0% found this document useful (0 votes)
6 views17 pages

Data Driven Approaches Toward Smarter Additive Manufacturing

The document reviews the integration of data-driven approaches in additive manufacturing (AM), highlighting their potential to enhance material and structural design. It discusses the advantages of AM, such as the ability to create complex geometries and optimize materials, while addressing challenges like material properties and process parameters. The authors emphasize the importance of leveraging large datasets and machine learning to improve the efficiency and effectiveness of AM processes, particularly in aerospace and biomedical applications.

Uploaded by

andrei sackser
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
6 views17 pages

Data Driven Approaches Toward Smarter Additive Manufacturing

The document reviews the integration of data-driven approaches in additive manufacturing (AM), highlighting their potential to enhance material and structural design. It discusses the advantages of AM, such as the ability to create complex geometries and optimize materials, while addressing challenges like material properties and process parameters. The authors emphasize the importance of leveraging large datasets and machine learning to improve the efficiency and effectiveness of AM processes, particularly in aerospace and biomedical applications.

Uploaded by

andrei sackser
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 17

REVIEW

www.advintellsyst.com

Data-Driven Approaches Toward Smarter Additive


Manufacturing
Chenxi Tian, Tianjiao Li, Jenniffer Bustillos, Shonak Bhattacharya, Talia Turnham,
Jingjie Yeo,* and Atieh Moridi*

products with complex geometries and, in


The latest industrial revolution, Industry 4.0, is driven by the emergence of digital some cases, with multiple materials. The
manufacturing and, most notably, additive manufacturing (AM) technologies. former enables lightweight designs of sin-
The simultaneous material and structure forming in AM broadens the material gle components that can be topologically
optimized as well as the possibility of con-
and structural design space. This expanded design space holds a great potential
solidating multiple discrete parts, typically
in creating improved engineering materials and products that attract growing designed for conventional assembly, into a
interests from both academia and industry. A major aspect of this growing single component.[5,6] The latter enables
interest is reflected in the increased adaptation of data-driven tools that accel- production of functionally graded parts or
erate the exploration of the vast design space in AM. Herein, the integration of composites.[7] Other complementary
data-driven tools in various aspects of AM is reviewed, from materials design in advantages of AM include reducing pro-
duction lead time through elimination of
AM (i.e., homogeneous and composite material design) to structure design for
tooling, being cost efficient for small or
AM (i.e., topology optimization). The optimization of AM tool path using machine custom production runs, and supply chain
learning for producing best-quality AM products with optimal material and transformation by skipping many tradi-
structure is also discussed. Finally, the perspectives on the future development of tional manufacturing steps.[8,9] Significant
holistically integrated frameworks of AM and data-driven methods are provided. advances in AM are steering AM away from
merely a rapid prototyping tool and toward
being a viable candidate for large-scale
manufacturing of components, especially
1. Introduction for the aerospace and biomedical industries.[10]
The broadened material and geometrical design space of AM
Additive manufacturing (AM) is a process in which feedstock requires more efficient methods for discovering optimal designs.
materials are joined layer by layer guided by a digital design. Some AM-specific challenges, such as anisotropic material prop-
Various AM processes can be categorized into different material erties and residual stress accumulation, require further develop-
classes (polymer, ceramic, and metal), feedstock types (slurry, ments in the materials used in AM as well as in fine tuning
powder, wire/filament, and sheet), or energy sources (kinetic process parameters and tool path design. The complex underlying
bonding and thermal fusion). Table 1 shows the most widely physics of AM processes also brings challenges in understanding
used AM processes. the relationship between process parameters and quality of the
While each process has different print resolution, print speed, product, thereby hindering industrial applications that harness
and feedstock type, they share some common advantages that set the full potential of AM. However, the emergence of a diverse
AM apart from traditional manufacturing methods.[1–4] range of modern data-driven approaches has a great potential
Arguably, the most significant advantages are in manufacturing in enabling the AM community to overcome these challenges.
By unifying the first three scientific paradigms of theory,
experiment, and computation/simulation, data-driven
C. Tian, T. Li, J. Bustillos, S. Bhattacharya, T. Turnham, Prof. J. Yeo, approaches are illuminating the way toward the fourth paradigm
Prof. A. Moridi of science: data-driven science.[11–13] Data-driven science is
Mechanical and Aerospace Engineering
Cornell University strongly galvanized by the enhanced ability to gather massive
124 Hoy Rd., Upson Hall, Ithaca, NY 14850, USA amounts of data, the rapid development of data infrastructures
E-mail: [email protected]; [email protected] where big data are stored, as well as advances in methods of data
The ORCID identification number(s) for the author(s) of this article mining that employ such data.[14] Recently, with extensive advan-
can be found under https://doi.org/10.1002/aisy.202100014. ces in both computational hardware and algorithms, data-driven
approaches are rapidly permeating many fields of science and
© 2021 The Authors. Advanced Intelligent Systems published by Wiley-
VCH GmbH. This is an open access article under the terms of the technology. For AM in particular, data-driven approaches are
Creative Commons Attribution License, which permits use, distribution increasingly being adopted. Large quantities of materials data
and reproduction in any medium, provided the original work is have already been generated for, or can be utilized by, AM.
properly cited. Some of these data are from benchtop experiments and
DOI: 10.1002/aisy.202100014 tests,[15,16] whereas a large portion of them are generated by

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (1 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

Table 1. List of different AM processes with their most commonly used material classes, feedstock types, and energy source.

Process name Material class Feedstock Energy source Reference


FDM Polymer Filament Resistance heating [167–169]
SLA Polymer Liquid Laser or light [169,170]
SLS Polymer, ceramic, or metal Powder Laser [171–173]
PBF Metal Powder Laser or electron beam [174–177]
Binder jetting Polymer, metal, or ceramic Powdera) Thermal heating [172,178]
DED Metal Powder or Wire Laser, electron beam, or plasma arc [179]
Cold spray Metal Powder Kinetic energy [180]
Wire arc additive mfg. (WAAM) Metal Wire Plasma arc [181]
Ultrasonic sheet lamination (UAM) Metal Sheet Ultrasound [182]

a)
Powder feedstock joint with liquid binder.

computer simulations and calculations.[17,18] Leveraging these


large quantities of data, various machine learning (ML) algo-
rithms can be used to extract the hidden correlations and pat-
terns in material characteristics and properties. In this way,
data-driven methods inform the design of materials and
manufacturing processes for AM. The datasets also enable
high-throughput methods for material mining/screening, in
which optimal materials are filtered according to a desired design
attribute or technical requirement.[14] In this way, data-driven
approaches allow cost-effective material selection for AM.
Thus, the application of such data-driven approaches in AM
has the potential to address the challenges mentioned earlier.
In this Review, we highlight different crucial aspects of cou- Figure 1. Graphical illustration of the structure of the current Review.
pling data-driven approaches and AM. Although efficient data-
driven approaches can simultaneously use various computational
techniques to produce data or process data, the specifics of each data-driven methods can accelerate the optimization of the tool
computational technique will not be covered in this Review. path and process parameter design. In addition, discussions on
Interested readers can learn more about these techniques from relevant databases as fundamentals of data-driven approaches are
relevant reviews.[19–21] We will focus on data-driven methods that included into each section of this Review.
are most applicable to AM of metallic products, while drawing
parallels to developments in the field of polymer AM when such 2. Homogeneous Material Design for AM
methods discussed are yet to be implemented in metallic AM. A
detailed review is available elsewhere on the state-of-the-art tech- The rapid pace of advances in manufacturing technologies across
niques in data-driven AM that are more specific to polymeric multiple industries requires significant enhancements in the
materials.[19] The data-driven methods discussed in our performance of engineering materials, such as metals with
Review include various digital methods of high-throughput sim- mechanical properties that exceed those of traditional alloys.[22,23]
ulation and optimization, ML, neural networks, and genetic algo- Yet, materials that are suitable for AM are limited. For instance, it
rithms (GA). For a more detailed review on current is estimated that merely 0.2% of commercial metallic alloys are
developments in ML specifically for metallic AM, we refer the printable.[22,24] This statistic emphasizes the need for robust
readers to another excellent review article.[21] The structure of computational tools that can aid in the design of alloys that
this Review is shown in Figure 1. Here, in the first portion, are suitable for the AM process. Efforts in computational mate-
we focus on data-driven approaches that advance the develop- rials design have been successful in screening promising new
ment of AM in material discovery. For homogenous materials, stable materials via density functional theory (DFT) and molecu-
we discuss how data-driven approaches can accelerate the discov- lar dynamics (MD) simulations.[23,25–27] Recently, along with
ery of materials most suited for AM’s process characteristics and demands from industry, materials science has seen a dramatic
intrinsic properties. For composite materials, we discuss the ben- change in the landscape of materials design and specifically alloy
efit of using data-driven tools to discover superior designs that design via data-driven approaches.[23,25–27] Data-driven
can achieve enhanced mechanical performance and realize the approaches based on the powerful advances in computational
full potential of AM. The second portion of our Review demon- modeling have already demonstrated their remarkable potential
strates how AI-aided structural design can take advantage of for various applications in materials research. Hence, greater
reduced constraints on component geometry in AM and how attention is needed from the AM community to harness the

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (2 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

data-driven research scheme to further expand the library of strength of the final product.[41] A recent study focused on radia-
applicable materials. tive cooling in DED-AM and obtained the dielectric constant of
the nickel-based superalloy Inconel-718 via experiments as well
as DFT calculations to determine the spectral reflectivity and
2.1. DFT and High-Throughput DFT emissivity.[42] Due to the generality of the DFT calculation, these
reported methods can be applied to a large variety of materials
From the standpoint of AM, first-principle atomistic modeling and can be readily turned into tools for materials design and
such as DFT and the corresponding data are highly beneficial quality control for AM.
in terms of teasing apart the complex physics underlying the Currently, DFT calculations are one of the most accepted
manufacturing process. DFT simulates the collective behavior approaches to assess material properties using their atomistic
of atoms using quantum mechanical calculations to unravel compositions and structures.[14] However, the unprecedented
material characteristics. These computations are vital for under- boom in computational resources and rapid advances in data sci-
standing the molecular properties that consequently affect the ences greatly increased the awareness and ability of the material
meso- and macroscale properties of materials. The core of atom- science community in storing and interpreting large quantities of
istic modeling lies in accurate and efficient techniques for data generated by these individual DFT calculations for identify-
numerically solving the Schrödinger equation for complex sys- ing higher-level correlations and patterns for material discovery
tems that consist of multiple atoms.[20,28] Examples of using and design.[14,43] Recently, aggregating tens and even hundreds
DFT in materials science, and more specifically in materials of thousands of DFT calculations in large databases combined
design, include studying the formation of defects in metallic with data-driven methods has intensified the ability to conduct
materials,[29–31] tailoring mechanical properties of alloys,[32–34] high-throughput DFT (HT DFT) simulations.[44] One such large
and polymeric materials design.[20,35,36] online database is the Open Quantum Materials Database
Using quantum mechanical calculation such as DFT to (OQMD)[17,44] that contains thermodynamic and structural prop-
directly inform material research for AM is shown as the blue erties of over 800 000 materials calculated using DFT and the
flow in Figure 2. For instance, in powder bed fusion (PBF) numbers are growing steadily. OQMD was successfully used
AM, the surface energies of different material phases drive to search for the most stable set of structures in Mg-based ternary
the melt penetration phenomenon and strongly affect the quality alloy systems at a given composition using grand canonical linear
of the resulting component.[37] The surface energies are difficult programming.[45] OQMD was also used to conduct HT DFT in
to measure experimentally, but DFT calculations offer a way to the search for strengthening precipitates that could be applied to
determine these energies in several metallic surfaces.[38,39] a variety of alloy matrices including Fe, Al, Mg, Ni, Co, and Ti.[46]
Dumontet et al. examined how the elastic properties of OQMD was also used as a source for large training datasets to
Ti 6Al 4V were altered by nonequilibrium structures derived train an ML model in a recent study for identifying quaternary
from the high cooling rates in PBF-AM.[40] Their DFT calcula- Heusler materials.[47] Besides OQMD, there are several efforts to
tions showed that the α’ martensitic structure had lower generate large-scale HT DFT databases, such as the Materials
Young’s modulus than the pure Ti-α phase, which was attributed Project,[48] AFLOWLIB,[49] Computational Materials
to the influence of Al and V substitutional elements.[40] In direct Repository, [50]
and NOMAD.[51] Recent advances of HT DFT
energy deposition (DED), the melt pool cooling rate is one of the database developments and their applications in data-driven
most important factors that influence the microstructure and methods can be found in several review articles.[14,26,52]

Figure 2. Quantum mechanical calculation (e.g., DFT) and MD simulations combined with data-driven approaches for AM materials design. Blue: using
quantum mechanical calculations directly to guide materials design. Green: materials design through HT DFT. Purple: using MD simulations to directly
guide materials design. Red: using machine-learned potential energies in MD simulations to guide materials design. The yellow shapes represent the
output/input which characterize the boxes.

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (3 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

Developing these HT DFT infrastructures will enable expeditious of polymer crosslink density, thereby guiding better control over
screening of materials in general, whereas currently, to our best print quality.[59] Such MD and CGMD simulations are important
knowledge, their application in AM materials research is scarce. sources of raw data for constructing databases for data mining to
Therefore, the AM community can strongly benefit by leveraging uncover hidden properties and correlations that convention
the advances in HT-DFT for the purpose of materials discovery observation of such data may not reveal. For example, The
and design research, shown as the green flow in Figure 2. Inorganic Crystal Structure Database (ICSD),[60] arguably the
world’s largest database of fully evaluated and published crystal
2.2. MD Simulations and ML Potentials structure data, was extended in 2017 and started accepting theo-
retical structure data, including those generated by MD simula-
In terms of AM, MD has demonstrated its strength in directly tions.[61] Taking a leaf out of these efforts, we propose that
tailoring the design of the materials’ nanostructure and guiding materials design and selection in AM will also benefit signifi-
the optimization of the manufacturing process, shown as the cantly from systematic storage, curation, and dissemination of
purple flow in Figure 2. In addition, MD simulation is known relevant experimental and computational data. Such databases
to be one of the very few methods available to provide informa- will become excellent platforms for data mining applications
tion of the nature of phase transformation of high-entropy alloys for materials discovery and design in AM.
(HEAs), which is emerging as an important material for AM, Other than data mining for hidden composition
given that the transformation process is very fast.[53] structure function correlations from databases with MD data,
Computational quantum mechanical methods are generally data-driven approaches such as ML can also change the scheme
accurate at the molecular scale but become extremely computa- of MD-aided materials discovery and design by introducing ML
tionally expensive at much larger spatiotemporal scales.[20] These potentials. MD is particularly potent for simulating systems of
costs limit computational quantum mechanical methods to relatively higher spatiotemporal scales, but it has lower accuracy
exploring phenomena that occur on the order of tens of picosec- compared with quantum mechanical methods. This is because
onds and a few thousand atoms. In contrast, MD methods pro- the central component of MD, the potential energy surfaces
vide an alternative for overcoming these limitations. Using the (PES), is usually fitted and applied empirically, which introduces
Born–Oppenheimer approximation, the quantum mechanics approximation errors.[20] To alleviate this drawback, there is a
laws governing the interactions between atoms can be approxi- growing interest in using ML methods to derive MD atomic
mated with the laws of Newtonian mechanics. Due to this potentials from large quantum mechanical datasets. This is done
approximation, electron-related processes cannot be simulated, in two steps: first, transform the reference atomic configurations
but the complexity of the calculation is significantly reduced. into structural descriptors and second, use ML algorithms to
The reduced complexity enables simulations of atomistic sys- learn the mapping between the descriptors and the correspond-
tems up to micrometers in length scale and microseconds in ing potential energy.[62] This process is shown as the red flow in
time scale. Furthermore, through dimensionality reduction Figure 2. Once trained, the ML potentials provide comparable
(while preserving the conformational details, thermodynamic accuracy with quantum mechanical calculations, while accelerat-
properties, and particle dynamics), coarse-grained MD ing the simulations by orders of magnitude.[63] Artificial neural
(CGMD) methods are able to simulate systems of even larger spa- networks (ANNs) are extensively used to achieve this.[64,65]
tiotemporal scales. Thus, MD methods are more suited for Specifically, ANNs are increasingly used to develop interatomic
designing larger material systems. potentials for a large variety of alloys.[66–69] Numerous frame-
Using MD simulations, the thermodynamic properties (size works are available to learn the PES from quantum mechanical
dependency of the melting point, heat of fusion, entropy of data using ANN, including SIMPLE-NN,[70] SchNet,[71,72]
fusion, and surface energy) of Ag nanoclusters during the heat- DeePMD,[73] and TensorAlloy.[69] Another class of ML method
ing and cooling process were quantified to inform the thermal used for this purpose is Gaussian processes (GP), a nonparamet-
and energy management for AM.[54] In another MD study of ric method with no fixed functional form. Thus, GPs are suitable
core/shell nanoparticles, the effects of the core volume fraction for modeling highly complex potential energy landscapes. A
on the interfacial melting of the core metal and the gradual alloy- representation of this method is the Gaussian approximation
ing of the molten shell were investigated.[55] This study revealed potentials (GAPs).[74–76] GAP was also used to derive the atomic
principles of designing powder materials and implied new ways potential from quantum mechanical datasets to model and sim-
of reducing the parts of manufacturing time. Using CGMD ulate metallic materials.[77,78] In general, ANNs require greater
methods such as dissipative particle dynamics (DPD) and its efforts in model training compared with GP, whereas making
extended version, the many-body dissipative particle dynamics predictions using a GP model is slower than in ANNs.[70]
(MDPD) enables modeling and simulating systems at the meso- Nevertheless, these ML potentials allow researchers to conduct
scale. For instance, the agglomeration of ultraviolet ink fluid and atomistic simulations for materials design and discovery with
the effects of surfactants in 3D nanoinkjet printing were exam- higher physical accuracy and computational efficiency. The mate-
ined with DPD and MDPD.[56,57] Similar methods were also used rials research community has already started expediting the
to study nanodroplet formation in 3D nanoinkjet printing.[58] application of ML potentials in a wide range of materials
CGMD was also used to simulate the 3D polymer printing pro- research, but specifically for purposes of AM, we find that such
cess known as continuous liquid interface production (CLIP) for applications are still not well documented in current literature.
identifying the factors influencing the fidelity during the Hence, an intensified focus on applying ML potentials can be
manufacturing process.[59] This study highlighted the effects anticipated to firmly bolster materials research in AM.

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (4 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

2.3. Computational Thermodynamics and postprocessing. This recent introduction of AM-enabled


HEA applications and the lack of extensive prior studies in
The cyclic thermal history, intrinsic to the additive process, HEA drives attention to the use of data-driven thermodynamic
entails a number of liquid-to-solid and solid-state phase transfor- tools like high-throughput CALPHAD to study the composi-
mations affecting the microstructure and properties of alloys.[22] tion structure mapping. While the data of a large amount of ter-
Therefore, the material design process for AM must consider a nary systems required for reliable prediction in exploring design
thorough understanding of the composition-dependent space of HEAs makes the trial-and-error approach infeasible,
thermo chemical and physical events associated with the HT-DFT has been shown to be a viable alternative to provide
far-from-equilibrium processing.[22,24] enthalpy of mixing data for solid solutions and enthalpy of for-
The calculation of phase diagram (CALPHAD) approach is mation data for intermetallic phases.[83] One particular example
widely used to predict thermodynamic and kinetic properties of used high-throughput CALPHAD modeling to conduct thermo-
new powder compositions.[79,80] The CALPHAD approach is a dynamic data-driven design of Al Co Cr Fe Ni multiprinci-
unique tool that describes a material’s thermodynamic properties pal element alloy. This study successfully predicted the formation
using empirical and/or theoretical data of its constituents at specific phases and microstructures of high-performing designs from
temperature ranges (room temperature—above melting tempera- 10 000 different compositions with 1% step size in the predeter-
tures).[80] By fitting the data to the correct thermodynamic model, mined alloying element atomic percent ranges.[84] This study
phase diagrams can be extracted to describe the thermo chemical illustrates the capability of CALPHAD modeling in comprehend-
behavior of the material as a function of temperature and composi- ing the solidification path of complex HEAs, as well as predicting
tion, various thermo chemical (i.e., heat of mixing, chemical mechanical performances qualitatively from phase volume
activity, vapor pressure), and thermo physical properties fractions.
(i.e., coefficient of thermal expansion, bulk modulus) based on However, accurate HEA phase diagram prediction is based on
the thermodynamic equilibrium states and chemical potentials having complete data for full binary and ternary subsystems of all
(i.e., regular solution behavior).[79,80] These properties at equilib- elements, but even the most comprehensive databases include
rium can then be used to analyze closer-to-process conditions such only a limited number (about 5%) of fully assessed ternary sys-
as nonequilibrium and rapid cooling rates that are encountered dur- tems.[85] Close attention by the users is required to the quality of
ing AM (Figure 3a). Readily available software packages, including data, as well as its applicability to the described thermodynamic
ThermoCalc and Factsage, use the CALPHAD approach to simulate problem. For instance, many of the now-matured thermody-
and predict the dynamic behavior of material.[79–81] namic databases were established, targeting the study of conven-
For example, HEA is a large category of AM-specific materials, tional single-principal alloys.[83] The aforementioned bias on
showing outstanding mechanical performance with intricate single principal element alloys in many thermodynamic data-
deformation mechanisms (among others), but has limited appli- bases makes them sometimes inadequate for interpolation of
cations due to poor machinability.[82] The uniqueness of AM properties for alloys with multiple principal elements, such as
makes manufacturing of HEAs feasible with reduced machining HEAs. The increased adoption of CALPHAD in AM-specific

Figure 3. Informed design process of higher-performing alloys for AM via a data-driven approach.

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (5 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

materials design correspondingly necessitates intelligent GA could suggest alternatives that outperform known composi-
improvement of its infrastructure and expansion of modeling tions based on the similarity hypothesis.[21] By evaluating the can-
to more properties of interests.[86,87] The capabilities of didate’s composition through a fitness function that measures a
CALPHAD are clearly summarized in a separate review.[86] composition-based property criteria, the algorithm can iteratively
While highlighting current developments in CALPHAD, that select the best-performing compositions based on the input data-
review also points out the lack of strategies for implementing set.[21,23,94] The qualifications of the alloy compositions are
new models and data to efficiently update the databases. evaluated with a model describing the composition and pro-
Furthermore, CALPHAD modeling can greatly benefit from a cess-dependent properties, and the success of the adaptive learn-
more sustainable data ecosystem, where individual data reposi- ing algorithm is based on this accelerated qualification.[94] In the
tories have mechanisms to communicate and form a joint data- case of AM alloys, these models can include computational mod-
base when needed.[87] els describing AM-specific phenomena as described throughout
In addition to aiding the design of AM-specific alloys, the use this section (DFT modeling, MD simulations, thermo chemical,
of CALPHAD modeling has also been extended to improve the and thermo physical calculations).[21,94]
manufacturing of powder alloy feedstock used in most AM pro- In addition to GAs, metamodeling as an ML approach can
cesses. A method to quantify the uncertainty between prealloyed swiftly screen metallic compositions based on datasets generated
and target powder composition is developed using a CALPHAD- solely via computational modeling.[95] An example of metamod-
based integrated computational materials engineering (ICME) eling to design alloys with enhanced properties was demon-
framework.[88] Uncertainty in AM part material composition, strated by the development of an ANN model that predicted
including feedstock variation between batches, localized compo- the sigma phase fraction in a Ni superalloy.[95] The sigma phase
sitional deviation from nominal feedstock, and contamination is known to precipitate in the Ni superalloy at high operating
from powder recycling, negatively affects the success rate of temperatures and this phase is detrimental to the component’s
the overall AM processes. Therefore, it is important to capture structural integrity. The model consists of a complex alloy system
this uncertainty in manufacturing during AM materials (mole fractions of Ni, Al, Co, Cr, Mo, Re, Ta, Ti, W) and temper-
design.[88] This study identified an optimized average composi- ature as input parameters and output the target property of the
tion of the high-strength low-alloy (HSLA) powder used in AM, sigma phase fraction. The speed of this ANN model is demon-
increasing the success rate of AM builds by 44.7%.[88] strated by taking merely 1.7 s to calculate the target property of a
Other computational thermodynamics methods are also being 40 000-point dataset on a personal computer. Being a neural net-
integrated with CALPHAD to understand complex solidification work model, this model has the advantage of producing multi-
events and phase formation in AM. An alternative diffusion mod- outputs simultaneously as well as the capability to be scaled
ule, DICTRA, enables a closer-to-practice prediction of solidifica- up for larger design spaces.[95]
tion parameters by accounting for diffusion in the liquid and These early alloy design studies demonstrate the potential of
solid phases.[79,80,89] Furthermore, the thermodynamics and screening a vast design space of virtual alloys with targeted com-
kinetics of the solidification process of the alloy can be predicted position and process-dependent properties without the need for
by extracting simulated melt pool conditions in laser PBF (L-PBF) experimentation. It is foreseen that integrating the complex
via finite element analysis (FEA) (Figure 3b). These thermody- AM-related thermo physical and thermo chemical phenomena
namic and kinetic computations effectively described the solidi- in a single data repository will enhance the accuracy and appli-
fication events of commercial alloys used in AM, such as Inconel cability of data-driven methods, aiding the design of alloys suit-
625 and Ti 6Al 4V, by closely predicting those found experi- able for the AM process.
mentally.[90–93] In the case of Inconel 625, a correct cooling rate
was predicted to result in the precipitation of carbides and Laves
phases in the dendritically segregated material, aiding the iden- 3. Composite Material Design for AM
tification of postprocessing conditions that will limit the growth
of precipitates, which affect the strength of the alloy.[90] In the Modern structural applications strongly demand materials with
case of Ti–6Al–4V, the simulated diffusion-controlled phase high strength and excellent toughness. However, the mutual
transformations were tracked to predict the final α-phase compo- exclusivity of these properties requires undesirable compromises
sition after AM processing.[92] Nevertheless, CALPHAD model- when designing structural components with a single homoge-
ing, DICTRA, and FEA simulations are excellent tools for neous material. Composite materials allow material scientists
populating the composition structure map of AM-specific mate- to pick multiple sets of mechanical properties, such as the mod-
rials, but their computational cost can still be a burden when the ulus, strength, ductility, and toughness, to achieve a target prop-
design space is too large to exhaustively search for optimal erty that normally does not exist in any single material. Major
solutions. approaches in designing composites include taking inspiration
The power of ML algorithms lies in their capability to from nature or simple stacking of disparate materials.[96–98] By
make informed predictions reflecting the proces- harnessing AM and, more specifically, multimaterial printing,
sing microstructure property relationship when exhaustive material scientists have greater freedom in determining the spa-
search of the design space is infeasible (Figure 3d). For instance, tial arrangement of the constituents, rather than being limited to
ML may be a solution to the costly combinatorial experiments arranging the constituents of composite material in simple lami-
and/or computationally expensive simulations required to screen nate structures or particle-/fiber-reinforcing matrix structures. A
the vast space of potential metallic alloys.[23] Without explicitly brief overview of current progress and challenges on AM of mul-
modeling each composition, an adaptive search algorithm like timaterial structures can be found in another review article.[99]

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (6 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

With increased design freedom, the vast design space of choos- Afterward, GA was used to conduct multiattribute optimization
ing material combinations and how they are arranged in a struc- to design composites with desired properties. In addition, a two-
ture can be difficult to explore without sufficient data that usually step process of feature extraction and ensemble-based regression
require expensive experimentation or extensive computational captured the elastic localization relationships in high-contrast
simulations. This difficulty can be addressed using composites (i.e., composites with large ratios between the elastic
data-driven methods to accelerate the discovery of moduli of the constituents).[101] The high-contrast composite was
material arrangement property mapping using relatively lim- represented in 3D voxels and the damage-inducing localized
ited information and expand the learned mapping to guide strain was captured in ML-aided prediction of the elastic
design decisions in this vast design space. As the material selec- response of a single voxel of interest within the whole micro-
tion (Figure 4a) and materials’ arrangement (Figure 4b,c) have structure. The relationship between this focal voxel and its neigh-
intertwined effects on the properties of the composite material bors was represented numerically and used as features in this
design (Figure 4d), this section of our Review will focus on study. Selected high-impact features were regressed with ran-
data-driven discovery of 1) different constituent materials and dom forests to achieve a good balance between prediction accu-
2) different constituent arrangements and their effects on com- racy and time consumption. This study showed the fundamental
posite properties. These two areas illustrate the utility of viability of using data-driven tools to make computationally effi-
data-driven methods in all aspects of composite design for AM. cient predictions in composition property linkage and pre-
sented strategies to improve prediction accuracy. Similarly, a
3.1. Material Selection for Composites 3D convolutional neural network (CNN) was used as a
data-driven tool to establish connections between the salient fea-
The emergence of easily accessible data-driven methods reduces tures of 3D microstructures and the effective material properties
the need for prohibitively large amounts of data in exploring the of interest.[102] Following this study, property prediction on
broad design space offered by AM. For example, ML was imple- high-contrast composites was also achieved using a deep learn-
mented to design an alumina-reinforced aluminum matrix com- ing architecture adopted from traditional CNN models.[103]
posite with improved tribomechanical properties.[100] First, an Homogenization linkages were established between the macro-
ANN model was trained with data extracted from the literature scale effective stiffness and the extracted higher-order neighbor-
to learn the mapping between composite properties and the com- hood information from the 3D microstructures. This
position, size, and morphology of reinforcing particles. homogenization linkage allowed the arrangement property

Figure 4. Compared with conventional methods in manufacturing, AM excels in realizing advanced composite designs due to the expanded design space
from: a) an expanded selection of materials for composite constituents; b) achieving conventionally infeasible constituent arrangements; and
c) introduction of structural hierarchy across length scales. Data-driven methods accelerate exploration of this vast design space through
d) efficient discovery of the mapping between composite design and performance.

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (7 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

linkage to be directly trained on the input 3D arrangement of clearly exemplified by the aerospace industry and is expected
constituent materials, thus eliminating the need for feature engi- to be seen in future AM composite databases. While developing
neering, and improving learning capability. The overall approach a standardized database will strongly supplement AM product
in this study was tested with generalized constituents having certification, the open-access nature of a common database
elastic modulus ratio of 50 to obtain high-contrast composites increases the availability of data to all stakeholders and racks
and therefore can be extended to study constituent pairs with up sizable savings when comparing with the cost of establishing
other elastic modulus ratios. Similar studies on the generalized localized databases separately.
purely elastic constituent identified optimal designs for material
properties such as the ratio of modulus, toughness, and volume
3.2. Optimization of Constituent Arrangement in Composites
ratio.[104] Compared with elastic behavior of materials, the predic-
tion of material properties with plasticity requires more complex
The bottom-up nature of AM processes inherently confers the
computations due to its nonlinearity and could benefit more
freedom to arrange constituents in ways that are different from
from data-driven methods. These benefits are demonstrated in
the simple stacking of layers or particle-/fiber-reinforced matrix
a pioneer study, in which the data-driven framework is used
arrangements. Taking advantage of this geometrical freedom of
to predict the constitutive model of the 2D hyperelastic compos-
AM is integral to achieving further breakthroughs in composite
ite with homogenization procedures and optimize the design of
3D inelastic composite material for toughness without costly design. Algorithm-driven optimizations of the constituent
numerical simulations.[105] Data-driven frameworks with further arrangements can help to unlock the full potential of geometrical
development are shown to be capable of predicting stress strain freedom in AM. A pioneering study in this area iteratively arrived
behavior beyond the elastic limit of the material with reasonable at near-optimized arrangements of constituents in a composite
computational cost.[106,107] A similar approach of predicting system from randomly generated starting points in a very broad
load-displacement curves was used for cotton fiber/PVC compo- design space.[114] The choice of the starting point could poten-
sites, confirming the applicability of this technique for any kind tially affect the optimized solution undesirably as an inadequate
of composite systems.[108] Other than mechanical properties, starting point that might trap the optimization at a local extre-
data-driven tools could predict thermal conductivities of compos- mum. This potential inability to explore the whole design space
ite materials with better accuracy than traditional analytical was addressed using data-driven methods that identified the opti-
models.[109] mal arrangements of the constituents for a composite system to
While several of the aforementioned data-driven methods are achieve high toughness and strength.[115] Although the ML
based on data acquired from physics-based simulations, simpli- model used in this study was trained with data from less than
fications made in these simulations for reducing computational 1% of the whole design space, it achieved high accuracy (around
costs cannot ignore the important problems that are specific to 90%) in predicting the resulting composite properties from geo-
multimaterial printing, such as the interfacial strength between metrical arrangements of constituents. Moreover, the data used
constituents. A proper strategy is needed for bonding different for ML model training were binary representations of the com-
constituents in a multimaterial print, especially functional and posite design properties (i.e., either good or bad). Without any
structural parts.[110] Ignoring the interfaces between constituents information on the overall ranking, the ML model was able to
will not only cause a potential performance decay in composite accurately predict rankings of all data after training and therefore
AM materials when compared with that of their individual con- produce the best-performing arrangements. With excellent toler-
stituents, but there may also be a potential deterioration over the ance to lower order data, data-driven methods are particularly
material’s lifetime due to noncompatible thermal expansion and useful for extracting meaningful guidance in composite design
cooling rates.[99,110] Proper characterization of interfacial bond- when higher-order data are unavailable or the uncertainty in
ing strengths is essential for successfully selecting constituents experiments compromises the quality of the data.
for AM composite materials.[111,112] Another key aspect of optimizing constituent arrangement in
Experimental validations and computational simulations can composite material is the structural hierarchy spanning different
provide greater value if these data can be stored in commonly length scales. Structural hierarchy in natural composite materials
accessible databases for different future studies, while avoiding positively influenced the macroscale property of the compos-
duplicated research efforts in the field. While the development of ite.[116] This hierarchical complexity is difficult to mimic synthet-
such databases is still in its infancy, it is important to emphasize ically due to the limitations of conventional manufacturing
the standardization needed to increase the broader acceptance of methods.
any composite material databases. By agreeing upon common Conventional methods are not capable of 3D spatial manipu-
materials testing and qualification standards for each application, lation of the material’s internal structures. Among established
a group of small aircraft companies greatly reduced the cost of methods, casting molten materials into ingots or bulk compo-
material certification, while simultaneously decreasing the devel- nents allows addition of reinforcing material into the composite,
opment time of a new material with needed properties.[113] The but the reinforcement distribution has to be uniform. In addition
reduced effort for certification is built on the basis of material to that, sheet lamination method introduces additional freedom
equivalence, that is, all users manufacturing the same material to fine-tune the intra- and interlayer parameters, but the design
under the exact conditions recorded in the database should pro- flexibility is still limited and insufficient for production of hier-
duce the same material property. AM composites are yet to archical materials. In contrast, the geometrical freedom of AM
achieve such a high level of consistency in the final product, enables the manufacturing of such hierarchy in composite mate-
but the benefit of a widely accepted, standardized database is rial designs. Guiding the design of hierarchical composites

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (8 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

usually requires a mapping between the constituent arrangement and obtain better prediction of reliability of aerospace composite
and the resulting composite property. Obtaining this mapping by structures, proving the improved relevance of complete data.[123]
solving a physics-based model typically requires extensive Furthermore, AM as a digital manufacturing method is also ben-
computational power, but data-driven tools can help obtain this eficial to streamlining data exchange between users and database.
mapping efficiently. This capability was demonstrated in an Despite the additional spatial freedom in composite material
experimentally validated identification of optimally arranged con- design brought by AM, once a complete and effective categoriza-
stituents of a hierarchical polymeric composite material.[117] A tion scheme is put in place, it can be expected that a centralized
CNN model was trained to learn the mapping between the hier- database storing valuable experimental and simulation data can
archical arrangements of the constituents and the toughness of be established to greatly facilitate the use of data-driven methods
the resulting composite design, thereby predicting the toughness in designing AM composites.
of composite designs with arbitrary constituent arrangements.
Furthermore, three unit cells were designed with the same pair
of polymeric constituent materials that had different mechanical 4. Structure Design for AM
properties to obtain varying overall in-plane stiffness properties.
By optimizing the arrangement of these three unit cells in a 4.1. Topology Optimization and Generative Design
lattice-like composite system, the CNN model reduced the order
of the design problem and discovered an optimal solution for a Topology Optimization (TO) and Generative Design (GD) are
composite system with multiple levels of hierarchy. ML-aided two potential design tools for developing novel architectures with
optimization of repetitive lattice composite materials can also desirable structural and functional properties. TO mathemati-
automatically generate structures of microscale representative cally optimizes the material layout in a given design domain
volume elements for desired macroscale elastic moduli.[118] based on a predefined set of loads and boundary conditions
While data-driven methods significantly advanced the AM of (mechanical, thermal, or both).[124] GD is an iterative design pro-
structurally complex polymer composites, this success is not rep- cess that generates multiple outputs which meet the predefined
licated in metal composites due to the challenge of preventing constraints.[125,126] Although both methods appear to be similar
the constituent materials from reacting with each other under and the nomenclatures are often used interchangeably, there are
high temperatures. Nonetheless, the success of AI-aided design subtle differences between the two methods. TO is a founda-
in AM-enabled polymer composites can still inspire the design of tional technology that aims to reduce the material of an existing
AM-enabled metal composite, as shown in the optimization of object, while retaining its functional ability, whereas GD is a
the constituents’ staggered arrangements with generalized mate- newer concept that builds on TO to create “imaginary designs”
rial properties that are applicable to both polymers and met- that satisfy the given boundary conditions. To put it simply, one
als.[119] Recent developments in particle-reinforced and could say TO is subtractive and GD is additive in nature. In both
functionally graded metallic composites are also very promising methods, the optimized outputs are often complex structures
for improving AM metallic composite materials.[120–122] that are difficult to manufacture using subtractive methods.
Data-driven methods can aid in improving the mechanical prop- The geometric freedom offered by AM increased the use of these
erties of metallic composites once AM is further enhanced to iterative tools in developing novel structures with superior per-
solve the problem of chemical reactions between the matrix formance. Based on mechanical, thermal, and fluidic boundary
and the reinforcements. conditions, these tools were used in designing novel structures
While AM-specific composite material is still an emerging that were manufactured subsequently.[127–129]
field, the digital nature of AM calls for strong data infrastructures The next step in improving and expanding the output of these
for more users to take advantage of the evolving data-driven tools. design tools is implementing data-driven approaches in creating
Composite material databases, due to composites’ heterogeneity outputs. Conventionally, iterative design tools such as TO and
and added complexity, have a shifted focus to the relationship GD generate outputs after a series of FEA of the structures,
between microscale constituent properties and macroscale com- which is very demanding computationally.[130] The goal of using
posite response. Using fiber-reinforced composite material as an data-driven approaches in combination with these tools is to
example, the added complexity compared with homogeneous potentially replace the demanding FEA aspects that come with
material includes volume fraction, orientation, and spatial distri- TO and GD. The most popular method of doing so is to apply
bution of reinforcing fiber, matrix material property, interfacial CNN to existing designs and determine an optimized structure
bonding characteristic, etc. This added complexity means greater for the given boundary conditions.[131,132] This process involves
variability in the data, posing challenges on relevance and accep- several steps. The first step is to create a dataset of previously
tance of data stored in databases. To increase the relevance of designed structures. The program can evaluate the performance
data, the completeness of the data must be emphasized. of new structures after this dataset trains the program on the per-
When variability in composite material property is not coupled formance of each existing design. The second step is to generate
with the data stored in database, users of such databases (i.e., optimized structures based on benchmark designs, together with
material scientist and designers) have to make much more con- TO. Subsequently, a similarity criterion is used to filter similar
servative design allowable based on the extracted data. It is shown designs from the database. This is an iterative process where the
that complete data on mechanical properties of a notched carbon goal is to get the number of similar designs below a certain
fiber laminate composite with statistical variance that is not threshold. If this threshold is not achieved, the CNN is used
included in typically reported average values for properties in to generate new structures based on the patterns learnt from
handbook-based database allow engineers to use statistical tool the benchmark designs and TO-generated structures. Step 2

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (9 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

and 3 is repeated iteratively to filter designs until the desired are categorized according to AM’s benefits, including functional
threshold is achieved. Once this iterative process is complete, improvement, and part consolidation requirement for assem-
the next step involves evaluating the design options generated blies, all features are generalized to be applicable to different
using a built-in loss function that is developed using the training products manufactured by various processes. To maximize the
dataset. The final step is a visual display of all the design options potential of AM, completely new redesigns are often required
for the designer to choose from that fit of the given application. rather than incrementally improving from designs made for sub-
Figure 5 shows the framework for one such model.[133] tractive manufacturing methods. However, a completely new
Well-designed data-driven models that aid structural design redesign usually requires designers to find optimal solutions
optimizations can possess good predictability with high accura- in a much broader design space. Data-driven methods like TO
cies and effectively reduce the computational costs of traditional and GD that are coupled with illustrative features stored in gen-
optimization techniques. For example, one such CNN model eralized databases for structure design will greatly accelerate the
generated topologically optimized 2D structures for a given set exploration, promoting more innovative structure designs for
of boundary conditions that were nearly identical to the geome- AM over simply improving existing designs.
tries produced by FEA-based models.[134,135] Another study used
a CNN-based model to generate optimized structures for auto- 4.2. Design with AM Constraints
mobile wheels.[133] Furthermore, data-driven TO models could
also generate multiclass microstructures (i.e., structures that AM-specific manufacturing constraints cannot be overlooked.
have more than one type of unit cell in their microstructure) Designing with AM constraints in mind is a major component
which have important applications in the biomedical and aero- of the DfAM revolution and it is essential to take advantage of the
space industry as they can be designed to have specific properties possibilities offered by AM. For detailed review on all aspects of
that cannot be found naturally.[136] DfAM, we direct the reader’s attention to this thorough
Integration of data-driven approaches with design tools like Review.[138] Here, we focus on managing supports, overhangs,
TO and GD allows designers to efficiently generate novel com- and minimum feature size in structure design.
plex structures with minimal computational cost. A crucial aspect Support structures are necessary in AM for long overhang fea-
is to construct a thorough reference dataset that can generate out- tures at a shallow angle. They not only increase material waste,
put designs similar to those generated by FE-based models. but also increase postprocessing time for support removal, espe-
Unlike material databases that contain data universal to almost cially for metallic AM products. To remedy the impact of reduced
all users, data types like geometry or structure designs make sys- manufacturability on complex AM structure designs, an iterative
tematic classification tremendously difficult due to the case-by- process is proposed to slightly modify the theoretically optimal
case nature of different design criteria and therefore needs to topology and avoid introducing support structures.[139] The
be generalized. A specific geometry design may not be useful advantage of this method is its universal applicability as it is a
for any other user of the database, whereas a generalized data digital postprocess after obtaining the optimal structure design
entry can be adapted for different design applications. This gen- from data-driven frameworks. This iterative process also works
eralization usually entails feature extraction from a geometry, for all build orientations of any 2D shape and therefore promotes
rather than the whole geometry or structure design. One example further optimization between structure design and proposed
of such a generalized database is demonstrated to be inspira- build orientation.[139] TO frameworks with integrated overhang
tional and relevant to the Design for AM (DfAM) process.[137] constraints can also successfully create self-supporting structures
While the design features stored in the aforementioned database that are validated experimentally in both polymer and metallic

Figure 5. Framework for the data-driven approach to structure design. Adapted with permission.[133] 2021, American Society of Mechanical Engineers (ASME).

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (10 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

AM products.[140,141] Data-driven methods can be incorporated


into the constraint function to achieve more complex optimiza-
tion output when compared with TO frameworks using fixed
overhang angles and feature size constraints.[142] With limited
experimental data, the dynamic data-driven design rule can be
created for determining the minimum feature sizes using over-
hang angles. This method allows different minimum features for
varying overhang angles, as well as incorporating horizontal
“bridging” features that connect two bodies in close proximity
to each other.[142]
Apart from topologically optimized structures specifically
designed for AM, data-driven methods can also facilitate
improvement of existing component designs to harness the ben-
efits enabled by AM. One example of such improvement is using
lattice structures. When using lattice unit cells from a predefined
unit-cell library, the large-scale multivariable problem to recreate
a completely new optimized structure is reduced to only two
design variables: unit-cell type and size.[143] This is vital for
decreasing the computational costs in data-driven structure opti-
mizations and it is only possible with an existing unit-cell library.
This highlights the need for a more extensive database of unit-
Figure 6. Various current challenges in AM: a) Support reduction; b) tool
cell designs and properties. When utilizing lattice structures in path-induced void in layers; c) scan strategy-induced material anisotropy;
AM structure design, an important constraint is that different d) nonuniform thermal history within and across prints; and e) thermal-
AM methods have varying print resolutions. Thus, the minimum induced residual stress and resulting print warping.
feature size constraint must be incorporated into the lattice struc-
ture optimization for AM to ensure manufacturability.[144] In
addition, when process imperfection causes optimal lattice sections detail recent developments in using data-driven tools to
design infeasible to manufacture, data-driven methods such as aid active alteration of AM deposition strategies in the prepro-
Bayesian ML is shown to be a robust design framework to adapt cessing step at the very beginning of the AM process hierarchy
to added manufacturing constraints and provide valuable before irreversible fabrication takes place.[147] In both polymer
insights in lattice structures that are commonly used in AM and metal AM processes, these active measures will improve
metamaterial design.[145] the quality of the prints by achieving full density, increasing
dimensional accuracy, reducing support material, and reducing
material anisotropy.
5. Tool Path and Process Parameter Design for
AM
5.1. Tool Path Planning to Achieve Full Density
The process of introducing, and subsequently removing support
materials, lowers the efficiency of AM in industrial applications. To achieve fully dense structures without unintentional voids,
In addition, in fusion-based AM, the materials undergo extremely several process parameters should be optimized in tool path
rapid heating and cooling which lead to nonequilibrium phase planning. These parameters include horizontal spacing between
transformation during deposition (e.g., DED, WAAM) or join- line scans (hatch spacing), layer height, and tool path adjust-
ing/sintering (e.g., PBF). As a result, several challenges arise, ments to prevent material void or pile-up at corners and intersec-
including thermally induced defects and material anisotropy in tions of line scans within a layer. A previous study showed that in
direction of the thermal gradient. Figure 6 shows the current chal- FDM, the relation between the occurrence of voids induced by
lenges in AM, namely, support reduction, voids elimination, reduc- insufficient material and hatch spacing in each layer could be
ing material anisotropy, as well as reducing thermal nonuniformity modeled with deep neural networks without requiring computa-
and its resulting problems, such as residual stress development. tionally heavy optimization of a multiattribute problem.[148] If the
Some of these challenges can be minimized by optimizing the tool geometry of the cross section and the uniformity of the deposited
path with either computational simulation or trial-and-error experi- line is known, the tool path can be optimized to eliminate voids
mentation, but these methods are resource intensive and data- between lines of deposited material. This knowledge depends on
driven tools are promising to reduce this cost. understanding the dynamics of the deposition process, as well as
The workflow of AM consists of three major steps: preprocess- the interactions between the deposited materials and the print
ing, processing, and postprocessing.[146] Data-driven frameworks surfaces or prior layers. Data-driven models can efficiently obtain
can establish the correlation between user input process param- such information. In an example of inkjet printing, such a model
eters and product quality without solving multiphysics models serves as a powerful predictor of the ejected droplets’ dynamics
that are computationally demanding. They can also guide users after deposition and its interactions with substrate, while
in adjusting the corresponding process parameters in the prepro- requiring minimal human interventions.[149] In this study, unsu-
cessing stage, such as laser power and tool path. The following pervised deep learning was used to examine droplet evolution

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (11 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

from video data of the droplet ejection. Video data, as opposed to strategies, and volume uncertainties should be considered. For
static image data, could be used due to the lower computational example, the optimal or near-optimal build orientation, deposi-
resource requirement and unsupervised nature of the method, tion direction, and slicing solution for a flat build platform were
which is a strong advantage as image data does not include tem- determined effectively using GA.[146] The GA-based optimization
poral domain information. This enhanced prediction of droplet was conducted over a vast range of orientations, while consider-
dynamics is vital for estimating the line track geometry as well as ing factors such as surface accuracy, number of turns, and
planning for adequate hatch spacing. The benefits of requiring start stops in the tool path. Alternatively, the optimal tool path
minimal human input were also examined in a recent study to reduce build time for a part with a predefined orientation was
using ANN in an adaptive void-filling strategy for arbitrary- obtained from multiattribute optimization with evolutionary
junction geometries in metal AM of rib-web structures.[150] An algorithms.[158] However, having the optimal tool path for
optimal training set of merely 46 different junctions joining reduced support material and build time does not guarantee a
three or four edges with various angles in between was created perfect build, due to inevitable uncertainties in AM processes.
to train the neural network. While such uncertainties cannot be effectively reduced yet,
In addition to designing the tool path trajectory, the print uncertainties in material deposition and solidification were pre-
parameters along the tool path must be optimized to avoid defects dicted using CNNs when specific process parameters and tool
such as keyholes or lack of fusion. High-fidelity models in differ- paths were provided.[159]
ent metal AM processes improved the fundamental understand-
ing of porosity defects, temperature distribution in melt pools, 5.4. Tool Path Planning to Address Anisotropy, Residual Stress,
and evolution across layers. However, it is infeasible to use such and Other Challenges
models to study a full component at the same resolution due to
the significant computational effort involved.[151,152] To facilitate In addition to all factors presented earlier, the crystalline nature
more efficient identifications of print parameters, a GP-based sur- of most materials used for metal AM, coupled with directional
rogate model was constructed and trained with only 139 datasets heat transfer during processing, may introduce thermally
to predict melt pool properties, resulting in low standard deviation induced anisotropy in printed parts.[160] The impact of print
from limited print parameters as inputs.[153] The experimental parameters on the anisotropy of the parts can be modeled and
cost of finding optimal print parameters along a tool path can should be considered for tool path planning in AM.[161] Cyclic
be further reduced by adequately combining data mining from temperature profile with high peak temperatures also causes
simple experiments or literature and using ML algorithms.[154,155] thermally induced residual stress in parts produced with
AM.[162] Local alterations to the tool path has been used to reduce
5.2. Tool Path Planning to Reduce Support Structure the overall peak temperature experienced by the parts during
printing, thus reducing the accumulation of residual stresses.[163]
Greater ease of producing geometrically complex components is a An unsupervised ML model established the strategy for identify-
major advantage that sets AM apart from traditional manufactur- ing areas that require adaptations in tool path for peak tempera-
ing methods. However, overhang structures are sometimes inev- ture reduction.[164] This model-clustered parts of the tool path
itable in complex components. While most AM processes require according to their thermal history and thus this model are appli-
adding support structures to produce overhang structures, the cable for comparing process regimes within single components
increase in build time and material waste as well as difficulties and across different components. While the tool path depen-
in removing the support material are significant problems that dency of AM can be minimized by data-driven methods, the dif-
need to be addressed. This problem can be remedied considerably ferences in material properties of the final product could still
by changing the orientation of the build-platform and adjusting exist. The inevitable process uncertainty is best minimized when
print parameters. Printing of free-form, thin shell parts without process-dependent property changes, such as distortional hard-
using support structures was achieved recently using a dynami- ening in composite materials, can be predicted with data-driven
cally reorienting build-platform.[156] An ANN-based context- methods.[165] However, the increasing number of factors to be
dependent compensation scheme was used to apply corrections considered for every point in the tool path as well as the increas-
to complex tool paths. Another strategy is changing print param- ing number of points in the tool path to model large-scale parts
eters adaptively, which is more applicable especially when large- are dramatically increasing the complexity of tool path planning
scale parts limit the capability to dynamically reorient the for metal AM. By reducing the order of the model as well as
build-platform. With less than 50 irregularly spaced data points establishing relationships between print parameters and print
from characterization tests, a biharmonic spline interpolation outcomes when numerical methods do not exist, ML helps users
model was used to adaptively alter print parameters along part make decisions in tool path planning more accurately and
of the tool path and achieved support-free, near-net-shape product efficiently.[166]
on flat build-platforms using metal wire-arc AM (WAAM).[157]
5.5. Tool Path Design Database Development to Increase AM
5.3. Tool Path Planning to Reduce Print Time and Improve Accessibility
Dimensional Accuracy
Tool path design is a complex subject with multiple parameters
For designing tool path trajectories that reduce build time while that are intertwined with each other, affecting the final quality of
achieving high-dimensional accuracy, the orientations, slicing AM prints. Its complex nature and high cost of trial and errors

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (12 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

demand data-driven methods for efficient optimization and the expanded design space of composite material enabled by
decision-making. Establishing a database for tool path designs AM, thus approaching the full potential of geometrical and mate-
is fundamental for incorporating data-driven methods and it rial freedom of AM. In addition to data-driven identification of
avoids unnecessary duplicated experimental validations for the optimal materials for AM, data-driven TO and GD further take
same AM methods. Different manufacturing processes have dif- advantage of the unique capability of AM to create complex struc-
ferent pros and cons, limiting their capabilities to realize some tures. These additional design freedoms in AM have proven to be
designs. For example, the resolution of a typical DED system is of great interest, and ML is demonstrated to be necessary for tool
worse than a typical PBF system, limiting its application in pro- path optimization to achieve efficient and highly accurate produc-
ducing parts that have very intricate features. In contrast, the tion of AM-enabled designs. The challenges of solving computa-
absence of a powder bed means increased versatility and scalabil- tionally expensive multiphysics models are addressed by
ity for DED systems, making them better candidates for building implementing data-driven tools in establishing parameter
large-scale components and repairing existing components when property mapping, and examples of optimizing tool path to
compared with PBF systems. Certain process parameters are achieve full density, dimensional accuracy, and thermal stability
only applicable to a certain type of process, like powder feed rate of print in situ are discussed.
in DED system, whereas the operating windows for certain Data-driven approaches are demonstrated to drive AM toward
parameters are dramatically different across processes, such as a smarter manufacturing method. However, data-driven meth-
the laser/electron beam scan speed in DED or PBF systems. ods are not infallible tools that handle every problem without lim-
These variabilities described earlier necessitate a classification itations and there is still significant room for improvements. For
or organization scheme for the database that is applicable to example, connections between different aspects of DfAM will
all different processes. One study presents a viable solution to need to be established in future works of data-driven AM to for-
address this need, using NASA-developed materials informatics mulate the design optimization problem in a global framework
system Materials and Processing Technical Information System where materials, structure, tool path, and process control are con-
(MAPTIS) as an example.[15] This system allows the user to filter sidered simultaneously. This article aims to showcase the
data using material type, AM processes, build parameters, and so state-of-the-art development of data-driven AM and inspire
on. The connectivity of data is achieved by their classification future work toward improved adaptation of data-driven methods
tags. For example, when the user aims to find optimal parame- in AM.
ters for printing Inconel 718, the build parameters for DED sys-
tems will differ from those of PBF systems, and the database
therefore requires classification tags to guarantee a targeted Acknowledgements
response to the users’ query.
As long as the important criteria for tool path design databases J.B gratefully acknowledges funding received by the 2020-2021 Knight @
KIC Engineering Graduate Fellowship.
discussed earlier are met, the opportunities of such databases are
invaluable to the growing community of AM users. Despite the
benefits, a mature database accessible to most companies and
researchers is yet to be established, hindered by International Conflict of Interest
Treaty of Arms Regulations (ITAR) restrictions and concerns The authors declare no conflict of interest.
about proprietary data. [15] We still need continued effort to
encourage incentives toward the establishment of a more acces-
sible AM structure and tool path design database, that not only
facilitates the increased use of data-driven methods but also Keywords
accelerates the development of AM field as a whole. additive manufacturing, artificial intelligence, machine learning, material
designs, tool paths, topology optimizations

6. Conclusions and Future Perspectives Received: January 21, 2021


Revised: June 27, 2021
Recent progress in implementing data-driven methods in Published online: July 23, 2021
designing for AM is discussed from different perspectives in this
Review. Starting from designing materials specifically optimized
for AM, we show examples of improvements achieved both in [1] T. Debroy, H. L. Wei, J. S. Zuback, T. Mukherjee, J. W. Elmer,
accuracy and in computational speed of physics-based models J. O. Milewski, A. M. Beese, A. Wilson-heid, A. De, W. Zhang,
Prog. Mater. Sci. 2018, 92, 112.
such as DFT and MD using data-driven methods. Coupled with
[2] J. Y. Lee, J. An, C. K. Chua, Appl. Mater. Today 2017, 7, 120.
the existing database of large quantum mechanical datasets, data-
[3] T. D. Ngo, A. Kashani, G. Imbalzano, K. T. Q. Nguyen, D. Hui,
driven methods enable more accurate prediction of the material’s
Compos. Part B Eng. 2018, 143, 172.
property from its composition and structure at the atomic level, [4] W. E. Frazier, J. Mater. Eng. Perform. 2014, 23, 1917.
as well as improve the speed of screening, accelerating the iden- [5] J. Liu, A. T. Gaynor, S. Chen, Z. Kang, K. Suresh, A. Takezawa, L. Li,
tification of optimal material suited for AM. On a larger scale, J. Kato, J. Tang, C. C. L. Wang, L. Cheng, X. Liang, A. C. To, Struct.
examples of data-driven methods efficiently establishing compo- Multidiscip. Optim. 2018, 57, 2457.
sition structure property mapping are presented. These [6] J. Schmelzle, E. V. Kline, C. J. Dickman, E. W. Reutzel, G. Jones,
high-throughput mappings bring us closer to full discovery of T. W. Simpson, J. Mech. Des. Trans. ASME 2015, 137, 1.

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (13 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

[7] G. H. Loh, E. Pei, D. Harrison, M. D. Monzón, Addit. Manuf. 2018, [42] E. B. Curry, S. Sahoo, C. Herrera, I. Sochnikov, S. Pamir Alpay,
23, 34. R. J. Hebert, B. G. Willis, J. Qi, J. N. Hancock, J. Appl. Phys. 2020,
[8] S. H. Huang, P. Liu, A. Mokasdar, L. Hou, Int. J. Adv. Manuf. Technol. 127, 245111.
2013, 67, 1191. [43] S. Curtarolo, G. L. W. Hart, M. B. Nardelli, N. Mingo, S. Sanvito,
[9] S. H. Khajavi, J. Partanen, J. Holmström, Comput. Ind. 2014, 65, 50. O. Levy, Nat. Mater. 2013, 12, 191.
[10] S. Bhattacharya, J. Bustillos, F. Gonzalez, J. Spector, A. Moridi, Am. J. [44] J. E. Saal, S. Kirklin, M. Aykol, B. Meredig, C. Wolverton, JOM 2013,
Biomed. Sci. Res. 2020, 7, 6. 65, 1501.
[11] A. Agrawal, A. Choudhary, APL Mater. 2016, 4, 053208. [45] J. E. Saal, C. Wolverton, Acta Mater. 2014, 68, 325.
[12] R. Jose, S. Ramakrishna, Appl. Mater. Today 2018, 10, 127. [46] S. Kirklin, J. E. Saal, V. I. Hegde, C. Wolverton, Acta Mater. 2016, 102,
[13] L. Himanen, A. Geurts, A. S. Foster, P. Rinke, Adv. Sci. 2019, 6, 125.
1900808. [47] K. Kim, L. Ward, J. He, A. Krishna, A. Agrawal, C. Wolverton, Phys. Rev.
[14] G. R. Schleder, A. C. M. Padilha, C. M. Acosta, M. Costa, A. Fazzio, J. Mater. 2018, 2, 1.
Phys. Mater. 2019, 2, 032001. [48] A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek,
[15] T. Prater, Prog. Addit. Manuf. 2017, 2, 11. S. Cholia, D. Gunter, D. Skinner, G. Ceder, K. A. Persson, APL
[16] H. Ko, P. Witherell, N. Y. Ndiaye, Y. Lu, IEEE Int. Conf. Autom. Sci. Mater. 2013, 1, 011002.
Eng. 2019, 2019, 648. [49] S. Curtarolo, W. Setyawan, S. Wang, J. Xue, K. Yang, R. H. Taylor,
[17] S. Kirklin, J. E. Saal, B. Meredig, A. Thompson, J. W. Doak, M. Aykol, L. J. Nelson, G. L. W. Hart, S. Sanvito, M. Buongiorno-Nardelli,
S. Rühl, C. Wolverton, npj Comput. Mater. 2015, 1, 15010. N. Mingo, O. Levy, Comput. Mater. Sci. 2012, 58, 227.
[18] A. van de Walle, C. Nataraj, Z. K. Liu, CALPHAD: Comput. Coupling [50] D. D. Landis, J. S. Hummelshøj, S. Nestorov, J. Greeley, M. Dulłak,
Phase Diagrams Thermochem. 2018, 61, 173. T. Bligaard, J. K. Nørskov, K. W. Jacobsen, Comput. Sci. Eng. 2012, 14,
[19] Z. Jin, Z. Zhang, K. Demir, G. X. Gu, Matter 2020, 3, 1541. 51.
[20] C. Zhai, T. Li, H. Shi, J. Yeo, J. Mater. Chem. B 2020, 8, 6562. [51] C. Draxl, M. Scheffler, J. Phys. Mater. 2019, 2, 036001.
[21] N. S. Johnson, P. S. Vulimiri, A. C. To, X. Zhang, C. A. Brice, [52] K. T. Butler, D. W. Davies, H. Cartwright, O. Isayev, A. Walsh, Nature
B. B. Kappes, A. P. Stebner, Addit. Manuf. 2020, 36, 101641. 2018, 559, 547.
[22] T. M. Pollock, A. Van der Ven, MRS Bull. 2019, 44, 238. [53] J. Li, Q. Fang, B. Liu, Y. Liu, Acta Mater. 2018, 147, 35.
[23] C. Wang, X. P. Tan, S. B. Tor, C. S. Lim, Addit. Manuf. 2020, 36, [54] T. Q. Vo, B. H. Kim, Int. J. Precis. Eng. Manuf. – Green Technol. 2017, 4,
101538. 301.
[24] T. M. Pollock, A. J. Clarke, S. S. Babu, Metall. Mater. Trans. A. Phys. [55] F. Rahmani, J. Jeon, S. Jiang, S. Nouranian, J. Nanoparticle Res. 2018,
Metall. Mater. Sci. 2020, 51, 6000. 20, 133.
[25] R. Liu, A. Kumar, Z. Chen, A. Agrawal, V. Sundararaghavan, [56] S. Aphinyan, K. R. Geethalakshmi, J. Yeo, A. Shakouri, T. Y. Ng, Polym.
A. Choudhary, Sci. Rep. 2015, 5, 11551. Adv. Technol. 2017, 28, 1057.
[26] J. Wei, X. Chu, X. Sun, K. Xu, H. Deng, J. Chen, Z. Wei, M. Lei, InfoMat [57] S. Aphinyan, E. Y. M. Ang, J. Yeo, T. Y. Ng, K. R. Geethalakshmi,
2019, 1, 338. J. Polym. Sci., Part B: Polym. Phys. 2018, 56, 1615.
[27] G. X. Gu, C. T. Chen, D. J. Richmond, M. J. Buehler, Mater. Horizons [58] S. Aphinyan, E. Y. M. Ang, J. Yeo, T. Y. Ng, R. Lin, Z. Liu,
2018, 5, 939. K. R. Geethalakshmi, Model. Simul. Mater. Sci. Eng. 2019, 27, 055005.
[28] W. Koch, M. C. Holthausen, A Chemist’s Guide to Density Functional [59] Z. Wang, H. Liang, A. V. Dobrynin, Macromolecules 2017, 50, 7794.
Theory, John Wiley & Sons, Weinheim 2015. [60] M. Hellenbrandt, Crystallogr. Rev. 2004, 10, 17.
[29] P. Yu, Y. Zhuang, J. P. Chou, J. Wei, Y. C. Lo, A. Hu, Sci. Rep. 2019, 9, [61] D. Zagorac, H. Muller, S. Ruehl, J. Zagorac, S. Rehme, J. Appl.
1. Crystallogr. 2019, 52, 918.
[30] S. K. Nayak, C. J. Hung, V. Sharma, S. P. Alpay, A. M. Dongare, [62] J. Behler, J. Chem. Phys. 2016, 145, 170901.
W. J. Brindley, R. J. Hebert, npj Comput. Mater. 2018, 4, 11. [63] J. Behler, Angew. Chem., Int. Ed. 2017, 56, 12828.
[31] T. T. Lau, C. J. Först, X. Lin, J. D. Gale, S. Yip, K. J. V. Vliet, Phys. Rev. [64] J. Behler, M. Parrinello, Phys. Rev. Lett. 2007, 98, 1.
Lett. 2007, 98, 98. [65] J. Behler, Phys. Chem. Chem. Phys. 2011, 13, 17930.
[32] H. Kindlund, D. G. Sangiovanni, L. Martínez-De-Olcoz, J. Lu, [66] L. Tang, Z. J. Yang, T. Q. Wen, K. M. Ho, M. J. Kramer, C. Z. Wang,
J. Jensen, J. Birch, I. Petrov, J. E. Greene, V. Chirita, L. Hultman, Phys. Chem. Chem. Phys. 2020, 22, 18467.
APL Mater. 2013, 1, 042104. [67] N. Xu, Y. Shi, Y. He, Q. Shao, J. Phys. Chem. C 2020, 124,
[33] M. Wen, X. Xie, Y. Gao, H. Dong, Z. Mu, F. Wu, C. Y. Wang, J. Alloys 16278.
Compd. 2019, 806, 1260. [68] C. M. Andolina, P. Williamson, W. A. Saidi, J. Chem. Phys. 2020, 152,
[34] S. Z. Han, E. A. Choi, S. H. Lim, S. Kim, J. Lee, Prog. Mater. Sci. 2020, 154701.
117, 100720. [69] X. Chen, X. Y. Gao, Y. F. Zhao, D. Y. Lin, W. D. Chu, H. F. Song,
[35] V. V. Ginzburg, Macromolecules 2013, 46, 9798. Comput. Phys. Commun. 2020, 250, 107057.
[36] M. Mandal, P. Borgohain, P. Begum, R. C. Deka, T. K. Maji, New J. [70] K. Lee, D. Yoo, W. Jeong, S. Han, Comput. Phys. Commun. 2019, 242,
Chem. 2018, 42, 2260. 95.
[37] R. J. Hebert, J. Mater. Sci. 2016, 51, 1165. [71] K. T. Schütt, H. E. Sauceda, P. J. Kindermans, A. Tkatchenko,
[38] N. E. Singh-Miller, N. Marzari, Phys. Rev. B – Condens. Matter Mater. K. R. Müller, J. Chem. Phys. 2018, 148, 241722.
Phys. 2009, 80, 1. [72] K. T. Schütt, P. Kessel, M. Gastegger, K. A. Nicoli, A. Tkatchenko,
[39] J. Du, A. Zhang, Z. Guo, M. Yang, M. Li, S. Xiong, ACS Omega 2017, K. R. Müller, J. Chem. Theory Comput. 2019, 15, 448.
2, 8803. [73] H. Wang, L. Zhang, J. Han, W. E, Comput. Phys. Commun. 2018, 228,
[40] N. Dumontet, D. Connétable, B. Malard, B. Viguier, Scr. Mater. 2019, 178.
167, 115. [74] A. P. Bartók, M. C. Payne, R. Kondor, G. Csányi, Phys. Rev. Lett. 2010,
[41] D. Hu, R. Kovacevic, Proc. Inst. Mech. Eng. Part B J. Eng. Manuf. 2003, 104, 1.
217, 441. [75] A. P. Bartõk, G. Csányi, Int. J. Quantum Chem. 2015, 115, 1051.

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (14 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

[76] A. P. Bartók, R. Kondor, G. Csányi, Phys. Rev. B – Condens. Matter [109] H. Wei, S. Zhao, Q. Rong, H. Bao, Int. J. Heat Mass Transf. 2018, 127,
Mater. Phys. 2013, 87, 1. 908.
[77] F. Maresca, D. Dragoni, G. Csányi, N. Marzari, W. A. Curtin, npj [110] L. R. Lopes, A. F. Silva, O. S. Carneiro, Addit. Manuf. 2018, 23, 45.
Comput. Mater. 2018, 4, 69. [111] K. Kowsari, S. Akbari, D. Wang, N. X. Fang, Q. Ge, 3D Print. Addit.
[78] F. C. Mocanu, K. Konstantinou, S. R. Elliott, J. Phys. D. Appl. Phys. Manuf. 2018, 5, 185.
2020, 53, 244002. [112] I. Q. Vu, L. B. Bass, C. B. Williams, D. A. Dillard, Addit. Manuf. 2018,
[79] J. O. Andersson, T. Helander, L. Höglund, P. Shi, B. Sundman, 22, 447.
CALPHAD: Comput. Coupling Phase Diagrams Thermochem. 2002, [113] P. Shyprykevich, J. S. Tomblin, M. G. Vangel, in ASTM Special
26, 273. Technical Publication, ASTM International, Philadelphia, PA 2002,
[80] J. Smith, W. Xiong, W. Yan, S. Lin, P. Cheng, O. L. Kafka, pp. 99–115.
G. J. Wagner, J. Cao, W. K. Liu, Comput. Mech. 2016, 57, 583. [114] G. X. Gu, S. Wettermark, M. J. Buehler, Addit. Manuf. 2017, 17, 47.
[81] I. H. Jung, M. A. Van Ende, Metall. Mater. Trans. B. Process Metall. [115] G. X. Gu, C. T. Chen, M. J. Buehler, Extrem. Mech. Lett. 2018, 18, 19.
Mater. Process. Sci. 2020, 51, 1851. [116] U. G. K. Wegst, H. Bai, E. Saiz, A. P. Tomsia, R. O. Ritchie, Nat.
[82] J. Kim, A. Wakai, A. Moridi, J. Mater. Res. 2020, 35, 1963. Mater. 2015, 14, 23.
[83] H. L. Chen, H. Mao, Q. Chen, Mater. Chem. Phys. 2018, 210, 279. [117] G. X. Gu, C. T. Chen, D. J. Richmond, M. J. Buehler, Mater. Horizons
[84] Y. Liu, S. Yen, S. Chu, S. Lin, M.-H. Tsai, Mater. Today Commun. 2018, 5, 939.
2021, 26, 102096. [118] T. Xue, T. J. Wallin, Y. Menguc, S. Adriaenssens, M. Chiaramonte,
[85] S. Gorsse, O. N. Senkov, Entropy 2018, 20, 1. Extrem. Mech. Lett. 2020, 41, 100992.
[86] C. E. Campbell, U. R. Kattner, Z. K. Liu, Integr. Mater. Manuf. Innov. [119] G. Costagliola, R. Guarino, F. Bosia, K. Gkagkas, N. M. Pugno, Phys.
2014, 3, 158. Rev. Appl. 2020, 13, 1.
[87] Z. K. Liu, J. Phase Equilibria Diffus. 2018, 39, 635. [120] B. Vrancken, L. Thijs, J. P. Kruth, J. Van Humbeeck, Acta Mater. 2014,
[88] X. Wang, W. Xiong, npj Comput. Mater. 2020, 6, 1. 68, 150.
[89] A. Borgenstam, A. Engström, L. Höglund, J. Ågren, J. Phase Equilibria [121] A. Levy, A. Miriyev, A. Elliott, S. S. Babu, N. Frage, Mater. Des. 2017,
2000, 21, 269. 118, 198.
[90] T. Keller, G. Lindwall, S. Ghosh, L. Ma, B. M. Lane, F. Zhang, [122] E. M. Parsons, Compos. Struct. 2019, 223, 110656.
U. R. Kattner, E. A. Lass, J. C. Heigel, Y. Idell, M. E. Williams, [123] P. D. Shah, C. A. Cimini, J. D. D. Melo, J. T. Fong, Am. Soc. Mech.
A. J. Allen, J. E. Guyer, L. E. Levine, Acta Mater. 2017, Eng. Press. Vessel. Pip. Div. PVP 2010, 6, 777.
139, 244. [124] G. I. N. Rozvany, Struct. Multidiscip. Optim. 2009, 37, 217.
[91] G. Lindwall, C. E. Campbell, E. A. Lass, F. Zhang, M. R. Stoudt, [125] H. Bohnacker, B. Gross, J. Laub, C. Lazzeroni, Generative Design:
A. J. Allen, L. E. Levine, Metall. Mater. Trans. A. Phys. Metall. Visualize, Program, and Create with Processing, Princeton
Mater. Sci. 2019, 50, 457. Architectural Press, New York 2012.
[92] R. Martukanitz, P. Michaleris, T. Palmer, T. DebRoy, Z. K. Liu, [126] M. McKnight, KnE Eng. 2017, 2, 176.
R. Otis, T. W. Heo, L. Q. Chen, Addit. Manuf. 2014, 1, 52. [127] C. Y. Lin, N. Kikuchi, S. J. Hollister, J. Biomech. 2004, 37, 623.
[93] Z. Liang, I. Zhirnov, F. Zhang, K. K. Jones, D. Deisenroth, [128] V. J. Challis, J. K. Guest, Int. J. Numer. Methods Eng. 2009, 79, 1284.
M. Williams, U. Kattner, K. won Moon, W. K. Liu, B. Lane, [129] T. Dbouk, Appl. Therm. Eng. 2017, 112, 841.
C. Campbell, Materialia 2020, 14, 100934. [130] S. Banga, H. Gehani, S. Bhilare, S. J. Patel, L. B. Kara, arXiv, 2018.
[94] E. Menou, G. Ramstein, E. Bertrand, F. Tancret, Model. Simul. [131] Y. Yu, T. Hur, J. Jung, I. G. Jang, Struct. Multidiscip. Optim. 2018, 59,
Mater. Sci. Eng. 2016, 24, 055001. 787.
[95] K. Paulus, Combined CALPHAD and Machine Learning for Property [132] S. Lee, H. Kim, Q. X. Lieu, J. Lee, Knowledge-Based Syst. 2020, 198,
Modelling, KTH, School of Industrial Engineering and 105887.
Management (ITM), Stockholm, Sweden 2020. [133] S. Oh, Y. Jung, S. Kim, I. Lee, N. Kang, J. Mech. Des. 2019, 141,
[96] A. Ghazlan, T. D. Ngo, P. Tran, Compos. Struct. 2016, 153, 278. 111405.
[97] G. X. Gu, M. Takaffoli, A. J. Hsieh, M. J. Buehler, Extrem. Mech. Lett. [134] H. T. Kollmann, D. W. Abueidda, S. Koric, E. Guleryuz, N. A. Sobh,
2016, 9, 317. Mater. Des. 2020, 196, 109098.
[98] G. X. Gu, M. Takaffoli, M. J. Buehler, Adv. Mater. 2017, 29, 1. [135] D. W. Abueidda, S. Koric, N. A. Sobh, Comput. Struct. 2020, 237,
[99] A. Bandyopadhyay, B. Heer, Mater. Sci. Eng. R Rep. 2018, 129, 1. 106283.
[100] T. Banerjee, S. Dey, A. P. Sekhar, S. Datta, D. Das, Trans. Indian Inst. [136] L. Wang, S. Tao, P. Zhu, W. Chen, J. Mech. Des. 2020, 143, 031708.
Met. 2020, 73, 3059. [137] S. Bin Maidin, I. Campbell, E. Pei, Assem. Autom. 2012, 32, 235.
[101] R. Liu, Y. C. Yabansu, A. Agrawal, S. R. Kalidindi, A. N. Choudhary, [138] M. K. Thompson, G. Moroni, T. Vaneker, G. Fadel, R. I. Campbell,
Integr. Mater. Manuf. Innov. 2015, 4, 192. I. Gibson, A. Bernard, J. Schulz, P. Graf, B. Ahuja, F. Martina, CIRP
[102] A. Cecen, H. Dai, Y. C. Yabansu, S. R. Kalidindi, L. Song, Acta Mater. Ann. – Manuf. Technol. 2016, 65, 737.
2018, 146, 76. [139] M. Leary, L. Merli, F. Torti, M. Mazur, M. Brandt, Mater. Des. 2014,
[103] Z. Yang, Y. C. Yabansu, R. Al-Bahrani, W. keng Liao, 63, 678.
A. N. Choudhary, S. R. Kalidindi, A. Agrawal, Comput. Mater. Sci. [140] Y. H. Kuo, C. C. Cheng, Struct. Multidiscip. Optim. 2019, 60, 1109.
2018, 151, 278. [141] A. T. Gaynor, J. K. Guest, Struct. Multidiscip. Optim. 2016, 54, 1157.
[104] C. T. Chen, G. X. Gu, Adv. Theory Simulations 2019, 2, 1. [142] B. M. Weiss, J. M. Hamel, M. A. Ganter, D. W. Storti, J. Manuf. Sci.
[105] M. A. Bessa, R. Bostanabad, Z. Liu, A. Hu, D. W. Apley, C. Brinson, Eng. Trans. ASME 2021, 143, 1.
W. Chen, W. K. Liu, Comput. Methods Appl. Mech. Eng. 2017, 320, [143] J. Nguyen, S. Park, D. Rosen, Int. J. Precis. Eng. Manuf. 2013, 14,
633. 1071.
[106] C. Yang, Y. Kim, S. Ryu, G. X. Gu, MRS Commun. 2019, 9, 609. [144] R. M. Gorguluarslan, U. N. Gandhi, Y. Song, S. K. Choi, Rapid
[107] C. Yang, Y. Kim, S. Ryu, G. X. Gu, Mater. Des. 2020, 189, Prototyp. J. 2017, 23, 305.
108509. [145] M. A. Bessa, P. Glowacki, M. Houlder, Adv. Mater. 2019, 31, 1.
[108] M. K. Kazi, F. Eljack, E. Mahdi, Compos. Struct. 2020, 254, 112885. [146] N. Ahsan, A. Habib, B. Khoda, Procedia Manuf. 2016, 5, 974.

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (15 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

[147] B. Khoda, 3D Print. Addit. Manuf. 2014, 1, 210. [167] B. N. Turner, R. Strong, S. A. Gold, Rapid Prototyp. J. 2014,
[148] J. Jiang, C. Yu, X. Xu, Y. Ma, J. Liu, Math. Biosci. Eng. 2020, 17, 3382. 20, 192.
[149] J. Huang, L. J. Segura, T. Wang, G. Zhao, H. Sun, C. Zhou, Addit. [168] B. N. Turner, S. A. Gold, Rapid Prototyp. J. 2015, 21, 250.
Manuf. 2020, 35, 101197. [169] S. C. Ligon, R. Liska, J. Stampfl, M. Gurr, R. Mülhaupt, Chem. Rev.
[150] L. Nguyen, J. Buhl, M. Bambach, Addit. Manuf. 2020, 35, 101265. 2017, 117, 10212.
[151] S. A. Khairallah, A. T. Anderson, A. Rubenchik, W. E. King, Acta [170] F. P. W. Melchels, J. Feijen, D. W. Grijpma, Biomaterials 2010, 31, 6121.
Mater. 2016, 108, 36. [171] E. O. Olakanmi, R. F. Cochrane, K. W. Dalgarno, Prog. Mater. Sci.
[152] V. Manvatkar, A. De, T. Debroy, J. Appl. Phys. 2014, 116, 124905. 2015, 74, 401.
[153] G. Tapia, S. Khairallah, M. Matthews, W. E. King, A. Elwany, Int. J. [172] S. F. S. Shirazi, S. Gharehkhani, M. Mehrali, H. Yarmand,
Adv. Manuf. Technol. 2018, 94, 3591. H. S. C. Metselaar, N. Adib Kadri, N. A. A. Osman, Sci. Technol.
[154] C. Kamath, Int. J. Adv. Manuf. Technol. 2016, 86, 1659. Adv. Mater. 2015, 16, 1.
[155] N. Hertlein, S. Deshpande, V. Venugopal, M. Kumar, S. Anand, [173] S. Kumar, JOM 2003, 55, 43.
Addit. Manuf. 2020, 32, 101089. [174] C. Y. Yap, C. K. Chua, Z. L. Dong, Z. H. Liu, D. Q. Zhang, L. E. Loh,
[156] P. M. Bhatt, R. K. Malhan, P. Rajendran, S. K. Gupta, Addit. Manuf. S. L. Sing, Appl. Phys. Rev. 2015, 2, 041101.
2020, 32, 101003. [175] T. M. Mower, M. J. Long, Mater. Sci. Eng., A 2016, 651, 198.
[157] T. F. Lam, Y. Xiong, A. G. Dharmawan, S. Foong, G. S. Soh, Int. J. [176] L. E. Murr, S. M. Gaytan, D. A. Ramirez, E. Martinez, J. Hernandez,
Adv. Manuf. Technol. 2020, 108, 1061. K. N. Amato, P. W. Shindo, F. R. Medina, R. B. Wicker, J. Mater. Sci.
[158] M. Murua, A. Suarez, D. Galar, R. Santana, IEEE Access 2020, 8, 91574. Technol. 2012, 28, 1.
[159] S. Korneev, Z. Wang, V. Thiagarajan, S. Nelaturi, CAD Comput. Aided [177] S. L. Sing, J. An, W. Y. Yeong, F. E. Wiria, J. Orthop. Res. 2016, 34,
Des. 2020, 127, 102852. 369.
[160] B. E. Carroll, T. A. Palmer, A. M. Beese, Acta Mater. 2015, 87, 309. [178] Binder Jetting | Additive Manufacturing Research Group |
[161] N. Raghavan, R. Dehoff, S. Pannala, S. Simunovic, M. Kirka, Loughborough University, https://www.lboro.ac.uk/research/
J. Turner, N. Carlson, S. S. Babu, Acta Mater. 2016, 112, 303. amrg/about/the7categoriesofadditivemanufacturing/binderjetting/
[162] P. Mercelis, J. P. Kruth, Rapid Prototyp. J. 2006, 12, 254. (accessed: January 2021).
[163] M. P. Sefidi, R. Israr, J. Buhl, M. Bambach, Procedia Manuf. 2020, 47, [179] A. Dass, A. Moridi, Coatings 2019, 9, 418.
1134. [180] A. Moridi, S. M. Hassani-Gangaraj, M. Guagliano, M. Dao, Surf. Eng.
[164] S. P. Donegan, E. J. Schwalbach, M. A. Groeber, Mater. Charact. 2014, 30, 369.
2020, 161, 110123. [181] D. Ding, Z. Pan, D. Cuiuri, H. Li, Int. J. Adv. Manuf. Technol. 2015, 81,
[165] M. Mozaffar, R. Bostanabad, W. Chen, K. Ehmann, J. Cao, 465.
M. A. Bessa, Proc. Natl. Acad. Sci. U. S. A. 2019, 116, 26414. [182] Ultrasonic Additive Manufacturing – Inside Metal Additive
[166] N. S. Johnson, P. S. Vulimiri, A. C. To, X. Zhang, C. A. Brice, Manufacturing, https://www.insidemetaladditivemanufacturing.com/
B. B. Kappes, A. P. Stebner, Addit. Manuf. 2020, 95, 101641. blog/ultrasonic-additive-manufacturing (accessed: January 2021).

Chenxi Tian received his bachelor’s degree from University of California, Berkeley. Currently, he is a
Ph.D. student, and he joined the Laboratory for Advanced Materials and Manufacturing (LAMM) in 2019
under the direction of Professor Atieh Moridi. His research interests focus on AM material
process property mapping, algorithm-driven optimization of AM composite material design, and
manufacturing.

Tianjiao Li received his bachelor’s degree from Xi’an Jiaotong University in 2017 and master’s degree
from Northwestern University in 2019. Presently, he is a Ph.D. student, and he joined the J2 Lab for
Engineering Living Materials in 2019 under the direction of Professor Jingjie Yeo. His research interests
focus on multiscale computational modeling, as well as data- and algorithm- driven materials
optimization and design.

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (16 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
26404567, 2021, 12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aisy.202100014 by CAPES, Wiley Online Library on [06/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.advancedsciencenews.com www.advintellsyst.com

Jenniffer Bustillos received her B.S. in mechanical engineering in 2018 and M.S. in materials science in
2019 from Florida International University (FIU, Miami, FL.) From 2014 to 2019, she worked as a
research assistant at the Plasma Forming Laboratory (FIU, Miami, FL). Currently, she is a doctoral
candidate at Cornell University (Ithaca, NY) in the Sibley School of Mechanical and Aerospace
Engineering under the advisement of Moridi, joining the Laboratory for Advanced Materials and
Manufacturing. Her research interests include the development of high-performance materials for
additive manufacturing and evaluation of mechanical, tribological, and thermal properties of
nanostructural materials.

Jingjie Yeo is an assistant professor in the Sibley School of Mechanical and Aerospace Engineering at
Cornell University. He heads the J2 Lab for Engineering Living Materials which focuses on improving the
prevention and treatment of diseases associated with aging and noncommunicable diseases. The lab
develops soft, adaptive, and responsive biomaterials through interdisciplinary studies of material and
biological phenomena by advancing multiscale, multiphysics computational methods and simulations.
Prior, he was a research scientist in the Institute of High Performance Computing, Singapore. He did his
postdoctoral research in the United States at both Tufts University and Massachusetts Institute of
Technology. He received his Ph.D. and his B.Eng. from Nanyang Technological University Singapore.

Atieh Moridi is an assistant professor of Mechanical and Aerospace Engineering and Director of the
Laboratory for Advanced Materials and Manufacturing at Cornell University. Prior to that, she was
postdoctoral fellow in the Departments of Mechanical Engineering and Materials Science and
Engineering at MIT. She received her Ph.D. Cum Laude (the highest institute honor) in mechanical
engineering from Politecnico di Milano, Italy. Her research interests lie in the areas of advanced
materials and manufacturing. She exploits the intrinsic properties as well as flexibility of additive
manufacturing to synthesize novel high-performance materials.

Adv. Intell. Syst. 2021, 3, 2100014 2100014 (17 of 17) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH

You might also like