, J., Liu, W., & Wu, C. (2020). Investigating
Francis, J., & Letters, L.B.-M. (2019). Deep learning for distortion the generation process of molten droplets and arc plasma in the
prediction in laser-based additive manufacturing using big data. confined space during compulsively constricted WAAM. Journal
Amsterdam: Elsevier. of Materials Processing Technology, 275, 116355. https://doi.org/
Frank, A. G., Dalenogare, L. S., & Ayala, N. F. (2019). Industry 4.0 10.1016/J.JMATPROTEC.2019.116355
technologies: Implementation patterns in manufacturing compa- Guo, N., & Leu, M. C. (2013). Additive manufacturing: Technology,
nies. International Journal of Production Economics, 210, 15–26. applications and research needs. Frontiers of Mechanical Engi-
https://doi.org/10.1016/J.IJPE.2019.01.004 neering, 8, 215–243. https://doi.org/10.1007/s11465-013-0248-8
Frazier, W. E. (2014). Metal additive manufacturing: A review. Jour- Hapfelmeier, A., & Ulm, K. (2014). Variable selection by Random
nal of Materials Engineering and Performance, 23, 1917–1928. Forests using data with missing values. Computing Statical Data
https://doi.org/10.1007/S11665-014-0958-Z/FIGURES/9 Analysis, 80, 129–139. https://doi.org/10.1016/j.csda.2014.06.017
Garbie, I. H. (2013). DFSME: Design for sustainable manufactur- Hoffmann, A. G. (1990). General limitations on machine learning.
ing enterprises (an economic viewpoint). International Journal ECAI, 90, 345–347.
of Production Research, 51, 479–503. https://doi.org/10.1080/ Hofmann, E., & Rüsch, M. (2017). Industry 4.0 and the current status
00207543.2011.652746 as well as future prospects on logistics. Computers in Industry, 89,
Garg, A., Tai, K., & Savalani, M. M. (2014). State-of-the-art in empir- 23–34. https://doi.org/10.1016/J.COMPIND.2017.04.002
ical modelling of rapid prototyping processes. Rapid Prototyping Hojjati, A., Adhikari, A., Struckmann, K., Chou, E. J., Ngoc, T.,
Journal, 20, 164–178. https://doi.org/10.1108/RPJ-08-2012-0072 Nguyen, T., et al. (2016). Leave Your Phone at the Door: Side
Ghobakhloo, M. (2020). Industry 4.0, digitization, and opportunities Channels that Reveal Factory Floor Secrets. In: Proceedings of
for sustainability. Journal of Cleaner Production, 252, 119869. 2016 ACM SIGSAC Conference on Computer Communications
https://doi.org/10.1016/J.JCLEPRO.2019.119869 Security. https://doi.org/10.1145/2976749.
Gibson, I., Rosen, D., Stucker, B., & Khorasani, M. (2021). Additive How IoT & Industry 4.0 Relate - and Why Manufacturers Should
manufacturing technologies. Additive Manufacturing Technology. Care n.d. https://lucidworks.com/post/how-are-iot-and-industry-
https://doi.org/10.1007/978-3-030-56127-7 4-related/ (accessed July 16, 2021).

123
Journal of Intelligent Manufacturing (2023) 34:21–55 51

Hu, Z., & Mahadevan, S. (2017). Uncertainty quantification and man- Khanzadeh, M., Rao, P., Jafari-Marandi, R., Smith, B. K., Tschopp,
agement in additive manufacturing: Current status, needs, and M. A., & Bian, L. (2018). Quantifying geometric accuracy with
opportunities. International Journal of Advanced Manufacturing unsupervised machine learning: Using self-organizing map on
Technology, 93, 2855–2874. https://doi.org/10.1007/s00170-017- fused filament fabrication additive manufacturing parts. Journal of
0703-5 Manufacturing Science and Engineering. https://doi.org/10.1115/
Hu, Z., Mahadevan, S., & Du, X. (2016). Uncertainty quantification of 1.4038598
time-dependent reliability analysis in the presence of parametric Knoll, D., Neumeier, D., Prüglmeier, M., & Reinhart, G. (2019). An
uncertainty. ASCE-ASME J Risk Uncertain Eng Syst Part B Mech automated packaging planning approach using machine learn-
Eng. https://doi.org/10.1115/1.4032307 ing. Procedia CIRP, 81, 576–581. https://doi.org/10.1016/j.procir.
Huang, S. H., Liu, P., Mokasdar, A., & Hou, L. (2012). Additive manu- 2019.03.158
facturing and its societal impact: A literature review. International Koltchinskii, V., Abdallah, C. T., Ariola, M., & Dorato, P. (2001). Statis-
Journal of Advanced Manufacturing Technology, 67, 1191–1203. tical learning control of uncertain systems: Theory and algorithms.
https://doi.org/10.1007/S00170-012-4558-5 Applied Mathematics and Computation, 120, 31–43. https://doi.
Hudson, C. A. (1982). Computers in manufacturing. Science, 215, org/10.1016/S0096-3003(99)00283-0
818–825. https://doi.org/10.1126/SCIENCE.215.4534.818 Kreutz, M., Ait-Alla, A., Varasteh, K., Oelker, S., Greulich, A., Freitag,
ISO/ASTM52900 - 15 Standard Terminology for Additive Manufactur- M., et al. (2019). Machine learning-based icing prediction on wind
ing – General Principles – Terminology n.d. turbines. Procedia CIRP, 81, 423–428. https://doi.org/10.1016/j.
Imran, M., Kang, C., Lee, Y. H., Jahanzaib, M., & Aziz, H. (2017). procir.2019.03.073
Cell formation in a cellular manufacturing system using simulation Kulkarni, P., Marsan, A., & Dutta, D. (2000). Review of process plan-
integrated hybrid genetic algorithm. Computers & Industrial Engi- ning techniques in layered manufacturing. Rapid Prototyp J, 6,
neering, 105, 123–135. https://doi.org/10.1016/j.cie.2016.12.028 18–35. https://doi.org/10.1108/13552540010309859
Jacobsmühlen, J. (2015). SK-I 2015-41st, 2015 undefined. Detection Kumar, S. (2016). Ultrasonic assisted friction stir processing of 6063
of elevated regions in surface images from laser beam melting aluminum alloy. Archives of Civil and Mechanical Engineering,
processes. IeeexploreIeeeOrg n.d. 16, 473–484. https://doi.org/10.1016/j.acme.2016.03.002
Jafari-Marandi, R., Khanzadeh, M., Tian, W., Smith, B., & Bian, L. Kumar, S., & Kar, A. (2021). A review of solid-state additive manu-
(2019). From in-situ monitoring toward high-throughput process facturing processes. Transactions on Indian Natational Academic
control: Cost-driven decision-making framework for laser-based Engineering, 6, 955–973. https://doi.org/10.1007/S41403-021-
additive manufacturing. Journal of Manufacturing Systems, 51, 00270-7
29–41. https://doi.org/10.1016/J.JMSY.2019.02.005 Kumar, S., & Kishor, B. (2021). Ultrasound added additive manu-
Jia, C. B., Liu, X. F., Zhang, G. K., Zhang, Y., Yu, C. H., & Wu, C. facturing for metals and composites: Process and control addit
S. (2021). Penetration/keyhole status prediction and model visu- subtractive manuf compos (pp. 53–72). Singapore: Springer.
alization based on deep learning algorithm in plasma arc welding. Kumar, S., & Wu, C. S. (2018). A novel technique to join Al and Mg
International Journal of Advanced Manufacturing Technology, alloys: Ultrasonic vibration assisted linear friction stir welding.
117, 3577–3597. https://doi.org/10.1007/s00170-021-07903-9 Materials Today Proceedings, 5, 18142–18151. https://doi.org/10.
Jin, Z., Zhang, Z., Demir, K., & Gu, G. X. (2020). Machine learning for 1016/j.matpr.2018.06.150
advanced additive manufacturing. Matter, 3, 1541–1556. https:// Kumar, S., & Wu, C. S. (2020). Suppression of intermetallic reaction
doi.org/10.1016/j.matt.2020.08.023 layer by ultrasonic assistance during friction stir welding of Al and
Johnsson, C., Brandl, D. (2006). K U. ISA 95 for Beginners, Report. Mg based alloys. Journal of Alloys and Compounds, 827, 154343.
2006. https://doi.org/10.1016/j.jallcom.2020.154343
Joshi, M. S., Flood, A., Sparks, T., Liou, F. W. (2019). Applications of Kumar, S., & Wu, C. (2021b). Eliminating intermetallic compounds
supervised machine learning algorithms in additive manufactur- via Ni interlayer during friction stir welding of dissimilar
ing: A review. Solid Free. Fabr. 2019 Proc. 30th Annu. Int. Solid Mg/Al alloys. Journal of Material Research and Technology, 15,
Free. Fabr. Symp. - An Addit. Manuf. Conf. SFF 2019. 4353–4369. https://doi.org/10.1016/J.JMRT.2021.10.065
Jurkovic, Z., Cukor, G., Brezocnik, M., & Brajkovic, T. (2018). A Kumar, S., & Wu, C. (2021a). Strengthening effects of tool-mounted
comparison of machine learning methods for cutting parameters ultrasonic vibrations during friction stir lap welding of Al and
prediction in high speed turning process. Journal of Intelligent Mg alloys. Metallurgical and Materials Transactions a, Physical
Manufacturing, 29, 1683–1693. https://doi.org/10.1007/s10845- Metallurgy and Materials Science, 52, 2909–2925. https://doi.org/
016-1206-1 10.1007/s11661-021-06282-w
Kamath, C. (2016). Data mining and statistical inference in selective Kumar, S., Wu, C. S., Padhy, G. K., & Ding, W. (2017). Application
laser melting. International Journal of Advanced Manufacturing of ultrasonic vibrations in welding and metal processing: A status
Technology, 86, 1659–1677. https://doi.org/10.1007/s00170-015- review. Journal of Manufacturing Processes, 26, 295–322. https://
8289-2 doi.org/10.1016/j.jmapro.2017.02.027
Kang, P., & Cho, S. (2008). Locally linear reconstruction for instance- Kumar, S., Wu, C. S., & Shi, L. (2020b). Intermetallic diminution
based learning. Pattern Recognition, 41, 3507–3518. https://doi. during friction stir welding of dissimilar Al/Mg alloys in lap con-
org/10.1016/J.PATCOG.2008.04.009 figuration via ultrasonic assistance. Metallurgical and Materials
Kang, H. S., Lee, J. Y., Choi, S., Kim, H., Park, J. H., Son, J. Y., et al. Transactions a: Physical Metallurgy and Materials Science, 51,
(2016). Smart manufacturing: Past research, present findings, and 5725–5742. https://doi.org/10.1007/s11661-020-05982-z
future directions. International Journal of Precision Engineering Kumar, S., Wu, C. S., & Song, G. (2020a). Process parametric depen-
Manufacturing - Green Technology, 3, 111–128. https://doi.org/ dency of axial downward force and macro- and microstructural
10.1007/s40684-016-0015-5 morphologies in ultrasonically assisted friction stir welding of
Khanzadeh, M., Chowdhury, S., Tschopp, M. A., Doude, H. R., Maru- Al/Mg alloys. Metallurgical and Materials Transactions a: Physi-
fuzzaman, M., & Bian, L. (2018). In-Situ Monitoring of Melt cal Metallurgy and Materials Science, 51, 2863–2881. https://doi.
Pool Images for Porosity Prediction in Directed Energy Depo- org/10.1007/s11661-020-05716-1
sition Processes, 51, 437–455. https://doi.org/10.1080/24725854. Kumar, S., Wu, C. S., Sun, Z., & Ding, W. (2019). Effect of ultrasonic
2017.1417656 vibration on welding load, macrostructure, and mechanical prop-
erties of Al/Mg alloy joints fabricated by friction stir lap welding.

123
52 Journal of Intelligent Manufacturing (2023) 34:21–55

International Journal of Advanced Manufacturing Technology, Manav, O., & Chinchanikar, S. (2018). Multi-objective optimization of
100, 1787–1799. https://doi.org/10.1007/s00170-018-2717-z hard turning: A genetic algorithm approach. Material Today Pro-
Kuric, I., Císar, M., Tlach, V., Zajačko, I., Gál, T., & Wi˛ecek, D. (2018). ceedings, 5, 12240–12248. https://doi.org/10.1016/j.matpr.2018.
Technical diagnostics at the department of automation and produc- 02.201
tion systems. Advances in Intelligent Systems and Computing, 835, Mavrikios, D., Papakostas, N., Mourtzis, D., & Chryssolouris, G.
474–484. https://doi.org/10.1007/978-3-319-97490-3_46 (2011). On industrial learning and training for the factories of the
Kwon, O., Kim, H. G., Ham, M. J., Kim, W., Kim, G.-H., Cho, J.- future: A conceptual, cognitive and technology framework. Jour-
H., et al. (2018). A deep neural network for classification of nal of Intelligence and Manufacturing, 24, 473–485. https://doi.
melt-pool images in metal additive manufacturing. Journal of org/10.1007/S10845-011-0590-9
Intelligence Manufacturing, 31, 375–386. https://doi.org/10.1007/ McFadden, K. L. (1997). Predicting pilot-error incidents of US airline
S10845-018-1451-6 pilots using logistic regression. Applied Ergonomics, 28, 209–212.
Le Calvé, A., & Savoy, J. (2000). Database merging strategy based https://doi.org/10.1016/S0003-6870(96)00062-2
on logistic regression. Information Process and Management, 36, Meng, L., McWilliams, B., Jarosinski, W., Park, H. Y., Jung, Y. G., Lee,
341–359. https://doi.org/10.1016/S0306-4573(99)00036-9 J., et al. (2020). Machine learning in additive manufacturing: a
Learned-Miller, E. G. (2014). Introduction to Supervised Learning. review. JOM Journal of the Minerals Metals and Materials Society,
Department of Computer Science, University of Massachusetts. 72, 1. https://doi.org/10.1007/s11837-020-04155-y
Lee, J., Lapira, E., Bagheri, B., & Kao, H. (2013). Recent advances and Monostori, L. (2003). AI and machine learning techniques for man-
trends in predictive manufacturing systems in big data environ- aging complexity, changes and uncertainties in manufacturing.
ment. Manufacturing Letters, 1, 38–41. https://doi.org/10.1016/j. Engineering Applications of Artificial Intelligence, 16, 277–291.
mfglet.2013.09.005 https://doi.org/10.1016/S0952-1976(03)00078-2
Lee, W. J., Mendis, G. P., Triebe, M. J., & Sutherland, J. W. (2020). Montazeri, M., & Rao, P. (2018). Sensor-based build condition moni-
Monitoring of a machining process using kernel principal compo- toring in laser powder bed fusion additive manufacturing process
nent analysis and kernel density estimation. Journal of Intelligent using a spectral graph theoretic approach. Journal of Manufactur-
Manufacturing, 31, 1175–1189. https://doi.org/10.1007/s10845- ing Science Engineering Transactions ASME. https://doi.org/10.
019-01504-w 1115/1.4040264/741453
Lee, S. H., Park, W. S., Cho, H. S., Zhang, W., & Leu, M. C. (2016). Morrison, A. (2015). Design issues and orientations in additive manu-
A neural network approach to the modelling and analysis of stere- facturing Steinar Killi*. William Lavatelli Kempton, 5, 289–307.
olithography processes. Proceedings of the Institution of Mechan- Mozaffar, M., Paul, A., Al-Bahrani, R., Wolff, S., Choudhary, A.,
ical Engineers Part B Journal of Engineering Manufacture, 215, Agrawal, A., et al. (2018). Data-driven prediction of the high-
1719–1733. https://doi.org/10.1177/095440540121501206 dimensional thermal history in directed energy deposition pro-
Li, L., McGuan, R., Isaac, R., Kavehpour, P., & Candler, R. (2021). cesses via recurrent neural networks. Manufacturing Letters, 18,
Improving precision of material extrusion 3D printing by in-situ 35–39. https://doi.org/10.1016/J.MFGLET.2018.10.002
monitoring & predicting 3D geometric deviation using conditional Multivariate Statistical Methods in Quality Management. n.d.
adversarial networks. Additive Manufacturing, 38, 101695. https:// Nascimento, D. L. M., Alencastro, V., Quelhas, O. L. G., Caiado, R.
doi.org/10.1016/J.ADDMA.2020.101695 G. G., Garza-Reyes, J. A., Lona, L. R., et al. (2018). Exploring
Lopez, F., Witherell, P., & Lane, B. (2016). Identifying uncertainty in Industry 4.0 technologies to enable circular economy practices in
laser powder bed fusion additive manufacturing models. J Mech a manufacturing context: A business model proposal. Journal of
Des Trans ASME. https://doi.org/10.1115/1.4034103 Manufuring and Technology Management, 30, 607–627. https://
Loughnane, G. (2015). A Framework for Uncertainty Quantification doi.org/10.1108/JMTM-03-2018-0071
in Microstructural Characterization with Application to Additive Nilsson, N. J. (1996). Introduction to Machine Learning. An early draft
Manufacturing of Ti-6Al-4V. Brows All Theses Dissertation, 2015. of a proposed textbook 1996.
Loyer, J. L., Henriques, E., Fontul, M., & Wiseall, S. (2016). Compar- Niu, X., Yang, C., Wang, H., & Wang, Y. (2017). Investigation of ANN
ison of Machine Learning methods applied to the estimation of and SVM based on limited samples for performance and emis-
manufacturing cost of jet engine components. International Jour- sions prediction of a CRDI-assisted marine diesel engine. Applied
nal of Production Economics, 178, 109–119. https://doi.org/10. Thermal Engineering, 111, 1353–1364. https://doi.org/10.1016/j.
1016/j.ijpe.2016.05.006 applthermaleng.2016.10.042
Lu, S. C. Y. (1990). Machine learning approaches to knowledge Noriega, A., Blanco, D., Alvarez, B. J., & Garcia, A. (2013). Dimen-
synthesis and integration tasks for advanced engineering automa- sional accuracy improvement of FDM square cross-section parts
tion. Computers in Industry, 15, 105–120. https://doi.org/10.1016/ using artificial neural networks and an optimization algorithm.
0166-3615(90)90088-7 International Journal of Advanced Manufacturing Technology, 69,
Lu, T. (2016). Towards a fully automated 3D printability checker. Pro- 2301–2313. https://doi.org/10.1007/S00170-013-5196-2
ceedings of IEEE International Conference in Industrial Technol- Okaro, I. A., Jayasinghe, S., Sutcliffe, C., Black, K., Paoletti, P., &
ogy, 2016, 922–927. https://doi.org/10.1109/ICIT.2016.7474875 Green, P. L. (2019). Automatic fault detection for laser powder-bed
Lucke, D., Constantinescu, C., Westkämper, E. (2008). Smart Factory fusion using semi-supervised machine learning. Additive Man-
- A Step towards the Next Generation of Manufacturing. Manuf. ufacturing, 27, 42–53. https://doi.org/10.1016/J.ADDMA.2019.
Syst. Technol. New Front (pp. 115–118). Springer. https://doi.org/ 01.006
10.1007/978-1-84800-267-8_23. Park, S. I., Rosen, D. W., Choi, S., & Duty, C. E. (2014). Effective
Ma, L., Fong, J., Lane, B., Moylan, S., Filliben, J., Heckert, A., et al. mechanical properties of lattice material fabricated by material
(2015). Using design of experiments in finite element modeling extrusion additive manufacturing. Additive Manufuring, 1, 12–23.
to identify critical variables for laser powder bed fusion. Austin: https://doi.org/10.1016/j.addma.2014.07.002
University of Texas. Paturi, U. M. R., & Cheruku, S. (2021). Application and performance
Manallack, D. T., & Livingstone, D. J. (1999). Neural networks in drug of machine learning techniques in manufacturing sector from the
discovery: Have they lived up to their promise? European Jour- past two decades: A review. Materials Today Proceedings, 38,
nal of Medicinal Chemistry, 34, 195–208. https://doi.org/10.1016/ 2392–2401. https://doi.org/10.1016/j.matpr.2020.07.209
S0223-5234(99)80052-X Peres, R. S., Barata, J., Leitao, P., & Garcia, G. (2019). Multistage
quality control using machine learning in the automotive industry.

123
Journal of Intelligent Manufacturing (2023) 34:21–55 53

IEEE Access, 7, 79908–79916. https://doi.org/10.1109/ACCESS. laser powder bed fusion additive manufacturing process. Additive
2019.2923405 Manufacturing, 24, 273–286. https://doi.org/10.1016/J.ADDMA.
Peters, J., De, B. B., Verhoest, N. E. C., Samson, R., Degroeve, S., De, 2018.09.034
B. P., et al. (2007). Random forests as a tool for ecohydrologi- Scime, L., & Beuth, J. (2019). Using machine learning to identify in-situ
cal distribution modelling. Ecological Modelling, 207, 304–318. melt pool signatures indicative of flaw formation in a laser powder
https://doi.org/10.1016/j.ecolmodel.2007.05.011 bed fusion additive manufacturing process. Additive Manufactur-
Pham, D. T., & Afify, A. A. (2005). Machine-learning techniques and ing, 25, 151–165. https://doi.org/10.1016/J.ADDMA.2018.11.010
their applications in manufacturing. Proceedings of Institute and Seibi, A., & Al-Alawi, S. M. (1997). Prediction of fracture tough-
Mechanical Engineering Part B, 219, 395–412. https://doi.org/10. ness using artificial neural networks (ANNs). Engineering Frac-
1243/095440505X32274 ture Mechanics, 56, 311–319. https://doi.org/10.1016/S0013-
Piro, P., Nock, R., Nielsen, F., & Barlaud, M. (2012). Leveraging k- 7944(96)00076-8
NN for generic classification boosting. Neurocomputing, 80, 3–9. Sharp, M., Ak, R., & Hedberg, T. (2018). A survey of the advancing
https://doi.org/10.1016/j.neucom.2011.07.026 use and development of machine learning in smart manufacturing.
Ponche, R., Kerbrat, O., Mognol, P., & Hascoet, J. Y. (2014). A novel Journal of Manufacturing Systems, 48, 170–179. https://doi.org/
methodology of design for additive manufacturing applied to 10.1016/j.jmsy.2018.02.004
additive laser Manufacturing process. Robot Computer Integrat- Shevchik, S. A., Kenel, C., Leinenbach, C., & Wasmer, K. (2018).
ing Manufacturing, 30, 389–398. https://doi.org/10.1016/J.RCIM. Acoustic emission for in situ quality monitoring in additive man-
2013.12.001 ufacturing using spectral convolutional neural networks. Additive
Qi, X., Chen, G., Li, Y., Cheng, X., & Li, C. (2019). Applying Manufacturing, 21, 598–604. https://doi.org/10.1016/J.ADDMA.
neural-network-based machine learning to additive manufactur- 2017.11.012
ing: Current applications, challenges, and future perspectives. Shinde, P. P., & Shah, S. (2018). A Review of Machine Learn-
Engineering, 5, 721–729. https://doi.org/10.1016/J.ENG.2019. ing and Deep Learning Applications. In: Proceedings - 2018
04.012 4th International Conference Computer Communication Con-
Raghunath, N., & Pandey, P. M. (2007). Improving accuracy through trol Autom ICCUBEA 2018. https://doi.org/10.1109/ICCUBEA.
shrinkage modelling by using Taguchi method in selective laser 2018.8697857.
sintering. International Journal of Machine Tools and Manu- Da Silva, C. E. S., Salgado, E. G., Mello, C. H. P., Da Silva, O. E.,
facture, 47, 985–995. https://doi.org/10.1016/j.ijmachtools.2006. & Leal, F. (2014). Integration of computer simulation in design
07.001 for manufacturing and assembly. International Journal of Produc-
Rai, R., Tiwari, M. K., Ivanov, D., & Dolgui, A. (2021). Machine tion Research, 52, 2851–2866. https://doi.org/10.1080/00207543.
learning in manufacturing and industry 4.0 applications, vol. 59, 2013.853887
pp. 4773–4778. https://doi.org/10.1080/00207543.2021.1956675 Singh, S. K., & Gupta, A. K. (2010). Application of support vec-
Ramachandran, S., Jayalal, M. L., Riyas, A., Jehadeesan, R., & tor regression in predicting thickness strains in hydro-mechanical
Devan, K. (2020). Application of genetic algorithm for opti- deep drawing and comparison with ANN and FEM. CIRP Journal
mization of control rods positioning in a fast breeder reactor of Manufacturing Science and Technology, 3, 66–72. https://doi.
core. Nuclear Engineering Design, 361, 110541. https://doi.org/ org/10.1016/j.cirpj.2010.07.005
10.1016/j.nucengdes.2020.110541 Singh, S., Ramakrishna, S., & Singh, R. (2017). Material issues
Rawat, S., & Shen, M. H. H. (2018). A novel topology design approach in additive manufacturing: A review. Journal of Manufacturing
using an integrated deep learning network architecture. Processes, 25, 185–200. https://doi.org/10.1016/j.jmapro.2016.
Razvi, S. S., Feng, S., Narayanan, A., Lee, Y. T. T., & Witherell, P. 11.006
(2019). A review of machine learning applications in additive Singh, R., Shah, D. B., Gohil, A. M., & Shah, M. H. (2013). Overall
manufacturing. Proceedings ASME Design Engineering Techni- Equipment Effectiveness (OEE) Calculation - Automation through
cal Conference. https://doi.org/10.1115/DETC2019-98415 Hardware & Software Development. Procedia Eng, 51, 579–584.
Rolf, B., Reggelin, T., Nahhas, A., Lang, S., & Müller, M. (2020). https://doi.org/10.1016/J.PROENG.2013.01.082
Assigning dispatching rules using a genetic algorithm to solve a Sivaram, V. (2018). Taming the Sun: Innovations to harness solar energy
hybrid flow shop scheduling problem. Procedia Manufuring, 42, and power the planet. Cambridge: MIT Press.
442–449. https://doi.org/10.1016/j.promfg.2020.02.051 Sosnovik, I., & Oseledets, I. (2019). Neural networks for topology
Rosenblatt, F. (1958). The perceptron: A probabilistic model for optimization. Russian Journal of Numerical Analysis and Mathe-
information storage and organization in the brain. Psychological matical Modelling, 34, 215–223. https://doi.org/10.1515/RNAM-
Review, 65, 386–408. https://doi.org/10.1037/h0042519 2019-0018
Said, M., Abdellafou, K. B., & Taouali, O. (2020). Machine learning Stojanovic, V., He, S., & Zhang, B. (2020). State and parameter joint
technique for data-driven fault detection of nonlinear processes. estimation of linear stochastic systems in presence of faults and
Journal of Intellegence Manufuring, 31, 865–884. https://doi.org/ non-Gaussian noises. International Journal of Robust and Nonlin-
10.1007/s10845-019-01483-y ear Control, 30, 6683–6700. https://doi.org/10.1002/RNC.5131
Schaaf, K. (1999). Uncertainty and Sensitivity Analysis of the Heat Sukumar, M. S., Ramaiah, P. V., & Nagarjuna, A. (2014). Optimiza-
Transfer Mechanisms in the Lower Head, No. NEA-CSNI- tion and prediction of parameters in face milling of Al-6061 using
R—1998-18. Taguchi and ANN approach. Procedia Eng, 97, 365–371. https://
Schreiber, M., Klöber-Koch, J., Bömelburg-Zacharias, J., Braunreuther, doi.org/10.1016/j.proeng.2014.12.260
S., & Reinhart, G. (2019). Automated quality assurance as an intel- Sun, Y., Qin, W., Zhuang, Z., & Xu, H. (2021). An adaptive fault
ligent cloud service using machine learning. Procedia CIRP, 86, detection and root-cause analysis scheme for complex industrial
185–191. https://doi.org/10.1016/j.procir.2020.01.034 processes using moving window KPCA and information geomet-
Scime, L., & Beuth, J. (2018a). Anomaly detection and classification in ric causal inference. Journal of Intelligent Manufacturing, 32,
a laser powder bed additive manufacturing process using a trained 2007–2021. https://doi.org/10.1007/s10845-021-01752-9
computer vision algorithm. Additive Manufacturing, 19, 114–126. Sutton, R. S., & Barto, A. G. (2015). Reinforcement Learning (2nd ed.).
https://doi.org/10.1016/J.ADDMA.2017.11.009 New York: The MIT Press.
Scime, L., & Beuth, J. (2018b). A multi-scale convolutional neural Swiler, L. P., Eldred, M. S., & Adams, B. M. (2017). Dakota: Bridg-
network for autonomous anomaly detection and classification in a ing advanced scalable uncertainty quantification algorithms with

123
54 Journal of Intelligent Manufacturing (2023) 34:21–55

production deployment. Handbook of Uncertainity Quantification Weiwen, X., Junqi, W., & Wansheng, Z. (2018). Break-out detection
(pp. 1651–1693). Berlin: Springer. for high-speed small hole drilling EDM based on machine learn-
Tapia, G., Elwany, A. H., & Sang, H. (2016). Prediction of porosity in ing. Procedia CIRP, 68, 569–574. https://doi.org/10.1016/j.procir.
metal-based additive manufacturing using spatial Gaussian pro- 2017.12.115
cess models. Additive Manufacturing, 12, 282–290. https://doi. Widodo, A., & Yang, B. S. (2007). Support vector machine in machine
org/10.1016/J.ADDMA.2016.05.009 condition monitoring and fault diagnosis. Mechanical Systems
Tapia, G., Khairallah, S., Matthews, M., King, W. E., & Elwany, and Signal Processing, 21, 2560–2574. https://doi.org/10.1016/
A. (2017). Gaussian process-based surrogate modeling frame- j.ymssp.2006.12.007
work for process planning in laser powder-bed fusion additive Widrow, B., & Lehr, M. A. (1990). 30 years of adaptive neural networks:
manufacturing of 316L stainless steel. International Journal of perceptron, madaline, and backpropagation. Proceedings of the
Advanced Manufacturing Technology, 94, 3591–3603. https://doi. IEEE, 78, 1415–1442. https://doi.org/10.1109/5.58323
org/10.1007/S00170-017-1045-Z Wiendahl, H. P., & Scholtissek, P. (1994). Management and con-
Thoben, K. D., Wiesner, S. A., & Wuest, T. (2017). “Industrie 4.0” and trol of complexity in manufacturing. CIRP Annals - Manufac-
smart manufacturing-a review of research issues and application turing Technology, 43, 533–540. https://doi.org/10.1016/S0007-
examples. International Journal of Automative Technology, 11, 8506(07)60499-5
4–16. https://doi.org/10.20965/ijat.2017.p0004 Wu, D., Jennings, C., Terpenny, J., Gao, R. X., & Kumara, S. (2017). A
Thompson, M. K., Moroni, G., Vaneker, T., Fadel, G., Campbell, comparative study on machine learning algorithms for smart man-
R. I., Gibson, I., et al. (2016). Design for additive manufactur- ufacturing: Tool wear prediction using random forests. Journal of
ing: Trends, opportunities, considerations, and constraints. CIRP Manufacturing Science and Engineering Transactions on ASME,
Annals - Manufacturing Technology, 65, 737–760. https://doi.org/ 2017, 139. https://doi.org/10.1115/1.4036350
10.1016/j.cirp.2016.05.004 Wu, M., Phoha, V. V., Moon, Y. B., & Belman, A. K. (2016b). Detecting
Tian, L., & Luo, Y. (2020). A study on the prediction of inherent defor- malicious defects in 3D printing process using machine learning
mation in fillet-welded joint using support vector machine and and image classification. ASME International Mechanical Engi-
genetic optimization algorithm. Journal of Intelligent Manufactur- neering & Congress and Exposition Proceedings. https://doi.org/
ing, 31, 575–596. https://doi.org/10.1007/s10845-019-01469-w 10.1115/IMECE201667641
Tizghadam, A., Khazaei, H., Moghaddam, M. H. Y., & Hassan, Y. Wu, D., Rosen, D. W., Wang, L., & Schaefer, D. (2015a). Cloud-based
(2019). Machine learning in transportation. Journal of Advanced design and manufacturing: A new paradigm in digital manufac-
Transportation. https://doi.org/10.1155/2019/4359785 turing and design innovation. Computer Design, 59, 1–14. https://
Vaidya, S., Ambad, P., & Bhosle, S. (2018). Industry 4.0 – A glimpse. doi.org/10.1016/J.CAD.2014.07.006
Procedia Manufacturing, 20, 233–238. https://doi.org/10.1016/J. Wu, M., Song, Z., & Moon, Y. B. (2017). Detecting cyber-physical
PROMFG.2018.02.034 attacks in CyberManufacturing systems with machine learn-
Wagner, T., Schertl, A., Elger, J., & Vollmar, J. (2008). Evaluation ing methods. Journal of Intelligence and Manufacturing, 30,
of effectiveness and impact of decentralized automation. IEEE 1111–1123. https://doi.org/10.1007/S10845-017-1315-5
International Conference Emerging Technology Facture Automa- Wu, H., Wang, Y., & Yu, Z. (2015b). In situ monitoring of FDM
tive ETFA, 2008, 1128–1136. https://doi.org/10.1109/ETFA.2008. machine condition via acoustic emission. International Journal
4638539 of Advanced Manufuring Technology, 84, 1483–1495. https://doi.
Wang, T., Kwok, T. H., Zhou, C., & Vader, S. (2018a). In-situ droplet org/10.1007/S00170-015-7809-4
inspection and closed-loop control system using machine learning Wu, H., Yu, Z., et al. (2016). A new approach for online monitoring of
for liquid metal jet printing. Journal of Manufacturing Systems, additive manufacturing based on acoustic emission. Asmedigital-
47, 83–92. https://doi.org/10.1016/J.JMSY.2018.04.003 collection. https://doi.org/10.1115/MSEC2016-8551
Wang, C., Tan, X. P., Du, Z., Chandra, S., Sun, Z., Lim, C. W. J., Wu, H., Yu, Z., & Wang, Y. (2016a). Real-time FDM machine condition
et al. (2019). Additive manufacturing of NiTi shape memory alloys monitoring and diagnosis based on acoustic emission and hidden
using pre-mixed powders. Journal of Materials Processing Tech- semi-Markov model. International Journal of Advanced Man-
nology, 271, 152–161. https://doi.org/10.1016/J.JMATPROTEC. ufacturing Technology, 90, 2027–2036. https://doi.org/10.1007/
2019.03.025 S00170-016-9548-6
Wang, C., Tan, X., Liu, E., & Tor, S. B. (2018b). Process parameter Wuest, T., Irgens, C., & Thoben, K. D. (2014). An approach to monitor-
optimization and mechanical properties for additively manufac- ing quality in manufacturing using supervised machine learning
tured stainless steel 316L parts by selective electron beam melting. on product state data. Journal of Intelligent Manufacturing, 25,
Materials and Design, 147, 157–166. https://doi.org/10.1016/J. 1167–1180. https://doi.org/10.1007/s10845-013-0761-y
MATDES.2018.03.035 Wuest, T., Weimer, D., Irgens, C., & Thoben, K.-D. (2016). Machine
Wang, C., Tan, X. P., Tor, S. B., & Lim, C. S. (2020). Machine learning learning in manufacturing: Advantages, challenges, and appli-
in additive manufacturing: State-of-the-art and perspectives. Addi- cations. Prod Manuf Res, 4, 23–45. https://doi.org/10.1080/
tive Manufacturing, 36, 101538. https://doi.org/10.1016/j.addma. 21693277.2016.1192517
2020.101538 Xin, X., Tu, Y., Stojanovic, V., Wang, H., Shi, K., He, S., et al. (2022).
Wang, L., Törngren, M., & Onori, M. (2015). Current status and Online reinforcement learning multiplayer non-zero sum games
advancement of cyber-physical systems in manufacturing. Journal of continuous-time Markov jump linear systems. Applied Mathe-
of Manufacturing Systems, 37, 517–527. https://doi.org/10.1016/ matics and Computation, 412, 126537. https://doi.org/10.1016/J.
J.JMSY.2015.04.008 AMC.2021.126537
Weiss, S. M., Dhurandhar, A., Baseman, R. J., White, B. F., Logan, R., Xing, C., Jia, C., Han, Y., Dong, S., Yang, J., & Wu, C. (2020).
Winslow, J. K., et al. (2014). Continuous prediction of manufactur- Numerical analysis of the metal transfer and welding arc behav-
ing performance throughout the production lifecycle. Journal of iors in underwater flux-cored arc welding. International Journal
Intelligent Manufacturing, 27, 751–763. https://doi.org/10.1007/ of Heat and Mass Transfer, 153, 119570. https://doi.org/10.1016/
S10845-014-0911-X J.IJHEATMASSTRANSFER.2020.119570

123
Journal of Intelligent Manufacturing (2023) 34:21–55 55

Yao, X., Moon, S. K., & Bi, G. (2017). A hybrid machine learning Zhang, M., Sun, C., Zhang, X., & Goh, P. (2019). High cycle fatigue
approach for additive manufacturing design feature recommenda- life prediction of laser additive manufactured stainless steel: A
tion. Rapid Prototyping Journal, 23, 983–997. https://doi.org/10. machine learning approach. Amsterdam: Elsevier.
1108/RPJ-03-2016-0041 Zhou, J., Jia, C., Guo, M., Chen, M., Gao, J., & Wu, C. (2021). Inves-
Ye, D., Hong, G. S., Zhang, Y., Zhu, K., & Fuh, J. Y. H. (2018a). Defect tigation of the WAAM processes features based on an indirect arc
detection in selective laser melting technology by acoustic signals between two non-consumable electrodes. Vacuum, 183, 109851.
with deep belief networks. International Journal Advanced Man- https://doi.org/10.1016/J.VACUUM.2020.109851
ufacturing Technology, 96, 2791–2801. https://doi.org/10.1007/ Zhou, L., Tao, H., Paszke, W., Stojanovic, V., & Yang, H. (2020).
S00170-018-1728-0 PD-type iterative learning control for uncertain spatially inter-
Ye, D., Hsi Fuh, J. Y., Zhang, Y., Hong, G. S., & Zhu, K. (2018b). In connected systems. Mathematics, 8, 1528. https://doi.org/10.3390/
situ monitoring of selective laser melting using plume and spatter MATH8091528
signatures by deep belief networks. ISA Transactions, 81, 96–104.
https://doi.org/10.1016/J.ISATRA.2018.07.021
Yi, L., Gläßner, C., Krenkel, N., & Aurich, J. C. (2020). Energy simula- Publisher’s Note Springer Nature remains neutral with regard to juris-
tion of the fused deposition modeling process using machine learn- dictional claims in published maps and institutional affiliations.
ing approach. Procedia CIRP. https://doi.org/10.1016/j.procir.
2020.01.002
Yin, Y., Stecke, K. E., & Li, D. (2017). The evolution of produc-
tion systems from Industry 2.0 through Industry 4.0. International
Journal of Production Research, 56, 848–861. https://doi.org/10.
1080/00207543.2017.1403664
Yu, L., Liu, H. (2003). Feature selection for high-dimensional data: A
fast correlation-based filter solution. Proceedings of 20th Inter-
nationsl Confernce and Machine Learning (ICML-03), 2003,
pp. 856–863.
Zhang, Y., Hong, G. S., Ye, D., Zhu, K., & Fuh, J. Y. H. (2018). Extrac-
tion and evaluation of melt pool, plume and spatter information for
powder-bed fusion AM process monitoring. Materials and Design,
156, 458–469. https://doi.org/10.1016/J.MATDES.2018.07.002
Zhang, H.-C., & Huang, S. H. (1995). Applications of neural networks
in manufacturing: A state-of-the-art survey. International Jour-
nal of Production Research, 33, 705–728. https://doi.org/10.1080/
00207549508930175
Zhang, Z., Liu, Z., & Wu, D. (2021). Prediction of melt pool temperature
in directed energy deposition using machine learning. Additive
Manufacturing, 37, 101692. https://doi.org/10.1016/J.ADDMA.
2020.101692

123